A Meta-analysis - Jogo Remoto

jogoremoto.com

A Meta-analysis - Jogo Remoto

Prevalence of Disordered Gamblingiv• Among adolescents sampled from thegeneral population—14.82%, within a95% confidence interval of 8.99% and20.66%.In addition to estimates of the extent of U.S. andCanadian gambling disorders, this meta-analysisproduced a variety of findings that are informativeabout the nature and distribution of disorderedgambling. These findings can be summarizedbriefly as follows:♦ During the past two decades, gambling disordershave evidenced an increasing rateamong adults sampled from the generalpopulation.♦ To date, prevalence research has not demonstratedan increase in the rate of gamblingdisorders among adolescents or adults sampledfrom treatment or prison populationsduring the past two decades.♦ Gambling disorders are significantly moreprevalent among young people than amongthe general adult population.♦ Gambling disorders are significantly moreprevalent among males than females withinevery population segment considered in thisstudy.♦ Individuals with concurrent psychiatricproblems display much higher rates of disorderedgambling than either adolescents oradults sampled from the general population.♦ There was no significant regional variationin the rates of gambling disorders identifiedacross regions of Canada and the UnitedStates.♦ To date, the overall methodological qualityof disordered gambling prevalence researchhas not improved during the past 20 years.♦ Methodological study quality did not influencethe magnitude of prevalence estimates.This study also provided the opportunity toidentify and consider a variety of conceptual andmethodological problems associated with estimatingand interpreting prevalence rates. Thisdiscussion includes an examination of the constructvalidity of disordered gambling. For example,one of the most important issues associatedwith the study of any psychiatric disorder isthe absence of a definitive “gold standard” fordetermining who has the disorder and who doesnot. Without a gold standard, it is impossible todetermine with confidence whether a screeninginstrument over- or under- estimates the problemin the general population. The validity ofany judgment about the prevalence of disorderedgambling must be evaluated by an independentmeans of assessment. This thorny conceptualand methodological problem reflects thematter of “construct validity.” Without independentvalidation, gambling researchers mustbegin to consider to what extent gambling disordersmay overlap with other psychiatric illness.This study also identified an interesting characteristicof research in the field of gamblingprevalence: overall, the methodological qualityof prevalence research did not meaningfully influenceestimates of prevalence. Further, accordingto the methodological quality index employedin this study, the quality of prevalenceresearch has not advanced significantly duringthe past two decades.To help advance the field of gambling studies,this meta-analysis encourages investigators toreport interval estimates routinely and not justoccasionally. This practice is different from themore common custom of representing a prevalencerate by providing a single index, or pointestimate, without an associated measure of confidence.Discussions about interval estimationshould take a more prominent role in studies ofdisordered gambling prevalence in general, andaround the reporting of point estimates of gamblingdisorders in particular. This practice willprovide legislators, health care planners, andother policy makers with an explicit standard ofconfidence about each prevalence estimate sothat they can better judge its value. In additionHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gamblingvto confidence intervals, we suggest that investigatorsreport specific prevalence rates (e.g.,male rates versus female rates) rather than aggregaterates for an entire population.In addition to methodological matters, the discussionalso considers implications of the presentfindings for future research, public policy,and treatment. This report concludes with a briefconsideration of suggested guidelines for theconduct of future prevalence research directed atdisordered gambling.AcknowledgementsA project of this magnitude involves a variety ofcontributions from many people. This undertakingwould not have been possible if it were notfor the hard work and thoughtfulness of manydifferent people. Thanks are extended to MaryFoley and Tom Murphy for their work on a pilotversion of this project. We appreciate the helpof Sheryl H. Bell, Frank Biagioli, Renee Cunningham-Williams,Jeffrey Derevensky, MarkDickerson, Rosa Dragonetti, Don Feeney, G.Ron Frisch, Dewey Jacobs, Henry Lesieur, EdLooney, Valerie Lorenz, Reece Middleton, JoeRinehart, Lee Robins, Jill Rush, Kathy Scanlan,Brian Steeves, Randy Stinchfield, Rachel Volberg,George Wandrak, Ken Winters, HaroldWynne, and Darryl Zitzow in our effort to identifyand locate studies for this project. Specialthanks are extended to Richard LaBrie, whoconsistently helped us to understand and applystatistical instruments in a most useful and elegantway. We appreciate the tireless efforts ofGreg O’Donohue and Blase Gambino, whojoined the team of coders who compulsivelyabstracted the data necessary for this study.Blase Gambino, Richard LaBrie, RichardMcGowan, Peter Reuter, Graham Colditz, JamesOppenheim, and Kathy Scanlan contributedwise and valuable comments about earlier versionsof this manuscript; we appreciate theirinsight and support. We also acknowledge andappreciate the careful administrative work ofJay Alves and Jennifer Wickham during the finalrounds of manuscript revision. ThomasCummings has been a guiding light, providingus with support and insight; we send him ourspecial thanks. We also send a note of appreciationto our friend and colleague William McAuliffe,whose important contributions to the fieldof substance abuse treatment needs assessmentand prevalence research helped to stimulatemany important conceptual and methodologicalideas that came to fruition during this project.Emily McNamara, who worked diligently toassure that the administrative details of thisproject were coordinated, deserves special distinction.She provided thoughtful input andguidance for everyone associated with this project.We owe Emily a special debt of gratitude.Finally, we want to thank the National Centerfor Responsible Gaming (NCRG); the NCRGhas our gratitude for providing financial support,and waiting patiently for our findings.Last, but certainly not least, we must acknowledgeour families and friends who made space intheir lives so we could complete this work. Theproject was vast, and so was their support.About the AuthorsHoward J. Shaffer currently is an associateprofessor at Harvard Medical School; in addition,he is the Director of the Division on Addictionsat Harvard Medical School. Recently, hehas been serving as principal or co-principalinvestigator on three government sponsored researchprojects: (1) Addiction Training Centerof New England; (2) A Science Education andDrug Abuse Partnership project, “AdvancingScience Through Substance Abuse Education;”and (3) a newly completed randomized clinicaltrial comparing Hatha yoga with psychodynamictreatment as an adjunct to methadone therapyfor narcotic dependence.Dr. Shaffer serves as an Associate Editor of TheJournal of Substance Abuse Treatment. He isthe current editor of The Journal of GamblingStudies. Dr. Shaffer also is an active memberof the editorial boards for the Journal of PsychoactiveDrugs, Advances in Alcohol and SubstanceAbuse, The Psychology of Addictive Behaviors,and Sexual Addiction and Compulsivity:The Journal of Prevention and Treatment.Dr. Shaffer serves as a reviewer for the NewHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingviEngland Journal of Medicine.Dr. Shaffer's major research interests include thesocial perception of addiction and disease, thephilosophy of science, impulse control regulationand compulsive behaviors, adolescent gambling,and the natural history of addictive behaviors.Dr. Shaffer has written extensively aboutthe treatment of addictive behaviors and the natureof addiction. His research and writing hasshaped how the health care field conceptualizesand treats the full range of addictive behaviors.Matthew Hall has worked for four years as aResearch Associate at Harvard MedicalSchool’s Division on Addictions. Mr. Hall is onthe Editorial Board of the Journal of GamblingStudies and is co-editor of the WAGER (theWeekly Addiction Gambling Educational Report),a weekly fax newsletter on gamblingrelatedresearch. With Dr. Shaffer, he authoredthe first meta-analysis of adolescent prevalencerates in the field of gambling disorders. He isco-author of Facing the Odds: The Mathematicsof Gambling, a middle-school mathematics curriculumunit on probability, statistics, and numbersense related to gambling. In addition, hehas published a variety of articles on addictionrelatedtopics in a number of leading professionaljournals.Joni Vander Bilt, Instructor of Public Health inthe Department of Neurobiology, HarvardMedical School, is a Research Associate at theDivision on Addictions. Ms. Vander Bilt is theeditor-in-chief of The WAGER, a Weekly AddictionGambling Educational Report. She alsoco-authored Facing the Odds, a mathematicscurriculum for middle school students teachingprobability, number sense, and critical thinkingskills within the context of gambling. In additionto disordered gambling research, Ms.Vander Bilt’s research interests include the areasof women and addictions and the efficacy ofalternative treatment modalities for addictivebehaviors. Ms. Vander Bilt holds a Master'sDegree in Public Health from Boston University.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingviiTable of ContentsEXECUTIVE SUMMARY ........................................................................................................................................IIACKNOWLEDGEMENTS .......................................................................................................................................VABOUT THE AUTHORS..........................................................................................................................................VLIST OF FIGURES....................................................................................................................................................XLIST OF TABLES.................................................................................................................................................... XIINTRODUCTION .......................................................................................................................................................1BACKGROUND: THE NEED FOR ESTIMATES OF DISORDERED GAMBLING...................................................................1United States ........................................................................................................................................................1Canada.................................................................................................................................................................2Risks of Gambling ................................................................................................................................................2On the Conduct of a National Prevalence Study..................................................................................................2ESTIMATING PREVALENCE USING META-ANALYSIS...................................................................................................3META-ANALYTIC METHODOLOGY.............................................................................................................................4MANUFACTURING PREVALENCE ESTIMATES..............................................................................................................5EPIDEMIOLOGICAL CONSIDERATIONS.........................................................................................................................6Prevalence versus Incidence................................................................................................................................6The Importance of Range Estimates ....................................................................................................................7Time frame ...........................................................................................................................................................7Semantics of Disordered Gambling: Conceptual Chaos .....................................................................................8A BRIEF HISTORY OF DISORDERED GAMBLING PREVALENCE STUDIES .....................................................................9CURRENT STATUS OF THE DISORDERED GAMBLING PREVALENCE FIELD: METHODOLOGICAL CONSIDERATIONS....11Coverage............................................................................................................................................................11Populations ........................................................................................................................................................11Special Populations ...........................................................................................................................................12Instrumentation ..................................................................................................................................................12Methodological Quality .....................................................................................................................................13MEASUREMENT, PREVALENCE ESTIMATION, AND SOCIAL SETTING ........................................................................14THE CENTRAL HYPOTHESES ....................................................................................................................................15METHODS.................................................................................................................................................................15IDENTIFYING PRIMARY STUDIES ..............................................................................................................................15ELIGIBILITY CRITERIA..............................................................................................................................................16DATA ABSTRACTION AND DATA ENTRY PROCEDURES ............................................................................................17QUALITY COMPONENT ANALYSIS: METHODOLOGICAL QUALITY SCORES ...............................................................19NOMENCLATURE & CLASSIFICATION: LEVELS OF DISORDERED GAMBLING SEVERITY...........................................21Applying the Level System to Meta-analysis ......................................................................................................21Level 2: In-transition Implies Bi-directionality.................................................................................................22ESTIMATION METHODS............................................................................................................................................22Weighting ...........................................................................................................................................................22Prevalence Estimates: A Multi-Method Approach.............................................................................................23A NOTE ON TIME FRAMES.......................................................................................................................................24RESULTS...................................................................................................................................................................24STUDY DEMOGRAPHICS...........................................................................................................................................24THE FILE DRAWER PROBLEM: PUBLICATION STATUS, SIGNIFICANT FINDINGS, & PREVALENCE RATES..................27METHODOLOGICAL QUALITY AMONG PREVALENCE STUDIES..................................................................................27Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingviiiAnalyzing the Methodological Quality Score.....................................................................................................29ORGANIZING THE RESULTS BY THE CENTRAL HYPOTHESES.....................................................................................30INITIAL OBSERVATIONS ON COLLINEARITY OF PRIMARY VARIABLES ......................................................................30Collinearity of Individual Variables ..................................................................................................................31TESTING THE CENTRAL HYPOTHESES ......................................................................................................................32Hypothesis 1: Rates of Disordered Gambling Prevalence Vary by Population Segment ..................................33Comparing The Rates Of Disordered Gambling Among Different Population Segments............................................... 34Estimating Population Prevalence Differences While Controlling for Instrument.......................................................... 36Gender-specific Prevalence Estimates ............................................................................................................................ 37Attributable Proportion................................................................................................................................................... 39Other Stratified Data ....................................................................................................................................................... 41Hypothesis 2: The Prevalence of Disordered Gambling is Shifting Over Time.................................................41Hypothesis 3: Differences Among Prevalence Rates Derived from Different Screening Instruments ...............43Hypothesis 4: Differences Among Prevalence Rates Derived from Different Regions......................................46Hypothesis 5: Differences Among Prevalence Rates Derived from Different Researchers ...............................46Hypothesis 6: Experience with Different Types of Gambling Activities Yield Different Rates of GamblingDisorder .............................................................................................................................................................46The Relationship between Specific Gambling Activities and Prevalence Rates ............................................................. 47REGRESSION ANALYSES: FACTORS THAT INFLUENCE PREVALENCE ESTIMATES OF DISORDERED GAMBLING.........48ESTIMATING THE NUMBER OF DISORDERED GAMBLERS IN THE UNITED STATES AND CANADA...............................50DISCUSSION.............................................................................................................................................................51CAN DIFFERENT PREVALENCE ESTIMATES BE INTEGRATED? ..................................................................................52FACTORS ASSOCIATED WITH DISORDERED GAMBLING ESTIMATES: SOURCES OF INFLUENCE .................................52THE CENTRAL HYPOTHESES: DRAWING CONCLUSIONS...........................................................................................54Hypothesis 1: Rates of Disordered Gambling Prevalence Vary by Population Segment .................................54Identifying Risk Factors for Disordered Gambling......................................................................................................... 55Hypothesis 2: The Prevalence of Disordered Gambling is Shifting over Time.................................................56Hypothesis 3: Differences among Prevalence Rates Derived from Different Screening Instruments................57Hypothesis 4: Differences among Prevalence Rates Derived from Different Regions .....................................57Hypothesis 5: Differences among Prevalence Rates Derived from Different Researchers...............................58Hypothesis 6: Experience with Different Types of Gambling Activities Yield Different Rates of GamblingDisorder .............................................................................................................................................................58HOW DOES THE PREVALENCE OF DISORDERED GAMBLING COMPARE WITH OTHER PROBLEMS?: MAKING SENSE OFTHE NUMBERS .........................................................................................................................................................59METHODOLOGICAL CONSIDERATIONS & PREVALENCE ESTIMATION .......................................................................60Prevalence Study Quality...................................................................................................................................60Comparing Lifetime and Past-Year Prevalence Rates.......................................................................................61Time Frame Caveats and Concerns................................................................................................................................. 62Calculating Prevalence Rates............................................................................................................................64The Different Purposes of Prevalence and Incidence Studies............................................................................65The Validity of Disordered Gambling Prevalence Estimates.............................................................................65Paradigms for Understanding: The Relative Nature of Disordered Gambling................................................................ 66Reconsidering Clinical Diagnoses as the “Gold Standard”............................................................................................. 67Validity as a Theory-Driven Construct ........................................................................................................................... 68Considering False Positives and False Negatives: The Need for Theory........................................................................ 69Is Pathological Gambling a Primary Disorder?............................................................................................................... 70TOWARD AN UNDERSTANDING OF DISORDERED GAMBLING IN CONTEXT: THE INTERACTION OF SOCIAL SETTINGAND PERSONALITY...................................................................................................................................................72The Impact of Social Setting on Gamblers and Gambling.................................................................................74Risk Factors and Social Sanctions .................................................................................................................................. 75The Natural History of Illicit Activities: The Shift from Socially Sanctioned to Socially Supported Activities .76TOWARD THE FUTURE: IMPLICATIONS FOR NEW PREVALENCE RESEARCH..............................................................77Applications of Prevalence Estimates................................................................................................................77Social Objectives ............................................................................................................................................................ 78Economic Applications................................................................................................................................................... 78Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingixPsychological Purposes................................................................................................................................................... 78Prevalence and Treatment Planning................................................................................................................................ 78Toward New Theoretical Maps: Guiding Future Research & Practice.............................................................79Nomenclature & Classification: A Conceptual Map for Future Prevalence Research .................................................... 80Collinearity and Its Implications for Future Prevalence Research ...................................................................83RECOMMENDED STANDARDS FOR FUTURE PREVALENCE RESEARCH ......................................................................83CAVEATS & LIMITATIONS........................................................................................................................................84META-ANALYSES OF PREVALENCE ESTIMATES: ADVANCING THE FIELD ................................................................85CONCLUSIONS..........................................................................................................................................................85PRIMARY PREVALENCE STUDIES....................................................................................................................86REFERENCES ..........................................................................................................................................................93APPENDIX 1: MULTI-METHOD ESTIMATORS............................................................................................102APPENDIX 2: STUDY CHARACTERISTICS ...................................................................................................106APPENDIX 3: A BRIEF GUIDE TO PLANNING PREVALENCE STUDY PROTOCOLS .........................111INTRODUCTION ......................................................................................................................................................111PURPOSE................................................................................................................................................................112LITERATURE REVIEW.............................................................................................................................................112SELECTING AN INSTRUMENT .................................................................................................................................113On Instrumentation ..........................................................................................................................................113Forms of Gambling ..........................................................................................................................................113TIME FRAME & SYMPTOM CLUSTERS FOR GAMBLING DISORDERS........................................................................114ELIMINATING OTHER PSYCHIATRIC DISORDERS: EXCLUSION CRITERIA ................................................................114SAMPLING DESIGN I: SELECTING RESPONDENTS....................................................................................................114SAMPLING DESIGN II: POWER ANALYSIS AND PLANNING FOR AN ADEQUATE REPRESENTATION OF THE POPULATIONOF INTEREST ..........................................................................................................................................................115Alternatives to Stratified Random Sampling: Optimal Allocation ...................................................................116Respondent Selection Method ..........................................................................................................................117Connection Attempts, Call-Back Procedures, and Conversion of Refusals .....................................................117RESPONSE AND COMPLETION RATES .....................................................................................................................117Calculating Response and Completion Rates ..................................................................................................118INTERVIEWER SELECTION AND TRAINING..............................................................................................................119DATA CODING AND ENTRY....................................................................................................................................119DATA ANALYSIS ....................................................................................................................................................119APPLYING THE RESULTS ........................................................................................................................................120State and Sub-state Approaches for Treatment Planning ................................................................................120SUMMARY .............................................................................................................................................................120REFERENCES..........................................................................................................................................................120Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingxList of FiguresFigure 1: Percentage of Study Population Types.........................................................................................................25Figure 2: History of Prevalence Estimates...................................................................................................................25Figure 3: Authors of Gambling Prevalence Studies.....................................................................................................25Figure 4: Trend of Quality Scores Over Time .............................................................................................................29Figure 5: History of Methodological Quality Scores ...................................................................................................30Figure 6: Lifetime Prevalence M-Estimates for Adult GP Level 2 Gambling .............................................................33Figure 7: Lifetime Prevalence M-Estimates for Adult GP Level 3 Gambling .............................................................34Figure 8: Lifetime Prevalence of Level 2 & 3 Among the Adult General Population ................................................34Figure 9: Past-Year Prevalence of Level 2 & 3 Among the Adult General Population ..............................................34Figure 10: Lifetime Prevalence Rates Among Population Segments...........................................................................35Figure 11: Adult Lifetime Level 2 & 3 Combined Prevalence ....................................................................................42Figure 12: Combined Lifetime Level 3 & 2 Trends....................................................................................................44Figure 13: Explained Unique & Shared Variance Among Prevalence Rates Due to Population Type........................49Figure 14: Elements of the Methodological Quality Index..........................................................................................49Figure 15: Sources of unique and shared variance among the explained sources of adult prevalence estimates.........50Figure 16: Sources of unique and shared variance among the explained sources of youth prevalence estimates........51Figure 17: Factors That Influence Prevalence Rates....................................................................................................53Figure 18: Is Pathological Gambling a Discrete Phenomenon?...................................................................................72Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingxiList of TablesTable 1: Literature Search Results...............................................................................................................................17Table 2: Level System for Meta-analysis....................................................................................................................21Table 3: Meta-analysis Population Segment Samples .................................................................................................26Table 4: Bivariate Correlations Among Primary Collinear Variables ........................................................................31Table 5: Mean Prevalence Rates (95% Confidence Intervals) for Four Study Populations........................................33Table 6: Significant Lifetime Prevalence Differences Among Population Groups.....................................................35Table 7: Relative Risk of Disordered Gambling by Population Types.......................................................................36Table 8: Comparing Prevalence Estimates & (95% Confidence Intervals) Associated With SOGS Studies ..............37Table 9: Significant SOGS Lifetime Prevalence Differences Between Population Groups........................................38Table 10: Estimates of Disordered Gambling by Population Group Stratified by Gender .........................................39Table 11: Gender-specific Relative Risks of Disordered Gambling...........................................................................39Table 12: Attributable Risks for Gender.....................................................................................................................40Table 13: Mean Adult Prevalence Rates for Pre-Median-Year and Post-Median-Year Groups.................................42Table 14: Comparing Screening Instruments Across Studies With Multiple Measures (Level 3 Rates) ....................45Table 15: Prevalence of Gambling Activity by Population Segment..........................................................................47Table 16: Estimated Number of Past-Year Disordered Gamblers in the United States & Canada .............................50Table 17: Comparing Lifetime and Past-year Prevalence Rates of Adult Psychiatric Disorders in the United States:Where Does Disordered Gambling Fit? .............................................................................................................59Table 18: Comparing Past-year Prevalence Rates of Psychiatric Disorders Among Adolescents: Where DoesDisordered Gambling Fit?..................................................................................................................................60Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Estimating the Prevalence of Disordered Gambling Behavior in theUnited States and Canada: A Meta-analysisIntroductionThis research report describes a conceptualand empirical review of the existing literaturewhich estimates the prevalence ofdisordered gambling in the United Statesand Canada. During the past decade, there hasbeen an increasing clarion call among researchers(Ladouceur, 1996; Sartin, 1988) and policymakers (e.g., National Gambling Impact andPolicy Commission Act of 1996) to developprecise estimates of gambling-related disordersamong both adults and adolescents throughoutthe United States and Canada. Since prevalenceestimates are a direct reflection of the researchmethods and strategies scientists develop andimplement to measure a particular phenomenon,debate and controversy are regular consequencesof prevalence estimation projects (e.g.,Nadler, 1985). This methodological debate resultsin confusion among the legislators, healthcare providers and public health program plannerswho use these estimates to make policy,funding, and treatment decisions (e.g., Eadington,1992). To minimize controversy and yieldthe most useful estimates of gambling-relatedproblems, this project employed a meta-analyticstrategy to develop estimates of gamblingrelateddisorders across an array of differingestimation methodologies and populations. Thisapproach provides the opportunity to assess andintegrate the variety of assumptions and strategiesused by the array of scientists who previouslyhave estimated disordered gamblingprevalence. Shaffer and Hall (1996) providedthe first meta-analytic estimate of gambling disordersamong an adolescent population. Thisresearch extends their earlier investigation andprovides new insight into the many gamblingrelatedfactors that require examination or additionalresearch.Background: The Need for Estimates ofDisordered GamblingUnited StatesWe are now in the midst of the thirdwave of widespread legal gamblingin the United States (Rose, 1986;1995). 1 The third wave began in the1930s, during the Great Depression, when Nevadare-legalized casinos and pari-mutuel gamblingspread across the country. The current eraof gambling expansion began in 1964, whenNew Hampshire initiated the first modern statelottery. Since the start of the New Hampshirelottery, the opportunity to gamble has explodedacross America. Between 1974 and 1996, thetotal amount of money legally wagered nationwideincreased from $17.3 billion to $586.5 billion(Christiansen, 1997; National Council onProblem Gambling, 1993). In 1996, the gamingindustry earned $47.6 billion, a 5.6% increasefrom the previous year (Christiansen, 1997).Between 1975 and 1996, the national per capitasales of lottery products alone increased from$20 to approximately $150 per year (Clotfelter& Cook, 1989; McQueen, 1996). To date, 37states in addition to Washington D.C. have legalizedlotteries and 26 states have NativeAmerican or independent casinos (Whyte,1997). All states with the exception of Utah andHawaii have legalized some form of gambling.1 According to Rose (1986), the first wave of gamblingin the United States began during the colonialperiod and did not end until the decades immediatelyprior to the Civil War. The second wave of legalgambling started with the Civil War and ended in aseries of scandals (e.g., Louisiana Lottery scandal) inthe last decade of the nineteenth century.


Prevalence of Disordered Gambling2CanadaAn expansive wave of gambling accessibilityhas similarly washed over Canada during thepast twenty years (Ladouceur, 1996). An estimated$20 to $27 billion a year is wagered onall forms of legal gambling in Canada (NationalCouncil of Welfare, 1996). The national total of$4.6 billion in gambling revenues represents2.7% of the total provincial and territorial revenues(National Council of Welfare, 1996). Currently,lotteries, bingo, and pari-mutuel wageringare available in every Canadian province(International Gaming & Wagering Business,1997). Casinos have been established in fiveprovinces: Quebec, Ontario, Nova Scotia, Manitoba,and Saskatchewan. In addition, charitycasinos have been established in British Columbia,Alberta, and Ontario.Risks of GamblingGambling is neither a financially nor a psychologicallyrisk-free experience. In addition to thepossibility that gamblers will lose their money,gamblers also risk experiencing a variety of adversepsychological, social, and biological consequencesfrom gambling. Given the increasingaccess to gambling during the latter half of the20 th century, public health researchers, clinicians,and policy makers have had both the opportunityand social obligation to study the impactof legalized gambling on adults as well aschildren and adolescents. As the popularity oflegalized gambling grows, society is directingmore attention toward the public health risksand the economic, legal, and social costs of expandedgambling (Eadington, 1994).Although most people view gambling as an entertainingrecreational activity, numerous studiesreveal the serious consequences of gamblingfor a segment of the population (e.g., Lesieur &Rosenthal, 1991; Shaffer, Hall, Walsh, &Vander Bilt, 1995). Gamblers experiencing adversereactions to gambling have become knownas “compulsive,” “problem,” or “pathological”gamblers. The American Psychiatric Association(APA) includes pathological gambling as animpulse disorder in their Diagnostic and StatisticalManual (APA, 1994). The essential characteristicof impulse disorders is the person’sinability to “…resist an impulse, drive, or temptationto perform an act that is harmful to theperson or to others” (APA, 1994, p. 609). Oftenpeople with impulse disorders feel an increasingtension prior to acting, and then a sense of relief,calm, or pleasure following the impulsiveact. Impulsive acts may or may not be followedby a sense of regret, guilt, or shame. The manualstates that “the essential feature of pathologicalgambling is persistent and recurrent maladaptivegambling behavior... that disrupts personal,family, or vocational pursuits. The diagnosisis not made if the gambling behavior isbetter accounted for by a manic episode...”(APA, 1994, p. 615).On the Conduct of a National PrevalenceStudyTo date, a range of studies have been conductedin the United States and Canada to determinethe prevalence of disordered gambling amongadults and adolescents. However, with the exceptionof Kallick, Suits, Dielman, and Hybels’(1979) national effort, which represents the firstand only national prevalence study, each ofthese studies restricted their sampling strategy toa limited geographic area. A variety of estimatesfrom a range of geographical areas hasprevented the gambling research field from producingprevalence rates that reflect the scopeand severity of disordered gambling within anational context. “There seems to be no realagreement among health professionals and governmentagencies as to the actual NUMBER ofAmericans afflicted with the subject at hand[i.e., disordered gambling]. Discrepancies invarious reports are too broad to make a viableaverage.... For a subject receiving so muchwidespread publicity, the disparity in thesenumbers is alarming” (Sartin, 1988, p. 371).The American Psychiatric Association’s FourthEdition of its Diagnostic and Statistical Manual(APA, 1994) notes that the prevalence of pathologicalgambling may be between 1% and 3%,stating that the vagueness of the statistic is dueto the limited availability of data.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling3Thus, the field of gambling research, the gamblingindustry, and national and state policymakers are in need of an accurate and reliableestimate of the extent of gambling problemsthroughout the United States and Canada.However, the time and expense required for anational prevalence study would be considerable,perhaps even prohibitive. For example,planning a national prevalence study would—and arguably should—entail a lengthy debateabout the propriety of methodological issues.For example, to date, at least 25 different surveyinstruments (including modifications) have beenused to measure gambling problems among adolescentsand adults. The selection of any one oreven several of these scales—or the creation ofa new scale—for use in a national study wouldrequire extensive research and scientific debate. 2For example, some researchers (e.g.,Christiansen/Cummings Associates Inc., PrincetonMarketing Associates, Inc., Spectrum AssociatesMarket Research Inc., & Volberg, 1992)have suggested that the South Oaks GamblingScreen (Lesieur & Blume, 1987), the mostwidely used instrument for estimating the prevalenceof disordered gambling, creates an overestimateof the disorder. In addition, investigatorsalso would have to resolve issues relating tothe reliability and validity of an instrument thatwould be applied across a diverse and representativenational sample. If a new instrument wasdeveloped, it might take years to determine theaccuracy and utility of its psychometric propertiesbefore it could be applied to a national sample.Further, the development of a valid samplingstrategy designed to yield a sample representativeof the entire country would be an extremelycomplicated methodological and administrativeundertaking. This sample would haveto represent every state or province and be basedon a stratified random selection strategy. If the2 Canadian researchers currently are engaged in thismethodological debate with an aim to adopt a standardizedproblem gambling screen for use in Canada(Measuring Problem Gambling in Canada: A Requestfor Proposals Issued by the Inter-ProvincialTask Force on Problem Gambling, June 6, 1997).study purpose included a meaningful effort tounderstand treatment needs and other socialcosts, then the actual administration of a surveyto a national sample would become prohibitivelyexpensive. 3 Under these circumstances it wouldbecome essential for the research to bear theadditional costs of contacting adolescents, hospitalizedpatients, homeless individuals, prisoninmates, and other groups who best representthe treatment-seeking population who are athigher risk for gambling disorders, and who oftenare difficult to contact by telephone.Finally, unless the sample size was considerable,a single national prevalence study wouldbe unable to compare geographic rates, chronologicaltrends, and the impact of shifting socialpolicies or events. Absent sufficient resourcesto ameliorate these matters and the other problemsdescribed earlier, scientists have not beenable to conduct an adequately designed nationalstudy of disordered gambling prevalence thatwould yield useful results.Estimating Prevalence Using MetaanalysisThere is an efficient, expeditious, and lessexpensive method available to developestimates of the extent of gambling problemsin the United States and Canada.This method entails a quantitative synthesis, ormeta-analysis, of previously conducted studieson the prevalence of gambling disorders. This3 Currently, for example, a single telephone interviewrequiring a 20- to 40-minute survey protocol costsabout $75.00 to administer. Most single state samplesrequire a minimum of approximately 7,000 initialinterviews to obtain a sample of disordered gamblersthat provides adequate power for the important comparisonsof interest; therefore, a national survey thatwould permit meaningful data analysis would requiremillions of dollars in survey administration costsalone. Further, the most time-consuming aspect ofmost prevalence studies is developing a samplingdesign and selecting and preparing survey materials;therefore, developing and conducting a nationalprevalence study even with limited objectives wouldrequire expenses of millions of dollars.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling4method is much more cost-effective than a nationalstudy, since it makes use of all of the researchalready conducted in the field. In addition,research in other fields (e.g., Cappelleri etal., 1996) has shown that the results of largestudies are similar to the results of metaanalysesof smaller studies. Shaffer and Hall(1996) conducted the first meta-analytic studyof the prevalence of youth gambling problems.However, at the time this research was conducted,only 11 youth gambling studies wereavailable for analysis. Currently, there are morethan 150 studies on the prevalence of gamblingproblems among youth and adult populations. Acomprehensive meta-analysis of adult gamblingprevalence studies has never been conducted.The study previously conducted by Shaffer andHall (1996) provides the conceptual andmethodological framework for this new broadbasedstudy. The original meta-analyticmethods employed by Shaffer and Hall (1996)will be expanded and enhanced so thatadditional comparisons of moderator variablescan be examined (e.g., evaluations of adultversus adolescent prevalence rates, high-qualitymethods versus less rigorous approaches,treatment populations versus generalpopulations, regional variations in prevalenceestimates).Meta-Analytic MethodologyMeta-analysis, a term coined in 1976 byGlass (Goodman, 1991), is a researchtechnique employed to review andsynthesize a body of research. Scientistsoften conduct meta-analyses when there is adisparity of results from individual studies in afield. This strategy allows for a quantitative,empirical integration of previously derived estimates(Rosenthal & Rosnow, 1991). Scientistshave applied meta-analytic techniques to betterunderstand, for example, the efficacy of varioustreatments or interventions across a broad rangeof health concerns. Scientists have used metaanalyticstrategies to study the efficacy of assessmentand psychotherapy with adults andchildren (Bowers & Clum, 1988; Brown, 1987;Hasselblad & Hedges, 1995; Lipsey & Wilson,1993; Smith & Glass, 1977; Stein & Lambert,1995; Weisz, Weiss, Alicke, & Klotz, 1987;Wampold, Mondin, Moody, Stich, Benson, &Ahn, 1997; Weisz, Weiss, Han, Granger, &Morton, 1995); gender differences (Bettencourt& Miller, 1996; Eagly, Karau, & Makhijani,1995); patterns of marital and friendship relationships(Erel & Burman, 1995; Newcomb &Bagwell, 1995); aggression (Bettencourt &Miller, 1996; Ito, Miller, & Pollock, 1996); depressionand memory impairment (Burt, Zimbar,& Niederehe, 1995); perceptual sensitivity andvigilance (See, Howe, Warm, & Dember, 1995);and educational achievement and ability (Slavin,1990).More specifically, within the field of addictions,meta-analysts have examined areas of drug prevention(Ennett, Tobler, Ringwait, & Flewelling,1994; Tobler, 1986); patterns of alcoholuse and alcohol treatment (Fillmore, Hartka,Johnstone, Leino, Motoyoshi & Temple, 1991;Hartka, Johnstone, Leino, Motoyoshi, Temple,& Fillmore, 1991; Longnecker, Berlin, Orza, &Chalmers, 1988; Longnecker, Orza, Adams,Vioque, & Chalmers, 1990; Miller, Brown,Simpson, Handmaker, Bein, Luckie, Montgomery,Hester, & Tonigan, 1995), and drug abusetreatment (Stanton & Shadish, 1997). In general,a small number of meta-analyses have beenconducted within the domain of prevalence estimation(Habermann-Little, 1991; Lipton &Stewart, 1997; Quigley & Vitale, 1997; Ritchie,Kildea & Robine, 1992; Stewart, Simon,Shechter & Lipton, 1995), and only Shaffer andHall (1996) have examined specifically theprevalence of gambling disorders using metaanalyticmethods. Consequently, this study ofdisordered gambling serves as one of the firstmeta-analytic models for synthesizing a pool ofprevalence estimates.In the field of gambling research, there is meaningfulscope of conceptual, regional, and methodologicaldifferences among the existing studiesthat estimate the prevalence of gambling disorders.A meta-analytic approach allows us tointegrate these disparate studies while respectingthe different strategies and assumptions employedby scientists. In their classic metaanalyticwork on psychotherapy outcome ef-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling5fects, Smith and Glass (1977) noted that, “Mixingdifferent outcomes together is defensible.First, it is clear that all outcome measures aremore or less related to ‘well being’ and so at ageneral level are comparable. Second, it is easyto imagine a Senator conducting hearings on theNIMH appropriations or a college president decidingwhether to continue funding the counselingcenter asking, ‘What kind of effect doestherapy produce—on anything?’ Third, eachprimary researcher made value judgments concerningthe definition and direction of positivetherapeutic effects for the particular clients heor she studied. It is reasonable to adopt thesevalue judgments and aggregate them” (p. 753).Like Smith and Glass, we also can imagine apublic official asking, “What is the prevalenceof gambling problems among young people andadults—the full range of gambling problems—as measured by a variety of estimates?” To answerthis question properly, we must considerthe influence of value judgments made by primaryresearchers and their respective conceptualizationsof disordered gambling. In addition,we must integrate different research methodologiesand gambling patterns that are evidentacross regions of the United States and Canada.A meta-analytic approach to these issues makesit possible to synthesize or pool a range of gamblingstudies to derive a more stable estimate ofgambling prevalence rates than could be obtainedfrom any single approximation (Shaffer& Hall, 1996).Manufacturing Prevalence EstimatesEstimating prevalence is a complex task.This task rests on a variety of conceptualissues. Casti (1989) noted that scientistsview the world through three differentframeworks: (1) realism, (2) instrumentalism,and (3) relativism. 4 The notion that there is a“true” prevalence rate to be identified reflects a“realistic” perspective on science and the scientificmethod. “Realists believe that there is anobjective reality ‘out there’ independent of ourselves.This reality exists solely by virtue ofhow the world is, and it is in principle discoverableby application of the methods of science...this is the position to which most working scientistssubscribe” (Casti, 1989, p. 24). Alternatively,instrumentalists cling “...to the belief thattheories are neither true nor false, but have thestatus only of instruments or calculating devicesfor predicting the results of measurements. Basically,this amounts to the belief that the onlythings that are genuinely real are the results ofobservations” (p. 25). Finally, relativism is becomingincreasingly popular: “...truth is nolonger a relationship between a theory and anindependent reality, but rather depends at leastin part on something like the social perspectiveof the person holding the theory. Thus for arelativist as one passes from age to age, or fromsociety to society, or from theory to theory,what’s true changes. In this view it’s not whatis taken to be true that changes: au contraire,what changes is literally truth itself” (pp. 25-26).4 Realists, instrumentalists, and relativists are similarto three baseball umpires. Like the realist, the firstumpire says, “Some are balls, and some are strikes,and I call them as they are.” Like the instrumentalists,the second umpire says, “Some are balls, andInstead of adopting a realistic or instrumentalistview of prevalence, we have taken a relativisticperspective on the concept of prevalence. Fromthis standpoint, scientists manufacture prevalenceestimates. Scientists adopt strategic andtactical plans, based upon the principles of thescientific method, to generate a prevalence estimate.Instead of simply assuming that a “true”prevalence estimate awaits our capacity to accuratelyidentify it, we believe that a dynamic interplayof factors influences and determinesevery prevalence estimate: which measurementinstrument, with which population, with whichsampling strategy, with which administrativeprocedure, at which historical point in time, unsomeare strikes, and I call them as I see them.” Thefinal umpire, like the relativist, says, “some are balland some are strikes, but they ain’t nothin’ till I sayso!”Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling6der the direction of which scientists all influencethe outcome of an effort to estimate prevalence.By adopting a relativistic posture, we can explicatethe manufacturing process responsible forgenerating prevalence estimates, which in turnwill allow us to improve the quality controlsassociated with these production activities.In addition to being guided by a relativist philosophy,we also think it vital to understand thespecific purpose for which a prevalence estimateis produced. For example, we could assume thatprevalence estimates simply exist and scientificcuriosity motivates the identification process—because it’s “out there.” While scientific interestis often sufficient to motivate an investigation,prevalence estimates have valuable applications.We can use prevalence estimates toallocate limited prevention or treatment resources,estimate social costs, inform diagnosticprotocols, or advise the development and implementationof social policy. With few exceptions(e.g., Sin, 1996; Steinberg, Kosten &Rounsaville, 1992; Thompson, Gazel & Rickman,1996; Vagge, 1996), scientists who haveestimated the prevalence of gambling disordersdo not inform their readers how, where, when,or by whom their newly developed prevalenceestimate will be utilized.Epidemiological ConsiderationsPrevalence versus IncidenceScientists usually investigate prevalence toestimate the public health consequencesof a specific disorder for the purpose ofinforming policy makers and treatmentplanners. Prevalence studies often are conductedwithin a field of inquiry that has yet todiscern the public health burden of a particularproblem. While prevalence estimates provide asnapshot of a disorder at one point or one periodin time, incidence rates estimate new cases ofthe disorder over a specified period of time bydetermining the change that occurs in a populationover time. Thus, incidence and prevalenceestimates answer different questions and servedifferent purposes. For example, an incidencerate can answer the following question: if youfollow a population for five years, what percentageof the population develops their first episodeof a gambling disorder? A related questionan incidence measure addresses is whether thereis a higher rate of disordered gambling followingthe development of a new avenue of gambling(e.g., a new casino, a new lottery game,legalization of a previously illegal gambling activity)than there was before this circumstance.Over the past two decades, researchers haveconducted virtually no incidence studies in thefield of disordered gambling. 5 Instead, the fieldof disordered gambling research has been fascinatedand absorbed by the attempt to establish aprecise and valid prevalence estimate. Prevalencestudies have been conducted on national,state and province, local, and other targetedpopulation levels. This effort has enabled thefield to grow increasingly confident of the baserate of disordered gambling, albeit within avariable range of estimates. One limitation ofsimply continuing to replicate studies designedto estimate prevalence is that two major factorsinfluence these estimates: (1) the scientific processof manufacturing prevalence rates; and (2)the socio-cultural factors (e.g., access to gambling,overall economy) associated with theacceptability of gambling. Moreover, a collectionof prevalence studies is unable to providethe same information revealed by a single incidencestudy.Collectively, researchers, policy makers, thegambling industry, and others appear to be increasinglyinterested in the factors responsiblefor shifting the prevalence rate of disorderedgambling over time. This question requires observationover time and can only be answered5 Winters et al. (1995) collected prevalence data fromthe same cohort at two different points in time. However,their study should be considered a prevalencestudy rather than an incidence study because they didnot focus on the onset and duration of disorderedgambling which would have allowed an estimate ofnew cases (i.e., an incidence rate).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling7with precision by conducting incidence studies.The most appropriate study design to assess incidencerates is a cohort study. Cohort studiesfollow a group of people prospectively overtime. In this type of research, investigators attemptto identify the rates of disordered gamblingacross the risk “exposure” categories. Researcherscan define the risk exposure to gambling,or availability of gambling, in a myriad ofconceptual and concrete ways. Pragmatically,the results of each cohort study will contributeunderstanding to the nature of the initiation, developmentand maintenance of disordered gambling.A series of incidence studies also canbegin to clarify temporal relationships. For example,do disordered gamblers who frequentcasinos develop more problems with alcoholthan the general population because they spendtime in an environment where alcohol is moreaccessible than usual? Or, alternatively, do disorderedgamblers become disordered gamblersbecause they like to drink alcohol and, therefore,are attracted to venues where alcohol is affordableand readily accessible? While prevalencestudies can only speculate about relationshipssuch as these, incidence studies can begin todecipher the “chicken or egg” questions of causality.The Importance of Range EstimatesInvestigators usually report prevalence estimatesof disordered gambling by providing a singlenumber. These estimates represent a distinctstatistic (e.g., 5.7%) designed to indicate theproportion of people who were at risk, or whohave experienced the outcome of interest (e.g.,disordered gambling) according to some criterion(e.g., a gambling screen). Single estimatesare the most simple—and often the most attractive—methodof providing an index of prevalence.However, a range estimate of prevalence(e.g., 5.3% [3.7-6.2]) is more informative than asingle estimate. Range estimates provide ameasure of confidence around the prevalenceindex. For example, a 95% confidence intervalindicates that if a particular prevalence estimateX% was generated by randomly drawing a set of100 population samples, this prevalence estimateX% would reside within the range establishedby the upper and lower bounds 95% ofthe time. Similarly, a 99% confidence rangeestablishes boundaries within which the estimatewould reside 99% of the time. To illustratefurther, for a 99% confidence interval, 99%of all the various sample proportions of size “n”would lie between the upper and lower boundariesestablished by the interval. In the case ofprevalence estimates, a confidence interval identifiesthe domain within which a distribution ofsample proportions reside. Finally, when populationsreside in different, non-overlapping intervals,investigators can claim that the populationsare “different” with increased assurance.Levels of confidence are very useful for determiningthe upper and lower limits of a phenomenon.Furthermore, range estimates providemore guidance than single estimates to policymakers and social planners who use these estimatesfor allocating limited resources. Rangeestimates permit planners to consider the upperand lower bounds of resource allocation; forexample, budget analysts can develop best andworst case scenarios for treatment programs. Inspite of the utility of range estimates, only10.6% of all the studies included in this metaanalysisreported confidence intervals. Thiscircumstance tacitly encourages the belief thatthere is a single accurate or “true” estimate ofgambling prevalence independent of the methodsand procedures employed to generate theseestimates.Time framePrevalence is a measure taken at one point intime that estimates the occurrence of a disorder.If a disorder tends to occur episodically, inwaves of active and inactive episodes, a pointprevalence rate will not necessarily be an accuraterepresentation of the extent of the problem.Pathological gambling, like most addictions, is adisorder that fluctuates in episodic waves (e.g.,Zinberg, 1984; Zinberg & Harding, 1982; Zinberg,Harding, & Winkeller, 1977). In the fieldof disordered gambling, most of the prevalencestudies have addressed this issue by capturing aHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling8period prevalence. Traditionally, disorderedgambling researchers have developed periodprevalence rates by asking individuals whetherthey have experienced a range of problemswithin the past year or during their lifetime.Identifying the time period is of great importancefor understanding the framework of a disorder.Furthermore, the value of lifetime rates isquestionable because of uncertainty about theclustering of lifetime symptoms within a particulartime frame, or confusion about otherpsychiatric disorders that may have stimulatedor sustained gambling behaviors across extendedtime periods (e.g., obsessive compulsivedisorder). In addition, since we can assume thatsome disordered gamblers resolve their problemsand recover, lifetime prevalence rates arenot a particularly reliable index of the currentscope and severity of disordered gambling(Dickerson, 1994). Finally, lifetime rates ofevery population sample are subject to shifts inthe population due to death, memory distortion,and forgetting. We will address issues related totime frames in more detail later in the Discussionsection.Semantics of Disordered Gambling:Conceptual ChaosConceptual and methodological chaos is commonamong emerging scientific fields (Cohen,1985; Shaffer, 1986a; Shaffer, 1997b). Thisdiscord encourages dialogue and debate amongworkers in the field. While researchers in thegambling field agree about the importance ofunderstanding the nature, scope, and severity ofdisordered gambling, there is much variation inthe terms that researchers use to designate variouslevels of disordered gambling. These termscommonly include “problem” gambling, “atrisk”gambling, “potential pathological” gambling,“probable pathological” gambling, and“pathological” gambling. Some authors haveused terms for adolescents that are differentfrom the terms generally used for adults (e.g.,Volberg, 1993a; Winters & Stinchfield, 1993).The particular term attached to a specific levelof disordered gambling severity is less importantthan terminology that allows researchers, clinicians,and others in the field to communicateprecisely and understand one another. Thevalue of any simple description of a complexand multidimension activity such as gambling“…is bought at the expense of a lack of precisionin meaning” (Walker & Dickerson, 1996, p.243). Precision is vital to the conduct of ameta-analysis, since research synthesis takesplace across an assortment of study designs.Therefore, when conducting a meta-analysis, itbecomes mandatory to find an effective classificationsystem that can make sense of the diversityof labels used by scientists to measure comparableconstructs. This classification systemmust permit comparisons among, for example,adolescent prevalence rates derived from studiesthat use different criteria and category labels. Inthe first meta-analysis of disordered gamblingstudies, Shaffer & Hall (1996) suggested a classificationnomenclature composed of disorderedgambling levels. To facilitate the necessarycomparisons, reconcile the divergent methods,and reduce the chaos of contemporary gamblingcategory terminology, we will employ theShaffer and Hall generic multi-level classificationscheme. We will discuss the details of thisclassification system later in the Methods section.To facilitate the conduct of a meta-analysis, investigatorsmust find common ideologicalground on which to integrate the findings of diverseresearch. The multi-level system proposedby Shaffer and Hall provides the conceptualarchitecture to integrate research findings.To use this system, we also require a semanticarchitecture. Therefore, we have chosen to employthe concept of “disordered” gamblingthroughout this report. 6 We have selected thephrase disordered gambling for two primaryreasons. First, the concept of disordered gamblingtranscends each of the existing constructs(e.g., excessive, problem, pathological, and6 Although we have chosen to use the term “disorderedgambling” in most cases, we will use otherterms where they are appropriate (e.g., the use of“pathological” in discussions of APA criteria).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling9compulsive gambling) by recognizing that eachof these categories represents, at various levelsof intensity, a lack of order in one of the majorsystems of human experience. For example,disordered gamblers can experience disruptionof their social, psychological, or biological system.Disordered gambling in its most intenseform can stimulate turmoil in all three of theserealms. Second, the notion of a disorder representsa continuum of experience. Not all gamblersexperience an excessive relationship withthe games they play; not all excessive gamblersexperience compulsive or pathological behaviors;not all pathological gamblers experienceimpairment in every part of their activities. Theconcept of disordered gambling encourages usto recognize the wide range of gradual shiftingof human experiences that can occur amonggamblers who make the transition from regulatedgambling to intemperate gambling.A Brief History of Disordered GamblingPrevalence StudiesIn the mid-1970s, Kallick et al. (1979) undertookthe daunting task of describing the natureand scope of gambling activities in theUnited States on behalf of the U.S. Commissionon a National Policy Toward Gambling.One of their objectives was to determine theextent of “compulsive” gambling among theirsample. While the national survey was beingconducted, Robert Custer was offering theAmerican Psychiatric Association Task Force adescription of compulsive gambling for use inthe Diagnostic and Statistical Manual (DSM)(Kallick et al., 1979, p. 73). Lacking an instrumentwith which to measure compulsive gambling,Kallick and her colleagues created a scaleof 18 items based on concepts from the extantliterature that seemed related to compulsivegambling. This first instrument became knownas the ISR (Institute for Social Research) test.Only one other researcher subsequently usedKallick et al.’s gambling scale (i.e., Culleton &Lang, 1985). Nevertheless, the process of attemptingto accurately measure the construct ofdisordered gambling officially had begun.The results of Custer’s and others’ advice andguidance to the APA on the subject of pathologicalgambling first surfaced in the third editionof the Diagnostic and Statistical Manual(DSM-III; APA, 1980). The appearance oromission of an illness, disorder, or syndrome inthe American Psychiatric Association’s Diagnosticand Statistical Manual is a reflection ofcultural norms, social perception, and currentmedical knowledge. For example, the symptomcluster “post-traumatic stress disorder” first appearedin the DSM-III in 1980, replacing diagnosessuch as “shell shock” and “combat fatigue”(Breslau & Davis, 1987). Conversely, in1973, “homosexuality” was removed from thesecond edition of the DSM (APA, 1973) reflectingthe medical profession’s shift toward viewingsexual orientation as something other than adisorder that needed to be treated (Bayer, 1981).In 1980, at the urging of Robert Custer, pathologicalgambling joined pyromania, kleptomania,and intermittent and isolated explosive disordersas an impulse disorder in the DSM-III(APA, 1980). Since 1980, many researchers andinstrument developers have opted to use theDSM-III or subsequent DSM-based instruments(e.g., DSM-III-R, DSM-IV) to assess and measurethe prevalence of pathological gambling.Kallick et al. (1979) designed The Institute forSocial Research scale to be used in populationbasedsamples by researchers. The DSM-III andsubsequent diagnostic manuals have offered cliniciansa guide to determining whether an individualwho presents with gambling-related problemshas a diagnosable disorder. In 1987, in aneffort to develop a “consistent, quantifiable,structured instrument that can be administeredeasily by nonprofessional as well as professionalinterviewers” (Lesieur & Blume, 1987, p. 1184),Henry Lesieur and Sheila Blume developed TheSouth Oaks Gambling Screen (SOGS). Lesieurand Blume used the DSM-III-R criteria to guideboth the development and validation of theSOGS (Culleton, 1989).The SOGS rapidly became and has continued tobe the instrument of choice among researchersestimating disordered gambling prevalence. Forexample, of the studies included in this meta-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling10analysis, 47.6% used the SOGS as at least oneof the study instruments. If subsequent derivativesof the SOGS (e.g., SOGS-RA, other modifiedSOGS for adolescent samples) are included,this percentage increases to 55.1%. One reasonfor the large proportion of SOGS-based prevalencestudies is the work of Rachel Volberg,who consistently has used the SOGS in hergambling research. Numerous provinces andstates have commissioned Volberg to estimatethe prevalence of disordered gambling in theirgeographic region. The work of Volberg andLesieur account for 17.2% and 11.9%, respectively,of all the prevalence studies identified inthis meta-analysis.Several observers of the disordered gamblingprevalence field have written about the evolutionof prevalence instruments and methodologicalissues in disordered gambling research(Culleton, 1989; Dickerson, 1994; Laundergan,Schaefer, Eckhoff & Pirie, 1990; Laventhol &Horwath, Guida, David Cwi & Associates, &Public Opinion Laboratory, 1990; Lesieur,1994; Lesieur & Rosenthal, 1991; Volberg,1996b; Walker & Dickerson, 1996). For example,Culleton raises the question of the appropriatenessof applying a screening test (e.g., theSOGS) to a population-based sample for thepurpose of establishing a prevalence rate (Culleton,1989). He criticizes this method on the basisof the low predictive value of a test thatscreens for a low base rate disorder within thegeneral population. That is, he believes theSOGS fails to account for the increase in falsepositives when used within a population withlow gambling pathology. Culleton recommendsestimating prevalence using the CumulativeClinical Signs Method (CCSM), a shortenedversion of the Inventory of Gambling Behaviour(Zimmerman, Meeland & Krug, 1985). Culletonconsiders the CCSM to be the best instrumentfor addressing the misclassification offalse positives and for estimating a preciseprevalence estimate. Despite this recommendation,only two studies in this meta-analysis usedthe CCSM, both of which were conducted byCulleton (Culleton & Lang, 1985; TransitionPlanning Associates, 1985). In spite of manyother issues surrounding the screening instrumentdebate sparked by Culleton, his concernwith estimating the prevalence of low base ratebehaviors represents a very important issue forinvestigators. Culleton introduced the importantmatter of positive predictive value to the gamblingliterature. Most simply, screening instrumentsappear most capable of identifying theproblem of interest given a positive score (i.e.,have high positive predictive value) when measuringa phenomenon that is common among thesample population; the accuracy of any screeninginstrument diminishes when investigatorsapply it to a sample where the base rate of thedisorder is low. Even when an instrument hasexcellent criterion validity, “…the actual predictivevalue of the instrument could be much morelimited, depending on the prevalence of the disorderof interest” (Goldstein & Simpson, 1995,p. 236).Dickerson has commented on the limitations ofboth the CCSM and the SOGS. He believes theCCSM is a poorly constructed test with littleface validity (Dickerson, 1994). He argues thatthe reliability of the SOGS is not well established,and that respondents with identicalscores could have entirely different characteristics.Dickerson also suggests that the use of theSOGS may result in an overestimation of prevalenceof pathological gambling (Dickerson,1994). Similarly, Volberg has suggested thatthe SOGS produces inflated estimates of thelifetime prevalence of pathological gambling(Volberg & Boles, 1995). Lesieur (1994) revisitedthe criticisms of the SOGS in his critique ofepidemiological surveys. He states that mostepidemiological surveys underestimate the extentof disordered gambling as a result of methodologicalflaws such as not including thehomeless or hospitalized populations and not“catching” gamblers at home in a telephone survey.While Lesieur is correct on methodologicalgrounds, investigators have still failed torecognize that scientists can identify over- orunder-estimates of a prevalence screening instrumentonly when a “gold” standard also existsto identify the attribute of interest. Theproper question, then, is not (1) whether theSOGS provides an overestimate or an underestimate,or (2) whether the methodological weak-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling11nesses of research protocols offset the uniquemeasurement characteristics of a screening instrument.Neither would it be correct to concludethat the SOGS yields a higher estimate ofdisordered gambling until scientists could assurethat comparison instruments do not underestimatethe prevalence of disordered gambling.Rather, the question is with what independentstandard can we compare the SOGS or any otherestimate of prevalence? Only by evaluating ascreening instrument against an independent andvalid standard can we decide about the precisionof its measurements. In the area of disorderedgambling prevalence, there is no epidemiological“gold” standard. We will discuss this matterin more detail later in the Discussion section.The history of the disordered gambling researchfield reflects the developmental process ofshifting scientific attempts to measure a singularphenomenon. Although various instruments areavailable to assess the prevalence of disorderedgambling, each instrument is best understood byviewing it through an evaluative lens which canfocus on the context of its origin, driving motivation,relationship to funding, and its inherentstrengths and weaknesses.Current Status of the Disordered GamblingPrevalence Field: MethodologicalConsiderationsAmeta-analysis allows a quantitative reviewof the current state of the disorderedgambling field. However, empiricalevidence always derives frommethodological considerations. In the followingsection, we will describe the current status offive central methodological issues associatedwith the development of prevalence estimates.These issues include coverage, populations,special populations, instrumentation, and theoverall methodological quality of prevalenceresearch.CoverageTo properly estimate the prevalence of any psychiatricdisorder among a population (e.g., residentsof the United States or Canada), researchersmust first determine how to represent thepopulation to be studied. To date, there are twomajor clusters of population studies. The firstof these investigations has focused attention onthe general adult or general youth populations.The adult studies typically have employed telephonesurveys. The youth studies rest primarilyon in-school and telephone surveys. However,adults and adolescents who have telephones orwho attend school provide only one perspectiveon gambling problems. Therefore, to better understandthe prevalence of gambling disorders,we must study representatives of the entirepopulation. This group includes many constituenciesoften ignored in general population surveys,for example, the full range of minoritygroups, the homeless, and those individuals whoreside within institutions (e.g., residential substanceabuse treatment programs, psychiatrichospitals, prisons). It is common for both thehomeless and residents of substance abusetreatment programs to experience a disproportionatenumber of psychiatric and other disorders(e.g., Greene, Ennett, & Ringwalt, 1997)compared with the general population. If thisphenomenon is evident with respect to gamblingdisorders, then limiting the assessment to generalpopulation studies will provide a conservativeestimate of the entire population’s gambling-relatedproblems and social costs.PopulationsPrevalence estimates have been derived frommany segments of society. These segments canbe classified into eight general groups: (1) adultgeneral population; (2) youth general population;(3) in-school youth; (4) college studentpopulation; (5) in-treatment adolescents; (6) intreatmentadults; (7) incarcerated adults; and (8)“special populations” (e.g., active bingo players,enlisted military personnel). Although researchers(e.g., Jacobs, 1989; Lesieur et al.,1991) historically have noted that the rate ofdisordered gambling among adolescents exceedsthe rate of disordered gambling among adults,no study has conducted the necessary investigationsto determine the magnitude of this difference.Similarly, we also would expect patientsHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling12with substance abuse or other psychiatric disorders,as well as prison inmates, to reflect higherrates of disordered gambling than the generalpopulation. In this study, we will compare thesevarious rates to determine whether there is ameaningful difference among the levels of gamblingdisorder found among adults, adolescents,patients, and inmates.Special Populations“Special populations” refers to an assorted categoryof prevalence studies that include sampleshaving at least one notable or unique characteristicthat distinguishes the group from the generalpopulation. These distinguishing attributesinfluence the likelihood of the sample to have arate of disordered gambling that is higher thanthe general population. Twenty three (17.2%)of the prevalence studies identified in this metaanalysisrepresent special populations. Thesestudies include samples of military personnel,video lottery terminal players, Native Canadiansand Native Americans, children of disorderedgamblers, and a range of other study samples.In addition to these populations that have beenstudied, there are other special populationswhose rates of disordered gambling either havenot been studied or have been studied in a preliminaryfashion (e.g., Castellani et al., 1996).Some of these special groups may evidencehigher rates of disordered gambling than thegeneral population. These population segmentsinclude senior citizens or the elderly, gay menand lesbians, and the homeless. To illustrate, asegment of senior citizens often have more timeand disposable income on hand; consequently,they are often targeted for gambling excursions.Gay men and lesbians have been shown to havehigher rates of substance abuse than the generalpopulation (Cabaj, 1992; Diamond & Wilsnack,1978; Lohrenz, Connelly, Coyne & Sparks,1978), due in part to the importance of gay barsand clubs within the gay community and culture,and perhaps even to the dysthymic consequencesof experiencing homophobia and discrimination(e.g., Glaus, 1989). The homeless,while seemingly having less disposable income,also may have a special attraction to gamblingbecause of the promise of a turn of good luckand seemingly few other opportunities to changetheir economic standing. The one study of disorderedgambling among the homeless (Castellaniet al., 1996) indicates that homeless individualswith gambling problems have significantlyfewer coping skills than homeless individualswithout gambling problems. We canhypothesize that there may be elevated rates ofdisordered gambling among the homeless, gaymen and lesbians, and the elderly. Future researchwill be instrumental in testing these hypotheses.During the early years of studying disorderedgambling prevalence, scientists tended to bemost interested in determining the scope andseverity of disordered gambling among the generalpopulation, both adults and adolescents. Asmore estimates became available describing thegeneral population, researchers turned to intreatmentgroups and special populations tomore closely investigate correlates of disorderedgambling and factors characterizing specificsubgroups of the population. In-treatment populationsand special populations (e.g., regularVLT gamblers, Gamblers Anonymous members)tend to have higher rates of disorderedgambling than the general population. Theshifting of this pattern of research interests overtime has resulted in a perception that rates ofdisordered gambling are increasing. That is,reported rates from more recent studies are oftenhigher than earlier reported rates. Complicatingmatters, general population studies usually havelarger sample sizes compared with smaller studiesof special populations. It is therefore critical,when observing patterns over time, to takeinto account the specific populations being studied.InstrumentationA total of 25 different instruments were used inthe collected pool of studies included in thismeta-analysis. Of these studies, 55.1% used theSOGS or derivatives of the SOGS for their estimates,11.1% used DSM criteria, 5.2% used theDiagnostic Interview Schedule (DIS), 1.7% usedHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling13the multi-factor method, 7 and 1.5% used theMassachusetts Gambling Screen (MAGS).Other instruments were used in 25.4% of theremaining studies included in this analysis.Some researchers have suggested that when usingthe South Oaks Gambling Screen (SOGS),“…estimates of the prevalence of problem andpathological gambling generated… are higherthan the estimates generated by different instruments…”(Christiansen/Cummings AssociatesInc. et al., 1992, p. 116). Consequently,Christiansen/Cummings Associates Inc. et al.(1992)—recognizing the relative nature ofprevalence estimates—suggest that one cannotcompare prevalence estimates of disorderedgambling across the use of different instruments.In addition to producing descriptive data on theuse of disordered gambling instruments to date,the use of meta-analytic methods permits an explorationof two related questions: (1) can differentinstruments’ estimates be compared, and(2) if yes, what are the differences in the estimatesbetween instruments?Some research on instrument comparison hasbeen conducted. 8 Within a study, researchersmay elect to use two or more instruments for thepurpose of comparing the instrument-based estimates.For example, when comparing theSOGS-RA, DSM-IV-J, and the GamblersAnonymous (G.A.) 20 questions, Derevenskyand Gupta (1997) found that the DSM-IV-Jyielded the most conservative estimates ofprevalence, even though the inter-correlationsamong all three instruments were moderatelyhigh (range .61 to .68). While reviewing theapproaches used to estimate disordered gambling,Dickerson commented “there is no doubtthat any prevalence work in this area stands orfalls according to the accuracy of the availablemeasurement instruments for detecting cases ofpathological gamblers in the general population”(Dickerson, 1994, p. 3). An approach that considersinstruments to hold various degrees ofaccuracy implies that there is a “true” prevalencethat can be measured. Our approach is, aswe stated earlier, that the choice of instrument isone of the important factors in the process ofmanufacturing prevalence estimates. The gamblingresearch field does not yet have an accepted“gold standard,” nor does it currentlyhave a biological marker that might diagnosepathological gambling with high “accuracy.”Until the field develops standardized tools, theability of an instrument to successfully determinewhether an individual is a pathologicalgambler remains dependent upon the method ofvalidation, interviewing technique, samplingdesign, and other methodological factors.Methodological QualityEpidemiologists think of “quality” as that which“gives us an estimate of the likelihood that theresults are a valid estimate of the truth” (Moher,Jadad, Nichol, Penman, Tugwell & Walsh,1995, p. 62). The cluster of factors that contributeto the methodological soundness of a studymay influence that study’s prevalence estimate.A meta-analytic research strategy allows us toevaluate the quality of the research methodsemployed by each principal investigator and tocompare the influence of these methodologicalattributes across the range of studies. Thiscomparison allows us to determine whethermore methodologically rigorous research designsyield significantly different estimates ofdisordered gambling (e.g., Miller et al., 1995).Intuitively, perhaps, investigators might thinkthat more rigorous methods would producehigher prevalence rates than less demandingresearch protocols. For example, rigorous researchdesigns could influence the sensitivity 97 The multi-factor method also is based on the SOGS,and represents a different method of calculating disorderedgambling.8 See Table 14, which lists each study included in themeta-analysis that reported rates on more than oneinstrument.9 Sensitivity refers to the probability that a test resultis positive among individuals with the characteristicof interest.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling14and specificity 10 of screening instruments forgambling disorders to yield either higher orlower prevalence estimates than designs withlimited methodological rigor. In the case oflowered test sensitivity and increased specificity,for example, prevalence screening researchwould generate conservative estimates. A carefulconsideration of methodological quality mayassist us in identifying the most appropriate researchdesigns for future prevalence studies.Measurement, Prevalence Estimation,and Social SettingReaders should bear in mind that gamblingproblems exist on a continuum,ranging from very minor problems atone end to very severe problems at theother end. To facilitate the delivery of appropriatetreatment, the estimation of prevalence,and the understanding of disordered gambling,clinicians and researchers have divided this continuuminto distinct categories (e.g., nonproblem,problem, pathological). However,these categories are simplifications of the existingcontinuum of gambling problems, and thedetermination of the division between one categoryand another is ultimately arbitrary. Thus,the threshold of, for example, level 3 gamblingcan shift from one study to another dependingon the authors’ respective conceptualizations oflevel 3 gambling and their purposes for estimatingprevalence.Researchers require a better understanding ofthe data needs in the field of disordered gamblingresearch in general and youth gamblingresearch in particular. As the previous paragraphsuggests, differences among instrumentsin criteria and category thresholds may result inthe generation of different prevalence rateswhen these instruments are used among thesame sample. In other words, some instrumentsconsistently provide more conservative esti-10 Specificity refers to the probability that a test resultis negative among individuals without the characteristicof interest.mates and other instruments consistently providemore liberal estimates. For example, in herstudy of Washington State adolescents, Volbergstates that “Our approach, while conservative, isintended to focus as clearly as possible on thoseadolescents who show incontrovertible signs ofproblematic involvement in gambling” (Volberg,1993a, p.17, emphasis added). Comparedto more sensitive standards, Volberg’s strategywill yield a lower estimate of level 3 gamblingprevalence. Indeed, this conservative adolescentinstrument yielded a level 3 estimate of 0.9%among this adolescent sample, while the “adult”scoring system yielded a level 3 estimate of1.5% among this sample. As this example illustrates,no prevalence estimate exists independentof the criteria used to determine the disorder.Investigators’ decisions about the nature of theirinstruments’ criteria (e.g., where to establish the“cutoff points” for the different group categories)determine the prevalence estimates thattheir instruments will provide.The process of redefining disorders is common.Both scholars and the public have recognizedthe need for shifting standards related to bothbiological and psychological disorders (e.g.,Centers for Disease Control and Prevention,1992; Flavin & Morse, 1991; Kleinman, 1987;Knox, 1997; Schuckit, Nathan, Helzer, Woody,& Crowley, 1991; Wakefield, 1992). For example,the Centers for Disease Control expandedthe surveillance case definition for AIDS to includepulmonary tuberculosis, recurrent pneumonia,and invasive cervical cancer as potentialindications of low CD4+ T-lymphocyte countsor percentages (Centers for Disease Control andPrevention, 1992). The case of diabetes is similar.Instead of defining diabetes as a conditionmarked by 140 milligrams of glucose per deciliterof blood, a shifting standard marked the disorderas a condition evidenced by a blood sugarthreshold of 126 milligrams of glucose per deciliterof blood (Knox, 1997). The shifting definitionsof both AIDS and diabetes served to increasethe numbers of people clinicians diagnoseand treat, compared with the numbers whowould have been cared for under the old criteria.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling15The decision to employ a more or less sensitivemeasure should take into account the purposesfor which the prevalence estimate was developed.For example, although a less sensitivemeasure would reduce the number of adolescentswho might be stigmatized by a “pathologicalgambler” label, it also would reduce the opportunityfor early identification of gamblingproblems and primary prevention and treatmentefforts. Conversely, a more sensitive measureof gambling problems would increase the magnitudeof the observed problems of adolescentgamblers but also would give advance warningof developing problems and create the opportunityfor focused education, prevention andtreatment programs.Finally, every attempt to estimate the prevalenceof a gambling disorder depends upon the interactionbetween (1) the measurement instrumentsand methods employed to construct a prevalenceindex and (2) the social context that influencesthe meaning and experience associated with averageday-to-day patterns of human behavior(Zinberg & Shaffer, 1985). Therefore, estimatesof gambling problems must consider the shiftingsocial milieu that is affecting those who gamble,their associated experiences, and the impact ofthis process on prevalence estimates. We willrevisit this theme later in the Discussion sectionof this report.The Central HypothesesAlthough more than 150 disordered gamblingprevalence studies have been conductedto date, there are a variety ofcentral hypotheses that have not yetbeen explored empirically. This meta-analysisof prevalence estimates will address the followingmajor hypotheses.1. Prevalence estimates of different populationsegments (e.g., youth and adult,males and females) will yield meaningfullydifferent estimates of disorderedgambling.2. The increased access and exposure togambling opportunities of the past 23years will be reflected in an increase inprevalence of gambling and the problemsand disorders associated with thisactivity. In addition, over the past 23years, the rates of gambling-associatedproblems will have increased or decreasedat different rates dependingupon the populations (e.g., in-treatmentpatients, prison inmates, youth) fromwhich the prevalence estimates were derived.3. The instruments used to generate prevalenceestimates will influence the rate ofobserved gambling problems.4. Different geographical regions will havedifferent rates of disordered gambling.5. Investigators will differ in the prevalenceestimates they generate as a consequenceof their characteristic researchmethods. In this hypothesis, we canview researchers as a proxy for studyprotocols, instrumentation, and an assortmentof other study attributes.6. Experience playing different types ofgambling activities (e.g., sports bettingor lottery playing) will influence disorderedgambling prevalence rates differentially.MethodsIdentifying Primary StudiesWe conducted a literature search toidentify the maximum number ofpublished and unpublished studies onthe prevalence of disordered gambling.We began by examining every issue ofthe Journal of Gambling Studies (formerly theJournal of Gambling Behavior) up to and includingthe Spring 1997 issue. This was an importantstep in optimizing the number of studiesavailable for analysis, since the Journal of GamblingStudies is not yet indexed in Medline, andwas not indexed in PsychInfo until 1985. Inaddition, to identify studies that received limitedHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling16distribution, we searched the library of the MassachusettsCouncil on Compulsive Gambling forunpublished articles.In May, 1996, we sent a letter to 10 prominentresearchers in the field of disordered gambling.In this letter, we included a list of the studies wealready had compiled and asked for assistance inidentifying and locating studies that were not onour list. In October, 1996, we repeated the initialmailing and expanded the list of recipientsto 60 individuals. On this occasion, we included40 researchers, clinicians, and program administratorsworking in fields related to gambling. Inaddition, we sent this letter to 17 state Councilson Compulsive Gambling and 3 representativesof the gaming industry. We received 18 responsesto these letters. In addition, we conductedsearches of standard research databases,including Medline, PsychInfo, and the HarvardOnLine Library Information System (HOLLIS).The Medline database indexes 3,600 journals.In Medline, keyword searches consistentlyyielded more entries than text word searches, sowe searched for the keyword “gambling.” InMedline, the keyword “gambling” is categorizedunder “risk-taking” and “impulse control disorders.”Combination searches of “gambling” and“prevalence” identified a total of 8 studiesamong the five Medline databases; therefore, webroadened our search to every “gambling” itemidentified by this strategy. Table 1 summarizesthe yield of prevalence studies from this comprehensivesearch strategy.Eligibility CriteriaTo be eligible for inclusion in this study, aprevalence study had to meet the followingseven criteria:1. the study had to be a report written inEnglish;2. the study had to be conducted in eitherthe United States or Canada;3. the study had to specify what instrumentor set of criteria was used to identifydisordered gambling;4. the study had to report the prevalence ofdisordered gambling among the sampleit studied;5. the study had to report the size of thesample it studied;6. the study had to be conducted betweenJanuary 1, 1974 and June 15, 1997;Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling17DatabaseYear of DatabaseTable 1: Literature Search ResultsTotal # of“Gambling”itemsTotal # identifiedstudies(overlapping)Total #distinct studiesTotal #eligible studies(ineligible removed)HOLLIS 1975-1997 873 1 1 1Journal of GamblingStudies(JGS)Spring/Summer1985 - Spring199726 9 6Medline♦ Medline 1966-1975 168 0 [0]♦ Medline 1976-1980 41 0 [0]♦ Medline 1981-1986 55 1 [1]♦ Medline 1987-1992 112 9 [9]♦ Medline 1993-June 1997 111 18 [15]♦ Citation appearsonly inMedline 1975-1997 3 3 2PsychInfo♦ Cited in PsychInfoonly 6 [6]♦ Cited in bothPsychInfo &JGS 18 [10]♦ Cited in bothPsychInfo &Medline 25 [23]♦ PsychInfoTotal 1984-1997 851 49 49 39Invisible College64 64 52On File Prior toStudy 12 26 26 20TOTAL 152 120 137. the study had to be available to the authorsfor review by June 15, 1997.11 Studies obtained through the “invisible college”(Rosenthal, 1994); studies in this category werebrought to our attention by colleagues or were sent toHarvard Medical School Division on Addictionswithout being requested.12 Studies that we “had on file” were in our librarybefore the present study.13 Bracketed numbers are not included in the total.Data Abstraction and Data Entry ProceduresThe first step of the data abstraction andentry procedure was to identify the variablesfor which we wanted data. Afterreviewing a sample of representative disorderedgambling prevalence studies and relevantmeta-analyses in other fields, we identified313 variables for which data would be collectedfrom each study. Multiple-choice coding optionswere assigned to each variable for which amultiple-choice format was appropriate. Wethen arranged the set of core variables (i.e.,those we thought should be reported in everystudy) on a six-page data collection codingHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling18form. Additional variables that we expectedonly a small percentage of the studies to reportwere arranged on “attachment forms.” When astudy presented data for these additional variables,we included an appropriate add-on datacollection form and attached it to the main form.We developed attachment forms for stratifiedprevalence estimates (gender, age, etc.), confidenceintervals, and illicit gambling.We then assigned four members of our researchteam to the job of “coding,” or identifying therelevant information in the studies. This groupof coders abstracted the information from eachstudy by completing a six-page data collectionform. The coders attended a comprehensivecoder training. Each of the coders had expertisein the substantive areas covered by the codingsystem. We pilot tested the coding protocol andthen revised it based on this experience. Toavoid any coding biases potentially associatedwith our research team members, we randomlyassigned each of the 152 studies to two of thefour coders. Data abstraction and entry thenproceeded according to the following eightsteps:1. Two randomly assigned coders abstractedinformation from each study.2. For each study, the two coding formswere compared and any discrepancieswere identified.3. We grouped the data abstracted fromeach study into one of the followingthree categories: (1) discrepancies onmajor data components; (2) discrepancieson individual variables; and (3) nodiscrepancies between the two coders.Major data components included thenumber of prevalence estimates providedin a study and the types of attachmentforms required completion.4. For studies with discrepancies on majorcomponents, we held a “ComponentDiscrepancy Resolution Session.” Duringthis session, each pair of coders metto discuss and resolve discrepancies relatedto major components. When aresolution required one coder to completeforms that the other coder alreadyhad completed, the coder would completethese forms independently, withoutreferring to the forms already completedby the other coder. In caseswhere coders could not agree on a resolution,the study was sent to the arbiter(a senior member of the research teamwho was not involved in coding) to beresolved. Once the two coding formsfor a study had equivalent components,the individual variables were compared.5. Coding forms with discrepancies on individualvariables were returned to coderswith the discrepant items marked.The coders reread the correspondingstudies and recoded these items withoutknowing how the other coder had respondedpreviously.6. Recoded studies were compared again.Discrepancies that remained weremarked and sent to the arbiter for finalresolution.7. Once all discrepancies had been resolved,studies were entered into anSPSS database. Every prevalence estimatewas double-entered by two differentstaff members to ensure accuracy.8. We assessed our data entry reliability byrandomly selecting 10% of the cases inour database and double-checking eachdata entry point. Of the 6,573 variablesentered in these 21 randomly selectedstudies, 13 data points, or .002% of thevariables entered for these studies, wereidentified as data entry errors. This percentageis well within the acceptablerange of less than 1% errors for data entry(e.g., McAuliffe et al., 1995).We assessed inter-rater reliability among thefour coders and the discrepancy rate of the totalnumber of coded judgments by taking a randomselection of 15.3% of each of the six coder-paircombinations from the total number of cases inthe data set. We defined a discrepancy as a dis-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling19agreement between the two coders that requiredresolution by the arbiter, as described in step 6above. The discrepancy rates for the six coderpaircombinations ranged from 1.17% to 5.38%with a mean of 3.36%. The discrepancy ratesfor each of the four coders evaluated across theirstudy samples were 2.34%, 2.45%, 3.91%, and4.59%. Finally, the total discrepancy rate for allcoders was 3.33%, yielding a study-wide ratefor coder reliability of 96.67%. 14Quality Component Analysis: MethodologicalQuality ScoresMethodological quality scores providean index of a study’s design and procedureattributes. Meta-analysts oftenuse quality scores to adjust for thewide range of methodological approaches chosento estimate similar constructs. In the field ofdisordered gambling prevalence research, methodologicalrigor varies enormously among studies.To evaluate the methodological quality ofstudies that provide estimates of disorderedgambling, we created a methodological qualityscore. We will briefly discuss our method ofcalculating the methodological quality score inthis section and later in more detail in the Resultssection. Now, we will describe some of14 To determine the percentage of absolute agreementamong raters, the simplest and most easily interpretedmeasure of inter-rater concordance is the percentageof ratings about which coders agree (Bangert-Drowns, Wells-Parker, & Chevillard, 1997). However,Bangert-Drowns et al. also note that we wouldexpect some of these agreements by chance. Toavoid over estimating inter-rater reliability, the kappastatistic (Cohen, 1960) provides a measure of theratio of agreements beyond those expected by chanceto the total number of observations minus agreementsexpected by chance. However, ultimately, inter-raterreliability is not an issue in this study, since all but thedata entry errors were resolved by arbitration. Therefore,we have not provided a kappa adjustment. Weinclude inter-rater reliability estimates to demonstratethat the arbitration process had minimal impact oncoding, and that coders were readily able to abstractdata objectively.the central methodological characteristics of theextant set of gambling prevalence studies. Forexample, among the studies included in thismeta-analysis, only 61.9% reported a responserate. Of these, the response rate ranged from11.39% to 100%. The mean response rate forthose that reported one was 71.89% (s.d. =22.1). Although 71.89% may seem to fallwithin an acceptable range for response rates,this finding is misleading due to several considerations.For example, consecutive admissionsto treatment within a specified time period werecoded as having a response rate of 100% so longas each admission agreed to participate in theresearch. Furthermore, only 44.9% of the studiesreporting a response rate used a denominatorwhich represented the entire pool of respondentseligible to participate in the study. 15 A numberof studies (11.1%) used a denominator that didnot include all eligible respondents, and 44% ofthe studies did not specify whether the denominatorused to calculate the response rate representedall of the eligible respondents. In addition,descriptions of age and gender of a studysample are essential because these are two factorsthat have been shown to influence rates ofdisordered gambling. Yet 9% of the studies didnot report which age group (e.g., adolescents,adults) their study examined, and a more alarming26.3% did not report the gender breakdownof their sample.These and other methodological factors providea rationale for considering the assignment of aweight to each included study corresponding toits methodological quality. This procedure assuresthat more methodologically sound studiesexert more influence on synthesized estimatesthan studies having more methodological weaknesses.This weighting system is based on theassumption that studies with better methodologyprovide more precise estimates. However, thisassumption is not inexorably accurate. For ex-15 When possible, if a study reported a response rateusing a denominator other than all eligible respondents,we calculated a new response rate with thegiven data.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling20ample, Blair et al. (1995) state that the use ofquality scores “…is controversial and approachesto it vary widely. Consensus was notreached on use of quality scores…. it has beenargued that quality scoring should be abandonedin favor of quality-component analysis…. Qualityscores, if used at all, should be specific tostudy characteristics such as study design ormethods of exposure assessment, rather thanproviding a single overall measure of quality”(Blair et al., 1995, p. 195). When quality scoresare used to weight studies, this procedure“...imposes subjective criteria into what purportsto be an objective procedure” (p.136). Alternatively,Hasselblad et al. (1995), Blair et al.(1995), and Olkin (1995) provide guidelines fordeveloping and applying quality indices in metaanalyticstudies and minimizing potential risksto objectivity. We have followed their suggestionsand avoided subjective rating scales tomeasure quality by constructing a quality componentindex that evaluates whether studies includevarious attributes; this method does notuse subjective rating scales for determiningquality. We developed the methodological indexfor this study as follows: First, a focusgroup of researchers convened to identify themethodological elements associated with highmethodological quality for prevalence estimationresearch. This procedure identified the followingnine methodological factors as essentialcomponents to any assessment of the internalvalidity of prevalence research:1. sample selection process (i.e., randomlyselected sites and/or respondents);2. response rate (including the appropriatenessof the method used to calculatethe response rate);3. survey anonymity (not applicable fortreatment or prison samples);4. whether the study underwent a peer reviewprocess (e.g., for publication in arefereed journal);5. whether the authors assessed the reliabilityof their data collection and entryprocedures;6. whether the authors varied the time ofday survey data was collected (forhousehold surveys);7. the number of respondents in the studysample;8. whether the authors took a multidimensionalapproach to measuring disorderedgambling (e.g., multiple dependentmeasures); and finally,9. whether the study was intended primarilyas a prevalence study.Rather than asking coders to apply their subjectivejudgment to assess the methodological qualityof a study, we asked coders to abstract objectivedata on these nine factors. These factorswere then combined into a single aggregate index.Since prevalence studies vary by settingand population, we calculated this methodologicalquality score differently for each study type.For example, factor 3 above (i.e., anonymity)was not used in the calculation of the qualityscore for treatment studies or prison studies,because conducting an anonymous survey wouldbe difficult given the nature of these settings. Inaddition, factor 6 above (i.e., varying the time ofday for collecting data) applied only to householdtelephone surveys. Since we used differentmethodological elements in the calculation ofthe composite index for different study types,we also standardized each composite index bycalculating it as a percentage of the maximumtotal possible. The standardized compositequality variable derived from this procedure iscontinuous and distributed normally. We willexamine the impact of quality scores on prevalenceestimates of disordered gambling. In addition,we will employ multiple statistical proceduresfor analyzing the moderator variables thatinfluence prevalence rates.In spite of our best efforts, we remain aware thatthere are many remaining threats to the integrityof quality scores. For example, Moher et al.(1995) warn that, “It is important to distinguishbetween assessing the quality of a trial and thequality of its report. We define the quality of atrial, our primary interest, as ‘the confidenceHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling21Table 2: Level System for Meta-analysisLevelDescription of Level3 Represents the most severe level of disordered gambling2 Represents gamblers with sub-clinical levels of gambling problems1 Represents the proportion of the population that does not experience gambling problemsthat the trial design, conduct, and analysis hasminimized or avoided biases in its treatmentcomparisons.’ This definition focuses on methodologicalquality. The quality of a trial reportcan be defined as ‘providing information aboutthe design, conduct, and analysis of the trial.’ Atrial designed with several biases that is wellreported can receive a high-quality score. Conversely,a well-designed and well-conductedtrial that is poorly reported would receive a lowqualityscore” (p. 63). Although ideally the objectof quality evaluation would be the trial (i.e.,the study itself) meta-analysts can evaluate onlywhat is included in the report. Therefore, a reportthat fails to reflect rigorous methodologywill yield a relatively low quality score in thismeta-analysis, even if investigators actuallywere meticulous about their design and protocols.This problem is somewhat existential: Ifscientists employ precise methods and fail toreport them accurately, will anyone know oftheir rigor?Nomenclature & Classification: Levelsof Disordered Gambling SeverityApplying the Level System to MetaanalysisAs Shaffer & Hall (1996) noted, disorderedgambling prevalence studies haveused a wide array of criteria and labelsto define and name the levels of disorderedgambling severity. However, while theclassification methods and labels are not identical,the majority of these designs are conceptuallyparallel. For example, what might be called“pathological” gambling in one study might becalled “compulsive” gambling or “probablepathological” gambling in another study or even“problem” gambling (e.g., in the MultifactorMethod and the SOGS-RA). Similarly, a groupcalled “problem” gamblers in one study mightbe conceptually equivalent to the group labeled“at-risk,” “in-transition,” or “potential pathological”gamblers in another study. Thus, forthe purposes of comparing conceptually equivalentgroups derived from different studies, wehave translated these categories into three genericlevels of disordered gambling.The first of these three levels, level 1, representsthe proportion of the population that does notexperience gambling problems. This group includesboth “non-problem” gamblers and nongamblers.The second level, level 2, representsgamblers with sub-clinical levels of gamblingproblems (e.g., “problem,” “at-risk,” “intransition,”“potential pathological”). The thirdlevel, level 3, represents the most severe categoryof disordered gambling (e.g., “pathological”gambling). Table 2 summarizes this conceptuallevel system. In many studies, level 3gamblers are those who meet established diagnosticcriteria for pathological gambling (e.g.,the DSM-IV criteria). In other studies, the establisheddiagnostic criteria have been modified,but the group remains conceptually equivalent.As stated previously, there was wide variation inhow the original authors classified respondentsinto these groups. In each case, we used theoriginal authors’ definitions of the levels describedabove. For example, some studiesamong adults defined level 2 as a score of 3 or 4on the SOGS, while other adult studies definedthis level as a score of 1 to 4 on the SOGS. Wecoded this data as it was reported, since wewanted to respect the research methods used byHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling22each original author and represent these methodsin our synthesis of the extant research.However, we found that studies that did notprovide an estimate of level 2 gambling hadsubstantially higher rates of level 1 gamblingthan studies that did provide level 2 estimates.In other words, if a study did not classify respondentsinto a level 2 group, those respondentswho would have been classified in thisgroup in another classification system were includedin the study’s level 1 group. Thus, tostandardize the definition of level 1 for the purposesof this study, we analyzed only estimatesof level 1 gambling that were derived from studiesthat also reported estimates of level 2 gambling.Level 2: In-transition Implies BidirectionalityLevel 2 gambling includes categories such as“in-transition,” “at-risk,” and the SOGS “problem”gambling. Although several of the diagnosticsystems reviewed in this paper, such asthe DSM, Pathological Gambling Signs Index,and the Gamblers Anonymous 20 Questions donot have a level 2 category, most of the classificationschemes currently in use do employ thiscategory.The matter of whether level 2 gamblers are “intransition” or “at risk” represents a very importantconceptual concern and practical problemthat has not yet been addressed adequately byempirical research. Most classification systemsreviewed in this paper organize the signs andsymptoms of gambling pathology into subclinical(i.e., symptoms do not meet diagnosticcode) and clinical categories. However, consideringthat gambling pathology likely resides on acontinuous rather than dichotomous dimension,and its intensity for any individual may changeover time, sub-clinical signs and symptoms maybe pre-clinical (i.e., have not yet reached clinicalstate) or post-clinical (i.e., are receding from aclinical state). With the exception of the notionof in-transition gamblers in the MAGS (Shafferet al., 1994), none of the extant classificationschemes has conceptually recognized the potentialpresence of level 2 gamblers who are movingaway from pathology. Most addiction modelsin general and gambling classification strategiesin particular imply that sub-clinical gamblersare simply moving in one direction, towarda more pathological state. However, since pastyearestimates of adolescent gambling pathologyconsistently exceed past-year estimates of adultgambling pathology, it also is possible that someyoung people are actively moving away from astate of pathological gambling (for relevant discussionsof stage change and addictive behavior,interested readers should see Prochaska, Di-Clemente and Norcross, 1992, and Shaffer,1992, 1997a).Estimation MethodsWeightingWeighting is a specific strategy thattakes into account the inevitable realityof heterogeneity among studysamples. That is, there are differencesamong studies with respect to sample size,sample population characteristics, and other factorsthat can influence the accuracy of a prevalenceestimate. To illustrate how sampling caninfluence estimation accuracy, consider a practicalapplication of the central limit theorem. Aprevalence study of 1,000 adults—sampled froma population of 10,000 people—will produce amore precise and representative estimate of disorderedgambling prevalence than a study of100 adults drawn from the same population of10,000. Stated differently, a smaller sample willtend to be more biased than a larger sample.When sampling bias is predictable, it is justifiableto assign appropriate weights (i.e., values)to study attributes to compensate for the inherentbias (Cooper & Hedges, 1994).We weighted the cases in our database to controlfor the following two factors: (1) multipleprevalence estimates derived from the use ofmultiple instruments within a single study population(e.g., using the SOGS and the DSM-IV astwo estimates of prevalence for a single adultsample); and (2) single studies published in multipleformats (e.g., the same study published as aHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling23book chapter and a journal article). For the firstfactor above, we created a weighting variablecalled Weight1. For this weighting variable,each prevalence estimate received a value equalto 1/(the number of instruments used to deriveprevalence estimates in the study). For example,if a study used three different instrumentsand provided three different prevalence rates fora single study sample, each of these estimateswould receive a weight of .33. For the secondfactor above, we created a second weightingvariable called Weight2. To determine a case’sweight on this variable, we identified and comparedidentical studies published in multipleformats. For studies that were published onmultiple occasions, we assigned a weight of 1 tothe publication that reported the most data; otherversions of the same study received weights of0. Finally, the two weighting variables weremultiplied; the product, a new weighting variablecalled Weight3, was used in all analysesunless otherwise specified.Prevalence Estimates: A Multi-MethodApproachOne of the primary objectives of this researchwas to establish more precise estimates of theprevalence of disordered gambling among thegeneral population of adults and youth in theUnited States and Canada than currently areavailable. Although estimating the central tendencyof a distribution of estimates (e.g., providinga meta-prevalence estimate for a population)appears straightforward, there are a variety ofcomplex issues associated with this process.Since the various estimates more distant fromthe mean of the set of estimates can be weightedmore or less, depending upon the statisticalstrategy, we have opted for a multi-methodstrategy (Campbell & Fiske, 1959). This strategyemploys a variety of statistical techniquesfor estimating the central tendency of the distributionof prevalence estimates. One of thesetechniques is the use of M-estimators, or robustmaximum-likelihood estimators (e.g., Barnett &Lewis, 1994; Norusis, 1993; Winer, 1971). M-estimators of central tendency differ in theweights they apply to the cases, or in this instancethe obtained prevalence estimates. Extremevalues receive less weight than valuescloser to the center of the distribution. Whenthe data is from a symmetric distribution withlong tails, or when the data set includes extremevalues, M-estimators provide better estimates ofthe location than do the traditional mean or median.The four estimators included in thisanalysis are Huber's M-estimator, 16 Andrew'swave estimator, 17 Hampel's redescending M-estimator, 18 and Tukey's biweight estimator. 19 Inaddition to the use of M-estimators, we haveincluded the unweighted mean, and otherweighted indicators (i.e., 5% trimmed mean, 2016 An M-estimator of location. Cases with standardizedvalues less than c receive a weight of 1 (c is amathematically determined standardized score valuethat serves as a “cutoff” point and has a differentvalue for each of the four M-estimators describedhere). Cases with absolute values larger than c haveweights that decrease as their distance from zero increases.17 A type of redescending M-estimator that does nothave abrupt changes in the weights assigned to cases.Instead, a smooth sine curve is used to determine caseweights. Standardized values in absolute value greaterthan c are assigned a weight of 0.18 A three-part redescending M-estimator that is characterizedby three constants (a, b, c). Standardizedobserved values with an absolute value greater than care assigned a weight of zero. Values between 0 and aare assigned a weight of 1, while values between aand b and between b and c are assigned weights thatdepend on their distance from zero.19 Tukey's biweight estimator assigns weights of zerofor observations with standardized values greater than4.685 and weights inversely proportional to the distancefrom the center for all other observations.20 The arithmetic mean calculated when the largest5% and the smallest 5% of the cases have been eliminated.Eliminating extreme cases from the computationof the mean results in a better estimate of centraltendency, especially when data are not distributednormally.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered GamblingWinsorized estimates, 21 and medians which representvalues that are 50% trimmed), and estimatesweighted by quality score for comparison.Appendix 1 provides the variety of data pointsassociated with various statistical strategies forcombining measures of central tendency. Finally,although the collinearity of prevalenceestimates with other factors discourages the developmentof a meta-point estimate, it is possibleto compute such an estimate. With this cautionin mind, for those interested, we have includedthe mean of these strategies as a “meta”point estimate.In addition to the measures of central tendencydescribed above, we calculated the 95% confidenceinterval around each central estimate,based on the distribution of prevalence estimates.Taken together, these multi-method strategiesyielded calculations of an assortment of valuesfor gambling levels 1, 2, and 3, in both lifetimeand past-year time frames. These calculationprocedure for these values is summarized accordingto the following six steps:1. the unweighted median;2. the mean and the 95% confidence intervalaround the mean;3. four different M-estimators (or maximumlikelihoodestimators);4. the mean with the 5% most extreme casesremoved;5. steps 1-2 above repeated with a Winsorizeddistribution;6. steps 1-4 above repeated with each caseweighted by its methodological qualityscore value.A Note on Time FramesThe majority of studies included in thismeta-analysis (62.7%) indicated whattime frame their prevalence rates represented(e.g., lifetime, past-year); however,over one third of the studies (37.3%) failedto indicate the time frame for their prevalencerates. Prevalence rates from these studies werere-coded to represent a lifetime time frame. Inaddition, studies that indicated that their prevalencerates represented a “current” time framebut did not provide more specific information onthe details of this time frame were re-coded torepresent a past-year time frame. Finally, threestudies provided prevalence rates in a six-monthtime frame; these rates were re-coded as pastyearrates to allow their inclusion in the categoriesestablished in this study. As a result ofthese modifications, the prevalence rates reportedin this study may reflect conservativeestimates. 22ResultsStudy DemographicsThe search strategy described earlier identified152 primary studies that wereavailable to test against our inclusioncriteria. Of the 152 primary studies identified,32 failed to satisfy the inclusion criteriaspecified previously in the Methods section, andwere deemed ineligible for analysis. The remaining120 studies satisfied the inclusion criteriaand were accepted into this study. Fiftysevendifferent primary authors conducted these120 studies and provided prevalence estimatesof disordered gambling from 134 distinct studypopulation samples. Of these 134 samples,73.9% were derived from studies conducted in2421 Winsorization is a technique in which a specifiednumber of extreme values (in this study, two) are removedfrom each end of the distribution; the nextmost extreme value on either end is then replicated anumber of times equal to the number of extreme casesthat have been removed (Winer, 1971).22 Six-month rates are more conservative than pastyearrates, just as past-year rates are typically moreconservative than lifetime rates. There simply is lessstatistical opportunity for people to develop a disorderwithin a six-month time frame compared with apast-year time frame.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling25Total Number of Studies4030201001977 1979 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997Adult GP (N=50)Adolescent (N=22)College (N=16)Adult Tx & Prison (N=18)Adolescent Tx (N=3) Special Pop. (N=25)Figure 2: History of Prevalence Estimatesstudy was published (i.e., Lesieur, Blume, &Zoppa, 1986). During the 1990s, like gamblingopportunities, estimates of disordered gamblingbecame widely available. Indeed, half of alldisordered gambling prevalence studies conductedin the United States and Canada to datehave been released since 1992.The variety of prevalence studies available havebeen conducted by an equally diverse group ofprincipal investigators. Figure 3 illustrates theseinvestigators and their relative contributions tothe gambling prevalence literature. As this figureindicates, Volberg and Lesieur are the twolargest individual contributors to the disorderedgambling prevalence literature, with 17% and12% of the studies, respectively. Furthermore,the United States (N= 99) and 26.1% were derivedfrom studies conducted in Canada (N=35). A comparison of the prevalence estimatesavailable from the United States and Canadarevealed no significant differences betweenthese countries for any of the population segments.Consequently, the remainder of theseanalyses describe pooled data.The 134 study samples represent the followingeight major segments of the population: (1)adult general population (Adult, GP) (N = 50);(2) youth general population (Youth, GP) (i.e.,telephone surveys; N = 9); (3) in-school youth(N = 13); (4) college students (N = 16); (5) adolescentsin treatment programs (N = 3); (6)adults in treatment programs (N = 16); (7) incarceratedadults (N = 2); and (8) “special populations”(e.g., Native Americans on reservations,lottery players in Connecticut; N = 25). Figure1 illustrates the percentage of studies associatedwith each population type.Figure 2 provides an illustration of the history ofprevalence estimation studies. Beginning withKallick and her colleagues, who conducted thenational study in 1975 (and published the reportin 1979), attempts to estimate the prevalence ofdisordered gambling have rested primarily onstudies of adults from the general population.During 1986, the first non-general-populationAuthors withsingle study: 37%VolbergLesieurShafferWintersLadouceurCriterion ResearchBaselineW allischSteinberg AdolescentCulleton SpecialInsight Canada Research PrisonTreatmentPopulationsZitzowFerris2% Govoni et al.Adult Treatment Nechi Training 2%Single17%Study12% Emerson/Laundergan WynneLaventhol & HorwathCollege12%School (K-12)10%Adolescent, GP7%Lesieur: 12%Adult, GP38%Figure 1: Percentage of Study Population TypesVolberg has conducted 36% of the adult generalpopulation prevalence study samples and servedas consultant to an additional 4 adult generalpopulation study samples. Consequently, Volberghas influenced a total of 44% of the adultgeneral population prevalence study samples.Lesieur has conducted 28% of the treatment(psychiatric treatment or prison) study samplesand 38% of the college student population studyVolberg: 17%Figure 3: Authors of Gambling Prevalence StudiesHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling26Table 3: Meta-analysis Population Segment SamplesPopulation SegmentsNumber of StudySamplesTotal Number ofRespondentsProportion of TotalRespondentsAdult general population 50 79,037 65%Adolescents 22 27,741 23%College Students 16 8,918 7%Adults in treatment 18 6,590 5%Totals 106 122,286 100%samples. Five separate research teams (i.e.,Ladouceur, Shaffer, Volberg, Wallisch, andWinters) each contribute 9% of the adolescentprevalence study samples. In addition, there are17 other primary authors or research teams thathave contributed two or more prevalence studiesto the field. Authors who have conducted onlyone disordered gambling prevalence study accountfor 38% of the studies in the field. AsFigure 3 shows, many different scientists haveestimated the prevalence of disordered gambling.Preliminary examination and analysis of thepopulation groups allowed us to modify theeight groups discussed above for further analyses.For example, youth studies employed twodifferent primary data collection methods (i.e.,school-based versus home-based). These investigationsprovided the opportunity to compareprevalence rates obtained from these methodologies.Comparisons of prevalence estimatesderived from general population youth studieswith those derived from in-school youth studiesrevealed no significant differences betweenthese two groups (lifetime level 3: χ 2 = 1.422, df= 1, p > .20; lifetime level 2: χ 2 = .135, df = 1, p> .50; past-year level 3: χ 2 = .031, df = 1, p >.80; past-year level 2: χ 2 = 1.741, df = 1, p >.10); therefore, we combined these groups foranalysis into a general youth group. Similarly,adults in treatment programs and adults inprison were combined after analyses revealed nosignificant differences between these twogroups (lifetime level 3: χ 2 = .000, df = 1, p >.95; lifetime level 2: χ 2 = .046, df = 1, p > .80).Thus, for the remainder of the report, the“treatment” group will refer to adults in treatmentfor substance abuse or psychiatric disordersas well as adults in prison. For the purposesof this report, we conceptualize psychiatricand substance abuse treatment as well as incarcerationas social responses to deviant behavior.Using the “special populations” category as a“study type” introduced considerable and misleadingvariation into the data analysis. Thisgroup of studies had no unifying characteristicand was more heterogeneous than any of theother population groups studied; therefore, toavoid analytic bias and misleading data fromwhich to draw integrative conclusions, we didnot enter this class of studies into the analyses.Finally, the “youth in treatment” group was notused in the following analyses because we werenot confident that the small number of studies inthis category (N = 3) represented this populationadequately. Thus, the analyses that follow classifiedthe population groups as follows: (1)adult general population (N = 50); (2) adolescents(N = 22); (3) college students (N = 16);and (4) adults in treatment (N = 18). Thesefour categories include 94 studies that provideprevalence estimates among 106 distinct studysamples. These 106 study samples represent anaggregate of 122,286 respondents. The adultgeneral population studies represent a total of79,037 respondents, the youth studies represent27,741 respondents, the college studies repre-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling27sent 8,918 students, and the adult treatmentstudies represent 6,590 respondents. Table 3summarizes the study samples and respondentsincluded in this meta-analysis. The mean numberof respondents per study sample across thesefour population segments was 1,154 (standarddeviation = 1,248); the median number was 880;the smallest sample size was 85 respondents andthe largest was 7,214. 23The File Drawer Problem: PublicationStatus, Significant Findings, & PrevalenceRatesRosenthal and Rosnow (1991) remind usthat researchers have long suspected thatpublished studies represent a biasedsample of the research that investigatorsconduct. The “file drawer problem” occurswhen published studies represent type I errors(finding significant results when there are in factnone), “…while the file drawers back at the labare filled with the 95 percent of the studies thatshow non-significant (e.g., p > .05) results”(Rosenthal & Rosnow, 1991, p. 509). Whileprevalence studies do not engage in “significancetesting” in the primary analyses, it is importantto consider whether published studiesthat examine the prevalence of gambling disordersdiffer from the body of unpublished research.In the field of gambling studies, mostprevalence studies remain unpublished. Of the106 cases included in this meta-analysis, only40.1% were peer-reviewed. Over half of thestudies (51.4%) were reports disseminated bythe author or through the funding organization(e.g., Insight Canada Research, 1994; Volberg,1996a, 1996b; Wallisch, 1996); these reports23 These descriptive statistics reveal data that ishighly skewed. Readers will note that most of theanalyses to follow employ non-parametric statisticswhich do not assume that the data are distributednormally. In addition, non-parametric statistics areless sensitive than parametric instruments. Typically,decreased statistical sensitivity encourages investigatorsto be conservative in their inferences and conclusions.must be considered unpublished studies, sincethey are not widely available to the public (i.e.,through libraries and other public channels).These organization reports represent a significantportion of the disordered gambling prevalenceliterature. A t-test analysis revealed thatthe methodological quality scores of peerreviewedstudies were not significantly differentfrom those of non-peer-reviewed studies (t =.178, df = 104, p = .859). 24 In addition, theprevalence rates of peer-reviewed and non-peerreviewed studies were not significantly different(lifetime level 3: t = -.674, df = 80, p = .502;lifetime level 2: t = -.073, df = 63, p = .942;past-year level 3: t = .488, df = 41, p = .628;past-year level 2: t = .240, df = 37, p = .812). 25As a result of these tests, we combined theprevalence estimates provided by these twogroups of studies for all of the remaining analysesin this report.Methodological Quality Among PrevalenceStudiesThe set of disordered gambling prevalencestudies identified for inclusion in thisinvestigation provides a reflection of themethodological quality of research in thisfield. The descriptive data that follows illustratesthe prevalence of each of the nine methodologicalquality criteria that protect the internalvalidity of scientific research:♦ Random Selection: Site selection. Half ofthe studies included in this meta-analysis(50.7%) used a random selection process tosample data collection sites. The secondmost prevalent strategy was an opportunisticsample of sites (36.1%); one study (0.9%)selected all of the eligible sites; and 3 stud-24 For this analysis, the variable indicating whether astudy was peer reviewed or not was removed from thecalculation of the methodological quality score, andquality scores were standardized within study type.25 For this analysis, prevalence rates were standardizedwithin study type.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling28ies (2.8%) used another sampling strategy.Ten studies (9.4%) failed to report any samplingstrategy for their site selection process.Respondent selection. Slightly morethan half of the studies included in thismeta-analysis (54%) randomly selectedstudy respondents. Ten percent (10.4%) selectedall of the eligible respondents; 6.6%selected respondents opportunistically; 9studies (8%) used another respondent selectionstrategy, and 22 (21%) studies failed toreport any strategy to select respondents.♦ Response Rate: The studies included in thisanalysis reported response rates that rangedfrom 25% to 100% with a mean of 72.1%(s.d. 19.8%). Thirty six (34%) of the studiesfailed to report a response rate. Of thestudies that reported response rates, only44.6% reported properly calculated rates(i.e., rates calculated using the entire pool ofeligible respondents as the denominator);13.1% reported improperly calculated responserates (i.e., rates calculated using lessthan the entire pool of eligible respondentsas the denominator), and 42.1% did notspecify the method used to calculate the responserate.♦ Peer Review: Forty percent (40.1%) of thestudies included in this meta-analysis weresubjected to and satisfied a formal peer reviewprocess. That is, of the study samplesincluded in this meta-analysis, those thatwere peer-reviewed also were published. Inmeta-analyses of controlled trials or otherstudies that produce measures of effect,publication bias can result because studieswith statistically significant findings aremore likely to be published (i.e., peerreviewed)than studies that did not reportstatistically significant findings (Cooper &Hedges, 1994). In the field of gamblingprevalence research, factors other than significantfindings also are responsible for determiningwhether a study is peer-reviewed(i.e., published) or not. For example, stateand provincial governments that contractscientists to conduct prevalence studies expecta written report; these groups, however,do not expect a published article in a peerreviewedacademic journal. The settingwithin which a researcher works (e.g., academic,private) influences whether publicationsare encouraged, for example, becauseof promotion and advancement criteria.Therefore, the institutional setting of prevalenceworkers may influence the format andpublication status of completed prevalencestudies. For example, of the 20 studies includedin this meta-analysis that were conductedby Volberg, who works in a privatesetting, 75% were not peer-reviewed. Nevertheless,readers should recall from the discussionabove that peer reviewed studies didnot report prevalence rates of disorderedgambling that were meaningfully differentfrom studies that were not peer reviewed.♦ Data Reliability: Data collection. Twelvestudies (11.1%) reported having assessedthe reliability of their collected data. Dataentry. Seven studies (6.6%) reported havingassessed the reliability of their data entryprocedure (e.g., by randomly checkingdata-entry accuracy).♦ Sample size: The number of respondents inthe studies included in this meta-analysisranged from 85 to 7,214, with a mean of1,154 (s.d. 1,248).♦ Multi-instrument methodology: The majorityof the studies (82.8%) used a single instrumentto assess the prevalence of disorderedgambling. Thirteen studies (11.8%)used two instruments, and six studies (5.4%)used three instruments.♦ Intention: The majority of studies includedin this meta-analysis (80.2%) were conductedwith the primary intention ofestimating prevalence rates of disorderedgambling.♦ Anonymity: Among all studies included inthis meta-analysis with the exception ofthose that sampled treatment populations,41.5% reported having administered thesurveys to respondents with an assurance ofHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling29anonymity. Three percent (3.4%) did notconduct an anonymous survey, and 55.1%did not report whether the survey wasanonymous or not.♦ Varied times: Among the studies that useda telephone survey methodology, 17.9% variedthe time of day at which the respondentswere called, and 1.9% reported no variationof time. Eighty percent (80.2%) did not reportwhether the times were varied or not.Analyzing the Methodological QualityScoreA one-way ANOVA revealed that there weresignificant methodological quality score differencesamong the four study population types (F= 7.001, df = 3, 102, p


Prevalence of Disordered Gambling30nificant differences between these groups (lifetimelevel 3: t = -.117, df = 31, p = .908; lifetimelevel 2: t = .231, df = 20, p = .819; past-yearlevel 3: t = .188, df = 13, p = .854; past-yearlevel 2: t = .041, df = 9, p = .968).Finally, linear curve estimation regressionanalyses did not identify any trend in the qualityscores of prevalence studies during the pasttwenty years (i.e., from the first study in 1977 to1997). Figure 4 illustrates the relatively flatpattern of quality scores over time across allstudy types. Figure 5 illustrates this pattern bydepicting the trend of quality scores over timefor all studies and for adult general populationstudies.Organizing the Results by the CentralHypothesesThe six central hypotheses presented earlierprovide a ready organizational strategyfor presenting the major results focusingon prevalence estimates. Thesehypotheses suggested that:1. prevalence estimates of different populationsegments (e.g., youth, adult) will yieldmeaningfully different estimates of theprevalence of disordered gambling;2. increased access to gambling opportunitiesof the past 15 years and exposure to theshifting social setting that exposes morepeople to gambling opportunities will be reflectedin an increase in gambling problems;over the past 23 years, the rates of gambling-relatedproblems will have increasedat different rates depending upon the populationsfrom which the prevalence estimateswere derived;3. instruments used to generate prevalence estimateswill influence the rate of observedgambling problems;4. different geographical regions will have differentrates of disordered gambling807060504030201005. researchers will differ in the prevalence estimatesthey generate as a consequence oftheir characteristic research methods; and6. experience playing different types of gamblingactivities (e.g., sports betting or lotteryplaying) will influence prevalence rates differentially.In addition, these results will include informationabout the various strategies applied toweight the data, including the methodologicalquality scores. Before presenting the results thatbear on each of these central hypotheses, wewill discuss important additional features of theoverall data set.All Studies197719791985198619871988198919901991199219931994199519961997Adult General PopulationFigure 5: History of Methodological Quality ScoresInitial Observations on Collinearity ofPrimary VariablesThe estimates of the prevalence of disorderedgambling derived from the 106study samples reveal a significant andunexpected empirical attribute: the populationtype, sample size, and prevalence rates areintercorrelated, or collinear. Cohen & Cohen(1975) define collinearity (also called multicollinearity)as “the existence of substantial correlationamong a set of IVs [independent variables]”(p.115). For example, in this data set,studies of the general adult population have thelargest samples, with youth, college, and treatmentfollowing in descending order. As the fol-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling31Table 4: Bivariate Correlations Among Primary Collinear VariablesSample SizeLifetimeLevel 3LifetimeLevel 2Past-YearLevel 3Past-YearLevel 2Quality ScoreStudy Type r = -.394,p


Prevalence of Disordered Gambling32New England region (r = .433, p < .01) andadolescent studies (r = .193, p < .05).♦ the work of Volberg (e.g., Volberg, 1996b)was significantly associated with adult studies(r = .414, p < .01), the use of the SOGS(r = .278, p < .01), and the use of 3 to 4items on the SOGS as the definition of level2 gambling (r = .355, p < .01).♦ the work of Winters (e.g., Winters, Stinchfield,& Fulkerson, 1993a, 1993b) was associatedsignificantly with the North CentralU.S. region (r = .400, p < .01) and the use ofthe SOGS-RA (r = .374, p < .01).The collinear characteristics of these results interferewith attempts to precisely identify theunique effect of any particular author, region,instrument, or moment in time on prevalenceestimates of disordered gambling. To illustrate:a significant difference between the prevalencerates of two regions may not be a direct result ofgeographic location. Instead, these apparentdifferences may be confounded by other factors.For example, these distinctions may be attributableto differences between the instruments associatedwith studies in those regions, or to differencesamong the methods of particular authorsworking in those regions. With collinearvariables such as these, it is not possible to distinguishthe causal influence of any one variablefrom the set of multiply correlated variables.Thus, although we have attempted to identifydifferences among instruments, regions, andauthors, and to identify patterns over time, wemust interpret these findings with caution. Wemust view with caution and skepticism aggregatedcomparisons of instruments, regions, andauthors. Throughout the results and discussionthat follow, we will revisit the issue of collinearityand illustrate its influence on our understandingof disordered gambling prevalence rates.Testing the Central HypothesesAlthough a meta-analytic strategy assumesthat there is a modicum of“truth” associated with each of the studiesand methodologies included in the data set,there have been concerns about how investigatorsintegrate their collected studies (e.g.,Goodman, 1991; Olkin, 1995; Mosteller &Colditz, 1996; Shadish, 1996). 32 Collinearitycan compromise the apparent truths associatedwith a complex data set. Therefore, the distinctivecollinearity of the present data set acts as alens through which the results must be viewedand evaluated. The following results addresseach of this study’s central hypotheses. Thereader should bear in mind, however, that whilefactors such as instrument, researcher, and regionappear as separate sections, as we havedescribed earlier, there are important underlyingcollinear relationships that reside among them.32 One methodological guideline from the PotsdamConsultation on meta-analysis (Cook, Sackett &Spitzer, 1995) is that scientists specify whether therandom effects model or the fixed effects model isused to statistically combine the study results. Therandom effects model allows for heterogeneity amongstudy results that may remain after stratification,while a fixed effects model assumes homogeneity ofresults across studies. Both models have been widelyused in meta-analyses that combine risk ratios, oddsratios, or other measures of effect. In this report, weare conducting statistical analyses under a mixedmodel of random and fixed effects assumption. Thetheoretical and practical implications of this assumptionare very important. For example, we can generalizethe prevalence rates identified in this study, andthe associated confidence intervals, to the populationat large. However, the moderator variables describedin these results and the respective influence of thesevariables on the identified prevalence rates can begeneralized only to the sample of studies that wereincluded in this meta-analysis. Future research willdetermine which model, if either, is most appropriatefor use with meta-analyses of prevalence estimates.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling33Table 5: Mean Prevalence Rates (95% Confidence Intervals) for Four Study Populations*Adult Adolescent College TreatmentLevel 3 Lifetime 1.60 (1.35-1.85) 3.88 (2.33-5.43) 4.67 (3.44-5.90) 14.23 (10.70-17.75)Level 2 Lifetime 3.85 (2.94-4.76) 9.45 (7.62-11.27) 9.28 (4.43-14.12) 15.01 (8.94-21.07)Level 1 Lifetime 94.67 (93.71-95.62) 89.56 (85.88-93.25) 86.66 (80.90-92.42) 71.54 (62.90-80.18)Level 3 Past year 1.14 (.90-1.38) 5.77 (3.17-8.37) -- --Level 2 Past Year 2.80 (1.95-3.65) 14.82 (8.99-20.66) -- --Level 1 Past Year 96.04 (95.04-97.04) 82.31 (75.59-89.03) -- --*Estimates are rounded to two decimal places.Hypothesis 1: Rates of DisorderedGambling Prevalence Vary by PopulationSegmentThere are a variety of methods for integratingprevalence estimates for any population. As wedescribed previously, methodologists suggestusing multiple methods to assure stability ofestimates (e.g., Campbell & Fiske, 1959; Olkin,1995). Therefore, we compared the prevalenceestimates reported in each study by a variety ofstatistical algorithms as described previously inthe Methods section. The results of these differenttechniques for each of thepopulation segments aresummarized in Appendix 1.The final estimate for eachpopulation segment in Appendix1 is the arithmetic4mean of these 16 alternativeestimates. For illustrativepurposes, Figures 6 and 7 revealthat there is little differenceamong these estimationtechniques for adult generalpopulation lifetime level 2and 3 data, respectively, suggestingthat the variety ofmethods for identifying centraltendency reflect considerablestability. This pattern ofstability is consistent across3210the various population segments. In addition,the range of adult lifetime level 3 prevalenceestimates across all of the weighting strategies is1.5% to 1.6%; similarly, the range of adult lifetimelevel 2 prevalence estimates is 3.0% to3.9%. These narrow ranges encourage confidencein prevalence estimates represented byunweighted data. Since unweighted data representsevidence that is independent from statisticalmodifications and assumptions that can influencethese methods, we will continue to usethe unweighted data represented in Table 5 fordiscussion. Table 5 summarizes these methodsby providing the unweighted mean prevalenceLevel 2M ean Median Huber TukeyHampell Andrews 5% Trimmed Winsorized MeanWinsorized Median Q Mean Q Median Q HuberQ Tukey Q Hampell Q Andrews Q 5% Trim m edFigure 6: Lifetime Prevalence M-Estimates for Adult GP Level 2GamblingHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling341.61.551.51.45Level 3Mean Median HuberTukey Hampell Andrews5% Trim m ed Winsorized M ean Winsorized M edianQ Mean Q Median Q HuberQ Tukey Q Hampell Q AndrewsQ 5% TrimmedFigure 7: Lifetime Prevalence M-Estimates for Adult GP Level 3 Gamblingrates and the confidence intervals associatedwith each of these means for the four populationsegments discussed above. Box and whiskerplots in Figures 8 and 9 illustrate the relativelystable distinctions of disordered gambling intensityacross levels 2 and 3. That is, level 3 estimatesof gambling disorders that satisfy diagnosticcriteria are almost uniformly lower thanlevel 2 estimates of sub-clinical gambling problems.These unweighted study estimates representthe general adultpopulation segment forlifetime and past-yeargambling disorders.An interesting findingemerges from this datafor adolescents: pastyearprevalence ratesexceed lifetime rates.This outcome would beimpossible within a singlesample; however,readers should note thatthe group of studies thatprovided past-year estimatesand the group ofstudies that providedlifetime estimates aremutually exclusive. Wewill examine this issue in more detail later in thesection on comparing lifetime and past-yearprevalence rates.Figure 9: Past-Year Prevalence of Level 2 & 3Among the Adult General Population2220Figure 8: Lifetime Prevalence of Level 2 & 3Among the Adult General PopulationPrevalence22201816141210864upper bound C.I.lower bound C.I.Lifetime Level 3Prevalence181614121086420-219921990199319931993Year study releasedupper bound C.I.lower bound C.I.Past Year Level 3upper bound C.I.lower bound C.I.Past Year Level 20-2199619961996199519951994199419931992199119901989198619851977Year study releasedlower bound C.I.Lifetime Level 2Comparing The Rates Of DisorderedGambling Among Different PopulationSegments2upper bound C.I.19961996199619961996199519951995199419941994The four primary population segments (i.e.,study types) described previously were com-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling35Table 6: Significant Lifetime Prevalence Differences Among Population Groups aGeneral PopulationAdultsAdolescentsCollege StudentsGeneral Population AdultsLevel 3 *AdolescentsLevel 2 *Level 3 *Level 3 nsCollege StudentsLevel 2 nsLevel 2 nsLevel 3 *Level 3 *Level 3 *Adults in Treatment or PrisonLevel 2 *Level 2 ns Level 2 nsa Where “*” indicates a significant statistical difference at p < .05 and “ns” indicates no significant differencepared to reveal differences among the groups’prevalence rates. For general population adults,adolescents, college students, and adults intreatment, Kruskal-Wallis tests (Siegel, 1956)revealed that there were significant group differencesfor lifetime level 3 and level 2 prevalencerates among these four groups (χ 2 =58.413, df = 3, p < .001 and χ 2 = 31.430, df = 3,p < .001, respectively). Figure 10 illustrates therelative levels of lifetime prevalence rates forthe four groups.Dunnett C tests, assuming unequal variance forpost-hoc analyses, revealed the following differencesbetween the groups: for lifetime level 3estimates, the rate of disorder among the adultgeneral population (M = 1.60) was significantlylower (p < .05) than adolescents (M = 3.88),college students (M = 4.67), and adults in treatmentor prison (M = 14.23). The adolescentgroup’s rate of level 3 lifetime gambling wassignificantly lower (p < .05) than adults intreatment or prison. College students also evidenceda meaningfully lower (p. < .05) level 3lifetime gambling rate than adults in treatmentor prison. For level 2 lifetime gambling rates,adults (M = 3.85) evidenced significantly lower(p < .05) prevalence than adolescents (M = 9.45)and adults in treatment or prison (M = 15.01).These findings are summarized in Table 6.For past-year rates, there was insufficient datafor the college and treatment groups to makecomparisons among the four groups. Thereforewe compared past-year prevalence rates amongthe remaining two groups, adults and adolescents,using Kruskal-Wallis tests. 33 For past-100%80%60%40%20%0%AdultsAdolescentsCollege StudentsPatients & PrisonersLevel 3Level 2Level 1Figure 10: Lifetime Prevalence Rates AmongPopulation Segments33 While using SPSS 7.5, we identified a problemwith the non-parametric algorithm. After consultationwith SPSS technicians, this difficulty was deemed tobe a software “bug.” This problem made analyses ofweighted data unreliable unless the weights were integers.To correct for this difficulty, we employed theKruskal-Wallis statistic instead of the more typicalMann-Whitney test because the former provides achi-square statistic and the latter does not. After multiplyingthe weight by a factor designed to produceinteger weighting, we divided the chi-square statisticby the same factor, therefore bypassing the “bug” andyielding a reliable statistic for interpretation. Wenotified SPSS and researchers should be assured thatthe corporation will repair the algorithm.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling36Table 7: Relative Risk of Disordered Gambling by Population Types(General Population Adults as Reference Group)Adolescents College TreatmentLifetime Level 3 2.43 2.92 8.89Lifetime Level 2 2.45 2.41 3.90Past-Year Level 3 5.06Past-Year Level 2 5.29year level 3 rates, adult rates were significantlylower than adolescent prevalence rates (χ 2 (1) =16.703, p < .001). Similarly, for past-year level2 rates, adult estimates were significantly lowerthan adolescent estimates (χ 2 (1) = 18.344, p


Prevalence of Disordered Gambling37Table 8: Comparing Prevalence Estimates & (95% Confidence Intervals) Associated With SOGS 34Studies 35Adult Adolescent College TreatmentLevel 3 LifetimeLevel 2 LifetimeLevel 3 Past YearLevel 2 Past Year1.71 (1.46 - 1.96)(n = 30)3.41 (2.81 - 4.0)(n = 27)1.12 (.945 - 1.30)(n = 26)2.16 (1.81 - 2.50)(n = 25)4.25 (1.91 - 6.59)(n = 6)8.58 (5.69 - 11.47)(n = 5)5.05 (3.55 - 6.56)(n = 14)7.0 (4.49 - 9.50)(n = 9)14.55 (10.60 - 18.50)(n = 16)8.83 (3.34 - 14.31)(n = 6)Kruskal-Wallis tests reveal similar patterns tothose presented earlier: for SOGS lifetime ratesof level 3 gambling, there were significant differencesamong the four study types (χ 2 =48.929, df = 3, p < .001). However, Dunnett’sC post-hoc tests revealed a somewhat differentpattern of results from those reported above: forSOGS lifetime rates of level 3 gambling, theprevalence among adult studies (M = 1.71) wassignificantly lower (p < .05) than the prevalenceamong college studies (M = 5.05) and treatmentstudies (M = 14.55) but was not significantlydifferent from adolescent studies (M = 4.25).However, we must exercise caution in interpretingthe finding that adult prevalence rates werenot significantly lower than adolescent prevalencerates in this analysis. Power analyses revealedthat the power to detect this differencewas 10%. This low level of power, causedmainly by the small number of studies availablefor analysis, indicates that any existing differencebetween these two groups most likelywould not be revealed by this analysis.Dunnett’s C tests also revealed the followingdifferences in lifetime level 3 rates: the adolescentrate was significantly lower (p < .05) thanthe treatment rate and the college rate was significantlylower (p < .05) than the treatmentrate. For SOGS lifetime rates of level 2 gambling,a Kruskal-Wallis test revealed significantdifferences among the four groups (χ 2 = 23.118,df = 3, p < .001). Dunnett’s C post-hoc analysesrevealed the following differences between thegroups: adult general population rates (M =3.41) were significantly lower (p < .05) thanadolescent rates (M = 8.58) and college rates (M= 7.00). There was not sufficient data to makecomparisons of past-year rates among thegroups. These comparisons are described inTable 9.34 This table represents studies that used the originalSOGS, the SOGS modified to reflect a past-year timeframe, and the SOGS modified minimally for usewith adolescent populations. This table does not includestudies that used the SOGS-RA, Multi-factorMethod, or other more substantial modifications ofthe SOGS.35 The level 2 prevalence rates in this table representstudies that defined level 2 gambling as a SOGS scoreof 3 or 4. These rates are conservative compared toGender-specific Prevalence Estimatesthose derived from the use of scores from 1 to 4 onthe SOGS as the definition of level 2 gambling. Weused the more conservative definition for this analysisbecause the use of this definition provided more datafor analysis than the use of the more liberal definitionwould have.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling38Table 9: Significant SOGS Lifetime Prevalence Differences Between Population Groups aGeneral PopulationAdultsAdolescentsCollege StudentsGeneral Population AdultsAdolescentsLevel 3 nsLevel 2 *College StudentsLevel 3 *Level 2 *Level 3 nsLevel 2 nsAdults in Treatment or PrisonLevel 3 *Level 2 nsLevel 3 *Level 2 nsLevel 3 *Level 2 nsa Where “*” indicates a significant statistical difference at p < .05 and “ns” indicates no significant differenceThis study confirms the empirical evidence oftenreported in the disordered gambling researchliterature that males exhibit higher rates of disorderedgambling than females. This does notpreclude the possibility that individual studiesoccasionally will find higher rates of disorderedgambling among females (e.g., New MexicoDepartment of Health, 1996). The languageused in the scientific literature to discuss genderdifferences in rates of disordered gambling isremarkably varied. Some reports refer to genderas a risk factor or a correlate of disordered gambling,while other studies illustrate gender differencesby presenting a table of the percentageof disordered gamblers (or combined level 2 andlevel 3 gamblers) who are male (e.g., Volberg,1992a, 1996a, 1996b; Ladouceur, 1991; Sommers,1988). Since prevalence studies essentiallyare epidemiological explorations of thepatterns and characteristics of a disorder or illnessof interest, presenting rates of genderspecificdisordered gambling prevalence areuseful for two primary reasons. First, stratifyingdata by gender allows a more precise examinationof the nature of the phenomenon of interest,unclouded by differences between males andfemales. For example, in studies reportinghigher rates of disordered gambling among menthan women, an unstratified rate underestimatesthe prevalence for men and overestimates theprevalence for women. Secondly, genderstratifiedrates are useful to calculate relativerisk ratios. A relative risk ratio compares themagnitude of difference in the risk for disorderedgambling that exists between two groups(e.g., male and female). That is, risk ratios allowthe observation that male adolescents are,for example, four times more likely to be pathologicalgamblers than female adolescents, insteadof simply stating that “males are morelikely than females to be pathological gamblers.”Risk ratios quantify the difference that gender—or any other factor on which stratified data isavailable—can make on the phenomenon of interest.For example, Bland et al. (1993) expressthe results of their epidemiological study ofEdmonton adults by observing a 3-to-1 ratio ofmale to female disordered gambling (more specifically,a relative risk of 3.1).Of the 134 estimates identified in this metaanalysis,56% (n = 75) provided gender-specificestimates of disordered gambling. Though thissample of 75 estimates is not precisely representativeof the larger sample of 134, it is representativeof all the prevalence estimates providedbetween 1975 and 1997 that include genderspecificdata.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling39The 75 estimates of gender-specific disorderedgambling use two different timeframes. Fifty-five (73.3%) represent lifetimeestimates and 26 (34.7%) are pastyearestimates. Some studies providedboth lifetime and past-year estimates,which is why these percentages add tomore than 100%. Table 10 displays thegender-specific and population-specificestimates of disordered gambling acrosstime frames and levels of gambling.Attributable ProportionTable 11: Gender-specific Relative Risks of DisorderedGamblingPopulation SegmentRelative Risk(females as referencegroup)General Population Adult♦ Lifetime Level 3 2.15♦ Lifetime Level 2 2.10♦ Past-year Level 3 1.78♦ Past-year Level 2 1.80Adolescents♦ Lifetime Level 3 3.03♦ Lifetime Level 2 2.75♦ Past-year Level 3 3.39♦ Past-year Level 2 1.67College Students♦ Lifetime Level 3 3.84♦ Lifetime Level 2 1.96Adults In-treatment or Prison♦ Lifetime Level 3 1.85♦ Lifetime Level 2 2.07Table 11 presents relative risks for therisk factor of gender across the fourpopulation segments (adult general population,adolescents, college students, andadults in treatment or prison). Whilethese relative risks reflect the comparativelikelihood of having level 3 or level2 gambling problems, a different measurereveals the level of influence that theidentified risk has on the population.This measure is an attributable proportion,which also is called attributable risk. Anattributable risk provides a measure of the publichealth impact of an exposure, assuming thatthe association is one of cause and effect (Hennekens& Buring, 1987). For example, in a studyon the association between smoking and cancerof the mouth and pharynx, the attributable proportionwas 72%. That is, 72% of the cases ofmouth or pharynx cancer in this study are attrib-Table 10: Estimates of Disordered Gambling by Population Group Stratified by Gender*Group Level 3Level 2Level 3Level 2LifetimeLifetimePast YearPast Yearf m f m f m f mGeneral Population Adult 1.24n = 11Adolescents 2.0n = 6College students 2.24n = 12Treatment 7.15n = 52.67n = 116.05n = 68.62n = 1213.22n = 53.34n = 115.13n = 510.86n = 114.48n = 47.03n = 1114.13n = 521.22n = 119.25n = 4.96n= 92.50n = 51.71n = 98.4n = 53.12n = 813.66n = 55.6n = 822.83n = 5*where f = female and m = maleHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling40utable to smoking as the causal factor(Ahlbom & Norell, 1990). Thenumerical value of the attributablerisk can be used to calculate theproportion of cases among the exposedsegment of the population thatcan be attributed to the exposureitself. The data available in thismeta-analysis provides few variablesor risk factors that we can consider“exposures” for level 2 or level 3gambling. However, gender is oneof these variables, or “exposures”which is associated with disorderedgambling. Given that there is a paucityof “exposure” data available inmost prevalence studies of disorderedgambling, and there is considerablegender data, gender is the“exposure” we will use to demonstratethe application of attributablerisk.Table 12 organizes the various attributableproportions of gender bypopulation segment, time frame, andlevel of disordered gambling. Forexample, the proportion of male lifetimelevel 3 gamblers for whom lifetimelevel 3 gambling is attributableto being male is 53%. In public health, attributableproportions usually convey a sense of theextent to which a particular illness or disordercan be prevented by blocking the effect of aspecific exposure (or eliminating it altogether)(Rothman, 1986). In the current example, sinceit is not possible to block the effect of beingmale, we are interested in understanding whatabout the state of “maleness” or “being male” iscontributing to or “causing” disordered gambling.That is, since we do not have empiricalevidence that the physiologic or genetic constitutionof males compared with females is anetiologic cause of disordered gambling, beingmale stands as a proxy for the causes of disorderedgambling that are yet to be identified. Forexample, males likely differ from females inhow they experience and relate to money, feelingstates (e.g., impulses), and regulatorymechanisms (e.g., coping with impulses).Table 12: Attributable Risks for GenderPopulation Segment Attributable Risk *(among males as“exposed” group”)Population AttributableRisk **(attributable to“maleness”)General Adult Population♦ Lifetime Level 3 .53 .358♦ Lifetime Level 2 .50 .350♦ Past-year Level 3 .38 .046♦ Past-year Level 2 .42 .269Adolescents♦ Lifetime Level 3 .67 .504♦ Lifetime Level 2 .64 .467♦ Past-year Level 3 .70 .530♦ Past-year Level 2 .40 .240College Students♦ Lifetime Level 3 .74 .559♦ Lifetime Level 2 .49 .312Treatment♦ Lifetime Level 3 .46 .355♦ Lifetime Level 2 .52 .420*The attributable risk percent is computed as: AR e= RR-1/RR**The population attributable risk (or proportion) is estimated by thefollowing formula: AR = p e(RR-I)/p e(RR-I)+1, where p e = % of exposedpeople in the populationIn addition to calculating the attributable riskamong an exposed (i.e., male) group, we alsocan calculate the attributable risk among theentire population. This measure has many differentnames in the epidemiological literature(e.g., total attributable risk, etiologic fraction).For the purposes of clarity, we will refer to themeasure of attributable risk among the entirepopulation as the population attributable risk.The population attributable risk is the excessrate of disease in the total study population thatis attributable to the exposure (Henneken &Buring, 1987). To determine the impact of anyone risk factor on public health, it is necessaryto know both the relative risk and the percentageof the population who have that risk factor(Kahn & Sempos, 1989). In the subset of studysamples that reported stratified data, the proportionof males ranged from 44.5% (adult generalpopulation level 3 lifetime) to 67.9% (adultHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling41treatment level 2 lifetime). Using the appropriateproportion for each population segment bytime frame and gambling level, we calculatedthe proportion attributable to the proxy of “beingmale.” For example, 50.4% of the rate oflevel 3 lifetime gambling among adolescents inthe general population is associated with beingmale and the risk factors that being male entails.Knowing the attributable risk due to gender doesnot mean that other factors also cannot be strongcauses of disordered gambling. Not only canother factors exist, but the attributable risks ofeach factor can add to more than 100%. Forexample, x% of cancer is caused by smoking,y% by diet, z% by alcohol, and zz% by otherfactors. Added together, these percentages canlegitimately exceed 100% because of the interactivenature of causality (Rothman, 1986). Thepopulation attributable risk caused by one factormay be shared by other influences. Therefore,since “maleness” represents other influences,these other factors also can “cause” a meaningfulproportion of disordered gambling. As theprevious example illustrates, these additionalfactors can cause the same or even higher proportionsof the disorder than “maleness.”Other Stratified DataStratified data was presented in 62.3% of theeligible prevalence studies. In addition to gender,these studies provided stratified data on anumber of other factors. A proportion of studiesprovided stratified data on age (36.3%), race orethnicity (25.7%), socio-economic status (SES)(25.2%), education level (30.6%), primary language(2.2%), family member(s) with gamblingproblems (18%), level of involvement in specificgambling activities (13.4%), and maritalstatus (29.1%). In addition, 34.9% of the studiesreported data stratified by a factor other thanthe ten reported here. In this results section, wewill report only the gender-stratified data andwill not investigate the other stratified factors.Some of these other factors are not conducive toa synthesis because studies classified data insuch a way that integration might yield misleadingconclusions. For example, socio-economicstatus can be approached in many differentways. Some investigators used annual householdincome as the marker for SES, but evenamong these investigators the sub-categoriesdiffered to such an extent that data pooling wasimpossible. Other factors did not yield reliabledata. For example, the existence of a familymember with gambling problems usually wasreported by someone other than that familymember. We believe this secondary data mustbe confirmed; yet, there were no studies thatverified these collateral reports by screening theappropriate family members directly. Therefore,we chose not to examine this factor basedon the subjective and potentially unreliable natureof the data. A future report will examinethis sub-set of the data in more thorough detail.Hypothesis 2: The Prevalence of DisorderedGambling is Shifting OverTimeWe examined the prevalence rates for all fourstudy types together in one grouping to identifyany significant changes in these rates over time.For this analysis, prevalence rates were standardizedwithin each study type, to control forthe fact that studies of populations with higherprevalence rates (e.g., adolescents, treatmentgroups) were more likely to have been conductedin recent years. This analysis revealed asignificant positive correlation between the yeara study was conducted and the rate of past-yearlevel 3 gambling (r = .448, p < .01). This relationshipalso existed between the year a studywas released and the rate of past-year level 3gambling (r = .382, p < .05).Although these correlations are significant,readers should note that the r 2 terms 36 are relativelysmall, indicating that the year a study wasconducted accounts for only a small percentageof the total variance associated with shiftingprevalence rates over time (i.e., 20%). Similarly,the year a study was released accounts for36 An r 2 term (in this case, the square of the correlation)represents the percentage of the variation in thedependent variable “explained” by the independentvariable(s).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling42Mean Prevalence7.576.565.554.544.264.5115% of the variance. Small r 2 values suggestthat other unidentified but meaningful factorsalso influence the increasing rates of disorderedgambling prevalence.Next, we examined prevalence rates within eachstudy type using the method described above toidentify any significant patterns over time.Analyses of adolescent, college, and treatmentstudies revealed no significant patterns overtime. Among adult studies, the past-year level 3rate showed a statistically significant pattern:this analysis revealed a significant positive correlationbetween the year a study was conductedand respondents’ past-year level 3 gamblingprevalence rate (r = .558, p < .01). This relationshipalso was revealed between the year astudy was released and past-year level 3 gambling(r = .415, p < .05). In addition, this analysisrevealed a significant positive correlationbetween the year a study was conducted andrespondents’ combined level 2 and level 3 gamblingprevalence rate (r = .338, p < .05). Similarly,this relationship was revealed between theyear a study was released and combined level 2and level 3 gambling (r = .353, p < .05).We validated this finding by using a second analyticstrategy. Here we compared the prevalencerates from studies released before the medianyear (i.e., 1993.5) for all adult studies with theprevalence rates from studies released after themedian year. Table 13 presents these meanprevalence rates. Kruskal-Wallis tests revealed6.421st 2nd 3rd 4th1977-1990 1991-1993 1994-1995 1996-1997n = 9n = 9n = 8n = 9GroupsFigure 11: Adult Lifetime Level 2 & 3 CombinedPrevalence7that, for studies among the adult general population,recent (i.e., post-median) studies had significantlyhigher prevalence rates than earlier(i.e., pre-median;) studies for lifetime level 2 (χ 2(1) = 5.792, p < .05), lifetime level 2 and level 3combined (χ 2 (1) = 7.5235, p < .01), and pastyearlevel 3 (χ 2 (1) = 4.033, p < .05).Table 13: Mean Adult Prevalence Rates for Pre-Median-Year and Post-Median-Year GroupsEarly Studies(1977-1993)Recent Studies(1994-1997)Lifetime Level 2 2.93 4.88*Lifetime Combined 4.38 6.72*Past-Year Level 3 .84 1.29**rates significantly higher than early studies’ rates, p < .05Next, we created approximately equally-sizedgroups (e.g., quartiles) of the range of adultstudies according to the year the studies werereleased. A Kruskal-Wallis test did reveal significantdifferences among the four quartiles forcombined lifetime level 3 and level 2 prevalencerates (χ 2 (3) = 8.102, p < .05). Figure 11 illustratesthe mean prevalence rates for these fourgroups. Kruskal-Wallis tests failed to identifysignificant differences among these four groupsfor lifetime level 3, lifetime level 2, past-yearlevel 3, and past-year level 2.Finally, we identified significant patterns inprevalence rates over time among adult studiesby conducting curve estimation regressionanalyses (i.e., trend analyses). These analysesrevealed significant linear patterns over time forlifetime combined level 3 and level 2 (r 2 =.22508, F (1,17) = 4.88948, p < .05) and pastyearlevel 3 (r 2 = .28231, F (1,17) = 6.62141, p< .05). In addition, for level 3 lifetime, the trendanalysis revealed a significant linear patternover time (r 2 = .36407, F (1,17) = 9.63720, p


Prevalence of Disordered Gambling43.01). Figure 12 illustrates one of the linear relationshipsand is representative of the two othersignificant findings. Similar to the results reportedbefore, each of these curve estimationregressions accounts for a relatively small percentageof the total variation associated withthese prevalence rates. Readers will recall thatsmall r 2 values indicate that other meaningfulfactors must be identified to account for the unexplainedvariance. The identification of theseinfluences will provide additional insight intothe nature of increasing rates of disorderedgambling prevalence. We will return to this issuein more detail later in the section that discussesregression analyses.Hypothesis 3: Differences AmongPrevalence Rates Derived from DifferentScreening InstrumentsHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling44Given the collinearity of the data set describedabove, the most sound method of identifyingdifferences among instruments is to compareprevalence rates derived from instruments thathave been used among the same study sample.For example, some authors have used two ormore instruments within a single study. The useof two or more instruments with the same studysample provides quasi-experimental conditionsin which all factors except the instrument (e.g.,the respondents, the method of survey administration,the survey administrator) remain constant.This strategy permits the difference inrates from different instruments to be examinedwith some confidence. This meta-analysis identifiedthirteen studies of adults, adolescents, orcollege students in which two or more instrumentswere used. Table 14 below shows thecomparisons of instruments derived from thesestudies. In this table, each horizontal row representsa study’s comparison of two instruments;each row indicates the author and year of thestudy, the study type, the time frame, the twoinstruments used among the study’s sample, andthe two level 3 rates derived from these instruments.The final cell in each row indicates theratio of the first instrument’s rate to the secondinstrument’s rate. For example, if instrument Aprovided a rate of 3.0% and instrument B provideda rate of 1.5% among the same sample,the ratio would read “instrument A = 2.0 instrumentB,” indicating that instrument A provideda rate that was two times the rate providedby instrument B.Table 14 illustrates four major comparisonsamong screening instruments (i.e., comparisonsthat have been replicated in three or more studies):(1) SOGS with versions of the DSM criteria;(2) SOGS with the Multifactor Method; (3)SOGS-RA narrow criteria with SOGS-RA broadcriteria 37 ; and (4) MAGS with DSM-IV.3020100Observed-10Linear1970198019902000Year study releasedFigure 12: Combined Lifetime Level 3 & 2 Trends37 As readers may recall, the SOGS-RA is a modifiedversion of the original SOGS instrument designed foruse with adolescents; it uses 12 scored items and apast-year time frame (Winters, Stinchfield, & Fulkerson,1993a). The SOGS-RA can be scored in twodifferent ways: the “narrow criteria” and the “broadcriteria” (Winters, Stinchfield, & Kim, 1995). Thesetwo scoring methods are scored as follows: The NarrowCriteria: “no problem” = 0-1; “at risk” = 2-3;“problem” = 4+ ; The Broad Criteria: “no problem”= no history of gambling, or gambling less than dailyand score of 0; “at risk” = weekly gambling andscore of 1, or less than weekly gambling and score of2+; “problem” = weekly gambling and score of 2+,or daily gambling and any score.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling45Table 14: Comparing Screening Instruments Across Studies With Multiple Measures (Level 3 Rates)Study Sample Time frame Instr. 1 Instr. 1 Instr. 2 Instr. 2 RatioRateRateOster & Knapp(1994)College lifetime SOGS 11.2 DSM-III-R 5.1 SOGS = 2.20DSM-III-ROster & Knapp(1994)College lifetime SOGS 8.0 DSM-III-R 5.7 SOGS = 1.40DSM-III-RFerris & Stirpe(1995)Adults lifetime SOGS .971 DSM-IV .485 SOGS = 2.00DSM-IVOster & Knapp(1994)College lifetime SOGS 11.2 DSM-IV 4.2 SOGS = 2.67DSM-IVVolberg (1996b) Adults past-year SOGS 1.4 DSM-IV .875 SOGS = 1.60DSM-IVVolberg (1996a) Adolescents lifetime SOGS 3.4 MultifactorMethod2.8 SOGS = 1.21M.M.Volberg (1993b) Adolescents lifetime SOGS 1.5 MultifactorMethod.9 SOGS = 1.67M.M.Wallisch (1993a) Adolescents lifetime SOGS 3.7 MultifactorMethod5.0 SOGS = .74M.M.Govoni et al. (1996) Adolescents past-year SOGS-RA“narrow”10.3 SOGS-RA“broad” criteria21.1 broad = 2.05narrowWinters et al. (1995) Adolescents past-year SOGS-RA“narrow”criteria2.9 SOGS-RA“broad” criteria8.2 broad = 2.83narrowWinters et al. (1995) Adolescents past-year SOGS-RA“narrow”criteria3.5 SOGS-RA“broad” criteria9.5 broad = 2.71narrowShaffer et al. (1994) Adolescents past-year 38 MAGS 8.5 DSM-IV 6.4 MAGS = 1.33DSM-IVVagge (1996) Adolescents past-year MAGS 4.3 DSM-IV 4.2 MAGS = 1.02DSM-IVShaffer & Hall(1994)Adolescents past-year MAGS 7.0 DSM-IV 8.0 MAGS = .875DSM-IVLadouceur & Adolescents lifetime PGSI 39 3.6 DSM-III 1.7 PGSI = 2.118Mireault (1988)DSM-IIIGovoni et al. (1996) Adolescents past-year SOGS 8.1 SOGS-RA “narrow”criteria10.3 narrow = 1.27SOGSGovoni et al. (1996) Adolescents past-year SOGS 8.1 SOGS-RA“broad” criteria21.1 broad = 2.60SOGSSteinberg (1997) Adolescents past-year MAGS 3.2 SOGS-RA“broad” criteria8.7 broad = 2.72MAGSThe four major comparisons among screeninginstruments can be summarized as follows:♦ Five studies provided the opportunity tocompare the SOGS with some version of theDSM criteria; in these studies, the SOGSprovided higher rates than the DSM criteriaby factors ranging from 1.4 to 2.67, with amean factor of approximately 2. A Wil-38 The rates from Shaffer et al. (1994) are calculatedamong respondents who have gambled in their lifetime.39 Pathological Gambling Signs Index.coxon signed ranks test revealed that theSOGS produced significantly higher estimatesthan DSM-based instruments (Z = -2.023, p < .05) in the five comparisons eligiblefor analysis. This set of five comparisonsincluded both college student and adultpopulation samples as well as lifetime andpast-year time frame prevalence estimates.♦ Three studies provided the opportunity tocompare the SOGS with the MultifactorMethod; in two of these studies the SOGSwas higher, and in one of these studies theMultifactor Method was higher. The meanratio of SOGS to Multifactor Method wasapproximately 1.2.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling46♦ The SOGS-RA broad criteria has been comparedwith the SOGS-RA narrow criteria inthree studies; in these studies, the broad criteriaprovided higher rates than the narrowcriteria by a mean factor of approximately2.5.♦ The MAGS has been compared with theDSM-IV criteria in three studies; in two ofthese studies the MAGS rate exceeded theDSM-IV rate, and in one of these studies theDSM-IV rate exceeded the MAGS rate.The mean ratio of MAGS to DSM-IV wasapproximately 1.In addition, there are four other comparisonsthat have been made; however, these comparisonsreside within a single study: (1) the ratio ofthe Pathological Gambling Signs Index (a precursorto the SOGS) to the DSM-III was approximately2; (2) the ratio of the SOGS-RAnarrow criteria to the SOGS was approximately1.3; (3) the ratio of the SOGS-RA broad criteriato the SOGS was approximately 2.6; (4) the ratioof the SOGS-RA broad criteria to the MAGSwas approximately 2.7.We strongly encourage readers to view theseratios as approximate, preliminary findings:these ratios should be considered as “ballpark”comparisons rather than precise estimates. Todate, there is insufficient data to draw confidentconclusions from these comparisons. Furthermore,the mean ratios provided above representthe aggregation of three or more studies; insome cases, these mean ratios represent the aggregationof different study types and/or timeframes. This methodology, although not ideal,is the only current procedure for aggregatingthis data. In the future, the addition of morestudies that provide the opportunity to comparemultiple screening instruments will allow metaanalyststo address each study type and timeframe individually. At this time, however, theseratios should be viewed as an early approximationof the potential rate differences betweeninstruments.Hypothesis 4: Differences AmongPrevalence Rates Derived from DifferentRegionsTo test the hypothesis that different regions ofthe United States and Canada would have meaningfullydifferent prevalence rates, we standardizedprevalence rates within study type by convertingthese rates to z-scores. Transformingrates to z-scores has the effect of removing theinfluence of study type on prevalence rates.This procedure permitted us to manage the collinearaspects of prevalence rates and study typeso that we could enter prevalence rates from allstudy types into a Kruskal-Wallis analysis. Thisanalysis examined whether there were any regionaldifferences in lifetime level 3, lifetimelevel 2, past-year level 3, or past-year level 2.This procedure revealed no significant prevalencerate differences among the regions of theUnited States and Canada on these four variables.Hypothesis 5: Differences AmongPrevalence Rates Derived from DifferentResearchersTo test the hypothesis that different primary researcherswould yield meaningfully differentprevalence rates, we followed the proceduredescribed above for Hypothesis 4 to develop z-scores. As we found with regional comparisons,Kruskal-Wallis analyses revealed no significantdifferences among individual researchers onstandardized rates of lifetime level 3, lifetimelevel 2, past-year level 3, or past-year level 2.Hypothesis 6: Experience with DifferentTypes of Gambling Activities YieldDifferent Rates of Gambling DisorderMany of the studies included in this metaanalysisreported the prevalence rates of participationin various gambling activities. Table 15illustrates the mean prevalence of participationin a variety of different gambling activities foreach of the three population segments within alifetime and past-year time frame. The treatmentpopulation segment was represented byHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling47Table 15: Prevalence of Gambling Activity by Population SegmentAdults (%) Adolescents (%) College (%)Lifetime Prevalence of Gambling 81.19 77.55 85.04Casino Games - Lifetime 32.32 7.74 40.59Casino Games - Past Year 14.95 12.56 60.83Lottery - Lifetime 61.25 34.89 50.29Lottery - Past Year 49.05 30.16 60.18Sports Gambling - Lifetime 26.83 38.17 28.45Sports Gambling - Past Year 14.76 30.69 30.5Pari-mutuel - Lifetime 25.11 10.88 27.17Pari-mutuel - Past Year 7.13 11.24 8.9Financial Markets - Lifetime 13.11 -- 16.65Financial Markets - Past Year 5.81 -- 4.2Non-Casino Card Games - Lifetime 28.16 53.46 47.37Non-Casino Card Games - Past Year 15.89 39.61 36.1Games of Skill - Lifetime 18.57 40.43 39.93Games of Skill - Past Year 10.25 31.61 23.93only three studies; therefore, it is not included inthe following table.As was the case with the aggregated prevalencerates presented earlier, readers should note thatin some cases past-year rates are higher thanlifetime rates (e.g., casino games for adolescentsand for college students). This phenomenonoccurs when the group of studies that providepast-year rates and the group of studies that providelifetime rates represent orthogonal datasets. That is, among this group of studies, thereare no investigations that provide both lifetimeand past-year rates. When there is a relativelysmall number of studies in each group, otherfactors (e.g., regional variations) are likely tobias the results. This phenomenon is illustratedin the college study estimates, in which regionalvariation may be a significant factor (e.g., of thefour study samples providing past-year rates,three are from Minnesota; the other is from AtlanticCity). In addition, there is a disproportionatenumber of studies conducted among studentsin colleges proximal to casinos; thus, thesecollege rates may be misleadingly high. Thisfinding reflects the robust collinear characteristicsof this data set and indicates that these ratesshould be interpreted with caution.The Relationship between SpecificGambling Activities and PrevalenceRatesCorrelations between study samples’ rates ofparticipation in seven different gambling activitiesand the rates of disordered gambling amongthese study samples were examined to identifyany significant relationships among these variables.We conducted these analyses separatelywithin each study type. Among adult studies,the analyses revealed the following relationships:there was a significant negative correlationbetween the lifetime rate of sports bettingand the lifetime rate of level 3 gambling (r = -.482, p < .05); lifetime rates of level 2 gamblinghad significant positive correlations with bothlifetime and past-year rates of gambling in financialmarkets (r = .737, p < .01 and r = .693, pHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling48< .05, respectively); there was a significantnegative correlation between rates of past-yearlevel 3 gambling and lifetime rates of parimutuelgambling (r = -.542, p < .05); finally,there was a significant positive correlation betweenrates of past-year level 2 gambling andlifetime rates of participation in the lottery (r =.476, p < .05).Although there were insufficient data points formost gambling activities to conduct these analysesamong the set of adolescent studies, a significantpositive relationship was revealed betweenrates of lifetime level 3 gambling andpast-year rates of gambling on games of skill (r= .920, p < .05). Similarly, insufficient dataprecluded most of these analyses for the set ofcollege studies. However, the following relationshipswere observed: rates of lifetime level 2gambling had significant positive correlationswith both lifetime rates of casino gambling (r =.755. p < .05) and lifetime rates of gambling onnon-casino-based card games (r = .804, p < .05).Insufficient data entirely precluded these analysesamong the set of in-treatment populationstudies.To evaluate whether the prevalence of participationin each of seven common gambling activities(casino games, lottery, sports betting, parimutuel,financial markets, cards, and games ofskill) was significantly different among particularsegments of the population, we again performedKruskal-Wallis tests. Of the fourteentests (seven activities within both lifetime andpast-year time frames), ten (all but lifetimesports betting and past-year pari-mutuel wagering)revealed significant differences among thefour population types (all tests significant at p


Prevalence of Disordered Gambling49lifetime level 2, past-year level 3, and past-yearlevel 2 (F = 91.324, df = 3, 78, p < .001; F =13.850, df = 3, 62, p < .001; F = 33.313, df = 3,39, p < .001; F = 49.145, df = 3, 36, p < .001,respectively). These regression analyses indicatedthat for lifetime level 3, lifetime level 2,past-year level 3, and past-year level 2, populationtype accounted for 77.7%, 40.1%, 71.9%,and 73.2% of the variance, respectively. Sincethe current data set does not permit us to distinguishthe unique variance accounted for bypopulation type, the amounts of variance explainedby population type given above areshared with other causal influences. Figure 13illustrates the extent to which population typeinfluences the variance associated with prevalenceestimates.In the area of prevalence estimates of disorderedgambling, there is no “gold standard” againstwhich to determine a measure of precision. Ifwe knew in advance what the prevalence of disorderedgambling was among a specific populationsegment, then we could determine how disparatevarious estimates are from that knownstandard. In the absence of such astandard, we must determine ameasure by which to value our setof prevalence estimates. Therefore,we elected to conduct ourweighted least squares regressionto explain prevalence estimatesusing a methodological qualityPast Year Level 2Past Year Level 3Lifetime Level 2Lifetime Level 3● Random selection ofrespondents (1 or 2 stage)● Response & CompletionRate (% & calculation)● Anonymity ofrespondents; can 3rd partycompromise datacollection● Published via peer reviewscore as the weighting variable.This strategy weights estimates ofprevalence as a function of themethodological characteristics ofthe study from which these estimatesare derived. Readers willrecall from the earlier discussionon the calculation of quality scoresthat we developed the methodologicalindex for this study by integratingdata about nine methodologicalfactors: (1) sample selectionprocess (i.e., randomly selected),(2) response rate (includingthe appropriateness of the method used tocalculate the response rate), (3) survey anonymity,(4) whether the study underwent a peer reviewprocess (e.g., for publication in a refereedjournal), (5) whether the authors assessed thereliability of their data collection and entry procedures,(6) whether the authors varied the timeof day survey data was collected, (7) the numberof respondents in the study sample, (8) whetherthe authors took a multidimensional approach tomeasuring disordered gambling (e.g., multipledependent measures), and (9) whether the studywas intended primarily as a prevalence study.0 20 40 60 80 100ExplainedUnexplained● Data integrity check● Varied time of calls(telephone survey only)● Study sample size● Multi-methodmeasurement strategy● Intended as prevalencestudyFigure 14: Elements of the MethodologicalQuality IndexFigure 13: Explained Unique & Shared Variance Among PrevalenceRates Due to Population TypeFigure 14 summarizes these elements of methodologicalquality.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling50These weighted least squares regression analyseswithin study type also were complicated bycollinearity; as we discussed previously, region,author, and instrument were highly collinear.As a result of this phenomenon, regressionanalyses with all of these variables entered simultaneouslydid not provide meaningful resultsfor interpretation. Thus, the nature of this dataset precluded the identification of the uniquevariance accounted for by an independent variable.Instead, we entered each variable individuallyinto a regression analysis for each dependentvariable and identified the relativeThe regression analyses yield the following relativeresults: Subject or population attributes(e.g., age, gender, psychiatric status) have themost powerful influence on prevalence estimates.Also important, but in descending orderof influence are measurement instruments, location,principal investigator of the study, and thehistorical moment of the research.Estimating the Number of DisorderedGamblers in the United States and Can-Table 16: Estimated Number of Past-Year Disordered Gamblers in the United States & Canada(in millions)United StatesCanadaAdolescents Adults Both Adolescents Adult BothLevel 3 Range: 1.2 - 3.2central estimate:2.2Range: 1.7 - 2.6central estimate:2.2Range: 2.9 - 5.8central estimate:4.4Range: 0.1 - 0.3central estimate:0.2Range: 0.2 -0.3central estimate:0.3Range: 0.3 - 0.6central estimate:0.5Level 2 Range: 3.4 - 7.9central estimate:5.7Range: 3.7 - 7.0central estimate:5.3Range: 7.1 - 14.9central estimate:11.0Range: 0.4 - 0.8central estimate:0.6Range: 0.4 -0.8central estimate:0.6Range: 0.8 - 1.6central estimate:1.2Combined Range: 4.6 - 11.1central estimate:7.9Range: 5.4 - 9.6central estimate:7.5Range: 10.1 - 20.7central estimate:15.4Range: 0.5 -1.2central estimate:0.8Range: 0.6 -1.1central estimate:0.9Range: 1.1 - 2.3central estimate:1.7Total: range of estimates = 11.2 - 23.0; central estimate = 17.1amount of variance explained; these figures representthe unique variance explained by an independentvariable plus the explained variance itshares with other variables. The relative degreeto which region, author, instrument, and yearinfluence prevalence estimates is depicted inFigure 15 for adults and Figure 16 for youth. Asthese figures illustrate, the year of the researchaccounts for a relatively small percentage of thevariance in prevalence rates compared to theother factors with the exception of past-yearlevel 3 among adults.Past Year Level 2Past Year Level 3Lifetime Level 2Lifetime Level 30% 20% 40% 60% 80% 100%Region Author Instrument Type YearFigure 15: Sources of unique and shared varianceamong the explained sources of adultprevalence estimatesHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling51Past Year Level 2Past Year Level 3Lifetime Level 2Lifetime Level 30% 20% 40% 60% 80% 100%Region Author Instrument Type YearFigure 16: Sources of unique and shared varianceamong the explained sources of youth prevalenceestimatesadaTo help people better understand theprevalence rates identified by this study,investigators often translate prevalencerates into estimates of the number of individualswho struggling with the illness or disorderin question. Policy makers and publichealth treatment planners also use these numbersto make decisions about, for example, bedsneeded in a treatment unit, or staff necessary totend to a compulsive gambling hotline. Table16 summarizes estimates of the number of pastyeardisordered gamblers based on the U.S.1997 census and the Canadian 1996 census. 40The range estimates in Table 16 reflect 95%confidence intervals; the central estimate representsan estimate of the middle of the confidenceinterval range. In this table, the adolescentgroups represent those people from thegeneral population 10-19 year-olds, and adultsrepresent people who are 20 years of age andolder. There was not sufficient data to estimatethe number of past-year disordered gamblersamong the treatment/prison or college popula-40 These statistics were retrieved from World WideWeb sites as follows:http://www.census.gov/population/estimates/nation/intfile2-1.txt (for U.S. figures);http://www.statcan.ca/english/Pgdb/People/Population/demo10a.htm (for Canadian figures).tions. Since we have not included these “treatment”and college populations in the Table 16estimates, readers should consider the estimatesto represent conservative approximations of thenumber of people with gambling associated disorders.Discussion“Any solution to a problem changes the problem.”- R. W. Johnson (1979) 41The results of this meta-analysis revealthat a number of different factors canexert influence on prevalence estimatesof disordered gambling. The results alsoreveal that study quality had little impact on observedprevalence rates. Taken together, theseobservations lead to the conclusion that disorderedgambling is a relatively robust phenomenonthat can withstand influence from a varietyof sources. In spite of these conclusions, thecollinearity of the present data set requires us toconsider all of the findings quite carefully andconservatively. These multiple correlationsamong a variety of study attributes do relatequite systematically to prevalence estimates.As a result of these observations, we have organizedthe following discussion into 6 majorsections that will (1) consider the primary factorsthat influence prevalence estimates, (2) examinethe original hypotheses that guided thisstudy, (3) consider methodological issues associatedwith prevalence estimation, (4) examinehow the interaction of social setting and personalitycan help to explain some of the currentfindings, (5) explore the implications of thesefindings for future research, and (6) recommendsome standards for conducting future prevalenceestimation research. Before we begin to discussthe findings directly, a brief digression is in order.Given the array of different analytical toolsemployed in this study, it is helpful to revisit a41 Washingtonian (1979, November).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling52basic issue associated with the conduct of metaanalyticresearch. Is it appropriate to integratedifferent estimates of disordered gamblingprevalence, gathered from different populations,using various methods?Can Different Prevalence Estimates BeIntegrated?Before examining the results presentedearlier, it is important to revisit a fundamentalissue underlying meta-analyticresearch in general and this synthesis inparticular: can different prevalence estimates,derived from different studies, using differentmethods, be integrated in a meaningful way?(Goodman, 1991; Hasselblad et al., 1995). Notin spite of, but precisely because of the diverseresearch strategies and methods used to generateestimates of disordered gambling, prevalenceestimates can, and should, be integrated. This isnot a conclusion reached after casual consideration.For example, we understand well that“...methodological quality can be related tostudy outcomes and thus confound the interpretationsone can draw from a body of research.However, the relation between quality and outcomeis not consistent among different researchdomains, and accounting for variation in studyquality is a thorny problem” (Bangert-Drowns etal., 1997, p. 424).The integration of prevalence estimates rests onthe very assumptions of meta-analytic researchdescribed by Smith and Glass (1977) in theirseminal work on psychotherapy outcomes.Smith and Glass encouraged meta-analysts tointegrate research findings specifically becausethe integration of evidence respected the varietyof assumptions associated with many differentinvestigators. Smith and Glass recognized thatthere was no single “true” psychotherapy outcomeevidenced by a single research undertaking.There are many different ways to examineimportant health-related questions that rangefrom treatment outcome to prevalence estimation.Since prevalence estimates are a directreflection of the variety of the research methodsand strategies that scientists develop and implementto measure this phenomenon, debateand controversy are common bedfellows ofprevalence estimation projects (e.g., Nadler,1985). These methodological disputes are particularlycommon among young scientific fields(Cohen, 1985).The primary task of a meta-analytic study ofprevalence estimates is to synthesize existingfindings across a range of research assumptions,methodologies, instruments, and results into amore precise and reliable estimate of the phenomenonunder investigation. This synthesis isjustifiable only if the various estimates aremeasuring essentially the same phenomenon.For example, we can measure body temperatureusing a variety of methods and instruments.Different instruments (e.g., thermometer or heatsensitivetape) and different routes of administration(e.g., oral, rectal, underarm, aural) donot deter observers from recognizing that resultsobtained by these various methods reflect anindex of body temperature. Although these differentmethods vary in precision, we accept eachmethod as an indication of a relatively stableunderlying construct (i.e., body temperature).Similarly, we can be confident that the variousinstruments used in the disordered gamblingfield measure essentially the same underlyingconstruct. Furthermore, since there is no “goldstandard” for the identification of disorderedgambling, we cannot determine the absolute accuracywith which any of these instrumentsidentifies the underlying construct of pathologicalgambling. As a result, aggregation of ratesderived from different instruments provides uswith a more stable estimate than would be availablefrom any single measurement instrument.We will discuss issues related to gold standardsand validity in more detail later.Factors Associated with DisorderedGambling Estimates: Sources of InfluenceWe began this report by suggesting thata scientific manufacturing process isresponsible for the production ofprevalence estimates. This relativistHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling53view (Casti, 1989) encourages us to consider thepossible factors that can influence the process ofprevalence estimation. The results reveal thatattributes of the individual, best interpreted asrisk factors (e.g., age, gender, psychiatricstatus), are the most potent influences on prevalenceestimates. Individual characteristics, representedby the four major population studytypes (i.e., adult, adolescent, college, and treatment),account for more variance associatedwith prevalence estimates than any other singlefactor.There are other important moderator variablesinfluencing prevalence estimates. In addition toindividual respondent “traits,” characteristics ofthe research process also influence estimates ofdisordered gambling prevalence. These influencesare more highly variable and account forless variance than do population characteristics.Research process influences include the followingfactors: who conducts the investigation, howthe investigator chooses to measure disorderedgambling (i.e., instrument), where the investigationis conducted (i.e., region) and when in historythe data is collected. Taken together, thesefactors combine to affect the rates of gamblingdisorder. Figure 17 provides an illustration ofthe primary factors that influence the estimationof disordered gambling prevalence rates.●Subject or Population Attributes– Age, Gender, Psychiatric Status●Measurement Instruments●Geography or Location●Principal Investigators●Historical Moment of Study (e.g., History ofAccess to Gambling)– Only among adults (level 2 or 3)Figure 17: Factors That Influence PrevalenceRatesSurprisingly, the quality of research methodshas exerted little influence on prevalence estimatesof gambling disorder. In this study, byusing a reliable coding procedure for identifyingthe presence or absence of study features associatedwith matters of internal and external validity(e.g., Bangert-Drowns et al., 1997), wehave determined that estimates of gambling disorderwere neither higher nor lower as a functionof research quality. Just as surprising as theabsence of study quality influence on estimates,but perhaps more ominous, is the finding thatresearch quality has not improved during thepast two decades of prevalence research. Observersof the history of science might have assumedthat scientists would have improved theirmethods of estimating disordered gamblingprevalence as they moved along a learningcurve. However, Bakan’s (1967) critical observationsabout the development of science mayapply here: scientific research rarely makes newdiscoveries, it usually confirms what we alreadyknow. In the field of gambling studies, in spiteof a growing body of empirical data, there isonly a modicum of theory. Theory is where importantscientific advances occur. The availabledata unfortunately is often disconnected fromorganized theory. Although it is possible tospeculate that during the past two decades thestudies of prevalence were of sufficiently highquality that there was little room for improvement,this does not appear to be the case. As wewill describe at the end of this discussion, weare of the opinion that it is time for investigatorsto adopt some basic standards for the conductand reporting of disordered gambling prevalenceestimates. Methodological advances from thefield of psychiatric epidemiology (e.g., Tsuang,Tohen, & Zahner, 1995) must be integrated intothe emerging research on disordered gambling.Unfortunately, until then, many conclusionsabout the extent and nature of disordered gamblingprevalence must be held as tentative.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling54The Central Hypotheses: Drawing ConclusionsHypothesis 1: Rates of DisorderedGambling Prevalence Vary by PopulationSegmentProducing a set of disordered gamblingprevalence estimates is at the heart of thismeta-analytic study. This project representsthe first attempt to integrate quantitativelythe array of prevalence estimates reportedin the published and unpublished gamblingliterature to date. The lifetime and pastyearestimates for each of the four populationgroups examined in this study demonstrate thatan individual’s risk of disordered gambling isprimarily dependent upon their age, clinicalsituation, and gender. Moreover, the results revealthat rates of disordered gambling do varyby population segment. The results of thismeta-analysis show that adolescents consistentlyhave higher rates of level 3 and level 2 timeframes gambling for both lifetime and past yearthan their adult general population counterparts,with only one exception. When examining onlythe subset of studies that used the SOGS instrumentsto measure disordered gambling, youngpeople in the general population did not evidencesignificantly higher level 3 lifetime prevalencerates compared with adults. However, thisfinding should be interpreted with caution. Thiscomparison had little statistical power to identifya statistically significant difference; a largernumber of studies using the SOGS among adolescentsmay reveal and confirm the more stableobservation that adolescents have higher rates oflevel 3 gambling than their adult counterparts.Notwithstanding this exception, youthful ageappears to be an important risk factor for developinggambling-related problems. One explanationfor this finding is that, compared to adults,youth have had more exposure to gambling duringan age when vulnerability is high and risktakingbehavior is a norm; consequently, theseyoung people have higher rates of disorderedgambling than their more mature and less vulnerablecounterparts. In addition to pre-collegeadolescents, there are other population segmentsthat have higher risk of experiencing gamblingdisorders. College students and adults in treatmentor prison consistently had significantlyhigher rates of lifetime level 3 gambling thanadults surveyed from the general population.The treatment/prison population evidenced thehighest rates of disordered gambling among allthe population groups studied. Using all thesurvey instruments, the results revealed that thein-treatment population had meaningfully higherrates of lifetime level 2 gambling than the adultgeneral population. 42Membership in youthful, college, treatment, orprison population segments must be consideredan important risk factor for the development ofgambling-related disorders. As we will discusslater in the section on the interaction of socialsetting and personality, we can understand therisks associated with each of these populationsegments as a function of the interaction betweenpersonality and social context. Thosemost at risk for disordered gambling are (1) indifferentor insensitive to the social pressures orsanctions against immoderate behavior (e.g., thesocial separation often experienced by peoplewith major mental illness, or the new independencecommonly experienced by college students),(2) extremely sensitive to the perceivedsocial pressures to participate in activities thatthey consider to be normative (e.g., as a result ofthe peer pressure often experienced by adolescents),or (3) in physical or emotional discomfortthat is ameliorated by the gambling experience(e.g., people who are depressed and findthat gambling relieves their discomfort). Disconnectionfrom the pressure of a social settingto behave in a particular way can derive from animmature stage of psychosocial development,psychiatric illness, personality disorder, or acombination of these circumstances.42 If the results were limited to only SOGS studies,then college students and not the treatment group hadsignificantly higher rates of lifetime level 2 gamblingthan the adult general population. This observationrelates to the limited sample of studies available andthe associated decrease in statistical power.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling55Lilly Tomlin humorously illustrated this relationshipbetween personality and social settingduring her one-woman play, The Search ForSigns of Intelligent Life in the Universe:“...reality is the leading cause of stress amongstthose in touch with it. I can take it in smalldoses, but as a lifestyle I found it too confining.It was just too needful; it expected me to bethere for it all of the time, and with all I have todo—I had to let something go. Now, since I putreality on a back burner, my days are jampackedand fun-filled” (Wagner, 1986, p. 18).Males often evidence higher rates of behavioraldisorders than females (e.g., fighting, crime,drunkenness). Across population segments,males similarly have higher rates of level 3 andlevel 2 lifetime and past-year gambling than females.These rates generally become less discrepantwith advancing age. Males tend to bemore insensitive to the pro-social forces exertedby the reality of their social setting than are theirfemale counterparts; when not insensitive, mencustomarily withdraw from social settings thatthey experience as behaviorally constraining.Often the result of this pattern of social separationis disproportionate behavioral excess. 43 Toillustrate, male college students are nearly fourtimes as likely to be lifetime level 3 gamblers asfemales, and male adolescents are three times aslikely to be lifetime level 3 gamblers as theirfemale counterparts. Male adults are twice aslikely as adult females to meet lifetime criteriafor pathological gambling; those in the generalpopulation are just over twice as likely, whilemale adults in treatment settings or prison arealmost twice as likely to be lifetime level 3gamblers as their female counterparts.This relationship also holds with lifetime andpast-year measures of sub-clinical levels ofgambling disorders. Being young, male, in college,having psychiatric co-morbidity, or a historyof antisocial behavior are factors that repre-43 For an interesting example of this behavior patternwith alcohol, interested readers should see Zinberg &Fraser (1979).sent meaningful risks for developing gamblingrelatedproblems. Maturing seems to provide aprotective device against behavioral excesses,unless dysthymia, diminishing health, or somepre-existing problem facilitates the developmentof an addictive behavior pattern. Winick (1962)speculated that people who experience narcoticdependence can “mature out” of their addiction.Additional research will clarify whetherWinick’s notion applies to disordered gamblingas well as narcotics abuse, or whether “maturingout” is better understood as an element of naturalrecovery (e.g., Shaffer & Jones, 1989). Testingthese unexplored hypotheses is the domainof new studies which we will discuss later inmore detail.Identifying Risk Factors for DisorderedGambling“Someone once asked me why women don’tgamble as much as men do, and I gave thecommon-sensical reply that we don’t have asmuch money. That was a true but incompleteanswer. In fact, women’s total instinct for gamblingis satisfied by marriage.” - Gloria Steinem(1983)Individual, social, and cultural factors (e.g.,gender, emotional temperament, access, availability,folkways and mores of a society) determinethe risk of becoming a disordered gambler.Gender represents a considerable risk factor fordisordered gambling. The results of this studyreveal that men are much more likely thanwomen to become disordered gamblers forevery population segment. This gender differenceis most apparent among college students,where men are nearly four times more likelythan women to become gamblers with level 3disorders during their lifetime. Youth alsorepresents a considerable risk factor for disorderedgambling. Within the general population,young people are almost 3 times more likelythan their adult counterparts to evidence a level3 gambling disorder during their lifetime and4.47 times more likely during the past year toexperience a level 3 disorder. Similarly, adultsin treatment are almost 9 times more likely thanadults in the general population to experience aHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling56level 3 gambling disorder during their lifetime.Taken together, these findings suggest thatyouth and psychiatric status represent importantrisk factors for gambling disorders. As we describedpreviously, youth with psychiatric comorbidityhave compound risk factors that interactto significantly elevate the rate of disorderedgambling prevalence over their non-impairedcohorts.Stratified data provides important informationthat can improve the utility, generalizability andmeaningfulness of prevalence estimates. Forexample, in addition to an aggregate or singleestimate (which provides an overestimate forfemales and an underestimate for males), separateestimates for males and females and a relativerisk ratio provide information that more accuratelydepicts the phenomenon for each gender.We recommend that, in the future, investigatorspresent stratified data for factors that mayserve as potential correlates of disordered gambling.That is, we recommend that genderspecific,age-specific or other risk-factorspecificrates be reported. In addition, the formatfor presenting these stratified rates is animportant consideration. Rather than reporting,for example, that males make up 80% of thegroup of pathological gamblers, we recommendthat investigators report gender-specific prevalencerates. For example, “4.3% of the males inthe sample were pathological gamblers comparedto 1.2% of the females in the sample.”The former approach is an indication of the proportionof the entire sample that is male, andconfounds an understanding of the relationshipbetween gender and disordered gambling; thelatter approach is a better index of a specific riskfactor (e.g., gender) for disordered gambling.Data reported in this manner will stimulate animproved understanding of the factors that contributemeaningfully to the phenomenon of disorderedgambling.Hypothesis 2: The Prevalence of DisorderedGambling is Shifting overTimeTemporal effects of rates of psychiatric disorders,including pathological gambling, canmanifest in several variations (Horwath &Weissman, 1995). Age effects refer to agespecificphases in life when individuals are athigher risk of developing pathological gambling.Period effects are specific eras in time associatedwith a higher prevalence of pathologicalgambling (e.g., a period effect might be seen ina country or a state that has just legalized gamblingfor the first time). Finally, a cohort effectindicates a specific group of individuals bornwithin the same time period (e.g., in the sameyear or decade) who have different rates ofpathological gambling as a cohort. Of thesethree variations of temporal effects, the metaanalysisdata only allowed for an exploration ofthe age effect and the period effect. The lack ofstudies that examine incidence data as well asthe lack of prospective research designs limit acritical review of cohort effects. These are importantareas of investigation, and new incidenceresearch initiatives are necessary to providebetter insight into the nature of cohort effectsand disordered gambling.We already have described the age effect evidentin rates of pathological gambling duringour exploration of Hypothesis 1: the data revealedthat adolescents have meaningfullyhigher rates of pathological gambling thanadults. As for the period effect, there is correlationalevidence supporting the notion that ratesof pathological gambling have increased duringthe two decades between 1977 and 1997. Thisevidence reveals that past-year level 3 gambling,lifetime level 2 gambling, and combined lifetimelevel 2+3 gambling is increasing over time.When we control for the influence of differingstudy types, the evidence becomes stronger thatthe rates of past-year pathological gamblingamong adults in the general population are increasingover time. In addition, rates amongadults in the general population are increasingfor combined lifetime level 2+3 gambling. Thedata also reveal a potential period effect amongHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling57adult lifetime level 3 gambling rates and lifetimelevel 2 gambling rates. As we suggested previously,this pattern of increasingly higher rates ofgambling disorder among the general adultpopulation seems to be the result of the interactionbetween personality and social setting.Adults in the general population are much moresensitive to the social sanctions against illicitbehaviors than are their adolescent, psychiatric,or criminal counterparts. As gambling has becomemore socially accepted and accessible duringthe past two decades, this population segmenthas started to gamble in increasing numbers.Adolescents, college students, psychiatricpatients, and criminals historically have notavoided gambling just because it was illicit;these groups have been relatively insensitive tosocial sanctions. Newly exposed to the gamblingexperience, adults in the general populationare having difficulty adjusting and, unlikethe other population segments who already evidencegambling problems, are beginning to reportincreasingly higher rates of gambling disorder.We can anticipate that, like the very slow adjustmentpeople have made to the informationabout tobacco-related dangers, or the repeal ofprohibition of beverage alcohol, the informaland formal social controls necessary to provideprotection against gambling problems willemerge slowly, perhaps only after decades andgenerations of social learning. Formal socialcontrols include law and other regulatorymechanisms; informal social controls rest moreon the folkways and mores of a given social setting.We will explore the idea of a shifting socialsetting and its impact on gambling in moredetail later in the section of the Discussion focusingon the interaction of personality and socialsetting.Hypothesis 3: Differences amongPrevalence Rates Derived from DifferentScreening InstrumentsThis meta-analysis included studies that used atotal of 25 different instruments to assess ratesof disordered gambling. It benefits researchersto know how these instruments compare to eachother. To compare rates derived from differentinstruments, we investigated studies that usedtwo or more different instruments among thesame sample. The ratio of the two estimateswithin a single study ranged from 1.02 (quitesimilar) to 2.83 (quite different). Though weview this observation as tentative, these analysesrevealed that the SOGS produces significantlyhigher estimates of pathological gambling thanversions of the DSM criteria. Studies that usedboth the SOGS-RA broad and the SOGS-RAnarrow criteria demonstrate that the broad criteriaproduce consistently higher rates than thenarrow criteria. As a result of the small numberof studies available for analysis, data comparingother screening instruments remains inconclusive.Hypothesis 4: Differences amongPrevalence Rates Derived from DifferentRegionsCharacteristics of the data set restricted theidentification of regional differences. First, althoughwe were able to standardize scores toremove the effect of study type, collinearityamong region, author, and instrument remainedand may have influenced this analysis. In otherwords, an existing regional difference may havebeen concealed by differences among authors orinstruments. In addition, statistical power forthis analysis was limited. For example, this designhad only 25% power to detect a moderateeffect size for lifetime level 3 rates with an averageof six estimates per region at a significancelevel of .05. To achieve adequate power(i.e., 80%) for this comparison, this analysiswould require 26 estimates per region to detect amoderate effect size at a significance level of.05. Given this limitation and protecting againstthe contaminating influence of study type byusing standardized scores, various analyses revealedno meaningful differences among regionsin the United States and Canada. Since therewas insufficient statistical power to reveal moderateregional differences, this finding does notpreclude the possibility that with additionaldata, regional differences will emerge. For now,Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling58however, we could not provide support for thishypothesis.Hypothesis 5: Differences amongPrevalence Rates Derived from DifferentResearchersAs with the tests of Hypothesis 4, there was limitedstatistical power for comparisons amongdifferent investigators. Given this limitationand by using standardized scores to protectagainst any contaminating influences from differentstudy types (as we did with regional differences),our analyses again revealed no meaningfuldifferences among different researcher’sestimates of disordered gambling prevalence.As with regional variation, there was insufficientstatistical power to reveal small differencesamong researchers. Therefore, as was thecase with the regional analysis, this finding doesnot preclude the possibility that with additionaldata, meaningful differences among investigatorswill become apparent. Although the casualobserver might detect higher rates produced byindividual researchers, the collinear nature ofthe current evidence demands caution when attemptingto interpret these observations. Thatis, researchers tend to conduct studies, be basedgeographically in certain regions, and use thesame screening instrument across their studies.These inter-relationships cannot be teased apartwithin the current collection of prevalence studies.Therefore, we conclude that the data in thismeta-analysis did not reveal any meaningfuldifferences among various researchers’ estimatesof disordered gambling.Hypothesis 6: Experience with DifferentTypes of Gambling Activities YieldDifferent Rates of Gambling DisorderAlthough we do not believe that any specificobject of addiction (e.g., heroin, cocaine, keno,lottery, or shopping) represents the necessaryand sufficient causes to produce addictive behavior,44 there is reason to examine the epidemiologicalrelationship between gambling disordersand the specific games people play. Therefore,this study examined the extent of participationin seven different common gambling activitiesamong the adult general population, adolescents,adults in treatment and prison populations,and college students. We found that, asexpected, adolescents participate significantlymore than adults in gambling activities that aremost socially accessible and do not require authorization.That is, adolescents are gamblingmore than adults on games of skill, non-casinocard games, and sports betting. Adolescents canparticipate in these three activities within agroup of school friends, with their families, orwith their friends’ families. Similarly, collegestudents are betting more than adults in the generalpopulation on non-casino card games andgames of skill; these represent activities whichare popular within a college setting. Unsurprisingly,adults in the general population are gamblingmore than adolescents on casino games,the lottery, and pari-mutuel wagering. Thoughthere are exceptions, the vendor of these adultactivities generally requires authorization from alicensing bureau or certification board. Althoughthere is evidence that adolescents areengaging in these three activities despite theirillegal status, the vast majority of individualswho participate in these “legal” forms of gamblingare adults.We also examined in this report the pattern ofrelationships among specific gambling activitiesand disordered gambling. Among adults, higherlevels of gambling in financial markets and onthe lottery were associated with level 2 disorders.On the other hand, lower levels of bettingon sports and pari-mutuel gambling was associatedwith more serious level 3 gambling. Foradolescents, higher levels of betting on games ofskill were associated with higher rates of level 3gambling. Finally, among college students,44 We encourage interested readers to review Shaffer(1996, 1997b) for a more complete discussion of thismatter.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling59higher levels of gambling at casinos and on noncasino-basedcard games were associated withhigher rates of level 2 gambling. These preliminaryrelationships must be evaluated with greatcaution, particularly the finding that lower levelsof betting on sports and pari-mutuels are associatedwith higher rates of diagnosable gamblingdisorders. This finding does not implythat participating in these two activities providesa “protective” factor; it may mean instead thatindividuals who choose not to gamble on sportsand pari-mutuels are gambling with greater intensityon other activities.Deciphering relationships among specific gamingactivities and disordered gambling requiressophisticated research that focuses on the natureof the relationships that exist between an individualand the object of their addiction; that is,their gambling activity of choice. The field ofgambling research would do well to emulatelines of inquiry within the substance abuse researchfield, which has discovered many importantand illuminating differences between varioussubstances and their substance-specificphysiological, psychological, and socioeconomicinfluences on their users. For example,Khantzian (1975, 1985, 1989, 1997) has foundthat alcohol has special “releasing” propertiesthat tend to dis-inhibit users. Cocaine has antidepressantstimulating properties. Khantzianhas suggested that certain personality types aremore attracted to each of these drug classes toproduce a self-medicating effect. Similarly, Jacobs(1989) suggests that certain gambling activitiescan produce dissociative effects that maybe differentially attractive to individuals withcertain personality attributes. Much remains tobe learned about the relationship between peopleand the games they choose to play.Table 17: Comparing Lifetime and Past-yearPrevalence Rates of Adult Psychiatric Disordersin the United States: Where Does DisorderedGambling Fit?L.T. P.Y.Gambling Disorder Level 3 1.6 1.1Anti-Social Personality 2.6 1.2Obsessive Compulsive 2.6 1.7Drug Abuse/Dependence 6.2 2.5Major Depressive Episode 6.4 3.7Generalized Anxiety 8.5 3.8Alcohol Abuse/Dependence 13.8 6.3How Does the Prevalence of DisorderedGambling Compare With Other Problems?:Making Sense of the NumbersWe can compare the prevalence of disorderedgambling among adults in theUnited States to the prevalence of avariety of other, better-known psychiatricdisorders. Although a range of epidemiologicstudies have produced a range ofprevalence estimates for each psychiatric disorder,Table 17 uses illustrative rates derived fromthe Epidemiologic Catchment Area study (Robins& Regier, 1991). Table 17 shows how estimatesof level 3 gambling derived from thisstudy compare with other disorders across bothlifetime and past-year time frames among adultsin the United States. Readers should note that,with the exception of the lifetime prevalencerate of alcohol abuse and dependence, all theprevalence rates are lower than 10%.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling60Prevalence estimates of psychiatric disordersamong adolescents also vary as a function ofmethodological considerations such as instrument,region, and year. For illustrative purposes,Table 18 shows our meta-analysis estimateof past-year level 3 gambling among adolescentsalong with 6-month estimates of wellknownpsychiatric disorders from two studies.These two studies are a New York study of adolescents11 to 20 years old (Velez, Johnson, &Cohen, 1989) and an Ontario study of youngpeople 4 to 16 years old (Offord et al., 1987).As Table 18 reveals, the rates of conduct andattention deficit disorders are similar to level 3gambling disorders.Methodological Considerations &Prevalence Estimation“I have yet to see any problem, however complicated,which, when you looked at it in the rightway, did not become still more complicated.” -Poul Anderson (1969) 45peer review. Taken together, the failure to improvemethods and the paucity of publishedprevalence studies encourages us to ask whatfactors may have hindered the evolution ofmethodological quality. For example, the peerreview process is one important means of improvingmethodological rigor among scientists.Since the majority of prevalence studies in thegambling field were absent this resource, it isnot surprising that, taken as a group, gamblingprevalence studies have not reflected improvedmethods over time. Furthermore, the largestsingle publication outlet for the published prevalencestudies was the Journal of Gambling Studies.While this is an excellent scholarly journal,it is not intended primarily as an epidemiologicaljournal. During the period that these prevalencestudies were published, to the best of ourknowledge, there were no epidemiologists onthe editorial review board. Therefore, given therelatively closed circle of investigators and reviewerswho were conducting and inspectingPrevalence Study QualityAccording to the methodological qualityindex developed for this study, the resultsreported earlier revealed that theoverall quality of disordered gamblingprevalence research has not improved significantlyduring the past 20 years. In addition,studies with higher quality scores did not generateprevalence rates that were meaningfully differentfrom those with lower quality indices.Among study types, adolescent and adult generalpopulation studies did evidence the highestoverall quality scores. This finding simply maybe due to the fact that these projects more oftenused larger sample sizes and random subjectselection than the special population studieswhich often relied on smaller samples of convenience.Table 18: Comparing Past-year Prevalence Ratesof Psychiatric Disorders Among Adolescents:Where Does Disordered Gambling Fit?Meta NY ONTGambling Disorder (Level 3) 5.8Conduct Disorder (All Types) 5.4 5.5Attention Deficit Disorder 4.3 6.2Anxiety Disorder 2.7 N.A.Major Depressive Episode 1.7 N.A.This study also revealed that only 40.1% of theavailable prevalence studies were subjected to45 New Scientist (1969, September 25).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling61gambling prevalence research, there was littlestimulation from independent psychiatric epidemiologistswho specialized in the developmentand implementation of prevalence research.It is imperative that the field of gamblingstudies attract new and seasoned investigatorsfrom other more mature fields of inquiry toadvance gambling research.Somewhat surprising was the finding that studyquality did not significantly influence prevalencerates. Since the methodological qualityindex was distributed normally, we might haveexpected that studies from the highest and lowestquartiles of the distribution would haveyielded meaningfully different prevalence rates.However, this was not the case. Since methodologicalquality apparently failed to influenceprevalence rates, we have concluded that disorderedgambling is a robust and reliable phenomenon.Disordered gambling appears to berelatively impervious to some of the weaknessesinherent in many of the research designs reviewedin this study. Nonetheless, readersshould consider the alternative explanation thatthe methodological quality index we have developedmay not accurately reflect the degree ofmethodological quality associated with a study.This failure to accurately reflect methodologicalquality could have occurred for two reasons.The first possibility is that our index of methodologicalquality may have failed to identify allof the appropriate variables that best representmethodological quality. Although we made effortsto include factors clearly related to the developmentof internal validity and methodologicalquality, other researchers might select asomewhat different set of factors; the possibilityremains that we omitted some important factorsand included some other factors less importantto study quality. The second possibility is thatsome of the researchers who did conduct highqualitystudies did not reflect these high-qualitymethods in their reports, and consequently receivedrelatively low methodological qualityscores. In addition, it is possible that researcherswho conducted studies with low overallquality actually were diligent about reportingthe details of their methodology. Under thesecircumstances, they would have received fullcredit for any individual components associatedwith study quality that appeared in their reports.In the cases where a study’s methodologicalquality score failed to reflect the true nature of astudy’s quality, any existing relationship betweenquality score and prevalence rate may nothave been revealed. More research is necessaryto investigate the relationship between prevalenceand study quality.The finding that methodological quality did notinfluence prevalence rates should not be takenas support for sloppy research methods or forthose who would avoid the expense associatedwith high-quality research protocols and designs.Scientists have a social responsibility:they are obliged ethically to develop and implementstudies that are rigorous, objective, andprecise. When a prevalence project includesmost of the elements of a quality research design,but compromises on some of the importantcomponents, the entire study is questionable.Questionable studies do not permit meaningfulinterpretation of results, and as a consequencecannot be generalized with precision. Sincepoor-quality studies demand resources and mayproduce misleading results, we encourage legislatorsand administrators to devote the resourcesnecessary to conduct high-quality studies. Researchersinterested in reading more about theessential components of high-quality prevalenceresearch should consult Appendix 3.Comparing Lifetime and Past-YearPrevalence RatesInspection of the aggregated lifetime and pastyearprevalence rates for adolescent studiessummarized in Table 5 reveals an unexpectedphenomenon for both level 3 and level 2 rates:the mean past-year rate is higher than the meanlifetime rate. There are a number of possibleexplanations for this curious phenomenon. Wewill consider two of the principal explanationshere. The first explanation is as follows: formost adolescents, past-year gambling experienceswill be comparable to lifetime gamblingexperiences; in other words, any gambling experiencesadolescents have had in their lifetimeHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling62are likely to be featured in their past-year experiences.Adolescents are fairly close in timeto the chronological moment when they startedexperimenting with gambling. Therefore, theyare unlikely both to have developed and alsorecovered from gambling problems. This notionthat adolescents’ lifetime and past-year rates aresimilar is supported by the fact that there is considerableoverlap between the confidence intervalsthat represent the lifetime and past-yearprevalence of disordered gambling.A second explanation is that very few adolescentstudies provided both lifetime and past-yearrates; those that did provide both types of ratesused different instruments for lifetime and pastyearrates. Thus, the studies that provided lifetimerates and the studies that provided pastyearrates represent two fairly distinct and exclusivesubsets of data. These two subsets ofstudies have other distinct characteristics. Forexample, lifetime studies were more likely torepresent Quebec and the South Central regionof the United States, while past-year studieswere more likely to represent Ontario, NewEngland, and the North Central Region. In addition,lifetime studies were more likely to use theSOGS, versions of the DSM, and the multifactormethod, while past-year studies were morelikely to use the MAGS and the SOGS-RA. Finally,lifetime studies were conducted earlier(mean = 1993) than past-year studies (mean =1995). Taken together, these factors may havecombined to yield higher past-year estimatesthan lifetime rates. An interesting example illustratesthis possibility. In Volberg’s (1993a)study of adolescents in Washington state, threeseparate prevalence rates are provided: a lifetimerate derived from the Multifactor Method,a lifetime rate derived from the SOGS, and apast-year rate derived from the SOGS-RA broadcriteria. A comparison of these three rates revealsthat, for level 3 gambling, the past-yearestimate is higher than either of the lifetime estimates(SOGS-RA broad criteria past-year rate= 3.0%; Multifactor Method lifetime rate =0.9%; SOGS lifetime rate = 1.5%). The samepattern exists for level 2 gambling (SOGS-RAbroad criteria past-year rate = 20.0%; MultifactorMethod lifetime rate = 9.0%; SOGS lifetimerate = 6.5%). This example suggests that instrumentationaccounts for most of the differencebetween lifetime and past-year rates.However, since we have identified an increasingrate of disordered gambling among the generaladult population, we should not dismiss the possibilitythat adolescents also are experiencing anincreasing prevalence of gambling disorders thatcurrently is not detectable using the existingmeta-analytic data set. Future research is necessaryto clarify this matter further.Time Frame Caveats and ConcernsThe results revealed that, compared to adultsfrom the general population, youth from thegeneral population were nearly three times aslikely to have experienced level 3 (i.e., pathological)gambling in their lifetime and approximately2.6 times as likely to have experiencedlevel 2 (i.e., “problem”) gambling in their lifetime.These findings raise interesting issues. Intheory, these findings could indicate that theprevalence of disordered gambling is increasingin the general population. That is, in theory,lifetime rates cannot decrease over time; thus,when the adolescent respondents represented inthis report reach adulthood, they should evidencehigher lifetime rates of disordered gamblingthan the current adult respondents representedin this report. This finding would provideevidence that there is a cohort effect that is influencingthe prevalence rates of level 3 gambling.In other words, assuming lifetime ratesare valid, these results indicate that the adultsrepresented in this report did not experience thesame degree of disordered gambling during theiradolescence as current adolescents are experiencing.Therefore, the higher rate of disorderedgambling found among contemporary adolescentsmay be attributable not simply to adolescencebut, rather, to some interaction of adolescenceand the current social setting (e.g., availabilityof gambling, changes in the social setting,cultural approval of gambling). If this isthe case, the rate of disordered gambling in thegeneral population will increase as these adolescentsgrow into adulthood and new generationsof adolescents repeat this pattern.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling63However, there are other possible explanationsfor these findings. One possibility is that theuse of liberal screening instruments used withadolescents and more conservative instrumentsused with adults is responsible for the differencebetween the prevalence rates of these twogroups. The present results, however, do notprovide support for this explanation. For example,a comparison of SOGS results for adultsand adolescents (using consistent definitions oflevel 2 and level 3 gambling throughout) indicatesthat, for both lifetime level 3 and lifetimelevel 2, adolescents do have higher rates (level3: adult mean = 1.71%, n = 29; adolescent mean= 4.83%, n = 4; level 2: adult mean = 3.38%, n= 27, adolescent mean = 9.26%, n = 3).Still another explanation for these relative riskfindings exists. Instruments using a lifetimetime frame may not measure what these instrumentspurport to measure. One reason that lifetimeprevalence rates may provide a less-thanvalidmeasure is that “poor recall is associatedwith advancing age” (Stewart, Simon, Shechter,& Lipton, 1995, p. 272). In other words, theadults represented in this study actually mayhave experienced a rate of disordered gamblingduring their adolescence that is similar to therate experienced by the adolescents representedin this study. However, with advancing timeand diminishing memories, adult respondentsforgot their adolescent experiences and underreportedtheir lifetime symptoms. This circumstancewould result in underestimates of lifetimedisordered gambling rates among adults. Thishypothesis will remain viable until scientistsconduct incidence studies that can clarify thewaxing and waning of gambling-related memoriesand support or refute this notion.Finally, lifetime time frames may provide overestimatesof disordered gambling for two verydifferent methodological reasons: symptomclustering and exclusionary requirements. First,we will consider symptom clustering. Whenscientists identify the prevalence of any psychiatricdisorder, it is important to measure the existenceof a determined number of symptomsduring a specified and meaningful period oftime. Instruments with lifetime time frames collectdata about lifetime symptom prevalence.However, the reported symptoms may have existedduring considerably different time periods,even separated by many years. Under this circumstance,a positive lifetime total score (e.g., 5or more positive responses on the SOGS) willnot reflect the existence of a disorder, just thewaxing and waning of subjective experiences.In other words, as Walker and Dickerson (1996)cautioned, survey respondents who report meetingthe necessary screening criteria may nothave experienced all of the reported symptomswithin the time frame necessary to identify acoherent disorder. To illustrate, with a lifetimetime frame, a respondent who experienced fiveSOGS symptoms simultaneously for an extendedperiod of time would receive the samescore as a respondent who experienced onesymptom during one year, a different symptomfour years later, two other symptoms the nextyear, and a fifth symptom three years later.Theoretically, the phenomenon of overestimatingprevalence as a result of “non-clustered”symptoms will increase as the age of respondentsincreases, since older respondents havemore opportunities to experience isolated symptoms;therefore, older respondents have moreopportunity to reach the threshold for lifetimepathological gambling. For this reason, we recommendthe use of past-year (or other “current”)time frames as the most accurate measureof the existence of clustered indicators of agambling disorder.The second potential methodological compromiseto lifetime time frames is that investigatorshave neglected to implement exclusionary requirements(Boyd et al., 1984). In virtuallyevery study of disordered gambling included inthis review, prevalence estimates fail to distinguishwhether the excessive gambling could bebetter explained by psychiatric illness (e.g.,manic episode, anti-social personality). Duringboth lifetime and past-year time frames, investigatorsmust determine whether the presence ofanother psychiatric illness could better explainintemperate gambling. Respondents with psychiatricillnesses that can confound estimates ofdisordered gambling must be excluded fromsurvey samples.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling64Cunningham-Williams, Cottler, Compton, &Spitznagel (in press) calculated odds ratios forthe likelihood of recreational and disorderedgamblers having higher rates of 13 psychiatricand substance abuse disorders than nongamblers.The association between gamblingand Anti-social Personality Disorder was thestrongest: the group of problem gamblers were6.1 times as likely as non-gamblers to meet diagnosticcriteria for Anti-social Personality Disorder.These findings and other preliminaryexplorations point to a reasonable hypothesisthat some percentage of people identified with agambling disorder may be suffering primarilywith another psychiatric disorder. We will discussthis matter in more detail later in the sectionof the Discussion that considers pathologicalgambling and exclusion criteria.To assist researchers in their attempt to managetime frame difficulties, we suggest that investigators—afterhaving excluded respondents whoqualify as having potentially confounding psychiatricillness—determine if the remaining eligiblesurvey respondents meet specific lifetimecriteria for disordered gambling that includes atime frame for symptom clustering. Investigatorsshould then determine whether this level ofgambling (e.g., pathological gambling) was presentduring the past year (i.e., 12 months). Investigatorsalso have the opportunity to determineif the respondent was in remission fromtheir gambling problem. Since DSM-IV doesnot include remission guidelines, and no standardhas been established in the area of gamblingdisorders, we suggest the following criteria.46 For full remission, respondents must havepreviously met criteria for a level 3 gamblingdisorder, and there should be no evidence ofgambling for the past 12 months; in addition,there must be no symptoms present for the past12 months. To establish partial remission, therespondent must have previously met criteria fora level 3 gambling disorder, but during the past12 months satisfies only level 2 criteria for subclinicalgambling. For respondents who havenever met criteria for a level 3 gambling disorder,the guidelines for a level 2 gambling problemplace them in the “problem” or “at-risk”group of gamblers; although these gamblers mayresolve their difficulties with no further progression,clinicians should consider them at risk fordeveloping a more serious disorder.Calculating Prevalence RatesNearly every study in this meta-analysis conceptualizedprevalence rates in the same way:Prevalence was calculated by dividing the numberof respondents experiencing disorderedgambling by the total number of respondents inthe study. Expressing prevalence rates as thepercentage of the “general population” that experiencesthe phenomenon in question is a standardpractice in epidemiological research.However, in the gambling research field, thereare benefits to including a second method ofcalculating prevalence rates: in this secondmethod, prevalence rates would be calculated bydividing the number of respondents experiencingdisordered gambling by the number of respondentswho are at risk for developing disorderedgambling (e.g., those in the total populationwho have gambled in their lifetime). Thisconceptualization of prevalence is based on thepremise that if one never gambles, there is noactive or practical risk of becoming a pathologicalgambler. 47 Similarly, if one never drinksalcohol, there is no risk of developing alcoholdependence. This approach is appropriate in anumber of prevalence-related research protocols.For example, when determining the prevalenceof adverse reactions to a prescriptionmedication or a toxic substance, only those exposedto the substance are considered at risk;thus, this group is used as the reference groupamong which prevalence is calculated.46 These criteria rest upon similar guidelines for researchfirst established by McAuliffe et al. (1995) andRobins et al. (1985).47 There always remains the theoretical risk that anon-gambler will begin to gambling and then becomea disordered gambler.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling65Unless the entire population has been exposedto gambling, prevalence rates of disorderedgambling are lower when using the epidemiologicalstandard of the entire population as thedenominator. Social observers might argue thatin contemporary America and Canada, everyoneis exposed to gambling because of the extent oflottery and casino advertising. However, beingexposed to gambling advertising is not the sameas being exposed to gambling experience. Exposureto gambling advertising can be considereda risk factor for determining the incidenceof new gamblers. In addition, this secondaryapproach to prevalence can be used to determinewhether different segments of the population(e.g., adolescents, treatment groups) have differentialsensitivity to the risk factor of exposure togambling. For example, two groups could havesimilar prevalence rates of disordered gamblingwhen prevalence is calculated in the standardmanner, but have significantly different rateswhen prevalence is calculated as the percentageof those who have gambled. In this hypotheticalscenario, one of the groups could have a lowerrate of gambling but a higher rate of disorderedgambling among those who have gambled. Werecommend that, in future studies, investigatorsreport their data using this method of calculatingprevalence as well as the standard method. Atthe very least, investigators should describe preciselyhow they calculate and operationally definetheir reported prevalence rates. This is aconceptual and methodological matter of considerableimportance, and it requires significantlymore attention and dialogue than it hasreceived.The Different Purposes of Prevalenceand Incidence StudiesAs we discussed earlier in this report, prevalenceand incidence measures are produced fromdistinct and different study methodologies anddesigns. Although several scientists whosework is included in this meta-analysis use theterms “prevalence” and “incidence” interchangeably(e.g., Allen, 1995; Kallick et al.,1979; Laventhol & Horwath et al., 1990), it isimportant to keep the differences between theseconcepts in mind. For example, although weidentified 120 prevalence studies of disorderedgambling, there have been virtually no incidencestudies conducted in this field.In lieu of conducting incidence studies, mostresearchers in the gambling field have opted toconduct cross-sectional baseline and replicationprevalence studies in an attempt to identifytrends. Under these circumstances, baseline andreplication studies will provide little useful informationabout trends unless sampling, methodologicaldesign, and study implementation arevery similar. Absent such duplication of methods,a replication study will simply provide anotherprevalence estimate that may evidenceerrors now compounded across two separatestudies. Although prevalence estimates are notoptimally suited to distinguish trends, theserates are useful to identify the extent of disorderedgambling and the potential treatmentneeds within a particular region or population.In addition, prevalence researchers have not yetdevoted attention to the prevalence of disorderedgambling among specific segments of thepopulation (e.g., homeless, senior citizens, gaymen and lesbians). If investigators are interestedin identifying trends in prevalence, theyshould conduct incidence studies. It is time forthe field of disordered gambling studies to conducttrue incidence research by prospectivelyexploring the factors and circumstances thatshift the scope and severity of disordered gamblingin the United States and Canada.The Validity of Disordered GamblingPrevalence Estimates“Most of the change we think we see in life is dueto truths being in and out of favor.”- Robert Frost (1914)This section of the Discussion brings our examinationof disordered gambling prevalenceestimates full circle. The results of this metaanalysissuggest that there is a relatively stableand robust phenomenon that we have called“disordered gambling.” These results also suggestthat estimates of disordered gambling remainstable in spite of various statistical maneuversand wide variation in study quality andHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling66characteristics. Although these results do encourageconfidence about the reliability of thisphenomenon, the stability and robustness of thisphenomenon should not be interpreted as aproxy for validity. To determine if the results ofthis meta-analysis provide a “valid” estimate ofdisordered gambling in the United States andCanada, we must first consider what the constructsof disordered gambling and validitymean within the context of American and Canadiansociety and contemporary scientific theory.At the outset of this report, we noted that scientiststend to view the world through three differentframeworks: (1) realism, (2) instrumentalism,and (3) relativism (Casti, 1989). We havetaken a relativistic perspective on the concept ofprevalence. From this standpoint, scientistsmanufacture prevalence estimates. This relativisticperspective provides room for culturallydiverse views of gambling. For example, unlikeAmericans and Canadians, Chinese do not recognizethe construct of pathological gambling,only “bad” gambling (Pathological gambling asa reflection of cultural norms, 1997). Certainly,some Chinese gamble to excess; however,within their view of the universe, this intemperategambling can be only bad or good; it cannotbe an illness. Within the Chinese culture, pathologicalgambling has no validity. This issuemay seem incredible to some, but validity is indeeda relative idea. Validity is fluid, dynamic,and not fixed by any single research finding ordata set. Validity is only as serviceable as thecurrent theory that provides it safe haven. Inthis section, we will explore some of the centralissues that relate to validity in general(Cochrane & Holland, 1971; Dohrenwend,1995; Malagady, Rogler, & Tryon, 1992; Robins,Helzer, Ratcliff, & Seyfried, 1982) and constructvalidity in particular (Cronbach & Meehl,1955).Paradigms for Understanding: TheRelative Nature of Disordered GamblingBurke (1985) reminds us that the universe isbest understood through a shifting theoreticallens known as a paradigm. Our social andscientific paradigms provide meaning to thesocial and natural order while simultaneouslyblinding us to alternative perspectives. “Whenwe observe nature we see what we want to see,according to what we believe we know about itat the time. Nature is disordered, powerful andchaotic, and through fear of the chaos we imposesystem on it. We abhor complexity, andseek to simplify things whenever we can bywhatever means we have at hand. We need tohave an overall explanation of what the universeis and how it functions. In order to achieve thisoverall view we develop explanatory theorieswhich will give structure to natural phenomena:we classify nature into a coherent system whichappears to do what we say it does... All communitiesin all places at all times manifest theirown view of reality in what they do. The entireculture reflects the contemporary model of reality.We are what we know. And when the bodyof knowledge changes, so do we” (Burke, 1985,p.11).When primitive peoples watched the sun moveacross the heavens, they saw evidence that thesun was rotating around the earth. Today, peoplewatching the same sun crossing the same skysee it as evidence for a very different phenomenon:now, the earth is revolving around the sun.Although the evidence is precisely the same, ourexperience of it is dramatically different. Ourtheory has changed, and therefore so has theuniverse. Today, we recognize excessive andintemperate gambling as the result of a confluenceof biological, psychological, and socialfactors (e.g., Shaffer, Stein, Gambino, & Cummings,1989). This understanding, or paradigm,provides the lens through which the construct ofdisordered gambling comes into focus. With theemergence of this gambling construct, we canaddress the matter of its validity.When considering the validity of any construct,scientists and policy makers must ask the primaryquestion: valid for what? A construct canhave considerable validity and utility, only tolose these attributes in a technological instantwhen a new finding shifts our understanding ofthe universe. For example, though modern ge-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling67netics had its roots in the—apparently concrete—constructof 48 chromosomes, the developmentof the electron microscope shifted ourparadigm and revealed the presence of only 46chromosomes. Instead of simply assuming thata “true” prevalence estimate awaits our capacityto identify it accurately, we believe that a dynamicinterplay of factors influences and determinesevery prevalence estimate: which measurementinstrument, with which population,with which sampling strategy, with which administrativeprocedure, at which historical pointin time, under the direction of which scientists;all of these factors influence the outcome of aneffort to estimate prevalence. This observationis particularly true when we try to estimate theprevalence of a latent class in the absence of agold standard (Faraone & Tsuang, 1994). Alatent class is the “actual” state of a person underconsideration, for example, a patient subjectedto psychological evaluation. “Simple examplesof latent classes from everyday life areseen in the human tendency to classify peopleaccording to personality attributes (honest ordishonest), emotional states (happy or sad), orintellectual ability (intelligent or unintelligent).We cannot directly observe honesty, happiness,or intelligence, but we can observe behaviorfrom which we infer these latent (i.e., unobservable)characteristics” (Faraone & Tsuang, 1994,p. 651). By adopting a relativistic posture to theanalysis of the state presumed to underlie disorderedgambling, we can explicate the manufacturingprocess responsible for generating prevalenceestimates; in turn, this posture will allowus to improve the quality controls associatedwith these production activities.Reconsidering Clinical Diagnoses asthe “Gold Standard”Investigators of gambling prevalence often haveassumed that clinicians provide the “gold standard”against which the accuracy of screeninginstruments should be measured (e.g., Lesieur &Blume, 1987; Volberg, 1996b; WEFA Group,ICR Survey Research Group, Lesieur, &Thompson, 1997). For example, Volberg(1996b) uses the term “probable pathologicalgambling” rather than “pathological gambling”and states that “the term probable distinguishesthe results of prevalence surveys, where classificationis based on responses to questions in atelephone interview, from a clinical diagnosis”(p. 3). Drawing a similar conclusion, the WEFAGroup et al. (1997) state that, “Because only aclinical evaluation using DSM-IV can diagnosepathological gambling, we have used the term‘probable’ pathological gambling” (p. 5-2).“Since the survey is not a clinical diagnosis, wecannot say that respondents can be ‘diagnosed’as pathological gamblers, rather we use the term‘probable’ pathological gamblers” (p. 5-5).However, clinicians who perform diagnosticevaluations are not as reliable as many peoplehave assumed. Meehl (1954; 1973) and others(e.g., Rosenhan, 1973; Ziskin, 1970) demonstratedlong ago that clinicians are extremelyvulnerable to biases in clinical judgment. Becauseof the many problems associated withclinical judgment, Kleinman (1987) remindedclinicians to sustain a tentative posture as theyconsidered diagnostic classification: “If [classification]...is applied as a tentative model, …with serious concern for its likely inadequacy tograsp the subtlety, ambiguity and obdurate humanityof the sick, if it is distrusted, received asa mere shorthand, regarded as one vision amongothers, then it is more likely to be adequate towhat should be the humane core of clinical work...” (pp. 51-52). In addition, Faraone and Tsuang(1994) emphasize the fact that psychiatricdiagnoses should not be considered a gold standardand that it is important to assess the adequacyof these diagnoses. Therefore, the assumptionthat gamblers should be grouped into atentative class, for example, probable pathologicalgamblers, partly because clinicians have notyet determined the accuracy of that categoricalassignment, is faulty. There is little evidencesuggesting that clinicians are more accurate instrumentsof classification than screening instruments.In fact, Lee Robins has concludedthat, “’clinical practice is not an adequate standardagainst which to measure the validity of aresearch instrument’” (cited in Malagady, Ro-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling68gler, and Tryon, 1992, p. 63). 48 Nevertheless, asKleinman encouraged, we suggest that all diagnosticclassification—whether clinician- or instrument-based—beheld as tentative, and notthe final word (Shaffer, 1986b).Validity as a Theory-Driven ConstructThe problems associated with determiningvalidity begin with its very definition. Validityis the capacity of an instrument to measure whatit purports to measure. Validity is neither astatic nor an inherent characteristic of ascreening instrument. As Goldstein andSimpson (1995) suggest, “validity refers to thequestions ‘for what purpose is the indicatorbeing used?’... and ‘how accurate is it for thatpurpose?’” (pp. 229-230). Determininginstrument validity is an unending and dynamicinvestigative process. We cannot simplyconclude that an instrument has been shown tobe valid for all purposes and all settings. “Anindicator (e.g., an instrument, such as a test, arating, or an interview) can be valid for onepurpose, but not for another” (Goldstein &Simpson, 1995, p. 230). Directed by theoreticaland ultimately practical purposes, validity is thedynamic consequence of applying an instrumentto a measurement task. In the field of gamblingstudies, theory is conspicuously absent frommost prevalence research. As we mentionedearlier, when conventional wisdom and theoryshift or change, the validity of an instrument canbe terminated abruptly. The history of theSOGS provides an example of the relativenature of validity. Although for some timeresearchers considered that the SOGS lifetimemeasure had “been found valid and reliable”(Volberg, 1994b, p. 238), some investigatorsnow suggest that the SOGS lifetime measures“…over-state the actual prevalence of pathologicalgambling” (Volberg, 1997, p. 41).In addition to a paucity of theory-driven gambling-relatedepidemiological data, there also isa shortage of objective-driven prevalence research.In the absence of clearly stated purposes,it is virtually impossible to determine thestandard against which the validity of the constructunderlying a prevalence estimate of disorderedgambling should be judged. “It is ordinarilynecessary to evaluate construct validityby integrating evidence from many differentsources. The problem of construct validationbecomes especially acute in the clinical fieldsince for many of the constructs dealt with it isnot a question of finding an imperfect criterionbut of finding any criterion at all” (PsychologicalBulletin Supplement, 1954, pp. 14-15; ascited in Cronbach & Meehl, 1955, p.285). Inthe absence of a criterion, or gold standard,“…many such tests have been left unvalidated,or a finespun network of rationalizations hasbeen offered as if it were validation. Rationalizationis not construct validation” (Cronbach &Meehl, 1955, p. 291).Given that the field of psychiatry in general, anddisordered gambling in particular, has no goldstandards (Faraone & Tsuang, 1994), there islittle rationale for placing one instrument, orclinical diagnosis, as the gold standard againstwhich another instrument is measured. AsFaraone & Tsuang (1994) note, “many studiesof psychiatric diagnosis compute accuracy statistics.However, these assess the accuracy ofone diagnostic approach (e.g., DSM-IV) withrespect to another (e.g., expert clinical diagnoses).They do not assess the accuracy of diagnosticprocedures with reference to the ‘true’but unobservable latent state of illness” (p. 651).Like the MAGS 49 (Shaffer et al., 1994), theSOGS has its roots in the DSM-based criteria(Lesieur & Blume, 1987). Similarly, clinicalassessment is, in theory, based on DSM criteria.Consequently, it is a tautological error to useclinical diagnosis or the DSM criteria as an indexof validity for the SOGS or the MAGS. In48 Interested readers should see Ziskin (1970) andFaraone & Tsuang (1994) for a more complete analysisof this problem.49 The MAGS also has its roots in the shortened versionof the Michigan Alcoholism Screening Test(Pokovny, Miller, & Kaplan, 1972).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling69each of the disordered gambling instrumentsderived from criteria established by the AmericanPsychiatric Association (e.g., SOGS,MAGS, and psychiatrists or other professionallytrained clinicians who diagnose disorderedgambling according to DSM criteria), DSM criteriado not provide an independent standardagainst which any of these clinical instrumentsor activities can be judged (Malagady et al.,1992). The following observations made byMalagady et al. (1992), although originallymade in reference to the development of the DiagnosticInterview Schedule (DIS), applyequally to the development of a valid instrumentfor disordered gambling: “The flaw in settingpsychiatrist’s clinical skills as the standard ofcriterion-related validity is the fact that psychiatricdiagnoses are not external to lay diagnosesbecause both come from the DIS ....in the absenceof empirical research linking DIS diagnosesto external (non-DIS) criterion-related standards,there is simply little that can be said regardingthe psychometric validity of the DIS atthe present time. Lay-psychiatric concordanceis praiseworthy because of the economic utilitygained in epidemiological research. On theother hand, lay-psychiatric concordance is notpraiseworthy when it is taken as a proxy for validity”(Malagady et al., 1992, pp. 62-63). Economic,emotional, and procedural inertia encouragescientists to continue using existingstandards of construct validation. Nevertheless,economic, personal, and administrative advantagesdo not necessarily provide a solution forthe formidable matters of validity.Absent a “gold standard” for determining pathologicalgambling, we do not know whether theSOGS over-estimates the prevalence rate ofgambling disorders or whether clinical assessmentand DSM-based instruments underestimatethe prevalence rate. This problem ofanchoring reveals itself often as scientists attemptto determine how best to frame prevalenceestimates. To illustrate, while discussingthe results of her recent New York replicationstudy, Volberg (1996b) suggests that “...the cutoffpoint for the DSM-IV Screen (5 + = pathologicalgambling) is too severe and should bemoved back to include individuals with less severegambling difficulties.” This adjustment“...would allow the screen to capture individualswhose pathology is well-developed but perhapsnot yet extreme” (p. 50). However, the practiceof using an instrument originally based on DSMcriteria (i.e., the SOGS) as the anchoring standardagainst which the DSM is held for comparativeevaluation is questionable. This conceptualconfusion has plagued epidemiologicalmeasurement for many decades, and psychologicalscreening instruments from their earliestdays (Cronbach & Meehl, 1955). We shouldnot be surprised that these issues also surroundthe construct of pathological gambling.In spite of Volberg’s notion that screening canidentify a “true” or “actual” prevalence rate ofdisordered gambling (Volberg, 1997, p. 41), shetacitly recognizes that prevalence rates areflexible and relative depending upon the purposefor which the estimates will be used. Forexample, in her study of Washington state adolescents,Volberg states that “Our approach,while conservative, is intended to focus asclearly as possible on those adolescents whoshow incontrovertible signs of problematic involvementin gambling” (Volberg, 1993a, p.17).Volberg uses this conservative approach, in part,because she is “…uncomfortable with a method[i.e., the SOGS] that classifies 8% of adolescentsas problem or probable pathological gamblers”(1993a, p. 34). In other words, the criteriafor identifying pathological gambling are notfixed, but vary depending on whether a given setof criteria yields an “acceptable” prevalence rateamong a particular group.Considering False Positives and FalseNegatives: The Need for TheoryThe often-repeated debate about false positivesand false negatives contributes little to our understandingof gambling disorders. An instrumentmay yield useful prevalence indices evenif it has only moderate sensitivity and specificity.So long as false positives are “balanced” byfalse negatives, an instrument will perform veryHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling70well as a general population screen. 50 Onlywhen these faulty classification assignments failto balance does a screening instrument begin todeteriorate as an index of population prevalence.The major problem with the debates surroundingfaulty classification is the same as the debatesabout validity: to determine a false assignment,scientists must invoke a standardagainst which we judge the classification system.Without a gold standard, there is the riskof an “infinite frustration”—always trying torefer to a higher standard for construct validity(Cronbach & Meehl, 1955). Consequently, theonly way out of this conceptual chaos is to developwhat Cronbach & Meehl called a nomologicalnetwork: an interlocking system oflaws that constitutes a theory.As we have described, with few exceptions theprevalence estimates reviewed in this studyseem to have been promulgated, at best, by thequestion of “let’s find out,” and, at worst, in aconceptual vacuum. The goals and objectives ofdisordered gambling prevalence estimation havenot been made sufficiently clear by the variousresearch teams who set out to generate such estimates.Without knowing whether investigatorswant the estimates of disordered gamblingprevalence to guide the development of publicpolicy, allocate limited public health resources,expand scientific knowledge, identify economicneeds, or inform some other activity, observersare left to project their own needs on prevalenceestimates. Absent a driving set of objectives,gambling epidemiologists have not providedpolicy makers with some of the basic informationessential to building an effective system oftreatment (e.g., estimates of the obstacles totreatment entry and how often people encounterthese difficulties). Similarly, because prevalenceresearch has not been driven by explicittheory, investigators have not identified the di-50 In clinical practice, however, quality patient carerequires a very different standard of accuracy; whenthere is only one case under consideration, there is noone to “balance” a misdiagnosis. Therefore, the consequencesof clinical errors may be quite severe.rection of movement among level 2 gamblerseither toward or away from more disorderedstates. Without this information, social cost estimatesare less than precise, and the challengesto conceptual validity continue without a foreseeableend.Is Pathological Gambling a PrimaryDisorder?The various versions of the DSM that have includedpathological gambling as a distinct disorderalso have drawn attention to the possibilitythat other disorders may coexist and negatethe diagnosis of pathological gambling. TheDSM-III (APA, 1980) instructed clinicians toavoid making the diagnosis of pathologicalgambling if the gambling was secondary to aprimary diagnosis of Antisocial Personality Disorder,hypomanic or manic episode, or socialgambling. The DSM-III-R (APA, 1987) revisedthe exclusion criteria by removing AntisocialPersonality Disorder. The DSM-IV (APA,1994) indicates that a diagnosis of pathologicalgambling should not include behavior patternsbetter explained by manic episodes. DSM-IValso includes a comment that pathological gamblingmust be distinguished from social and professionalgambling, and that “problems withgambling may occur in individuals with AntisocialPersonality Disorder; if criteria are met forboth disorders both can be diagnosed” (p. 617).In addition, DSM-IV notes that pathologicalgamblers “may be prone to developing generalmedical conditions that are associated withstress… Increased rates of Mood Disorders,Attention-Deficit Hyperactivity Disorder, SubstanceAbuse or Dependence, and Antisocial,Narcissistic, and Borderline Personality Disordershave been reported in individuals withPathological Gambling” (p. 616).Although clinicians may be heeding these exclusioncriteria for purposes of treatment planning,for the most part, researchers simply have ignoredthe implications of exclusion criteria(Boyd et al., 1984). One of the research implicationsof these clinical recommendations is thatprevalence estimates of pathological gamblingmay be overestimates when other disorders suchHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling71as manic episodes are not identified by the surveyinstrument. Another implication is that, asattested to by the shifting of the exclusion criteriafrom antisocial personality disorder to manicepisode within a seven-year period, investigatorsof disordered gambling phenomenon are stillstruggling to develop a clear definition of pathologicalgambling. Much remains unknownabout the nature of the overlap among antisocialpersonality, manic episodes, and pathologicalgambling. Future research that measures theprevalence of related psychiatric disorders alongwith pathological gambling will provide importantinsight into these questions.Ultimately, the field of gambling studies is inneed of research that can provide additionalconstruct validity. While the evidence from thismeta-analysis provides considerable support fora recognizable and identifiable pattern of behaviorsthat we have considered to be disorderedgambling, there are important conceptual questionsthat still remain. A primary construct validityquestion requires scientists to focus onwhether disordered gambling is a primary or asecondary disorder. For example, according tothe DSM-IV, a person meeting all of the criteriafor pathological gambling is not considered apathological gambler if he or she also meets thecriteria for a Manic Episode, and the Manic Episodeis responsible for excessive gambling(APA, 1994). In this case, pathological gamblingis not considered a unique disorder, butrather a cluster of symptoms associated withanother disorder. If pathological gambling representsa primary disorder, then it can emerge inthe absence of other co-morbidity and causesequelae independent of any other condition.However, if it is a secondary disorder, subordinateto other dysfunctional behavior, thenpathological gambling will only exist as aconsequent of another condition (e.g., manicepisode, anti-social personality, alcohol abuse,obsessive-compulsive disorder or adolescence;Jessor & Jessor, 1977). Although this metaanalysissuggests that researchers of disorderedgambling have measured a relatively stable androbust phenomenon, the field of gambling studieshas not yet established with ample certaintythat this phenomenon represents a unique construct.Does disordered gambling reflect a unique primaryor dominant psychiatric disorder? For example,consider the following hypothetical scenario:A survey investigates pathological gambling,depression, and alcoholism by using theSOGS, a depression inventory, and the CAGE(Ewing, 1984) with a general population sample.The results of this survey indicate that respondents’scores on the depression inventory andthe CAGE successfully discriminate pathologicalgamblers from non-pathological gamblers(i.e., these scores not only correlate, but alsopredict a respondent’s level of gambling problemson the SOGS). In this scenario, pathologicalgambling may not be a discrete syndrome,but rather may reflect a cluster of symptomsassociated with alcoholism and/or depression.In other words, disordered gambling may be theexpression of other underlying factors whichalso are common to alcohol abuse and depression.Rosenthal and Rosnow (1975) remind us that wemust be sensitive to the relationship betweenappropriate and inappropriate criteria during theprocess of establishing construct validity.Adapting an example originally provided byRosenthal and Rosnow (1975), suppose we findthat the SOGS correlates .85 with a criterion ofthe pooled judgment of expert clinicians. Whilethis would appear to provide concurrent validationfor the SOGS, suppose further that we alsoadminister the CAGE questionnaire and an indexof depression. If we also learned that thetotal SOGS score correlated significantly withthe composite CAGE and depression index,would the SOGS be a reasonably valid test ofgambling, alcoholism, depression, all three, orof none of these? As Rosenthal and Rosnow(1975) conclude in their analysis about a similarexample of test validation, “That question isdifficult to answer, but we could not claim onthe basis of these results to understand our[SOGS] test very well. It was not intended, afterall, to be a measure of [alcoholism or depression].In short, our test has good concurrentvalidity but poor differential validity. It doesHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling72not correlate differentially with criteria for differenttypes of observation” (p. 70).Manic EpisodeOther UnknownDisorderSince gambling researchers have paid very littleattention to this important conceptual issue ofdiscrete and comorbid phenomena 51 —and theassociated matter of differential validity—thepossibility remains that pathological gambling isnot a discrete primary disorder. Alternatively, italso is possible in some cases that pathologicalgambling will represent a discrete and primarydisorder (i.e., it will exist independent of anyother disorder), but in the majority of casesother primary disorders will provide better explanationsof excessive gambling (i.e., disorderedgambling will be subordinate to otherprimary disorders). Figure 18 illustrates thispossibility. Currently, according to the DSM-IV, pathological gambling can have either primaryor secondary status: in some cases it isconsidered to be a primary disorder (i.e., independentof other diagnoses), and in other casesit is considered to be the sequelae of anotherdisorder. The implications of this issue are potentiallysignificant for both the development oftreatment prescriptions and social policy initiativesdesigned to ameliorate or regulate gambling-relatedproblems.Whether we view disordered gambling as primaryor secondary, intemperate gambling inflictshuman suffering. If pathological gamblingis a primary disorder, it often will require professionalassistance; if it is a disorder secondary51 Briggs, Goodin, & Nelson (1996) have conductedpreliminary research on this issue. Their "findingsindicate no significant cross over of addictions betweenthe two samples. This would seem to indicatethat alcoholism and pathological gambling are independentaddictions. One might infer from this thatthey also involve independent processes" (Briggs etal., pp. 517-518). In spite of these conclusions, theBriggs et al. study employed a unique subject sample(e.g., use of self-help group members) that likelyrepresents the tails of two special self-selected distributions;they also employ a small sample size. Takencollectively, these factors encourage us to view theirresults as tentative and their conclusions as uncertain.PathologicalGamblingAlcoholismDepressionFigure 18: Is Pathological Gambling a DiscretePhenomenon?to another problem, it still may require specializedcare or assistance focusing on gamblingissues in addition to the problems related to theprimary disorder. Future research will help clarifythese theoretical, research, and clinical issues.Toward an Understanding of DisorderedGambling in Context: The Interactionof Social Setting and Personality“Excess on occasion is exhilarating. It preventsmoderation from acquiring the deadening effectof habit.”- W. Somerset Maugham (1938)It may seem very difficult to understand andintegrate the shifting patterns of disorderedgambling prevalence. Levine (1978) remindsus that our perception of addictivebehaviors in general and drunkenness in particularreadily shifts over time; our conception of adisorder changes as new experiences color understanding.Social learning provides a similarconceptual model to provide insight into theprevalence findings reported earlier. NormanZinberg adapted the traditional three-part modelof host, agent, and environment used by publichealth workers to understand infectious diseaseprocesses so that he could gain better insightinto drug effects (Zinberg, 1974; 1984). Duringthe middle 1980s, this model was modified toexplain the social psychology of intoxicant usein general (Zinberg & Shaffer, 1985; Shaffer &Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling73Zinberg, 1985). Zinberg’s (1984) model suggestedthat a drug and its effects were not isomorphic;in other words, to understand the effectsof any psychoactive drug, it would be necessaryto understand the drug, set (i.e., psychologicalexpectations), and setting as an interactivewhole. Similarly, to study the psychologicaland social effects of gambling, and patternsof disordered gambling in particular, it is necessaryto consider gambling, set, and setting as aninteractive whole. This three-factor model providesinsight into the effects of any activity thatholds the potential to shift subjective states(Shaffer, 1996). Now, we will adapt this modelto provide a conceptual context for understandingthe prevalence patterns of disordered gambling.Gambling represents the specific activities associatedwith playing a particular game (e.g.,determining what numbers to play in a lotteryand buying a ticket; determining what version ofpoker to play [5 or 7 card] and participating inthe game; playing a slot machine). Set refers tothe personality structure of the individual, includingtheir attitudes toward the experience andany values that might be associated with the activity.Setting is considered to be the influenceof the physical and social environment withinwhich gambling (or any intoxicant use) takesplace.To understand an individual’s decision to initiateor sustain gambling, we must consider thegambling/set/setting interaction, since these factorsdirectly influence the gambling experience.The relationships among gambling, personality,and social structure seem straightforward initially.For example, most observers of humanbehavior know that psychological states varygreatly and are influenced by the environmentand, of course, that gambling can make an impacton these states. Gambling can excite thespirit, temporarily reduce feelings of depression,and help people dissociate from day-to-dayproblems (e.g., Jacobs, 1989). While easy tograsp in the abstract, surprisingly, these issuesare very difficult to understand and accept inpractice. Most of us have become so accustomedto thinking about gambling as an excitingrecreational activity that we too often assumethe effect of gambling to (1) be the same foreveryone and (2) remain relatively constant overtime for each person. The operators and promotersof gaming establishments and activitiesare not eager to dispel this belief. To maximizethe impact of gambling experiences, advertisementsof the full array of gaming activities avoidreminding gamblers that the experience mayhave different effects on different people and,further, that the effects of any particular activityon the same person may vary over time.In spite of the overwhelming aura of hope andexcitement that gaming promoters have encouragedpeople to expect when they gamble, mostgaming experts have come to accept the influenceof set and setting on gambling experiences.For those not experienced with the gambling,set, and setting interaction, consider anothereven more common illustrative situation: alcohol,set, and setting. Most people recognize,either from observing their own behavior or thebehavior of others, that the effects of alcoholvary from person to person, and vary for thesame person over time. For example, Zinbergand Shaffer (1985) remind us of the range ofdrinkers: happy drinkers, morose drinkers, belligerentdrinkers, and flirtatious drinkers.Sometimes alcohol can be a relaxant (e.g., themartini after a hard day at the office), and sometimesit can be a stimulant (e.g., the first drink ata party). At times alcohol can release inhibitions,and at other times those who already haveput aside their inhibitions will take a drink ortwo to provide themselves with a socially acceptablealibi. Often, alcohol is a mood accelerator,deepening depression or heightening elation,depending on the preexisting conditions.From the pharmacological and neurophysiologicalperspectives, alcohol suppresses the actionof certain inhibiting centers in the brain and canhave no result inconsistent with this action. Yetthe range of actual effects observed, both behavioraland psychological, is extremely wide. Itmay be precisely this wide range of possibilitiesthat makes alcohol such a popular drug. Themultidimensional effects of alcohol serve toemphasize the importance of the interaction ofdrug, set, and setting.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling74Similar to alcohol, the interaction of gambling,set, and setting yields multifaceted effects. Wehave observed exhilarated, ecstatic, timid, angry,and depressed gamblers. Like alcohol,gambling is a mood accelerator that can deependepression or facilitate elation. Like alcohol,gambling can encourage some people to separatefrom their consciousness into another psychologicalrealm where time passes rapidly andfew worries exist (Jacobs, 1989). Similar toKhantzian’s self-medication hypothesis for intoxicantuse (e.g., Khantzian, 1975; 1985; 1989;1997) and Wagner’s (1986) humorous ideaabout the value of separating from reality sinceit is a primary source of stress, Jacobs (1989)also suggested that the experience of gamblingcan serve as an anodyne to painful emotionalstates. Most “recreational” activities provideemotional diversion, relief, and opportunity foraffective renewal; this is the sine qua non ofrecreation. While gambling may be work for theprofessional gambler, the amateur gambler expectsto experience a positive shift in their subjectiveexperience when they participate ingambling (Shaffer, 1996).The Impact of Social Setting on Gamblersand GamblingAs a result of changes in the social setting duringthe 1960s and 1970s, people quite differentfrom their earlier gambling counterparts beganto gamble. To illustrate, consider Zinberg andShaffer’s (1985) description of a pattern ofshifting events associated with the history ofdrug use. Weil, Zinberg, and Nelson (1968)were the first to notice that chronic marijuanausers were more anxious, antisocial, and likelyto be dysfunctional than the naïve subjects whowere just beginning to use marijuana in 1968.These early users were closer in spirit to the fewdisenchanted musicians and other groups thatused marijuana before the drug revolution of the1960s. By the late 1960s, drug use was beingexperienced as a more normative choice than ithad been before 1965. Later, during the early1970s, marijuana use was sufficiently widespreadthat it was not possible to describe thesesmokers as driven to drug use by deep-seated,self-destructive, unconscious motives. If we haddrawn conclusions about the effects of marijuanabased only upon people’s early experiencewith the drug and failed to consider the historicalmoment of its use, then we would have assumedincorrectly that marijuana alone causesprofound psychiatric and personality disturbance.In the 1990s, gamblers are quite different fromtheir earlier 1970 counterparts. The gamblerswho first tested the opportunities to gamble legallywere different from those who began togamble only when playing these games was sufficientlywidespread that it was normative. Beforecasino gambling became widely accessibleduring the last quarter of the 20 th century, legalgamblers had to have the motivation, time, andfinancial means to travel to Nevada or AtlanticCity to gamble. The relative isolation of mostpeople from these settings provided an informalsocial control over who would gamble, and howmuch they could gamble in casinos. Similarly,before New Hampshire resurrected legal numbersgaming by implementing a state lottery,gambling on numbers was limited strictly to illicitbookmaking. Today, numbers gaming is socommon that many people do not even considerlottery playing to be gambling.The data presented earlier suggests that disorderedgambling prevalence can be explained inlarge measure by knowing something about thegambler. If we assume that the something weneed to know about the gambler is their personality,then we would be adopting a psychodynamicperspective on gambling behavior.However, it is incorrect to assume that theprevalence of disordered gambling is simply afunction of personality patterns. This inaccurateassumption is common among clinician theoristswho tend to see only one very small segment ofdisordered gamblers—that is, help-seekinggamblers. Help-seeking gamblers likely arevery different from their non-help-seeking counterparts.To illustrate, when researchers examinepatients who enter treatment for addictivedisorders, they tend to find very high rates oftrauma and psychiatric disorder. This relationshipoften leads to the conclusion that these veryHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling75serious problems cause addiction. However,when researchers first identify people with highrates of addiction (independent of help-seekingbehavior), and then identify the pattern of theirpsychiatric disorders and trauma experience, thestrength of this apparent causal relationship diminishesor disappears (e.g., Gendreau & Gendreau,1971). There are many people with mentalillness and traumatic experiences who do notdevelop addictive behaviors; therefore, thesecomplex conditions cannot be the necessary andsufficient causes of addiction, though these conditionsqualify as meaningful risk factors.Risk Factors and Social SanctionsYouthful gamblers represented among the generalpopulation and college students have whatVaillant once quipped was a “time-limited disorder.”This “disorder”—best thought of as arisk factor for gambling-related problems and avariety of other social problems—is adolescence.The prevalence rates of disordered gamblingreflected among the treatment and prisonpopulations also distinguish groups with higherrates of psychiatric disorder than observedamong the general public. Like adolescence,psychiatric disorders represent a cluster of riskfactors for disordered gambling. In addition,adolescents, treatment, and prison groups areamong the least likely to avoid gambling simplybecause it is illicit, unavailable, or socially proscribed.These groups engage in more sociallydeviant or anti-social behavior—including gambling—thanthe general adult population. Thesegroups will tend to gamble regardless of theavailability of legalized gambling. Consequently,for these groups, the prevalence rates ofdisordered gambling will be influenced lessmeaningfully by the growth of legally availablegambling. In fact, it is possible that as gamblingbecomes even more mainstream—if indeed itdoes—then these more deviant groups mightbegin to move away from any form of licit gamblingbecause it will not offer the opportunityfor the expression of anti-social motives. 52 If weexamine only these atypical groups, absent anyinfluence from the social setting (e.g., a historicalperspective), we might conclude that higherrates of gambling experience caused their psychologicalproblems. However, this conclusionwould be faulty: these groups, traditionally anddevelopmentally associated with anti-social motivesand behavior patterns, are the least likelyto be influenced by the legal availability ofgambling (Shaffer & Zinberg, 1985; Zinberg,1984).Alternatively, the general adult population, havingrelatively low rates of psychiatric comorbidityand having already matured beyondtheir anti-social, independence-seeking stage ofdevelopment, is the most likely population segmentto be influenced by the availability of sociallyapproved gambling opportunities. Thisgroup is the least likely to gamble illicitly, andthe most likely to begin gambling only whengaming is perceived as mainstream preciselybecause of the absence of risk factors. Likewearing bell-bottoms or having pierced ears—activities of different eras that once were limitedonly to small counterculture groups with antisocialmotives and later adopted by the largersociety—gambling, which once was limited togroups with anti-social behavior patterns,gradually is becoming a mainstream activity inthe United States and Canada. Consequently,more people from the cultural mainstream willcontinue to be exposed to gambling experiencesas gaming becomes more accessible. The firstpeople to try a new social activity will be thosewho are more deviant than their successors. Wewill describe this phenomenon in more detail inthe next section of this discussion.Interestingly, if gambling falls out of social favor,which already has happened during at leasttwo other historical eras in the United States(Rose, 1986; 1995), then the gamblers who remainactive likely will have attributes similar totheir deviant counterparts who first pioneered52 This observation should not be taken as implicitsupport for the “legalization” of underage gamblingsince there are many other social issues that bear onthis question.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling76each of the preceding social movements thatbrought gambling into favor. Those who aremost pro-social will be the first to stop gamblingwhen the activity falls out of social favor. Thispattern is similar to the current observation—now that cigarette smoking is unpopular inAmerica—that smoking will become a markerfor psychiatric disturbance. This observationasserts that only those with psychiatric disturbancewill continue to smoke in the face ofmounting social pressures not to smoke. Theattitude that a culture has toward gambling willdetermine which segments of the population arewilling to “gamble” with gambling. The moreunfavorably society views gambling, the moredeviant groups will be represented disproportionatelyamong those who participate in theactivity.This theory raises an interesting question: willthe higher-than-adult prevalence rates of disorderedgambling reported by young people continueto be identified years from now when thiscohort evolves into the adult respondents ingeneral population studies, or will they convenientlyexperience memory distortion (Schacter,1995) that will lower the rate of lifetime disorderedgambling prevalence?The Natural History of Illicit Activities:The Shift from Socially Sanctionedto Socially Supported ActivitiesObservers of gambling history often quip thatthe church and the state once were the two mostardent opponents of gambling; now, these twoinstitutions are the two most active supporters ofgambling. Shaffer and Zinberg (1985) and Zinbergand Shaffer (1985) described the naturalhistory of an illicit activity and its evolution towidespread popularity. When an illicit activity—marijuanause or gambling—is (1) newlyintroduced into a social setting, (2) modified bystatute so that it shifts in status from illicit tolicit, or (3) modified by the folkways or moresof society so that it is more commonly experienced,the previously illicit activity is still perceivedas deviant by the larger culture. Generally,those who seek out new experiences withpreviously illicit or currently illicit activitieshave strong motives for doing so, and are thereforeconsidered by the larger social group to bemisfits or psychologically disturbed. First experiencesof the new activity usually areaccompanied by high anxiety becauseparticipants fear society’s disapproval, and areaccompanied by little knowledge of theactivity’s effects. Gradually, as the new ordeviant behavior becomes more prevalent andpopular—as marijuana use did during the mid-1960s and gambling did during the 1980s and1990s—knowledge about the activity and itseffects increases. Slowly, as experience withgambling grows, gamblers correctmisconceptions. New misunderstandings maydevelop, however: for example, the idea that allexcessive gamblers are very intelligent orhyperactive. In the midst of the inevitablecontroversy and questioning that developsbetween the first-generation gamblers and the“mainstream” culture, a second generation ofgamblers appears. Instead of attempting tobreak with the larger society as did the firstgenerationgambler, this second-generationgambler, motivated by curiosity or an objectiveinterest in the effect of gambling, stimulates amore comprehensive cycle of information acquisitionabout gambling and its consequences.As a result of this social evolution, gambling hasbecome a relatively average and expectable activityamong adolescents (e.g., Shaffer, 1993;Zinberg, 1984). As a cohort, young people nowparticipate in gambling just as they consistentlyhave used tobacco and alcohol in spite of theillicit status of this group of activities (Shaffer,1993). When this second generation of gamblerssupports the experiences of the first generation,gamblers are more likely to be heard bythe larger culture. These new gamblers aregreater in numbers, more diverse in background,and their motives less antagonistic than those ofthe first generation; therefore, this social groupis more acceptable to the mainstream. This secondgeneration of gamblers has the opportunityto explode many of the stereotypical myths thathave existed about gamblers and gambling.As the second generation of gamblers matures,the larger society tends to move away from itsformerly rigid position toward gambling. As aresult, society often becomes more confusedHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling77about its attitudes. This confusion tends to motivatenon-gamblers and communities or stateswithout gambling—those not energized by socialrebellion—to experiment with gambling.The experience of this group has more influenceon the larger social setting than either of the firsttwo groups of gamblers described earlier. Finally,as increasingly diverse groups of peopleparticipate in gambling and represent activegamblers, it becomes more probable that variousgambling styles will work better and cause lessdifficulty. Nevertheless, Rose (1986; 1995) remindsus that the growth of gambling can recedewith short notice when corruption and scandaloccur, as was the case in each of the previoustwo waves of gambling expansion in the UnitedStates.In sum, as each generation of gamblers acquiresmore knowledge about the activity and as thisinformation is disseminated, there is less likelihoodthat users who have disturbed personalitieswill predominate. A very few years from now,large numbers of people who were born after themodern advent of widespread legalized gamblingwill have children entering adolescence;the roles these parents may play in socializingtheir children about gambling may be very differentfrom the roles played by their own parentsin socializing their gambling behaviors.Toward the Future: Implications forNew Prevalence ResearchApplications of Prevalence EstimatesThe vast majority of studies reviewed inthis meta-analysis failed to describe howprincipal investigators, sponsors, healthproviders, or policy makers intended touse the newly constructed estimates of disorderedgambling prevalence once these appraisalsbecame available. Most studies simply generatedprevalence estimates because peoplewanted to know. Without a specific, welldefinedproject purpose—in advance of developinga study protocol—prevalence estimatesprovide little more than a sumptuary rejoinder toan active curiosity. While prevalence rates provideinteresting information for inquiring minds,there are more important applications availablefor this information. For example, will theprevalence data be used to plan for prevention,treatment, or law enforcement activities?This section of the discussion will briefly describesome of the important ways that investigatorscan use prevalence estimates to guide theallocation of limited resources. In each instance,however, the objectives of prevalenceresearch will require that investigators adjusttheir study methods and instruments to providethe specific information necessary and relevantto their designated purpose. Absent specificproject objectives, prevalence researchers willbe left to study aimlessly, unable to produce theessential findings, or the level of precision intheir findings, that policy makers require if theyare going to decide judiciously how to allocatelimited financial, human, and other resources.In the absence of explicit goals, investigatorshave not implemented the rigorous methodsnecessary to generate the representative samples,ample sample sizes, adequate numbers ofdisordered gamblers, or a set of overall findingsthat permit health policy makers to buildprevention and treatment systems for disorderedgamblers by allocating resources in a definitemanner. For example, after conducting a statewideand state-funded study to identify the impactof legalized gambling in Connecticut, theWEFA Group et al. (1997) concluded that thesample of pathological gamblers identified bytheir telephone survey was inadequate to allowthem to address issues related to (1) the socialcosts of gambling or (2) any of the importantpsychological, social or treatment characteristicsof intemperate gamblers in Connecticut. Thisdeficiency is remarkable, since the project’sprimary purpose was to complete “A study concerningthe effects of legalized gambling on thecitizens of the state of Connecticut.”Health policy makers need to know not onlyhow many disordered gamblers there are in agiven jurisdiction, but also what the obstacles totreatment may be for this group, how many peoplewould use treatment if it were available, andwhat level of treatment (e.g., inpatient or outpa-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling78tient) people in their catchment area might require.Social ObjectivesThere are a number of social objectives thatprevalence research can achieve. For example,estimates of disordered gambling prevalence canhelp monitor socio-cultural trends. These trendsassist policy makers, city planners, and others toforecast the future behavior of broad segmentsof the population. An understanding of thesetrends is essential to building an adequate systemof public health services. To prepare formoving people to and from leisure time activities,transportation, recreation, and other servicesthat have the capacity to assure customercomforts also can make use of this data.The identification of risk factors for disorderedgambling is another benefit of properly conductedprevalence studies. Identified risk factorswill allow social scientists to develop programsthat minimize the risks for populationswith specific characteristics (e.g., young males,psychiatric patients with major depression).Knowledge of population attributable risks hasimportant public health implications. For example,a measure of population attributable riskfor “maleness” addresses the question, “Out ofall risk factors, how much of disordered gamblingcan be attributed to maleness?” The importanceof this measure is twofold: First, it allowsadministrators and public health programmanagers to prioritize the allocation of limitedresources by identifying the more importantcauses from the set of known possible causes;further, it provides insight about where investigatorsshould look for additional informationabout risk factors that exert meaningful influenceon disordered gambling. Second, an attributablerisk assumes more importance if, inaddition to a specific risk factor being attributablefor a meaningful percentage of the problem,it identifies a risk factor that is mutable. Underthese conditions, public policy and public healthinterventions would be of benefit to those mostharmed by the risk factor. Technically, underthese circumstances, we can interpret populationattributable risks to represent the percentage ofthe gambling disorder that could be reduced ifthe risk factor were eliminated. For example, ifgambling after 1 A.M. was identified as havinga total attributable risk of 25% for a level 3gambling disorder, then we will be able to reducelevel 3 disordered gambling by 25% if apublic policy effectively limited late night andearly morning gambling. Similarly, if drinkingalcohol while gambling was identified as havingan attributable risk of 40% for level 3 gambling,then we will be able to reduce level 3 gamblingby restricting the use of beverage alcohol whilegambling.Economic ApplicationsEconomically, prevalence data can be used asone essential component of a mathematical algorithmto determine the social costs or socialbenefits of gambling. Prevalence data can helpto determine the nature of illicit gambling markets,and the impact of these markets on lawabidingindustries. Estimates of disorderedgambling behaviors can help the banking industryto better understand how people save, invest,and borrow money, goods, and services.Psychological PurposesFinally, from a psychological perspective, estimatesof disordered gambling prevalence help toprovide important insight into the relationshipsbetween risk-taking and risk-avoiding, immediategratification and delayed gratification,treatment needs and treatment seeking, naturalrecovery and treatment-based recovery, hopeand hopelessness, mindfulness and mindlessness,and euphoria and dysphoria. Epidemiologicalstudies of disordered gambling also providethe opportunity to study how depressionemerges, develops, and passes.Prevalence and Treatment PlanningWhile many prevalence investigators tout buildinga treatment system as the ultimate utilizationof disordered gambling rates, none have deter-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling79mined what are the obstacles to using treatment,what is the likelihood of using treatment, orwhat is the history of treatment utilizationacross various segments of the provider networkfor gamblers with various levels of disorder.Prevalence research in the field of gamblingrarely has contributed direct guidance to establishinga treatment system.The relationships among social, economic, andclinical forces can influence how individual cliniciansdiagnose and prescribe a course oftreatment for patients struggling with addictivedisorders. The confluence of contemporary socialparadigms (e.g., Schlesinger, Dorwart, &Clark, 1991; Shaffer, 1986a; Shaffer & Robbins,1991) can influence both individual clinicians(Shaffer, 1986b) and social policy makers(Schlesinger & Dorwart, 1992); as a result, bothtreatment prescriptions and treatment resourceallocations are sometimes made on the basis ofsocial perception instead of empirically determinedneed. For example, Schlesinger et al.(1991) examined the availability of treatmentresources in United States cities; they noted thatclinical resources were not always proportionatelyrelated to need. The availability of clinicalservices can reflect media and political emphasisrather than current substance abuse trends.Instead of examining what clinicians, politicians,and social policy makers speculate or assumethat people with excessive behavior patternsneed, it is imperative to examine the bestavailable evidence. Only on a few rare occasionsdoes a pound of anecdote yield an ounceof proof. Since gambling prevalence studiesoften reflect more emotional heat than lightwhen it comes to understanding the causal influenceof gambling and the development of disorderedgambling, we must examine the availableevidence carefully and conservatively. Thiscaution will protect us from drawing prematureconclusions about a complex social matter.Currently, we can match only a limited numberof treatment modalities to the array of treatmentneeds. This problem exists because the availableclinical resources are quite limited. Onceresearchers can identify the extent of treatmentneed by conducting empirically derived “needs”assessment, the development of a full range oftreatment resources will become possible andthe clinical matching process can proceed withprecision. To design optimal needs assessmentdata models, program planners should have aclear understanding of the data requirements andneeds assessment purposes. Clinical considerationslargely determine these requisites. Afterdata is obtained from a needs assessment, plannersmust decide how to allocate clinical treatmentservices to meet these identified needs.Unfortunately, some gambling treatment plannershave strong statistical backgrounds, but lessknowledge of the clinical aspects of gamblingand addiction treatment.Toward New Theoretical Maps: GuidingFuture Research & Practice“Categories are the outcomes of historical development,cultural influence, and political negotiation.Psychiatric categories – though mental illnesswill not allow us to make of it whatever welike – are no exception.”- Arthur Kleinman (1988)Similar to the shifting definition of AIDS anddiabetes (Centers for Disease Control and Prevention,1992; Knox, 1997), the concept ofpathological gambling likely will undergo transformationsas scientists refine gambling-relatedtheory, instrumentation, and research findings. 53As science advances our understanding of gamblingdisorders, a “gold” standard likely willemerge. This development will permit cliniciansto improve the diagnostic sensitivity andspecificity of screening instruments; in turn, thisadvance will make available improved opportunitiesfor more effective treatment matching.Similarly, researchers should become better ableto distinguish people most in need of clinicalservices from those who do not require intervention.In addition, screening instruments for adolescents,psychiatric patients, inmates, and vari-53 Similar conceptual shifts and compromises havebeen observed with regard to definitions of hypertensionand elevated cholesterol levels.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling80ous ethnic groups will improve as scientists developmore focused techniques for estimatingthe impact and consequences of gambling onthese population segments. To accomplish theseobjectives, we must understand the relationshipbetween theory and research.Theory guides research. Prevalence research isno exception to this rule. Research without theoryis like a person who has a very large checkin their possession—a check capable of shiftinghow the owner experiences life on a daily basis—butno banking mechanism available todeposit or cash it. If this person with latentwealth cannot transform the check into cash, thecheck’s financial potential will have little practicalconsequence. To advance scientific understandingand public welfare, new gamblingrelatedresearch initiatives will require theoreticalmaps. In addition, scientific investigators,working as explorers, will need to identifywhere to apply their theoretical maps; this terrainwill become the new territory of futureprevalence research.Nomenclature & Classification: AConceptual Map for Future PrevalenceResearch“Just as the largest library, badly arranged, isnot so useful as a very moderate one that is wellarranged, so the greatest amount of knowledge,if not elaborated by our own thoughts, is worthmuch less than a far smaller volume that hasbeen abundantly and repeatedly thought over.”- Arthur Schopenhauer (1893)Meta-analytic research requires a conceptualarchitecture that permits the synthesis of differentand sometimes quite disparate study results.Without trans-theoretical concepts to provide ahigher-level system than individual researchfindings provide, it would be virtually impossibleto integrate diverse data. The level system(Shaffer & Hall, 1996) provided the necessaryintellectual bridgework to complete this integrativeresearch without adding any stigmatizinglanguage. While there is little magic in the continuumof levels suggested by Shaffer and Hall(1996), this scheme emerged as a central cog inthis meta-analysis. Consequently, we suggestthat researchers who promulgate estimates ofdisordered gambling prevalence begin to reporttheir findings using the level system. By employingthe level system, investigators do nothave to abandon their theoretical preferences. Itstill remains possible to refer to disorderedgambling as pathological, probable pathological,compulsive, or excessive gambling. However,by recognizing that prevalence estimates disseminatedby one study inevitably will be comparedwith rates from other studies, investigatorscan provide the common currency to relatethese rates with more precision than is availablenow. Rather than leave the level assignmentand interpretation of their results to others, wesuggest that investigators report prevalence rates(1) using nomenclature of their choosing as wellas (2) a specific level estimate of their results.Estimates of level 0 (i.e., people who have nevergambled) and level 1 (i.e., people who havegambled but do not experience any adversesymptoms associated with their gambling)prevalence allow social policy makers the opportunityto employ primary prevention programsthat delay or prevent the onset of activitiesthat can lead to gambling and its potentiallyharmful consequences. Absent estimates oflevel 0 and level 1 gambling, it is not possible toevaluate the effectiveness of prevention programsdesigned to interrupt the initiation ofgambling.Several of the studies reviewed in this reportincluded categories resembling level 1 gambling.However, the majority of these studies(e.g., Volberg, 1993a; Winters et al., 1993b)integrated respondents who had never gambled(level 0) with non-problem gamblers (level 1) inthis category; by failing to distinguish these twogroups, these findings cannot be employed toimprove our understanding of gambling incidenceand primary prevention efforts designedto delay or stop the onset of gambling. Furthermore,by identifying level 1 prevalencerates, future studies will allow public policymakers and clinical service providers the opportunityto apply and evaluate secondary preventionprograms that hold the promise of minimiz-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling81ing the likelihood of gamblers developing problemsrelated to gambling.The level system also encourages scientists todevelop new instruments that can distinguishbetween sub-clinical gamblers (i.e., level 2) whoare moving toward pathological gambling states(i.e., level 3) and those who are moving awayfrom level 3 difficulties. By making these sophisticateddistinctions among gamblers andtheir gambling disorders, scientists permit healthcare planners to begin developing clinical servicesthat employ a treatment matching strategy.These distinctions among gamblers who experiencedifferent levels of adversity in their relationshipto gambling holds the potential to havebroad impact on the development of new socialpolicies and estimates of the clinical servicesnecessary to treat the next generation of pathologicalgamblers. For example, armed with theability to distinguish patients who are progressingtoward more disordered levels of gamblingfrom those who are moving in the direction ofless disordered levels, clinicians can begin toidentify those who require early treatment programsto arrest the progression of gambling relatedproblems and those who require relapseprevention programs to facilitate and sustainrecovery. Future research must begin to identifyand attend to the needs of level 2 gamblers.Similarly, by improving their ability to recognizelevel 3 gambling, policy makers and treatmentproviders can plan a full range of clinicalprograms, as well as tertiary prevention programsthat can minimize the harm alreadycaused gambling activities.The final category of this scheme, level 4 gambling,is actually a subset of level 3 gambling.This group includes respondents who meet level3 criteria and also seek treatment for their gambling-relatedproblems. We could not measurethe extent of level 4 gamblers in this study sincethere is a paucity of this type of prevalence data.Estimates of level 4 prevalence will be moreuseful to social policy makers and clinical serviceproviders than level 3 estimates alone,since level 4 rates provide an estimate of “caseness”(i.e., how many people with the observedsymptoms and signs might actually enter andbenefit from treatment; Vaillant & Schnurr,1988).The recognition of level 2 gamblers raises anotherimportant question: the estimates of gamblingdisorders reported in this article and others(e.g., Wallisch, 1993; Lesieur et al., 1991; Jacobs,1989) suggest that the rate of youth gamblingproblems exceeds that of adults. Duringthe next decades, either these young people willcontinue to evidence higher lifetime rates ofdisordered gambling than their adult counterparts,or the measurement of lifetime gamblingprevalence contains flaws and fails to reflectaccurately the level of gambling that adults experiencedduring their youth. If young peoplelower their past-year rate of disordered gamblingwithout treatment as they age, this mayprovide another example of natural recoveryfrom addictive behavior (e.g., Shaffer & Jones,1989). Existing studies have not determinedwhether young people have used self-helpgroups, clergy, school counselors, friends, ormany other resources to help them withdrawfrom gambling activities.Some prevalence researchers have stated thatthe difference between lifetime and currentprevalence rates represent individuals who canbe regarded as disordered gamblers in naturalrecovery (e.g., Volberg, 1995b) or “spontaneousremission” 54 (e.g., Wynne Resources Ltd., 1994,p. 51). This conjecture rests on the observationthat there are few treatment services for disorderedgamblers in the United States. However,Wallisch (1993) offers several explanations forhigher lifetime than current prevalence rates ofdisordered gambling; these explanations applyequally to research that identifies a lower rate oflifetime gambling or gambling disorders duringa follow-up or replication study. Groups that54 Shaffer & Jones (1989) demonstrated that naturalrecovery is anything but spontaneous. People whorecover from addictive disorders naturally actuallymove through a series of well-defined but overlappingstages. Readers interested in these stage changesshould review Shaffer (1992, 1994, 1997a) for acomprehensive and detailed analysis of these events.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling82evidence this curious pattern actually may (1)still be disordered gamblers who just do not reporta current problem, (2) be recovering gamblersor gamblers in remission who have undergonetreatment and no longer think of themselvesas they had previously, or (3) truly bedisordered gamblers who really have quit or cutback on their own.Scientists need to determine whether the differencesbetween adolescent and adult prevalencerates result from comparisons of meaningfullydifferent groups of cohorts. There remains stillanother complex possibility: if adults are moredefensive about gambling activities than adolescentsand, consequently, disguise or minimizetheir self-reported levels of gambling and itsconsequences, then prevalence estimates ofpathological gambling among adolescents andadults are more similar than current evidencereveals. Given the array of alternative explanations,suggestions about natural recovery rates inthe current literature must be considered highlyspeculative. Shaffer and Jones (1989) were thefirst investigators to document the process ofnatural recovery from cocaine addiction. Theirresearch reveals some of the major difficultiesassociated with making a precise determinationof recovery without treatment. New, carefullycrafted research must address these questions ofnatural recovery from gambling disorders withthe same level of rigor as investigators have exertedin other areas of addiction studies (e.g.,Cunningham, Sobell, Sobell, & Kapur, 1995;Sobell, Cunningham, & Sobell, 1996). Only bydeveloping precise estimates of natural recoverycan economists accurately estimate the socialcosts of disordered gambling.Understanding the behavior of level 2 gamblersalso holds considerable potential to lower thesocial costs associated with gambling disorders.“The common risk factors for many diseases arepresent in a large proportion of the population,and therefore, most of the cases of disease arisefrom the intermediate- and low-risk groups.Relatively small changes in risk among the middle-riskgroup can result in a greater overall reductionin disease burden than do greaterchanges in the high-risk group” (Brownson,Newschaffer, & Ali-Abarghoui, 1997, p. 736).To illustrate, although level 3 gamblers representpeople with a more intense and potentiallydestructive relationship with disordered gamblingthan their level 2 counterparts, this groupis considerably smaller in numbers. In spite ofthe more moderate nature of the problems experiencedby level 2 gamblers, their larger numbersare likely responsible for producing moreadverse impact on American and Canadian societythan their more disordered level 3 counterparts.This circumstance is very similar to theobservation that problem drinkers are responsiblefor more aggregate social problems thantheir alcohol dependent counterparts (e.g., Sobell& Sobell, 1993). As was the case with researchon problem drinkers, future gamblingresearch likely will reveal that level 2 gamblersare more responsive to treatment and social policyinterventions than level 3 gamblers.Finally, future research must continue to examinethe prevalence of gambling disorders amongyoung people to determine if the prevalence ofgambling problems increases as gambling opportunitiesbecome more readily available andmore socially sanctioned. Jacobs (1989) speculatedthat the prevalence of these problems willincrease; however, it is possible that while gamblingproliferates, the prevalence rate of disorderedgambling will remain constant or actuallybegin to diminish after people have sufficientexperience with gaming activities to beginadapting to the experience and protecting themselvesfrom the potential adversities that accrueto some gamblers. This social learning processis similar to the experience of hallucinogen usersduring the 1970s (Bunce, 1979). Unfortunately,both of these rather positive social possibilitiesinclude the likelihood that the rate oflevel 2 gambling will continue to increase untilinformal and formal social controls emerge tohelp young people better regulate their gamblingbehavior (e.g., Zinberg, 1984). Even if adverseconsequences from gambling do diminish, overtime, inter-generational forgetting occurs, andnew generations of young people will tend todevelop problems with self-destructive activitiesthat previously had been better controlled. Forexample, after many years of declining rates,Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling83and after significant education and preventionefforts, drug abuse rates among young peopleare beginning to rise again (e.g., Johnston,O'Malley, & Bachman, 1996). Just as we mighthave expected from the previous discussion,scientists observed this increase in substanceuse and abuse about a generation ago.Collinearity and Its Implications forFuture Prevalence ResearchThe existing set of disordered gambling prevalenceestimates revealed a consistent pattern ofcollinearity. The largest and most methodologicallysound studies evidenced the lowest overallrates of disordered gambling among broad populationsegments. The smallest and often mostmethodologically weak studies focused on narrowpopulation groups and yielded the highestestimates of disordered gambling prevalence.This pattern of multiple correlations can confuseobservers of the gambling prevalence literature,public health workers and social policy makersalike. For example, the common impression thatgambling disorders have increased proportionallyamong the young and often disenfranchisedsocial groups (e.g., those suffering with majorpsychiatric illness) is revealed to be a functionof the pattern of new emerging prevalence researchduring the 1990s, and not a characteristicof disordered gambling trends. Similarly, if weexamine only these atypical population groups,failing to recognize their unique psychologicalattributes, and ignoring the important influencesfrom the social setting (e.g., a historical perspective),we might conclude that higher rates ofgambling experience caused their psychologicalproblems. The confusion that collinearity cancause should serve as a clarion call to scientists.To better understand the nature of disorderedgambling prevalence, investigators must beginto conduct larger scale studies of special populations,and smaller but prospective studies of theadult and youth segments of the general population.Scientists will need to improve their samplingmethods and evaluate how these methodscan influence and bias estimates of prevalence.Policy makers likely will balk at this suggestion,claiming that the proportionally small segmentsof the total population with the highest rates ofdisordered gambling contribute relatively littleto the overall population prevalence rates.While this may be true mathematically, it missesthe purpose of having prevalence rates. Havinga purpose is critical to utilizing prevalence estimateseffectively. Policy makers need to understandthe rates and characteristics of small segmentsof the population who will use resourcesdisproportionately. To plan an effective healthcare system, planners will need to know whichgamblers will enter treatment and which willnot. They also will need to know the obstaclesthat exist to treatment entry for each of theseoften ethnically diverse groups. To facilitate theimpact of these services, planners must learnhow to engage level 2 and level 3 members ofthese groups into treatment, and then how tokeep them there long enough for the therapeuticexperience to be effective.To assist investigators as they implement prevalenceresearch, we recommend a minimum setof standards to guide future research. The followingsection will describe these suggestions,and Appendix 3 includes a brief guide to conductingprevalence research.Recommended Standards for FuturePrevalence ResearchTo accomplish the import tasks and objectivesof estimating the prevalence rates ofgambling and its associated disorders,future investigators must develop improvedstandards for research. To expedite thisprocess, we suggest that scientists observe atleast the following eleven basic standards toprepare and conduct investigations of disorderedgambling prevalence.1. Establish a precise statement of purposeand objectives;2. Select, modify, or create an instrumentthat can satisfy the study objectives;Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling843. Determine a time frame for the gamblingdisorder as well as its associatedcluster of signs and symptoms;4. Determine how best to eliminate alternativepsychiatric disorders;5. Determine a sampling design to selectrespondents;6. Use a power analysis to plan for adequaterepresentation of the population(s)of interest;7. Properly calculate response and completionrates;8. Provide a protocol for selecting andtraining interviewers;9. Supervise the collection of data;10. Check the integrity of data coding anddata entry;11. Apply the results.A full review of these issues is beyond the scopeof this report. However, in Appendix 3 we providea brief guide to these principles as well asan examination of these methodological principles.Caveats & LimitationsSimilar to Miller et al. (1995) in theirgroundbreaking meta-analysis of alcoholtreatment methods, we wish to emphasizethat this analysis of the prevalence ofgambling-related problems should be regarded a“first approximation” to summarizing the literaturewhile taking into account the methodologicalquality of studies. Some would differ withthe logic of our prevalence coding system. Otherswould disagree by adding or subtractingitems or algorithms to our methodological ratingsystem. Still others would quarrel with our dataweighting strategies or multi-method approachto drawing conclusions about the point values orconfidence intervals of our prevalence estimates.Miller et al.’s (1995) caveat applies tothis project as well: despite our multi-step“…review process to minimize errors, it is likelythat in any project of this size there are overlookeddetails, and surely judgment calls forspecific studies on which reasonable colleagueswould disagree” (p. 31).In addition to the strategic caveats describedabove, there are some specific sampling limitationsthat require consideration. With the exceptionof some treatment studies and some inschooladolescent studies, the vast majority ofthe study samples included in this meta-analysiswere unable to survey the entire population ofinterest. Consequently, prevalence studies employsampling strategies; in turn, samplingstrategies can introduce bias to research findings(e.g., as a result of a weak sampling strategy oran incomplete implementation of a sound strategy)(Walker & Dickerson, 1996). Walker andDickerson note that sources of sampling biascan include (1) excluding particular groups fromthe sample, (2) under-sampling specific ethnicor cultural groups, and (3) under-representingpathological gamblers among the selected sample.Although we examined many other potentialsources of bias that could compromise theinternal validity of research included in thisstudy (e.g., response rates, coverage issues, dataintegrity, sample size), this meta-analysis wasunable to explore every potential source of researchbias (e.g., whether the study samplesadequately represented the racial or ethnic distributionof the populations from which thesamples were taken).This meta-analysis was limited by the breadthand depth of the prevalence studies that met thecriteria for inclusion in this study. Collinearityof the data is also a result of the nature of theseincluded studies. It is striking that investigatorsconducted the large studies in this meta-analysisonly among general populations, and the treatmentpopulation studies were all small. Thismay be a reflection of the limited sources offunding for gambling research; it also may indicatethat researchers often are attracted to studytopics and employ methods that can be accomplishedwith the most convenience. It wouldbenefit the field of gambling research to examinethe gaps in knowledge that have resultedHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling85from these research patterns, and begin to addressthese gaps. For example, a large-scalestudy of psychiatric patients would have sufficientpower to inform clinicians, researchers,and policy makers regarding the health careneeds and treatment needs of pathological gamblerswho are among the psychiatric patientpopulation.Meta-Analyses of Prevalence Estimates:Advancing the FieldThis meta-analysis has both similarities toand differences from the growing body ofmeta-analyses currently being publishedin the scientific literature. All metaanalyses,including this one on the prevalence ofdisordered gambling in the United States andCanada, attempt to synthesize and integrate aselect body of scientific evidence into a morestreamlined, quantitative summary. This reporthas produced synthesized estimates of disorderedgambling prevalence across segments ofthe population. However, like all meta-analysts,we encountered conceptual and procedural elementsthat made this integration complex.These elements included outliers, missing information,implicit assumptions in the includedstudies, and conceptual and methodological inconsistenciesacross the included studies. Inaddition, the collinear nature of the data complicatedthe exploration of specific influences (e.g.,region, researcher) on the prevalence estimates.Throughout the scientific literature, the pool ofprevalence-related meta-analyses is still quitelimited. New research likely will contributeinnovative methodologies that can advance ourunderstanding of prevalence and its potentiallyshifting patterns.ConclusionsOne of the benefits of a meta-analysis isthe ability to answer and pose newquestions that may not be raised in eachof the included studies (Colditz, Berkey,& Mosteller, 1997). For example, there is indirectevidence suggesting that, like legalizationof psychoactive substances in other countries(e.g., MacCoun & Reuter, 1997), increases ingambling may be due more to advertising thansimply legalization (WEFA Group et al., 1997).Future analyses of prevalence estimates acrossjurisdictions that are matched for differentialexposure to advertising may provide additionalinsight into this situation. The implications ofthis research for future social policy is considerable.Regardless of the scope of future studies thatawait inevitable scientific inquiry, this metaanalysisprovides empirical evidence that disorderedgambling is a relatively reliable and robustphenomenon, capable of resisting a rangeof methodological variation. Given the stabilityof prevalence estimates across a wide array ofstudies, some final conclusions are in order.Adolescents and college students have higherrates of level 3 gambling than adults in the generalpopulation. Males have higher rates oflevel 3 gambling than females.Keeping the caveats that were described earlierin this discussion in mind, some additional conclusionscan be drawn. Both gambling andgambling studies have proliferated during thepast 20 years: as gaming in the United Statesand Canada expanded rapidly during the 1980sand 1990s, so did the scientific investigation ofgambling disorders. To date, half of the gamblingprevalence studies conducted in the UnitedStates and Canada were completed after 1992.In addition, the newer prevalence studies havefocused more on youth and treatment populations;these groups evidence higher relative riskfor gambling disorders than adults from the generalpopulation. Consequently, casual observersof these recent findings are likely to perceive anincrease in the prevalence of gambling-relateddisorders as a consequence of expanded gaming(e.g., lotteries, keno, casinos, charitable gaming,etc.). Nevertheless, the findings reported earlierfailed to provide evidence that disordered gamblinghas increased among young people fromthe general population, college students, oradults who are receiving some form of treatmentor incarceration. Despite these results, prevalencestudies examining adults from the generalpopulation do reveal an increasing pattern ofHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling86gambling-related problems. For example, theprevalence of level 3 gambling appears to haveincreased over time among adults in the generalpopulation.A number of important factors influence theestimation of disordered gambling prevalence.These elements include individual risk factors orpersonal attributes (e.g., age, gender, psychiatricstatus), and the components of the researchprocess. It is time for epidemiological scienceand public health demands to exert a strongerinfluence on the nature of this methodology andthe future of disordered gambling research studies.Primary Prevalence Studies 55Abbott, D. A. & Cramer, S. L. (1993). Gamblingattitudes and participation: A Midwesternsurvey. Journal of Gambling Studies, 9(3),247-263.*Abrahamson, M. & Wright, J. N. (1977).Gambling in Connecticut. Storrs, CT: ConnecticutState Commission on Special Revenue.*Allen, T. F. (1995). The incidence of adolescentgambling and drug involvement. Providence,RI: Rhode Island College School ofSocial Work.*Angus Reid Group. (1996). Problem gamblingsurvey 1996: Final report. (Submitted to theBritish Columbia Lottery Corporation). Vancouver,British Columbia: Author.*Baseline Market Research Ltd. (1992). Finalreport: Prevalence study: Problem gambling.(Prepared for Department of Finance, Provinceof New Brunswick). New Brunswick:Author.*Baseline Market Research Ltd. (1996a). Finalreport: 1996 prevalence study on problemgambling in Nova Scotia. (Prepared for Nova55 The Primary Prevalence Studies list includes the152 studies identified by our literature search. Ofthese 152 studies, 120 satisfied our inclusion criteriaand were included in this meta-analysis. These 120studies are marked with asterisks.Scotia Department of Health). Halifax, NovaScotia: Author.*Baseline Market Research Ltd. (1996b). Finalreport: Prevalence study: Problem gambling:Wave 2. (Prepared for Department of Finance).Fredericton: New Brunswick Departmentof Finance.*Baseline Market Research Ltd. (1996c). Finalreport: Test-re-test measures: Prevalencesurvey data. (Prepared for Nova Scotia Departmentof Health). Halifax, Nova Scotia:Author.*Bland, R. C., Newman, S. C., Orn, H., & Stebelsky,G. (1993). Epidemiology of pathologicalgambling in Edmonton. CanadianJournal of Psychiatry, 38, 108-112.*Bray, R. M., Kroutil, L. A., Luckey, J. W.,Wheeless, S. C., Iannacchione, V. G., Anderson,D. W., Marsden, M. E., Dunteman, G. H.(1992). 1992 worldwide survey of substanceabuse and health behaviors among militarypersonnel. Research Triangle Park, NC: ResearchTriangle Institute (National TechnicalInformation Service Publication No. ADA-264721; US Department of Commerce).Briggs, J. R., Goodin, B. J., & Nelson, T.(1996). Pathological gamblers and alcoholics:Do they share the same addictions? AddictiveBehaviors, 21(4), 515-519.Browne, B. A. & Brown, D. J. (1994). Predictorsof lottery gambling among American collegestudents. Journal of Social Psychology,134(3), 339-347.Buchta, R. M. (1995). Gambling among adolescents.Clinical Pediatrics, 34, 346-348.*Cardi, N. A. (1996). A study of adolescentgambling patterns October 1995 - May 1996.Providence, RI: Department of Counselingand Educational Psychology, School ofGraduate Studies, Rhode Island College.*Castellani, B., Wootton, E., Rugle, L.,Wedgeworth, R., Prabucki, K., & Olson, R.(1996). Homelessness, negative affect, andcoping among veterans with gambling problemswho misused substances. PsychiatricServices, 47(3), 298-299.*Christiansen/Cummings Associates Inc,Princeton Marketing Associates, Inc., SpectrumAssociates Market Research Inc., &Volberg, R. A. (1992). Legal gambling inHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling87Connecticut: Assessment of current statusand options for the future (Volume one). (Reportto the State of Connecticut Division ofSpecial Revenue). New York:Christiansen/Cummings Associates Inc.*Ciarrocchi, J. W. (1993). Rates of pathologicalgambling in publicly funded outpatient substanceabuse treatment. Journal of GamblingStudies, 9(3), 289-293.*Cook, D. R. (1987). Self-identified addictionsand emotional disturbances in a sample ofcollege students. Psychology of Addictive Behaviors,1(1), 55-61.*Criterion Research Corporation. (1993). Problemgambling study: Final report. (Report tothe Manitoba Lotteries Foundation). Winnipeg,Manitoba: Author.*Criterion Research Corporation. (1995). Problemgambling study: Final report. (Reportprepared for the Manitoba Lotteries Corporation).Winnipeg, Manitoba: Author.*Culleton, R. P. & Lang, M. H. (1985). Theprevalence rate of pathological gambling inthe Delaware Valley in 1984. (Report to PeopleActing To Help, Philadelphia). Camden,NJ: Forum for Policy Research and PublicService, Rutgers University.*Cunningham-Williams, R. M., Cottler, L. B.,Compton, W. M., & Spitznagel, E. L. (inpress). Taking chances: Problem gamblersand mental health disorders - results from theSt. Louis epidemiologic catchment area(ECA) study. American Journal of PublicHealth.*Daghestani, A. N., Elenz, E., & Crayton, J. W.(1996). Pathological gambling in hospitalizedsubstance abusing veterans. Journal of ClinicalPsychiatry, 57(8), 360-363.Derevensky, J. L., Gupta, R., & Cioppa, G. D.(1996). A developmental perspective of gamblingbehavior in children and adolescents.Journal of Gambling Studies, 12(1), 49-66.*Devlin, A. S. & Peppard, D. M. (1996). Casinouse by college students. Psychological Reports,78, 899-906.Dielman, T. E. (1979). Gambling: A social problem?Journal of Social Issues, 35(3), 36-42.*Donwood Problem Gambling Program. (1996).Pilot project: Final Report. Toronto, Ontario:Author.Dragonetti, R. & Faveri, A. (1995). A provincewide needs assessment of service providers:A summary report on compulsive gambling inOntario. Toronto, Ontario: Addiction ResearchFoundation.*Elia, C. & Jacobs, D. F. (1993). The incidenceof pathological gambling among NativeAmericans treated for alcohol dependence.International Journal of the Addictions,28(7), 659-666.*Emerson, M. O. & Laundergan, J. C. (1996).Gambling and problem gambling among adultMinnesotans: Changes 1990 to 1994. Journalof Gambling Studies, 12(3), 291-304.*Emerson, M. O., Laundergan, J. C., & Schaefer,J. M. (1994). Adult survey of Minnesotaproblem gambling behavior; a needs assessment:Changes 1990 to 1994. (Report to theMinnesota Department of Human Services,Mental Health Division). Duluth: Universityof Minnesota Center for Addiction Studies.*Feigelman, W., Kleinman, P. H., Lesieur, H.R., Millman, R. B., & Lesser, M. L. (1995).Pathological gambling among methadone patients.Drug and Alcohol Dependence, 39, 75-81.*Ferris, J. & Stirpe, T. (1995). Gambling in Ontario:A report from a general populationsurvey on gambling-related problems andopinions. Toronto, Ontario: Addiction ResearchFoundation, Problem and CompulsiveGambling Project.Frank, M. L. (1988, May 19-20). Casino gamblingand college students: Three sequentialyears of data. Paper presented at the ThirdNational Conference on Gambling Behavior,New York, NY.*Frank, M. L. (1990). Underage gambling inAtlantic City casinos. Psychological Reports,67, 907-912.*Frank, M. L. (1993). Underage gambling inNew Jersey. In W. R. Eadington & J. A. Cornelius(Eds.), Gambling behavior and problemgambling (pp. 387-394). Reno, NV: Institutefor the Study of Gambling and CommercialGaming.*Gaboury, A. & Ladouceur, R. (1993). Evaluationof a prevention program for pathologicalgambling among adolescents. Journal ofPrimary Prevention, 14(1), 21-28.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling88*Gambino, B., Fitzgerald, R., Shaffer, H., Renner,J., & Courtnage, P. (1993). Perceivedfamily history of problem gambling andscores on SOGS. Journal of Gambling Studies,9(2), 169-184.*Gambino, B., Shaffer, H. J., & Cummings, T.N. (1992). Compulsive gambling: An overlookedproblem. EAP Digest, 13(1), 32-35,46-47.*Gemini Research & Angus Reid Group.(1994). Social gaming and problem gamblingin British Columbia (Report to the BritishColumbia Lottery Corporation). RoaringSpring, PA: Gemini Research.*Gfellner, B. M. (1994). A profile of VLT gamblersin Brandon, Manitoba. Brandon, Manitoba:Brandon Crime Prevention Committee.*Govoni, R. & Frisch, G. R. (1996). The impactof the Windsor casino on adult gambling inthe city of Windsor. Windsor, Ontario: Universityof Windsor Department of Psychology,Problem Gambling Research Group.*Govoni, R., Rupcich, N., & Frisch, G. R.(1996). Gambling behavior of adolescentgamblers. Journal of Gambling Studies,12(3), 305-317.*Griffin-Shelley, E., Sandler, K. R., & Lees, C.(1992). Multiple addictions among dually diagnosedadolescents. Journal of AdolescentChemical Dependency, 2(2), 35-44.Hraba, J. (1989). Summary report to Iowa Departmentof Human Services: Research onthe Iowa Lottery and gambling. Ames: IowaState University, Department of Sociology.Hraba, J. & Lee, G. (1996). Gender, gamblingand problem gambling. Journal of GamblingStudies, 12(1), 83-101.Hraba, J., Mok, W., & Huff, D. (1990). Lotteryplay and problem gambling. Journal of GamblingStudies, 6(4), 355-377.*Insight Canada Research. (1993). Prevalenceof problem & pathological gambling in Ontariousing the South Oaks gambling screen.Toronto, Ontario: Author.*Insight Canada Research. (1994). An explorationof the prevalence of pathological gamblingbehaviour among adolescents in Ontario.Willowdale, Ontario: Canadian Foundationon Compulsive Gambling.Jacobs, D. F. (1989). Illegal and undocumented:A review of teenage gambling and the plightof children of problem gamblers in America.In H. J. Shaffer, S. A. Stein, B. Gambino, &T. N. Cummings (Eds.), Compulsive gambling:Theory, research, and practice (pp.249-292). Lexington, MA: Lexington Books.*Jacobs, D. F., Marston, A. R., Singer, R. D.,Widaman, K., Little, T., & Veizades, J.(1989). Children of problem gamblers. Journalof Gambling Behavior, 5(4), 261-268.*Kallick, M., Suits, D., Dielman, T., & Hybels,J. (1979). A survey of American gambling attitudesand behavior (Research report series,Survey Research Center, Institute for SocialResearch). Ann Arbor: University of MichiganPress.Kearney, C. A., Roblek, T., Thurman, J., &Turnbough, P. D. (1996). Casino gambling inprivate school and adjudicated youngsters: Asurvey of practices and related variables.Journal of Gambling Studies, 12(3), 319-327.Kennedy, H. G. & Grubin, D. H. (1990). Hotheadedor impulsive? British Journal of Addiction,85, 639-643.Kuley, N. B. & Jacobs, D. F. (1988). The relationshipbetween dissociative-like experiencesand sensation seeking among social andproblem gamblers. Journal of Gambling Behavior,4(3), 197-207.*Ladouceur, R. (1991). Prevalence estimates ofpathological gambling in Quebec. CanadianJournal of Psychiatry, 36, 732-734.*Ladouceur, R. (1993). Prevalence estimates ofpathological gamblers in Quebec, Canada. InW. R. Eadington & J. A. Cornelius (Eds.),Gambling behavior and problem gambling(pp. 379-384). Reno: University of NevadaPress.Ladouceur, R. (1996). The prevalence of pathologicalgambling in Canada. Journal of GamblingStudies, 12(2), 129-142.*Ladouceur, R., Arsenault, C., Dubé, D., Freeston,M. H., & Jacques, C. (1997). Psychologicalcharacteristics of volunteers in studieson gambling. Journal of Gambling Studies,13(1), 69-84.*Ladouceur, R. & Dubé, D. (1995). Prevalenceof pathological gambling and associatedproblems in individuals who visit non-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling89gambling video arcades. Journal of GamblingStudies, 11(4), 361-365.Ladouceur, R., Dubé, D., & Bujold, A. (1994a).Gambling among primary school students.Journal of Gambling Studies, 10(4), 363-370.*Ladouceur, R., Dubé, D., & Bujold, A.(1994b). Prevalence of pathological gamblingand related problems among college studentsin the Quebec metropolitan area. CanadianJournal of Psychiatry, 39, 289-293.*Ladouceur, R. & Mireault, C. (1988). Gamblingbehaviors among high school studentsin the Quebec area. Journal of Gambling Behavior,4(1), 3-12.*Laundergan, J. C., Schaefer, J. M., Eckhoff, K.F., & Pirie, P. L. (1990). Adult survey ofMinnesota gambling behavior: A benchmark,1990. St. Paul: State of Minnesota Departmentof Human Services, Mental Health Division.*Laventhol & Horwath, David Cwi & Associates,& Survey Research Associates, Inc.(1986). The effects of legalized gambling onthe citizens of the state of Connecticut. Newington:State of Connecticut Division of SpecialRevenue.*Laventhol & Horwath, Guida, F. V., DavidCwi & Associates, & Public Opinion Laboratory.(1990, November). A study of problemand pathological gambling among citizens ofIndiana associated with participation in theIndiana state lottery. Indianapolis, IN:Laventhol & Horwath.*Lesieur, H. R. (1988a). Altering the DSM-IIIcriteria for pathological gambling. Journal ofGambling Behavior, 4(1), 38-47.Lesieur, H. R. (1988b). Report on pathologicalgambling in New Jersey . Trenton, NJ: Governor'sAdvisory Commission on Gambling.Lesieur, H. R. (1989). Experience of employeeassistance programs with pathological gamblers.Journal of Drug Issues, 19(4), 425-436.Lesieur, H. R. (1995). Black and white collegestudents equally bitten by the gambling bug.Journal of Blacks in Higher Education,8(Summer), 83.*Lesieur, H. R. & Blume, S. B. (1987). TheSouth Oaks gambling screen (SOGS): A newinstrument for the identification of pathologicalgamblers. American Journal of Psychiatry,144(9), 1184-1188.*Lesieur, H. R. & Blume, S. B. (1990). Characteristicsof pathological gamblers identifiedamong patients on a psychiatric admissionsservice. Hospital and Community Psychiatry,41(9), 1009-1012.*Lesieur, H. R., Blume, S. B., & Zoppa, R. M.(1986). Alcoholism, drug abuse, and gambling.Alcoholism: Clinical and ExperimentalResearch, 10(1), 33-38.*Lesieur, H. R., Cross, J., Frank, M., Welch, M.,White, C. M., Rubenstein, G., Mosely, K., &Mark, M. (1991). Gambling and pathologicalgambling among university students. AddictiveBehaviors, 16, 517-527.*Lesieur, H. R. & Heineman, M. (1988). Pathologicalgambling among youthful multiplesubstance abusers in a therapeutic community.British Journal of Addiction, 83, 765-771.*Lesieur, H. R. & Klein, R. (1987). Pathologicalgambling among high school students. AddictiveBehaviors, 12, 129-135.Lesieur, H.R. & Rosenthal, R. J. (1991). Pathologicalgambling: A review of the literature(prepared for the American Psychiatric AssociationTask Force on DSM IV Committee onDisorders of Impulse Control Not ElsewhereClassified). Journal of Gambling Studies, 7,5-39.Lesieur, H. R. & Rothschild, J. (1989). Childrenof Gamblers Anonymous members. Journalof Gambling Behavior, 5(4), 269-281.*Lieberman, L., Cuadrado, M., & Fowler, P.(1996). Summary data on Florida high schoolstudent gambling in five counties. AltamonteSprings: Florida Council on CompulsiveGambling, Inc.Lorenz, V. C. (1983). An annotated bibliographyon pathological gambling with a profileof gambling behavior of students from theUniversity of Nevada at Las Vegas andGeorgia State University in Atlanta (Doctoraldissertation, University of Pennsylvania).Ann Arbor, MI: University Microforms International.*Lumley, M. A. & Roby, K. J. (1995).Alexithymia and pathological gambling. Psy-Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling90chotherapy and Psychosomatics, 63, 201-206.*McCormick, R. A. (1993). Disinhibition andnegative affectivity in substance abusers withand without a gambling problem. AddictiveBehaviors, 18, 331-336.*Miller, M. A. & Westermeyer, J. (1996). Gamblingin Minnesota. American Journal ofPsychiatry, 153(6), 845.*Nechi Training and Research & Health PromotionsInstitute. (1994). Spirit of bingoland: Astudy of problem gambling among Albertanative people. Edmonton, Alberta: Author.*Nechi Training, Research & Health PromotionsInstitute. (1995). Firewatch on aboriginaladolescent gambling. Edmonton, Alberta:Author.*New Mexico Department of Health & Universityof New Mexico Center on AlcoholismSubstance Abuse and Addictions. (1996).New Mexico survey of gambling behavior.Santa Fe, NM: Author.*Nova Scotia Department of Health Drug DependencyServices Division & DalhousieUniversity. (1996). Nova Scotia student druguse 1996 . Halifax: Nova Scotia Departmentof Health, Drug Dependency Services.*Omnifacts Research Limited. (1993). An examinationof the prevalence of gambling inNova Scotia (Report #93090). Halifax: NovaScotia Department of Health, Drug DependencyServices.*Oster, S. & Knapp, T. J. (1994, June 2). Casinogambling by underage patrons: Two studiesof a university student population. Paper presentedat the Ninth International Conferenceon Gambling and Risk-Taking, Las Vegas,NV.*Reilly, P. & Guida, F. (1990). Pathologicalgambling prevalence in New Jersey 1990 finalreport (Report to the New Jersey Departmentof Higher Education). Piscataway,NJ: University of Medicine and Dentistry.Rosenstein, J. & Reutter, R. (1980). Gambling:An adolescent activity. Journal of AdolescentHealthcare, 1, 180.*Rounsaville, B. J., Anton, S. F., Carroll, K.,Budde, D., Prusoff, B. A., & Gawin, F.(1991). Psychiatric diagnoses of treatmentseekingcocaine abusers. Archives of GeneralPsychiatry, 48, 43-51.*Shaffer, H. J. & Hall, M. N. (1994). The emergenceof youthful addiction: The prevalenceof pathological gambling among adolescentsat Agawam high school (MassachusettsCouncil on Compulsive Gambling: TechnicalReport Series 031594-HS100). Boston: MassachusettsCouncil on Compulsive Gambling.*Shaffer, H. J., LaBrie, R., Scanlan, K. M., &Cummings, T. N. (1994). Pathological gamblingamong adolescents: Massachusettsgambling screen (MAGS). Journal of GamblingStudies, 10(4), 339-362.*Shepherd, R. M. (1996). Clinical obstacles inadministering the South Oaks GamblingScreen in a methadone and alcohol clinic.Journal of Gambling Studies, 12(1), 21-32.*Sin, R. (1996). Gambling and problem gamblingamong Chinese adults in Quebec: Anexploratory study. Montreal, Quebec: ChineseFamily Service of Greater Montreal.*Smart, R. G. & Ferris, J. (1996). Alcohol,drugs and gambling in the Ontario adultpopulation, 1994. Canadian Journal of Psychiatry,41(1), 36-45.Smith, G. J. (1992). Compulsive gambling:General issues, treatments, and policy considerations(Report prepared for Alberta Lotteriesand Gaming). Edmonton: University ofAlberta.*Sommers, I. (1988). Pathological gambling:Estimating prevalence and group characteristics.International Journal of the Addictions,23(5), 477-490.*Spunt, B., Lesieur, H., Hunt, D., & Cahill, L.(1995). Gambling among methadone patients.International Journal of the Addictions,30(8), 929-962.*Spunt, B., Lesieur, H., Liberty, H. J., & Hunt,D. (1996). Pathological gamblers in methadonetreatment: A comparison between menand women. Journal of Gambling Studies,12(4), 431-449.*Steinberg, M. A. (1988). Unpublished researchreport . Hamden: Connecticut Council onCompulsive Gambling.*Steinberg, M. A. (1997). Connecticut highschool problem gambling surveys 1989 &Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling911996 . Guilford: Connecticut Council onProblem Gambling.*Steinberg, M. A., Kosten, T. A., & Rounsaville,B. J. (1992). Cocaine abuse andpathological gambling. American Journal onAddictions, 1, 121-132.Stinchfield, R., Cassuto, N., Winters, K., &Latimer, W. (1997). Prevalence of gamblingamong Minnesota public school students in1992 and 1995. Journal of Gambling Studies,13(1), 25-48.Taira, R. R. (1993). Legal gambling amongJapanese Americans (Master's thesis, CaliforniaState University, Long Beach). AnnArbor, MI: UMI Dissertation Services.Task Force on Gambling Addiction in Maryland.(1990). Final report: Task force ongambling addiction in Maryland . Baltimore:Maryland Department of Health and MentalHygiene, Alcohol and Drug Abuse Administration.*Templer, D. I., Kaiser, G., & Siscoe, K. (1993).Correlates of pathological gambling propensityin prison inmates. Comprehensive Psychiatry,34, 347-351.*Thompson, W. N., Gazel, R., & Rickman, D.(1996). The social costs of gambling in Wisconsin.Wisconsin Policy Research InstituteReport, 9(6), 1-44.*Transition Planning Associates. (1985). A surveyof pathological gamblers in the state ofOhio. Philadelphia: Author.*Ubell, E. (1991, June 2). Are you a gambler?Parade Magazine, 8.*Vagge, L. M. (1996). The development ofyouth gambling. Unpublished Honors Thesis,Harvard-Radcliffe Colleges, Cambridge, MA.*Volberg, R. A. (1992a). Gambling involvementand problem gambling in Montana. Albany,NY: Gemini Research.Volberg, R. A. (1992b). Treatment of pathologicalgamblers in Montana: Past, present andfuture. Albany, NY: Gemini Research.*Volberg, R. (1993a). Gambling and problemgambling among adolescents in Washingtonstate. Albany, NY: Gemini Research.Volberg, R. A. (1993b). Estimating the prevalenceof pathological gambling in the UnitedStates. In W. R. Eadington & J. A. Cornelius(Eds.), Gambling behavior & problem gambling(pp. 365-378). Reno: University of Nevada,Reno, Institute for the Study of Gamblingand Commercial Gaming, College ofBusiness Administration.*Volberg, R. A. (1993c). Gambling and problemgambling in Washington state (Report tothe Washington state lottery). Albany, NY:Gemini Research.*Volberg, R. A. (1994a). Gambling and problemgambling in Saskatchewan (Report to theMinister's Advisory Committee on Social Impactsof Gaming). Northampton, MA: GeminiResearch.*Volberg, R. A. (1994b). The prevalence anddemographics of pathological gamblers: Implicationsfor public health. American Journalof Public Health, 84, 237-241.*Volberg, R. A. (1995a). Gambling and problemgambling in Iowa: A replication study(Report to the Iowa Department of HumanServices). Roaring Spring, PA: Gemini Research.*Volberg, R. A. (1995b). Wagering and problemwagering in Louisiana (Report to theLouisiana Economic Development & GamingCorporation). Roaring Spring, PA: GeminiResearch.*Volberg, R. A. (1996a). Gambling and problemgambling among Georgia adolescents:Report to the Georgia Department of HumanResources. Roaring Spring, PA: Gemini Research.*Volberg, R. A. (1996b). Gambling and problemgambling in New York: A 10-year replicationsurvey, 1986 to 1996 (Report to theNew York Council on Problem Gambling).Roaring Spring, PA: Gemini Research.*Volberg, R. A. (1997). Gambling and problemgambling in Mississippi: A report to the MississippiCouncil on Compulsive Gambling(Social Research Report Series 97-1). MississippiState: Mississippi State University, SocialScience Research Center.*Volberg, R. A. & Boles, J. (1995). Gamblingand problem gambling in Georgia (Report tothe Georgia Department of Human Resources).Roaring Spring, PA: Gemini Research.*Volberg, R. A. & Glackman, W. (1995). Bingoplay in Vancouver: Report on a telephoneHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling92survey of active bingo players in Vancouver.Roaring Spring, PA: Gemini Research.*Volberg, R. A. & Precision Marketing Inc.(1993). Gambling and problem gamblingamong Native Americans in North Dakota(Report to the North Dakota Department ofHuman Services, Division of Mental Health).Albany, NY: Gemini Research.*Volberg, R. A. & Silver, E. (1993). Gamblingand problem gambling in North Dakota (Reportto the North Dakota Department of HumanServices, Division of Mental Health).Albany, NY: Gemini Research.*Volberg, R. A. & Steadman, H. J. (1988a). Minoritygambling survey. Albany: New YorkState Office of Mental Health, Bureau ofSurvey & Evaluation Research.*Volberg, R. A. & Steadman, H. J. (1988b). Refiningprevalence estimates of pathologicalgambling. American Journal of Psychiatry,145(4), 502-505.*Volberg, R. A. & Steadman, H. J. (1989a).Problem gambling in Iowa. Delmar, NY:Policy Research Associates, Inc.Volberg, R. A. & Steadman, H. J. (1989b).Compulsive gambling study. Delmar, NY:Policy Research Associates, Inc.*Volberg, R. A. & Steadman, H. J. (1989c).Prevalence estimates of pathological gamblingin New Jersey and Maryland. AmericanJournal of Psychiatry, 146(12), 1618-1619.*Volberg, R. A. & Stuefen, R. M. (1994). Gamblingand problem gambling in South Dakota:A follow-up survey. Vermillion: Universityof South Dakota, Business ResearchBureau.*Volberg, R. A., Stuefen, R. M., & Madden, M.K. (1991). Gaming in South Dakota: A studyof gambling participation and problem gamblingand a statistical description and analysisof its socioeconomic impacts. Vermillion:University of South Dakota, Business ResearchBureau.*Wallisch, L. S. (1993a). Gambling in Texas:1992 Texas survey of adolescent gamblingbehavior. Austin: Texas Commission on Alcoholand Drug Abuse.*Wallisch, L. S. (1993b). Gambling in Texas:1992 Texas survey of adult gambling behavior.Austin: Texas Commission on Alcoholand Drug Abuse.*Wallisch, L. S. (1996). Gambling in Texas:1995 surveys of adult and adolescent gamblingbehavior. Austin: Texas Commissionon Alcohol and Drug Abuse.*Walters, G. D. (1997). Problem gambling in afederal prison population: Results from theSouth Oaks gambling screen. Journal ofGambling Studies, 13(1), 7-24.*Westphal, J. R. & Rush, J. (1996). Pathologicalgambling in Louisiana: An epidemiologicalperspective. Journal of the Louisiana StateMedical Society, 148, 353-358.*Westphal, J., Rush, J., & Stevens, L. (1997).Gambling behavior and substance use among“high risk” adolescents in northwest Louisiana.Shreveport: Louisiana State UniversityMedical Center.*Winters, K. C., Bengtson, P. L., Stinchfield, R.D., & Dorr, D. (1996). 1995 college gamblingsurvey: University of Minnesota Twin Cities,University of Minnesota Duluth, and MoorheadState University (Report to the MinnesotaDepartment of Human Services GamblingProgram, Mental Health Division).Minneapolis: University of Minnesota Centerfor Adolescent Substance Abuse.*Winters, K. C., & Stinchfield, R. D. (1993).Gambling behavior among Minnesota youth:Monitoring change from 1990 to 1991/1992.Minneapolis: University of Minnesota Centerfor Adolescent Substance Abuse.*Winters, K. C., Stinchfield, R., & Fulkerson, J.(1990). Adolescent survey of gambling behaviorin Minnesota: A benchmark. Minneapolis:University of Minnesota Center forAdolescent Substance Abuse.*Winters, K. C., Stinchfield, R. D., & Fulkerson,J. (1993a). Toward the development ofan adolescent gambling severity scale. Journalof Gambling Studies, 9(1), 63-84.*Winters, K. C., Stinchfield, R., & Fulkerson, J.(1993b). Patterns and characteristics of adolescentgambling. Journal of Gambling Studies,9(4), 371-386.*Winters, K. C., Stinchfield, R. D., & Kim, L.G. (1995). Monitoring adolescent gamblingin Minnesota. Journal of Gambling Studies,11(2), 165-183.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling93Wittchen, H.-U., Robins, L. N., Cottler, L. B.,Sartorius, N., Burke, J. D., Regier, D., & Participantsin the Multicentre WHO/ADAMHAField Trials. (1991). Cross-cultural feasibility,reliability and sources of variance of thecomposite international diagnostic interview(CIDI). British Journal of Psychiatry, 159,645-653.*Wynne Resources Ltd. (1994). Gambling andproblem gambling in Alberta (Report preparedfor Alberta Lotteries and Gaming).Edmonton, Alberta: Author.*Wynne Resources Ltd. (1996). Adolescentgambling and problem gambling in Alberta.Edmonton: Alberta Alcohol & Drug AbuseCommission.*Zitzow, D. (1996a). Comparative study ofproblematic gambling behaviors betweenAmerican Indian and non-Indian adolescentswithin and near a northern plains reservation.American Indian and Alaskan Native MentalHealth Research, 7(2), 14-26.*Zitzow, D. (1996b). Comparative study ofproblematic gambling behaviors betweenAmerican Indian and non-Indian adults in anorthern plains reservation. American Indianand Alaskan Native Mental Health Research,7(2), 27-41.ReferencesAllen, J. P, Eckardt, M. J., & Wallen, J. (1988).Screening for alcoholism: Techniques and issues.Public Health Reports, 103, 586-592.Ahlbom, A., & Norell, S. (1990). Introductionto modern epidemiology. Stockholm: EpidemiologyResources, Inc.American Psychiatric Association. (1973).DSM-II: Diagnostic and statistical manual ofmental disorders. (2 nd ed.). Washington,D.C.: Author.American Psychiatric Association. (1980).DSM-III: Diagnostic and statistical manualof mental disorders. (3 rd ed.). Washington,D.C.: Author.American Psychiatric Association. (1987).DSM-III-R: Diagnostic and statistical manualof mental disorders. (3 rd ed., revised).Washington, D.C.: Author.American Psychiatric Association. (1994).DSM-IV: Diagnostic and statistical manualof mental disorders. (4 th ed.). Washington,D.C.: Author.Bakan, D. (1967). On method: Toward a reconstructionof psychological investigation. (1 stedition). San Francisco: Jossey-Bass.Bangert-Drowns, R. L., Wells-Parker, E., &Chevillard, I. (1997). Assessing the methodologicalquality of research in narrative reviewsand meta-analyses. In K. J. Bryant, M.Windle, & S. G. West (Eds.), The science ofprevention: Methodological advances fromalcohol and substance abuse research (pp.405-429). Washington, D.C.: American PsychologicalAssociation.Barnett, V., & Lewis, T. (1994). Outliers in statisticaldata. (3 rd ed.). Chichester, England:John Wiley & Sons Ltd.Bayer, R. (1981). Homosexuality and Americanpsychiatry: The politics of diagnosis. NewYork: Basic Books, Inc.Bettencourt, B. A. & Miller, N. (1996). Genderdifferences in aggression as a function ofprovocation: A meta-analysis. PsychologicalBulletin, 119(3), 422-447.Blair, A., Burg, J., Foran, J., Gibb, H., Greenland,S., Morris, R., Raabe, G., Savitz, D.,Teta, J., Wartenberg, D., Wong, O., &Zimmerman, R. (1995). Guidelines for applicationof meta-analysis in environmental epidemiology.Regulatory Toxicology andPharmacology, 22, 189-197.Bowers, T. G. & Clum, G. A. (1988). Relativecontribution of specific and nonspecifictreatment effects: Meta-analysis of placebocontrolledbehavior therapy research. PsychologicalBulletin, 103, 315-323.Boyd, J. H., Burke, J. D., Gruenberg, E., Holzer,C., E., Rae, D., S., George, L. K., Darno, M.,Stoltzman, R., McEvoy, L., & Nestadt, G.(1984). Exclusion criteria of DSM-III: a studyof co-occurrence of hierarchy-free syndromes.Archives of General Psychiatry,41(October), 983-989.Breslau, N. & Davis, G. C. (1987). Posttraumaticstress disorder: The stressor criterion.The Journal of Nervous and Mental Disease,175(5), 255-264.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling94Brown, J. (1987). A review of meta-analysesconducted on outcome research. ClinicalPsychology Review, 7, 1-23.Brownson, R. C., Newschaffer, C. J., & Ali-Abarghoui, F. (1997). Policy research for diseaseprevention: Challenges and practicalrecommendations. American Journal of PublicHealth, 87(5), 735-739.Bunce, R. (1979). Social and political sourcesof drug effects: The case of bad trips on psychedelics.Journal of Drug Issues, 9, 213-233.Burke, J. (1985). The day the universe changed.Boston: Little, Brown and Company.Burt, D. B., Zimbar, M. J., & Niederehe, G.(1995). Depression and memory impairment:a meta-analysis of the association, it pattern,and specificity. Psychological Bulletin,117(2), 285-305.Cabaj, R. P. (1992). Substance abuse in the gayand lesbian community. In J. H. Lowinson, P.Ruiz, R. B. Millman, & J. G. Langrod (Eds.),Substance abuse: A comprehensive textbook(pp. 852-860). Baltimore: Williams & Wilkins.Campbell, D. T. & Fiske, D. W. (1959). Convergentand discriminant validation by themultitrait-multimethod matrix. PsychologicalBulletin, 56(2), 81-105.Cappelleri, J. C., Ioannidis, J. P. A., Schmid, C.H., de Ferranti, S. D., Aubert, M., Chalmers,T. C., & Lau, J. (1996). Large trials vs. metaanalysisof smaller trials: How do their resultscompare? Journal of the AmericanMedical Association, 276(16), 1332-1338.Casti, J. L. (1989). Paradigms lost: Images ofman in the mirror of science. New York: WilliamMorrow & Company.Centers for Disease Control and Prevention.(1992). 1993 revised classification systemfor HIV infection and expanded surveillancecase definition for AIDS among adolescentsand adults. Morbidity and Mortality WeeklyReport, 41/No. RR-17.Christiansen, E. M. (1997). The U.S. 1996gross annual wager. International Gaming &Wagering Business (Special Supplement).Clotfelter, C. T. & Cook, P. J. (1989). Sellinghope: State lotteries in America. Cambridge,MA: Harvard University Press.Cochrane, A. L., & Holland, W. W. (1971).Validation of screening procedures. BritishMedical Bulletin, 27, 3-8.Cohen, B. I. (1985). Revolution in science.Cambridge, MA: Harvard University Press.Cohen, J. (1960). A coefficient of agreementfor nominal scales. Educational and PsychologicalMeasures, 20, 37-46.Cohen, J. (1988). Statistical power analyses forthe behavioral sciences. (2 nd ed.). Hillsdale,NJ: Lawrence Erlbaum.Cohen, J. & Cohen, P. (1975). Applied multipleregression/correlation analysis for the behavioralsciences. Hillsdale, NJ: LawrenceErlbaum Associates.Colditz, G.A., Berkey, C. S., & Mosteller, F.(1997). Reading and interpreting metaanalysis.Manuscript submitted for publication.Cook, T. D., Cooper, H., Cordray, D. S., Hartmann,H., Hedges, L. V., Light, R. J., Louis,T. A., & Mosteller, F. (1992). Meta-analysisfor explanation. New York: Russell SageFoundation.Cook, D. J., Sackett, D. L., & Spitzer, W. O.(1995). Methodological guidelines for systematicreviews of randomized control trialsin health care from the Potsdam consultationon meta-analysis. Journal of Clinical Epidemiology,48(1), 167-171.Cooper, H. & Hedges, L. V. (Eds.). (1994a). Thehandbook of research synthesis. New York:Russell Sage Foundation.Cronbach, L. J., & Meehl, P. E. (1955). Constructvalidity in psychological tests. PsychologicalBulletin, 52(4), 281-302.Culleton, R. P. (1989). The prevalence rates ofpathological gambling: A look at methods.Journal of Gambling Behavior, 5(1), 22-41.Cunningham, J. A., Sobell, L. C., Sobell, M. B.,& Kapur, G. (1995). Resolution from alcoholproblems with and without treatment: reasonsfor change. Journal of Substance Abuse, 7(3),365-72.Derevensky, J. L. & Gupta, R. (1997, June).Prevalence estimates of adolescent gambling:A comparison of the SOGS-RA, DSM-IV-J,and the G.A. 20 Questions. Paper presentedat the Tenth International Conference onHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling95Gambling and Risk-Taking, Montreal, Quebec.Diamond, D. & Wilsnack, S. (1978). Alcoholabuse among lesbians: A descriptive study.Journal of Homosexuality, 4, 123-142.Dickerson, M. (1994, June). Alternative approachesto the measurement of the prevalenceof pathological gambling. Paper presentedat the Ninth National Conference onGambling and Risk-Taking, Las Vegas, NV.Dohrenwend, B. P. (1995). "The problem ofvalidity in field studies of psychological disorders"revisited. In M. T. Tsuang, M. Tohen,& G. E. Zahner (Eds.), Textbook in psychiatricepidemiology (pp. 3-20). New York:Wiley-Liss.Eadington, W. R. (1992, October 27). Emergingpublic policy challenges from the proliferationof gaming in America. Paper presented atthe Second Annual Australasian Conferenceon Casinos and Gaming, Sydney, Australia.Eadington, W. R. (1994, May). Understandinggambling. Keynote Address, Ninth InternationalConference on Gambling and RiskTaking, Las Vegas, NV.Eagly, A., Karau, S. J., & Makhijani, M. G.(1995). Gender and the effectiveness of leaders:A meta-analysis. Psychological Bulletin,117(1), 125-145.Ennett, S. T., Tobler, N. S., Ringwalt, C. L., &Flewelling, R. L. (1994). How effective isdrug abuse resistance education? A metaanalysisof project DARE outcome evaluations.American Journal of Public Health,84(9), 1394-1401.Erel, O. & Burman, B. (1995). Interrelatednessof marital relations and parent-child relations:A meta-analytic review. Psychological Bulletin,118(1), 108-132.Ewing, J. A. (1984). Detecting alcoholism: theCAGE questionnaire. Journal of the AmericanMedical Association, 252(14), 1905-1907.Faraone, S. V. & Tsuang, M. T. (1994). Measuringdiagnostic accuracy in the absence of a"gold standard." American Journal of Psychiatry,151, 650-657.Fillmore, K. M., Hartka, E., Johnstone, B. M.,Leino, E. V., Motoyoshi, M., & Temple, M.T. (1991). A meta-analysis of life coursevariation in drinking. British Journal ofAddiction, 86(10), 1221-1267.Flavin, D. K. & Morse, R. M. (1991). What isalcoholism? Current definitions and diagnosticcriteria and their implications for treatment.Alcohol Health & Research World, 15,266-271.Fleiss, J. L. (1981). Statistical methods for ratesand proportions. New York: John Wiley &Sons.Ford, D. E. & Kamerow, D. B. (1990) Screeningfor psychiatric and substance abuse disordersin clinical practice. Journal of GeneralInternal Medicine, 5, (September/OctoberSupplement), S37-S41.Frankel, J. R. (1982). On the definition of responserates: A special report of the CASROtask force on completion rates. Port Jefferson,NY: The Council of American SurveyResearch Organizations.Frankel, M. (1983). Sampling theory. In P. H.Rossi, J. D. Wright, & A. B. Anderson (Eds.),Handbook of Survey Research (pp. 21-67).New York: Academic Press, Inc.Frost, R. (1914). North of Boston (1 st Ed.).London: D. Nutt.Gendreau, P. & Gendreau, L. P. (1971). Researchdesign and narcotic addiction proneness.Canadian Psychiatric AssociationJournal, 16, 265-267.Glaus, K. O. (1989). Alcoholism, chemicaldependency and the lesbian client. Women &Therapy, 8, 131-144.Goldstein, J. M. & Simpson, J. C. (1995). Validity:Definitions and applications to psychiatricresearch. In M. T. Tsuang, M. Tohen, &G. E. Zahner (Eds.), Textbook in psychiatricepidemiology (pp. 229-242). New York:Wiley-Liss.Goodman, S. N. (1991). Have you ever metaanalysisyou didn’t like? Annals of InternalMedicine, 114, 244-246.Greene, J. M., Ennett, S. T., & Ringwalt, C. L.(1997). Substance use among runaway andhomeless youth in three national samples.American Journal of Public Health, 87(2),229-235.Habermann-Little, B. (1991). An analysis ofthe prevalence and etiology of depression inHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling96Parkinson’s disease. Journal of NeuroscienceNursing, 23(3), 165-169.Hartka, E., Johnstone, B., Leino, E. V., Motoyoshi,M., Temple, M. T., & Fillmore, K. M.(1991). A meta-analysis of depressive symptomatologyand alcohol consumption overtime. British Journal of Addiction, 86(10),1283-1298.Hasselblad, V. & Hedges, L. V. (1995). Metaanalysisof screening and diagnostic tests.Psychological Bulletin, 117(1), 167-178.Hasselblad, V., Mosteller, F., Littenberg, B.,Chalmers, T. C., Hunink, M. G. M., Turner,J. A., Morton, S. C., Diehr, P., Wong, J. B., &Powe, N. R. (1995). A survey of currentproblems in meta-analysis: Discussion fromthe agency for health care policy and researchinter-PORT work group on literature review/meta-analysis.Medical Care, 33(2),202-220.Hennekens, C. H. & Buring, J. E. (1987). Epidemiologyin medicine. Boston: Little,Brown & Co.Horwath, E. & Weissman, M.M. (1995). Epidemiologyof depression and anxiety disorders.In M. T. Tsuang, M. Tohen, & G. E. P.Zahner (Eds.), Textbook in psychiatric epidemiology(pp. 317-344). New York: Wiley-Liss.International Game Technology. (1995). Gamingin the United States. [Brochure]. Las Vegas,NV: Author.International Gaming & Wagering Business.(1997, July). North American gaming report.International Gaming & Wagering Business,18(7), S3-S38.Ito, T. A., Miller, N., & Pollock, V. E. (1996).Alcohol and aggression: a meta-analysis onthe moderating effect of inhibitory cues, triggeringevents, and self-focused attention.Psychological Bulletin, 120(1), 60-82.Jacobs, D. F. (1989). A general theory of addictions:Rationale for and evidence supportinga new approach for understanding and treatingaddictive behaviors. In H. J. Shaffer, S.Stein, B. Gambino, & T. N. Cummings(Eds.), Compulsive gambling: theory, research& practice (pp. 35-64). Lexington,MA: Lexington Books.Jessor, R. & Jessor, S. L. (1977). Problem behaviorand psychosocial development: A longitudinalstudy of youth. New York: AcademicPress.Johnston, L. D., O'Malley, P.M., & Bachman, J.G. (1996). National survey results on druguse from the Monitoring the Future study,1975-1995. Volume I: Secondary school students.(NIH Pub. No. 97-4139). Rockville,MD: National Institute on Drug Abuse.Kahn, H. A., & Sempos, C. T. (1989). Statisticalmethods in epidemiology. New York:Oxford University Press.Khantzian, E. J. (1975). Self selection and progressionin drug dependence. Psychiatry Digest,36, 19-22.Khantzian, E. J. (1985). The self-medicationhypothesis of addictive disorders: Focus onheroin and cocaine dependence. AmericanJournal of Psychiatry, 142(11), 1259-1264.Khantzian, E. J. (1989). Addiction: Selfdestructionor self-repair? Journal of SubstanceAbuse Treatment, 6(2), 75.Khantzian, E. J. (1997). The self-medicationhypothesis of substance use disorders: A reconsiderationand recent applications. HarvardReview of Psychiatry, 4(5), 231-244.Kish, L. (1965). Survey sampling. New York:John Wiley & Sons, Inc.Kleinman, A. (1987). Culture and clinical reality:Commentary on culture-bound syndromesand international disease classifications, Culture,Medicine and Psychiatry, 11, 49-52.Kleinman, A. (1988). Rethinking psychiatry:From cultural category to personal experience.New York: The Free Press.Knox, R. A. (1997, June 24). Panel calls forbroader screening for diabetes. The BostonGlobe, p. A01.Levine, H. G. (1978). The discovery of addiction:Changing conceptions of habitualdrunkenness in America. Journal of Studieson Alcohol, 39, 143-174.Lipsey, M. W. & Wilson, D. B. (1993). The efficacyof psychological, educational, and behavioraltreatment: Confirmation from metaanalysis.American Psychologist, 48, 1181-1209.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling97Lipton, R. B. & Stewart, W.F. (1997). Prevalenceand impact of migraine. NeurologicClinics, 15(1), 1-13.Lohrenz, L., Connelly, J., Coyne, L., & Sparks,K. (1978). Alcohol problems in several midwesternhomosexual communities. Journalof Studies on Alcohol, 39, 1959-1963.Longnecker, M. P., Berlin, J. A., Orza, M. J., &Chalmers, T. C. (1988). A meta-analysis ofalcohol consumption in relation to risk ofbreast cancer. Journal of the AmericanMedical Association, 260(5), 652-656.Longnecker, M. P., Orza, M. J., Adams, M. E.,Vioque, J., & Chalmers, T. C. (1990). Ameta-analysis of alcoholic beverage consumptionin relation to risk of colorectal cancer.Cancer Causes and Control, 1, 59-68.MacCoun, R., & Reuter, P. (1997). InterpretingDutch cannabis policy: reasoning by analogyin the legalization debate. Science, 278, 47-52.Malagady, R. G., Rogler, L. H., & Tryon, W. W.(1992). Issues of validity in the diagnositc interviewschedule. Journal of Psychiatric Research,26(1), 59-67.Maugham, W. S. (1938). The summing up.New York: Literary Guild of America.McAuliffe, W. E., LaBrie, R. A., Mulvaney, N.,Shaffer, H. J., Geller, S., Fournier, E. A., Levine,E. B., Wang, Q., Wortman, S. M., &Miller, K. A. (1995). Assessment of substancedependence treatment needs: A telephonesurvey manual and questionnaire (RevisedEd.). Cambridge, MA: National TechnicalCenter on Substance Abuse Needs Assessment.McQueen, P.A. (1996, October). North Americanlottery sales report. International Gaming& Wagering Business, 17(10), 73-76.Meehl, P. E. (1954). Clinical versus statisticalprediction; a theoretical analysis and a reviewof the evidence. Minneapolis: Universityof Minnesota Press.Meehl, P. E. (1973). Psychodiagnosis; selectedpapers. New York: Norton.Miller, W. R., Brown, J. M., Simpson, T. L.,Handmaker, N. S., Bein, T. H., Luckie, L. F.,Montgomery, H. A., Hester, R. K., & Tonigan,J. S. (1995). What works? A methodologicalanalysis of the alcohol treatment outcomeliterature. In R. K. Hester & W. R.Miller (Eds.), Handbook of alcoholism treatmentapproaches: Effective alternatives (2 nded., pp. 12-44). Boston: Allyn and Bacon.Moher, D., Jadad, A. R., Nichol, G., Penman,M., Tugwell, P., & Walsh, S. (1995). Assessingthe quality of randomized controlledtrials: An annotated bibliography of scalesand checklists. Controlled Clinical Trials,16, 62-73.Mosteller, F. & Colditz, G. A. (1996). Understandingresearch synthesis (meta-analysis).Annual Review of Public Health, 17, 1-23.Nadler, L. B. (1985). The epidemiology ofpathological gambling: Critique of existingresearch and alternative strategies. Journal ofGambling Behavior, 1, 35-50.National Council on Problem Gambling. (1993,November 1). The need for a national policyon problem and pathological gambling inAmerica. New York: Author.National Council of Welfare. (1996). Gamblingin Canada. Ottawa, Ontario: Author.National Gambling Impact and Policy CommissionAct of 1996, Pub. L. No. 104-169.Newcomb, A. F. & Bagwell, C. L. (1995). Children'sfriendship relations: A meta-analyticreview. Psychological Bulletin, 117(2), 306-347.Norusis, M. J. & SPSS Inc. (1993). SPSS forwindows: Professional statistics, release 6.0.Chicago: SPSS Inc.Offord, D. R., Boyle, M. H., Szatmari, P., Rae-Grant, N. I., Links, P. S., Cadman, D. T., Byles,J. A., Crawford, J. W., Blum, H. M.,Byrne, C., Thomas, H., & Woodward, C.A.(1987). Ontario Child Health Study II. Sixmonth prevalence of disorder and rates ofservice utilization. Archives of General Psychiatry,44, 832-836.Olkin, I. (1995). Statistical and theoretical considerationsin meta-analysis. Journal ofClinical Epidemiology, 48(1), 133-146.Pathological gambling as a reflection of culturalnorms. (1997, October 28). The WAGER,2(43), 1.Pokovny, A. D., Miller, B. A., & Kaplan, H. B.(1972). The brief MAST: a shortened versionof the Michigan alcohol screening test.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling98American Journal of Psychiatry, 129, 342-345.Prochaska, J. O., DiClemente, C.C., & Norcross,J. C. (1992). In search of how people change:Applications to addictive behaviors. AmericanPsychologist, 47, 1102-1114.Quigley, H. A. & Vitale, S. (1997). Models ofopen-angle glaucoma prevalence and incidencein the United States. InvestigativeOphthalmology & Visual Science, 38(1), 83-91.Ritchie, K., Kildea, D., & Robine, J. M. (1992).The relationship between age and the prevalenceof senile dementia: A meta-analysis ofrecent data. International Journal of Epidemiology,21(4), 763-769.Robins, L. N. & Regier, D. A. (Eds.). (1991).Psychiatric disorders in America: The epidemiologiccatchment area study. New York:The Free Press.Robins, L. N., Helzer, J. E., Orvaschel, H., Anthony,J. C., Blazer, D. G., Burnam, A., &Burke, J. D., Jr. (1985). The diagnostic interviewschedule. In W. W. Eaton & L. G.Kessler (Eds.), Epidemiologic catchmentarea program (pp. 143-170). New York:Academic Press.Robins, L. N., Helzer, J. E., Ratcliff, K. S., &Seyfried, W. (1982). Validity of the diagnosticinterview schedule, version II: DSM-IIIdiagnoses. Psychological Medicine, 12, 855-870.Rose, I. N. (1986). Gambling and the law. (1sted.). Los Angeles, CA: Gambling TimesRose, I. N. (1995). Gambling and the law: Endlessfields of dreams. Journal of GamblingStudies, 11, 15-33.Rose, G. & Barker, D. J. P. (1978). What is acase? Dichotomy or continuum? BritishMedical Journal, 2, 873-874.Rosenhan, D. L. (1973). On being sane in insaneplaces. Science, 179, 250-258.Rosenthal, M. C. (1994). The fugitive literature.In H. Cooper & L. V. Hedges (Eds.), Thehandbook of research synthesis (pp. 85-94).New York: Russell Sage Foundation.Rosenthal, R. (1984). Meta-analytic proceduresfor social research. Beverly Hills: Sage Publications.Rosenthal, R. & Rosnow, R. L. (1975). Primerof methods for the behavioral sciences. NewYork: John Wiley & Sons, Inc.Rosenthal, R. & Rosnow, R. L. (1991). Essentialsof behavioral research: methods anddata analysis. (2 nd ed.). New York: McGraw-Hill Publishing Company.Rothman, K. J. (1986). Modern epidemiology.Boston: Little, Brown & Co.Sartin, H. G. (1988). WIN therapy: An alternativediagnostic and treatment procedure forproblem gamblers. In W. R. Eadington (Ed.),Gambling Research. Proceedings of the 7thInternational Conference on Gambling andRisk Taking. (Vol. 5: Research on PathologicalGambling, pp. 365-391). Reno: Universityof Nevada.Schacter, D. L. (Ed.). (1995). Memory distortion:How minds, brains, and societies reconstructthe past. Cambridge, MA: HarvardUniversity Press.Schlesinger, M. & Dorwart, R. A. (1992). Fallingbetween the cracks: Failing nationalstrategies for the treatment of substanceabuse. Daedalus, 121, 195-237.Schlesinger, M., Dorwart, R. A., & Clark R.(1991). Treatment capacity for drug problemsin the U.S.: Public policy in a fragmentedservice system. NIDA Drug AbuseServices Research Series No. 1 (pp. 16-57).DHHS Publication [ADM 91-1777].Schopenhauer, Arthur. (1893). Parerga andParalipomena. New York: Macmillan.Schuckit, M. A., Nathan, P. E., Helzer, J. E.,Woody, G. E., & Crowley, T. J. (1991). Evolutionof the DSM diagnostic criteria for alcoholism.Alcohol Health & Research World,4, 278-283.See, J. E., Howe, S. R., Warm, J. S., & Dember,W. N. (1995). Meta-analysis of the sensitivitydecrement in vigilance. Psychological Bulletin,117(2), 230-249.Shadish, W. R. (1996). Meta-analysis and theexploration of causal mediating processes: Aprimer of examples, methods, and issues.Psychological Methods, 1(1), 47-65.Shaffer, H. J. (1986a). Conceptual crises and theaddictions: A philosophy of science perspective.Journal of Substance Abuse Treatment,3(4), 285-296.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling99Shaffer, H. J. (1986b). Assessment of addictivedisorders: The use of clinical reflection andhypotheses testing. Psychiatric Clinics ofNorth America, 9(3), 385-398.Shaffer, H. J. (1992). The psychology of stagechange: The transition from addiction to recovery.In J. H. Lowinson, P. Ruiz, R. B.Millman, & J. G. Langrod (Eds.), Substanceabuse: A comprehensive textbook (2 nd ed., pp.100-105). Baltimore: Williams & Wilkins.Shaffer, H. J. (1993). The emergence of youthfuladdiction: The prevalence of underagelottery use and the impact of gambling. TechnicalReport (121393-100). Boston: MassachusettsCouncil on Compulsive Gambling.Shaffer, H. J. (1994). Denial, ambivalence andcountertransference hate. In J. D. Levin & R.Weiss (Eds.), Alcoholism: Dynamics andtreatment (pp. 421-437). Northdale, N.J.: JasonAronson.Shaffer, H. J. (1996). Understanding the meansand objects of addiction: Technology, theInternet, and gambling. Journal of GamblingStudies, 12(4), 461-469.Shaffer, H. J. (1997a). The psychology of stagechange. In J. H. Lowinson, P. Ruiz, R. B.Millman, & J. G. Langrod (Eds.), Substanceabuse: A comprehensive textbook (3 rd ed., pp.100-106). Baltimore: Williams & Wilkins.Shaffer, H. J. (1997b). The most important unresolvedissue in the addictions: conceptualchaos. Substance Use & Misuse, 32(11),1573-1580.Shaffer, H. J. & Hall, M. N. (1996). Estimatingthe prevalence of adolescent gambling disorders:A quantitative synthesis and guide towardstandard gambling nomenclature. Journalof Gambling Studies, 12(2), 193-214.Shaffer, H. J., Hall, M. N., Walsh, J. S., &Vander Bilt, J. (1995). The psychosocial consequencesof gambling. In R. Tannenwald(Ed.), Casino development: How would casinosaffect New England's economy? SpecialReport No. 2 (pp. 130-141). Boston: FederalReserve Bank of Boston.Shaffer, H. J. & Jones, S. B. (1989). Quittingcocaine: The struggle against impulse. Lexington,MA.: Lexington Books.Shaffer, H. J., & Robbins, M. (1991). Manufacturingmultiple meanings of addiction: Timelimitedrealities. Contemporary Family Therapy,13, 387-404.Shaffer, H. J., Stein, S. A., Gambino, B., &Cummings, T. N. (1989). Compulsive gambling:Theory, research, and practice. Lexington,MA: Lexington Books.Shaffer, H. J. & Zinberg, N. E. (1985). The socialpsychology of intoxicant use: The naturalhistory of social settings and social control.Bulletin of the Society of Psychologists in theAddictive Behaviors, 4, 49-55.Siegel, S. (1956). Nonparametric statistics forthe behavioral sciences. New York:McGraw-Hill Book Company, Inc.Slavin, R. J. (1990). Achievement effects ofability grouping in secondary schools: A bestevidencesynthesis. Review of EducationalResearch, 60, 471-499.Smith, M. L. & Glass, G. V. (1977). Metaanalysisof psychotherapy outcome studies.American Psychologist, 32, 752-760.Sobell, L. C., Cunningham, J. A., & Sobell, M.B. (1996). Recovery from alcohol problemswith and without treatment: Prevalence intwo population surveys. American Journal ofPublic Health, 86(7), 966-72.Sobell, M. B. & Sobell, L. C. (1993). Treatmentfor problem drinkers: A public health priority.In J. S. Baer, G. A. Marlatt, & R. J.McMahon (Eds.), Addictive behaviors acrossthe life span: Prevention, treatment, and policyissues (pp. 138-157). Newbury Park, CA:Sage Publications, Inc.Stanton, M. D. & Shadish, W. R. (1997). Outcome,attrition, and family-couples treatmentfor drug abuse: A meta-analysis and reviewof the controlled, comparative studies. PsychologicalBulletin, 122(2), 170-191.Stein, D. M. & Lambert, M. J. (1995). Graduatetraining in psychotherapy: Are therapy outcomesenhanced? Journal of Consulting andClinical Psychology, 63(2), 182-196.Steinem, G. (1983). Outrageous Acts and EverydayRebellions. New York: Holt, Rinehart& Winston.Stewart, W.F., Simon, D., Shechter, A., & Lipton,R. B. (1995). Population variation inmigraine prevalence: A meta-analysis. Journalof Clinical Epidemiology, 48(2), 269-280.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling100Sudman, S. (1983). Applied sampling. In P. H.Rossi, J. D. Wright, & A. B. Anderson (Eds.),Handbook of survey research (pp. 145-194).New York: Academic Press.Tobler, N. S. (1986). Meta-analysis of 143 adolescentdrug prevention programs: Quantitativeoutcome results of program participantscompared to a control or comparison group.Journal of Drug Issues, 16, 537-567.Tsuang, M. T., Tohen, M., & Zahner, G. E.(Eds.). (1995). Textbook in psychiatric epidemiology.New York: John Wiley & Sons,Inc.Vaillant, G. E. & Schnurr, P. (1988). What is acase? A 45-year study of psychiatric impairmentwithin a college sample selected formental health. Archives of General Psychiatry,45(4), 313-319.Velez, C. N., Johnson, J., & Cohen, P. (1989).A longitudinal analyses of selected risk factorsfor childhood psychopathology. Journalof American Academy of Child & AdolescentPsychiatry, 28, 861-864.Wagner, J. (1986). The search of signs of intelligentlife in the universe. New York: Harper& Row.Wakefield, J. C. (1992). The concept of mentaldisorder: on the boundary between biologicalfacts and social values, American Psychologist,47, 373-388.Walker, M. B. & Dickerson, M. G. (1996). Theprevalence of problem and pathological gambling:A critical analysis. Journal of GamblingStudies, 12(2), 233-249.Wampold, B. E., Mondin, G. W., Moody, M.,Stich, F., Benson, K., & Ahn, H. (1997). Ameta-analysis of outcome studies comparingbona fide psychotherapies: empirically, “allmust have prizes.” Psychological Bulletin,122(3), 203-215.WEFA Group, ICR Survey Research Group,Lesieur, H., & Thompson, W. (1997). A studyconcerning the effects of legalized gamblingon the citizens of the state of Connecticut.Newington: State of Connecticut Departmentof Revenue Services, Division of SpecialRevenue.Weil, A., Zinberg, N. E., & Nelson, J. (1968).Clinical and psychological effects of marihuanain man. Science, 162, 1234-1242.Weisz, J. R., Weiss, B., Alicke, M. D., & Klotz,M. L. (1987). Effectiveness of psychotherapywith children and adolescents: A metaanalysisfor clinicians. Journal of Consultingand Clinical Psychology, 55, 542-549.Weisz, J. R., Weiss, B., Han, S. S., Granger, D.A., & Morton, T. (1995). Effects of psychotherapywith children and adolescents revisited:A meta-analysis of treatment outcomestudies. Psychological Bulletin, 117(3), 450-468.Whyte, K., (1997, November 21). AmericanGaming Association, personal communication.Winer, B. J. (1971). Statistical Principles inExperimental Design. (2 nd ed.). New York:McGraw-Hill Book Company.Winick, C. (1962). Maturing out of narcotic addiction.United Nations Bulletin on Narcotics,14, 1-7.Zimmerman, M. A., Meeland, T., & Krug, S.(1985). Measurement and structure of pathologicalgambling behaviour. Journal of PersonalityAssessment, 49, 76-81.Zinberg, N. E. (1974). High states: A beginningstudy. Drug Abuse Council Publication No.SS-3. Washington, D.C.: The Drug AbuseCouncil, Inc.Zinberg, N. E. (1984). Drug, set, and setting:The basis for controlled intoxicant use. NewHaven, CT: Yale University Press.Zinberg, N. E. & Fraser, K. M. (1979). The roleof the social setting in the prevention andtreatment of alcoholism. In J. Mendelson &N. Mello (Eds.), The Diagnosis & Treatmentof Alcoholism (pp. 359-385). New York:McGraw-Hill Book Company.Zinberg, N. E., & Harding, W. M. (Eds.).(1982). Control over intoxicant use. NewYork: Free Press.Zinberg, N. E., Harding, W. M., & Winkeller,M. (1977). A study of social regulatorymechanisms in controlled illicit drug users.Journal of Drug Issues, 7, 117-133.Zinberg, N. E. & Shaffer, H. J. (1985). The socialpsychology of intoxicant use: the interactionof personality and social setting. In H.B. Milkman & H. J. Shaffer (eds.) The Addictions:Multidisciplinary perspectives andHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling101treatments (pp. 57-74). Lexington, MA: LexingtonBooks.Ziskin, J. (1970). Coping with psychiatric andpsychological testimony. (2 nd ed.). BeverlyHills: Law and Psychology Press.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling102Appendix 1: Multi-method EstimatorsThe following tables present multi-method estimators within each of the four populations segments.“LT” refers to lifetime estimates; “PY” indicates past-year estimates.AdultsLevel 3 LT Level 2 LT Level 3 PY Level 2 PYmedian 1.54167 3.32750 .95000 2.28333mean 1.59639 3.85156 1.13876 2.79570M: Huber 1.58588 3.26593 1.01987 2.30772M: Tukey 1.58564 3.09290 .95781 2.18396M: Hampell 1.59756 3.17876 1.00113 2.20707M: Andrews 1.58553 3.09403 .95691 2.184155% trimmed mean 1.59375 3.49565 1.08240 2.48452Winsorized median 1.51667 3.39650 .91667 2.31667Winsorized mean 1.58354 3.63091 1.08832 2.79605median weighted by quality score 1.5 3.08600 .9 2.3mean weighted by quality score 1.58646 3.78426 1.13036 2.80753weighted M: Huber 1.56500 3.16798 .99686 2.30635weighted M: Tukey 1.56384 2.96199 .94338 2.17917weighted M: Hampell 1.57920 3.08130 .96918 2.20437weighted M: Andrews 1.56342 2.95418 .94377 2.17935weighted 5% trimmed mean 1.58133 3.42929 1.07381 2.49188Mean of 16 estimation methods 1.57037 3.29992 1.00433 2.37674Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling103AdolescentsLevel 3 LT Level 2 LT Level 3 PY Level 2 PYmedian 3.46667 9.15 4.485 14.7mean 3.87931 9.44875 5.76959 14.82303M: Huber 3.70154 9.26592 5.46617 13.78063M: Tukey 3.73990 9.15895 5.27702 12.98932M: Hampell 3.83458 9.31347 5.40209 13.85322M: Andrews 3.73971 9.15921 5.27565 12.893315% trimmed mean 3.85 9.45417 5.46699 14.49226Winsorized median 3.4 8.7 4.69850 15.11950Winsorized mean 3.59 9.01023 5.96668 16.25383median weighted by quality score 3.4 9.9 4.363 13.4mean weighted by quality score 3.89493 9.62105 5.93010 14.64040weighted M: Huber 3.69682 9.65738 5.22038 13.52199weighted M: Tukey 3.73975 9.60731 5.12656 12.86780weighted M: Hampell 3.83267 9.63580 5.27337 13.35033weighted M: Andrews 3.73974 9.60688 5.12426 12.86618weighted 5% trimmed mean 3.86919 9.64561 5.55479 14.28933Mean of 16 estimation methods 3.71093 9.39592 5.27501 13.99007Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling104CollegeLevel 3 LT Level 2 LT Level 3 PY Level 2 PYmedian 4.93333 6.41667 -- --mean 4.67370 9.27896 -- --M: Huber 4.51701 7.39346 -- --M: Tukey 4.48378 6.63890 -- --M: Hampell 4.56345 7.31493 -- --M: Andrews 4.47721 6.64025 -- --5% trimmed mean 4.58946 8.63101 -- --Winsorized median 5 6.25 -- --Winsorized mean 4.91331 8.11435 -- --median weighted by quality score 5 6 -- --mean weighted by quality score 4.64466 8.84256 -- --weighted M: Huber 4.51641 7.07916 -- --weighted M: Tukey 4.48653 6.44433 -- --weighted M: Hampell 4.55570 6.92359 -- --weighted M: Andrews 4.48618 6.44173 -- --weighted 5% trimmed mean 4.55681 8.14612 -- --Mean of 16 estimation methods 4.64985 7.28475 -- --Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling105TreatmentLevel 3 LT Level 2 LT Level 3 PY Level 2 PYmedian 13.45550 13.00000 4.69350 --mean 14.22567 15.00508 4.69350 --M: Huber 13.51374 13.78262 4.69350 --M: Tukey 12.95594 13.56683 4.69350 --M: Hampell 13.24120 14.10941 4.69350 --M: Andrews 12.95979 13.57589 4.69350 --5% trimmed mean 13.72296 14.59403 -- --Winsorized median 13.45550 13 -- --Winsorized mean 13.63111 13.90964 -- --median weighted by quality score 13.13100 14 2.3 --mean weighted by quality score 14.38404 15.09467 4.04073 --weighted M: Huber 13.41216 14.17917 -- --weighted M: Tukey 12.71613 14.21022 -- --weighted M: Hampell 13.18334 14.60859 -- --weighted M: Andrews 12.72056 14.21373 -- --weighted 5% trimmed mean 13.89893 14.69357 3.96820 --Mean of 16 estimation methods 13.41297 14.09647 -- --Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling106Appendix 2: Study CharacteristicsThe following table lists and provides information on all of the studies included in this meta-analysis.Some of these studies reported prevalence rates for multiple study samples (e.g., Volberg 1994b); forthese studies, each separate study sample is listed. This table organizes study samples in groups accordingto population type (i.e., adult, adolescent, college, and treatment); within each population type, studysamples are organized first by region, then by state or province, then by instrument. Data for some of thestudy samples included in this meta-analysis was published in more than one source (e.g., Emerson &Laundergan, 1996 and Emerson, Laundergan, & Schaefer, 1994); although data for these samples wasweighted in our data set to avoid overrepresentation of these samples, each publication is listed in thistable for readers’ reference.RegionState/ProvinceInstrumentTime N Author Yearframe 56 ReleasedADULTNew England Connecticut SOGS lifetime 1000 Christiansen/ Cummings et al. 1992New England Connecticut DIS lifetime 1224 Laventhol & Horwath et al. 1986New England Connecticut 3-item scale lifetime 568 Abrahamson & Wright 1977New England Massachusetts SOGS lifetime 750 Volberg 1994bMiddle Atlantic Maryland SOGS lifetime 750 Volberg 1994bMiddle Atlantic Maryland SOGS lifetime 750 Volberg & Steadman 1989cMiddle Atlantic New Jersey SOGS lifetime 1000 Volberg 1994bMiddle Atlantic New Jersey SOGS lifetime 1000 Volberg & Steadman 1989cMiddle Atlantic New York SOGS lifetime 1000 Volberg & Steadman 1988bMiddle Atlantic New York SOGS both 1829 Volberg 1996bMiddle Atlantic New York DSM-IV past year 1829 Volberg 1996bMiddle Atlantic Combination IGB scale lifetime 534 Culleton & Lang 1985Middle Atlantic Combination ISR scale lifetime 534 Culleton & Lang 1985Middle Atlantic Combination IGB scale lifetime 534 Sommers 1988Southeastern Georgia SOGS both 1550 Volberg & Boles 1995Southeastern Mississippi SOGS both 1014 Volberg 1997North Central Indiana DSM-IV mod. lifetime 1015 Laventhol & Horwath et al. 1990North Central Iowa SOGS lifetime 750 Volberg 1994bNorth Central Iowa SOGS lifetime 750 Volberg & Steadman 1989aNorth Central Iowa SOGS both 1500 Volberg 1995aNorth Central Minnesota SOGS past year 1028 Emerson & Laundergan 1996North Central Minnesota SOGS past year 1028 Emerson et al. 1994North Central Minnesota SOGS past year 1251 Laundergan et al. 1990North Central Missouri DIS lifetime 2954 Cunningham-Williams et al. In pressNorth Central North Dakota SOGS both 1517 Volberg & Silver 1993North Central Ohio Custer criteria lifetime 801 Transition Planning Associates 1985North Central Ohio clinical signs lifetime 801 Transition Planning Associates 1985North Central South Dakota SOGS both 57 1767 Volberg & Stuefen 199456Time frames are listed as lifetime, past year, or both lifetime and past year unless otherwise noted. Readers should note that, aswe described in the Methods section of this report, prevalence rates for which no time frame was indicated were included in thelifetime category.57 Six-month and lifetime ratesHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling107N Author YearRegion State/Instrument TimeProvinceframe 56 ReleasedNorth Central South Dakota SOGS both 58 1560 Volberg et al. 1991North Central Wisconsin DSM-IV mod. lifetime 1000 Thompson et al. 1996South Central Louisiana SOGS lifetime 1818 Westphal & Rush 1996South Central Louisiana SOGS both 1818 Volberg 1995bSouth Central Texas SOGS both 6308 Wallisch 1993bSouth Central Texas SOGS both 7015 Wallisch 1996Northwestern Montana SOGS both 1020 Volberg 1992aNorthwestern Washington SOGS both 1502 Volberg 1993cSouthwestern California SOGS lifetime 1250 Volberg 1994bSouthwestern Nevada ISR scale lifetime 296 Kallick et al. 1979Southwestern New Mexico DSM-IV mod. past year 1279 New Mexico Dept. of Health 1996British Columbia British Columbia SOGS both 810 Angus Reid Group 1996British Columbia British Columbia SOGS both 1200 Gemini Rsrch. & Angus Reid 1994Grp.Prairie Provinces Alberta SOGS both 1803 Wynne Resources Ltd. 1994Prairie Provinces Alberta DIS both 59 7214 Bland et al. 1993Prairie Provinces Saskatchewan SOGS both 1000 Volberg 1994aPrairie Provinces Manitoba SOGS past year 1212 Criterion Research Corp. 1993Prairie Provinces Manitoba SOGS past year 1207 Criterion Research Corp. 1995Ontario Ontario SOGS lifetime 1030 Ferris & Stirpe 1995Ontario Ontario SOGS past year 2682 Govoni & Frisch 1996Ontario Ontario SOGS past year 3843 Govoni & Frisch 1996Ontario Ontario mod. SOGS past year 1200 Insight Canada Research 1993Ontario Ontario DSM-IV both 1030 Ferris & Stirpe 1995Ontario Ontario Life areas past year 1030 Ferris & Stirpe 1995Ontario Ontario gambling past year 1737 Smart & Ferris 1996problems scaleQuebec Quebec SOGS lifetime 1002 Ladouceur 1991Quebec Quebec SOGS lifetime 1002 Ladouceur 1993Atlantic ProvincesNew Brunswick SOGS both 800 Baseline Mkt. Research 1992Atlantic ProvincesNew Brunswick SOGS both 800 Baseline Mkt. Research 1996bAtlantic ProvincesNova Scotia SOGS both 801 Baseline Mkt. Research 1996aAtlantic ProvincesNova Scotia mod. SOGS lifetime 400 Baseline Mkt. Research 1996cAtlantic ProvincesNova Scotia mod. SOGS lifetime 810 Omnifacts Research Ltd. 1993Combined regionsCombination ISR scale lifetime 1736 Kallick et al. 1979does not specify does not specify 2-item scale lifetime 900 Ubell 1991ADOLESCENTSNew England Connecticut MAGS past year 3886 Steinberg 1997New England Connecticut DSM-III-R lifetime 1592 Steinberg 1997New England Connecticut SOGS-RA past year 3886 Steinberg 1997broadNew England Massachusetts MAGS past year 856 Shaffer et al. 199458 Six-month and lifetime rates59Six-month and lifetime ratesHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling108N Author YearRegion State/Instrument TimeProvinceframe 56 ReleasedNew England Massachusetts MAGS past year 854 Shaffer & Hall 1994New England Massachusetts MAGS past year 466 Vagge 1996New England Massachusetts mod. MAGS past year 1500 Allen 1995New England Massachusetts DSM-IV past year 854 Shaffer & Hall 1994New England Massachusetts DSM-IV past year 856 Shaffer et al. 1994New England Massachusetts DSM-IV past year 466 Vagge 1996Middle Atlantic New Jersey Path. Gamb. lifetime 892 Lesieur & Klein 1987Signs IndexSoutheastern Florida SOGS-RA past year 1882 Lieberman et al. 1996other 60Southeastern Georgia SOGS lifetime 1007 Volberg 1996aSoutheastern Georgia Multi-factor lifetime 1007 Volberg 1996amethodNorth Central Minnesota SOGS-RA past year 532 Winters & Stinchfield 1993narrowNorth Central Minnesota SOGS-RA past year 532 Winters et al. 1995narrowNorth Central Minnesota SOGS-RAnarrowpast year 532 Winters et al. 1995 (follow-up)North Central Minnesota SOGS-RA past year 1094 Winters et al. 1990broadNorth Central Minnesota SOGS-RA past year 702 Winters et al. 1993bbroadNorth Central Minnesota SOGS-RA past year 532 Winters et al. 1995broadNorth Central Minnesota SOGS-RAbroadpast year 532 Winters et al. 1995 (follow-up)North Central Minnesota SOGS-RA past year 460 Winters et al. 1993aotherNorth Central Minnesota mod. adolescentlifetime 277 Zitzow 1996aSOGSSouth Central Texas Multi-factor lifetime 924 Wallisch 1993amethodSouth Central Texas Multi-factor lifetime 3079 Wallisch 1996methodSouth Central Texas mod. adolescentlifetime 924 Wallisch 1993aSOGSNorthwestern Washington SOGS lifetime 1054 Volberg 1993aNorthwestern Washington SOGS-RA past year 1054 Volberg 1993abroadNorthwestern Washington Multi-factor lifetime 1054 Volberg 1993amethodPrairie Provinces Alberta mod. adolescentpast year 972 Wynne Resources Ltd. 1996SOGSOntario Ontario SOGS past year 935 Govoni et al. 1996Ontario Ontario SOGS-RA past year 935 Govoni et al. 1996narrowOntario Ontario SOGS-RAbroadpast year 935 Govoni et al. 199660 Winters, Stinchfield, & Kim (1995) describe two methods of scoring the SOGS-RA, the narrow criteria and the broad criteria.In addition to these two methods, Winters, Stinchfield, and Fulkerson (1993a) describe another method of scoring theSOGS-RA. In this table we will refer to this third, less common method as the “SOGS-RA other.”Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling109N Author YearRegion State/Instrument TimeProvinceframe 56 ReleasedOntario Ontario mod. adolescentpast year 400 Insight Canada Research 1994SOGSQuebec Quebec SOGS lifetime 289 Gaboury & Ladouceur 1993Quebec Quebec DSM-III lifetime 1612 Ladouceur & Mireault 1988Quebec Quebec Path. Gamb. lifetime 1612 Ladouceur & Mireault 1988Signs IndexAtlantic ProvincesNova Scotia mod. SOGS lifetime 300 Omnifacts Research Ltd. 1993Atlantic ProvincesNova Scotia Self- assessmentpast year 3857 Nova Scotia Dept. of Health 1996COLLEGENew England Connecticut SOGS-Plus lifetime 238 Devlin & Peppard 1996Middle Atlantic New Jersey SOGS lifetime 636 Frank 1990Middle Atlantic New Jersey SOGS lifetime 636 Frank 1993Middle Atlantic New Jersey SOGS lifetime 227 Lesieur et al. 1991Middle Atlantic New York SOGS lifetime 446 Lesieur et al. 1991North Central Michigan SOGS lifetime 1147 Lumley & Roby 1995North Central Minnesota SOGS lifetime 529 Winters et al. 1996North Central Minnesota SOGS lifetime 868 Winters et al. 1996North Central Minnesota SOGS lifetime 373 Winters et al. 1996North Central Wisconsin Self- assessmentpast year 604 Cook 1987South Central Oklahoma SOGS lifetime 583 Lesieur et al. 1991South Central Texas SOGS lifetime 299 Lesieur et al. 1991Southwestern Nevada SOGS lifetime 219 Lesieur et al. 1991Southwestern Nevada SOGS lifetime 544 Oster & Knapp 1994Southwestern Nevada SOGS lifetime 350 Oster & Knapp 1994Southwestern Nevada DSM-III-R lifetime 544 Oster & Knapp 1994Southwestern Nevada DSM-IV lifetime 544 Oster & Knapp 1994Southwestern Nevada DSM-III-R lifetime 350 Oster & Knapp 1994Quebec Quebec SOGS lifetime 1471 Ladouceur et al. 1994bCombined regionsCombination SOGS lifetime 1771 Lesieur et al. 1991does not specify does not specify SOGS lifetime 384 Lesieur & Blume 1987does not specify does not specify DSM-III-R lifetime 384 Lesieur 1988adoes not specify does not specify DSM-III-R lifetime 384 Lesieur & Blume 1987TREATMENTNew England Connecticut DSM-III-R lifetime 298 Steinberg et al. 1992New England Connecticut DSM-III-R both 298 Rounsaville et al. 1991New England Massachusetts SOGS lifetime 85 Gambino et al. 1993New England Massachusetts SOGS lifetime 93 Shepherd 1996Middle Atlantic Maryland SOGS lifetime 467 Ciarrochi 1993Middle Atlantic New York SOGS lifetime 220 Feigelman et al. 1995Middle Atlantic New York SOGS lifetime 297 Lesieur & Blume 1987Middle Atlantic New York SOGS/ SOGS- lifetime 117 Spunt et al. 1995PlusMiddle Atlantic New York SOGS lifetime 105 Lesieur & Blume 1990Middle Atlantic New York SOGS/ SOGS- lifetime 462 Spunt et al. 1996PlusMiddle Atlantic New York clinical judgmentlifetime 297 Lesieur & Blume 1987Middle Atlantic Pennsylvania SOGS lifetime 363 Walters 1997Middle Atlantic does not specify Path. Gamb. lifetime 458 Lesieur et al. 1986Signs IndexNorth Central Illinois SOGS lifetime 276 Daghestani et al. 1996Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling110N Author YearRegion State/Instrument TimeProvinceframe 56 ReleasedNorth Central Minnesota SOGS lifetime 201 Miller & Westermeyer 1996North Central Minnesota SOGS lifetime 211 Miller & Westermeyer 1996North Central Ohio SOGS lifetime 2171 McCormick 1993North Central Ohio SOGS lifetime 154 Castellani et al. 1996North Central South Dakota SOGS lifetime 85 Elia & Jacobs 1993Southwestern Nevada SOGS lifetime 136 Templer et al. 1993Ontario Ontario SOGS both 508 Donwood Problem Gamb.Prog.1996Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling111Appendix 3: A Brief Guide to Planning Prevalence Study ProtocolsAll problems are finally scientific problems.- George Bernard Shaw (1911)IntroductionDeveloping and implementing a researchprotocol that will yield reliable andmeaningful prevalence estimates of disorderedgambling among a general adultor youth population can be a formidable task forboth new and experienced investigators. Thereare common elements to this endeavor thatshould be present in a research design regardlessof the population being studied. By applyingtested research design principles, investigatorscan ease the difficulties associated with theseprojects (Rosenthal & Rosnow, 1991). Whendeveloping a research protocol, experienced investigatorshave found that there is no substitutefor detail, clarity and precision. All too often,when things can go wrong during the conduct ofclinical research, things will go wrong. Researchersinterested in developing and implementingresearch protocols for the conduct of ageneral population telephone-based prevalenceresearch can use this brief guide to minimize thepossibility of problems during both the conductof a prevalence survey and the application of thesurvey results.There are many basic characteristics of a researchprotocol that investigators often overlook.At the most fundamental level, for example,protocols and research guides should bereadable so the research team and others can usethese guides effectively. It is advisable to havecolleagues examine research protocols for organizational,logical, and written errors beforeeither submitting the document for funding reviewor going into the field to collect data.Similarly, each aspect of the prevalence survey—fromthe purpose of the study, to how respondentswill be identified and their data analyzed—deservescareful attention. In this verybrief guide, we will consider the cardinal aspectsof a disordered gambling prevalence surveyprotocol. Many of these principles derivefrom the review of existing prevalence studiesoriginally described in this meta-analysis. Acomprehensive analysis of epidemiological surveyresearch is well beyond the scope of thisbrief guide. However, we believe this documentcan serve as a model to help administrators andinvestigators alike move toward more usefulstudies of disordered gambling. In addition, thisbrief guide includes a description of methodologicalprinciples that will help investigatorsavoid the casualties of epidemiological research—forexample, discovering after muchhard work and expense that your response rateprecludes a scientific analysis, or having identifiedtoo few respondents with the target attributes.We suggest that there are 11 basic or core elementsnecessary for investigators to develophigh-quality population prevalence estimates.These elements are as follows:1. Establish a precise statement of purposeand objectives;2. Select, modify, or create an instrumentthat can satisfy the study objectives;3. Determine a time frame for the gamblingdisorder as well as its associatedcluster of signs and symptoms;4. Determine how best to eliminate alternativepsychiatric disorders;5. Determine a sampling design to selectrespondents;Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling1126. Use a power analysis to plan for adequaterepresentation of the population(s)of interest;7. Properly calculate response and completionrates;8. Provide a protocol for selecting andtraining interviewers;9. Supervise the collection of data;10. Check the integrity of data coding anddata entry;11. Apply the results.As we begin to address these essential methodologicalissues in more detail, we also will considersome of the important organizational andstrategic concerns that set the stage for determiningthe details of a research design and itsaccompanying protocol.In the remainder of this brief guide, we will introduceand highlight the central principles andactivities that must be addressed in planningprevalence research; readers should note onceagain that this appendix is not offered as a definitiveguide, just a blueprint for the core issuesthat await those interested in entering the precariousworld of prevalence research.PurposeAgeneral population gambling surveyresearch project should begin with aprotocol that includes a very precisestatement of purpose. A statement ofpurpose begins with a precise statement of thedata needs. This statement may answer some ofthe following questions: What do you want theprevalence estimate to do for you, and how doyou intend to use it? How will the developmentof a detailed assessment of prevalence direct theplanning, development and implementation ofexisting or new prevention, education, or treatmentservices? How will the assessment ofprevalence affect the availability of gaming inthe region? We cannot overstate the importanceof carefully determining the data needs of a project.While few estimates of disordered gamblingto date have a stated a specific purpose,one of the primary reasons for conducting aprevalence survey in this and other fields is togather the information essential to guide stateand sub-state resource allocation (e.g., treatmentplanning). This activity requires research that isdifferent from the survey traditionally designedto provide prevalence estimates of gambling andgambling related disorders; this endeavor ismore similar to substance abuse needs assessmentsurveys which typically have been conductedto determine “caseness” (e.g., Rose &Barker, 1978; Vaillant & Schnurr, 1988).A “case” represents someone who qualifies fortreatment services (i.e., meets the treatment eligibilityrequirements), or who would qualify fortreatment services if these services were available.Although many people who meet diagnosticstandards do not enter treatment, and manywho fail to satisfy such standards do entertreatment, the long-standing assumption of disorderedgambling prevalence investigators hasbeen that the estimates of those who meet somediagnostic standard represent the treatment needof a particular area. Investigators also shouldkeep in mind that, in addition to those who currentlysatisfy diagnostic criteria, there exists aproportion of the population who already havemet their treatment need—either through naturalrecovery, self-help, or professional treatmentservices. To date, no study has examined theextent of “met need” among those who haveexperienced disordered gambling. A prevalenceestimate for resource allocation would takethese factors into consideration and would assistregional, state, and local planners to determinethe clinical resources that must be allocated tomeet the demands of a disorder that exists at aparticular level in the community.Literature ReviewAcareful review of the literature is necessarybefore conducting a prevalencestudy so the conceptual and practicalpitfalls of this research can be identifiedand addressed. The literature review also willHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling113prove helpful as the research team and their administrativegroup select the measurement instrumentor instruments. A literature reviewshould be relevant to, and make a cogent andsuccinct argument for, the necessity to conductthe prevalence research. Further, the literaturereview should address the specific data needs orinterests of the study. Authors should present abrief review of the pertinent prevalence literatureand the methods for gathering sensitivegambling-related information. This literaturereview also should summarize any existing regionaldata. Have any relevant regional, state,or local surveys been completed? In addition, ifprevious statewide data is unavailable, are thereregional or national gambling prevalence ratesthat bear directly on your state, region or community?Are there any observed national trendsthat concern your state treatment planners?Selecting An InstrumentAgambling prevalence research protocolmust consider the full range of availablemeasurement instruments and determinewhich specific instrument provides thebest fit for the study’s stated purpose. There area variety of instruments available for investigatorsto employ that provide estimates of disorderedgambling prevalence (e.g., Diagnostic InterviewSchedule, DSM-IV, South Oak GamblingScreen, Massachusetts Gambling Screen).The use of an established instrument will expeditethe survey process and assure the inclusionof key variables. In addition, researchers alwaysshould choose instruments that reflect their project’sparticular purpose and data needs. Thechoice of instrumentation may differ dependingupon whether the prevalence study targetshouseholds, schools, treatment, prison, or other“special populations.”On InstrumentationWe urge investigators to select instruments withgreat care and caution. If you select an existinginstrument, do not make significant modificationsto the survey; instead, consider addingquestions relevant to your particular data needs.In this way, the psychometric properties of theoriginal survey instrument will be maintained.However, if any additions or deletions are madeto existing survey instruments, these changesshould be noted in the research protocol with adetailed description of the reasons for the adjustments—aftersome time, the research teammay forget their original motivations. If investigatorsare interested in integrating the gamblingprevalence survey with other surveys, thisprocedural scheme also should be noted in theprotocol and evaluated with great care. Cautionshould be taken to clarify in advance the manypossible relationships that can exist among thesurveys so that the data collectors can measurethe necessary variables in each of the pertinentsurveys with the least burden for the respondentand lowest risk to the integrity of the collecteddata.Forms of GamblingThe protocol must determine which forms ofgambling the survey will examine; for example,lottery, pari-mutuel, charitable, casino-basedgames, cock fighting, etc. If the prevalence surveytargets games of chance that are not includedon the original survey (e.g., pogs, noncasino-basedcard playing), investigators shoulddescribe these targets in detail and the rationalefor adding these forms of gambling to those alreadyincluded. For example, perhaps yourstate, province, or community currently providestreatment services for a certain number of personswith a specific pattern of gambling disorder;in this case, you may want to survey thisgroup to determine their pattern of gambling,and the games they play. Alternatively, thereare social indicators (e.g., arrest statistics) suggestingthat a particular pattern of gambling mayexist among arrestees; you may want to surveythis group and their pattern of gambling if it isnot already included in the research instrumentselected for use. Both of these examples demonstratea rationale for the need to examine specificforms of gambling that may not have beenincluded in the set of games originally associatedwith a particular research instrument.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling114Time Frame & Symptom Clusters forGambling DisordersEstimates of disordered gambling derivetraditionally from lifetime and past-yeartime frames. Absent a time frame, it isdifficult to make sense of a prevalenceestimate. Perhaps just as important, lifetimeestimates of gambling are suspect because researchershave failed to assure that the cluster ofsymptoms and signs necessary to identify a respondentas a disordered gambler coexisted.For example, during the early prevalence researchwith the South Oaks Gambling Screen(SOGS; Lesieur & Blume, 1987), investigatorsdid not assure that the 5 positive signs necessaryto meet the SOGS criteria occurred concurrently.To illustrate, consider a 40-year-old manwho has been gambling for 25 years. It is quitepossible—though admittedly unlikely—thateach positive sign existed in isolation duringseparate life eras (e.g., every 5 years). This manwould be classified according to the SOGS as aprobable pathological gambler, yet would neverhave had more than a single sign or symptomevery five years. Past-year estimates are notsubject to this type of error, since the time frameautomatically requires symptom clustering.Therefore, to secure precise data, researchersmust assure the clustering of lifetime symptomswithin a specified period.Eliminating Other Psychiatric Disorders:Exclusion CriteriaIf researchers fail to exclude psychiatric disordersthat can mimic or influence intemperategambling behaviors, there will be an absenceof precision among the existing estimatesof disordered gambling prevalence(Bland, Newman, Orn, & Stebelsky, 1993; Boydet al., 1984). “There are several problems withall of these [prevalence] studies.... DSM-IIIuses antisocial personality disorder as an exclusioncriterion for pathological gambling. Noneof the studies… appear to have even checked fora diagnosis of antisocial personality disorder,and presumably their prevalence figures includedpeople with this disorder” (Bland et al.,1993, pp. 108-109). Although Bland et al.’simportant comments predated DSM-IV (AmericanPsychiatric Association, 1994)—which excludesthose with manic disorders from receivinga diagnosis of pathological gambler—theydid not correct the exclusionary diagnostic problemsthat they first accurately identified in theirown prevalence research. Future prevalenceresearch must become more precise about distinguishingand then excluding respondents whohave other psychiatric disorders that better explainexcessive gambling from those identifiedas having primary gambling disorders.Sampling Design I: Selecting RespondentsThe sampling design is crucial to the developmentof any household telephonesurvey protocol (Frankel, 1983; Kish,1965; Sudman, 1983). The research protocolmust specify how households and respondentswill be selected (e.g., random digit dialingand most recent birthday), and how sites andindividual respondents will be excluded fromsurvey participation. We encourage investigatorsto employ random digit dialing to minimizethe problems associated with failing to accessunlisted telephone numbers. In addition, McAuliffeet al. (1995) recommend three major principlesfor determining respondent eligibility.Applying these principles will help secure theinternal validity of the survey.1. All eligible people should have the sameprobability of being sampled. For example,households with more than onetelephone line have an increased chanceof being contacted and present problemsfor telephone surveys. Problems of thisnature can be addressed by developingprecise eligibility standards.2. To be eligible, contacted persons must bein a setting that permits administrationof the interview in a manner that will allowvalid data to be collected. Thus, ashared living situation with many residents(e.g., more than five is a reasonablecutoff) who all access a singleHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling115phone line is an example of an inappropriatesetting for a moderately privateinterview that could require 20-40 minutesof uninterrupted time. Too oftenthese settings interfere with data collectiondespite a respondent’s best intentionto comply fully. In addition, sharedliving situations hold more potential forcontaminated data because of socialpressure on the respondent that mightnot be evident to the interviewer.3. For a treatment needs or resource allocationprevalence study, investigatorsshould exclude individuals who couldnot be served by the state's or agency’streatment resources. To illustrate, outof-stateor foreign visitors and militarypersonnel represent two groups of individualswho might utilize emergencyservices but who are not eligible forstate-funded treatment programs. Sincethis principle does not cover every situation,protocols should provide an operationaldefinition of eligibility that interviewerscan apply both easily and consistently(McAuliffe et al., 1995).Since these suggestions represent only a briefset of guidelines, for interested readers moreinformation about inclusion and exclusionguidelines is available from Boyd et al. (1984),Kish (1965), and McAuliffe et al. (1995).Sampling Design II: Power Analysisand Planning for an Adequate Representationof the Population of InterestAll research protocols should include astatistical power analysis (e.g., Cohen,1988; Fleiss, 1981; Kish, 1965). Apower analysis will yield an estimate ofthe number of respondents necessary to achievethe specific research purposes specified in theprotocol. For example, if we expect a pathologicalgambling prevalence rate of 1.5%, then,under typical circumstances, a survey sample of2000 will yield only 30 pathological gamblers.This sample is woefully inadequate for treatmentplanning or other resource allocation purposes,since we would not want to generalizefrom 30 people to the entire population of disorderedgamblers. Therefore, a larger total samplewill be required. To assure that an adequatesample of pathological gamblers is obtained, inthe following section, we recommend an optimalallocation strategy. Before turning to thissection, however, we want to encourage investigatorsand others interested in prevalence researchto consider with care the smallest comparisonsof interest. To illustrate one of themost common power issues that emerges fromthese comparison, consider the following example:If the purpose of a prevalence research projectincludes the need to determine whether disorderedgamblers who play instant lottery gamesdisplay more or less intense symptoms than disorderedgamblers who limit their wagering toextended state lottery games (e.g., weekly drawings),then we will need to compare a sample ofrespondents who represent these two primarygroups. However, if the investigators also areinterested in examining any gender differencesthat might exist between these two groups ofgamblers, than there really are four groups ofinterest (i.e., male and female instant and extendedlottery players). To compare these fourgroups of interest in a 2 x 2 analysis of variancedesign (ANOVA), at the .05 level of statisticalsignificance, assuming a moderate effect size of.25 for the gender, gambling type, and interactioneffects, and a power level of .80, we wouldrequire 33 disordered gamblers in each of thefour groups, or a total of 132 disordered gamblers.However, if the effect size was small instead ofmoderate for these comparisons, then the numberof disordered lottery gamblers required forthis analysis would become dramatically different.For example, to compare the four groups ofinterest in the same analysis described above,but now assuming an effect size of .25 for thegender comparison, .10 for the gambling typecomparison, and .10 for the interaction effects,and a power level of .80, we would now require198 disordered gamblers in each of the fourgroups, or a total of 792 disordered gamblers.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling116To randomly identify 792 adult disordered gamblersusing a general population telephonescreen, assuming the requirement of a 70% orgreater survey response rate and a level 3 gamblingprevalence rate of 1.5%, would requiremore than 72,000 telephone calls taking commonrates of refusals and conversions into consideration.61 This research strategy is likely tobe prohibitively expensive for most fundingsources. Therefore, investigators must consideralternative designs to randomly stratifying andselecting samples to compare the attributes andcharacteristics of disordered gamblers. In thenext section we will consider one of thesestrategies: optimal allocation. However, fornow, the examples above provide a simple illustrationof how the purpose and power analysisof a study can dramatically influence the planningand implementation of a research project.Alternatives to Stratified RandomSampling: Optimal AllocationIf the purpose of the research is to understandthe attributes or clinical needs of disorderedgamblers, we suggest a survey sampling strategythat is different from the traditional stratifiedrandom sampling approach. This strategy doesnot simply concentrate on respondents selectedat random from the general population. Instead,this strategy encourages investigators to focuson selecting respondents who most likely willrepresent disordered gamblers. This approachto respondent selection is optimal allocation(Kish, 1965; McAuliffe et al., 1995). The optimalallocation strategy yields a sampling planthat directly reflects the goals of a disorderedgambling treatment needs assessment project byestablishing a respondent sample that best repre-61 Actually, the number of telephone calls would beconsiderably larger, since the disordered gamblers weare targeting must have either instant or extendedlottery gambling as their main problem. Therefore,we would have to exclude gamblers who have problemswith other forms of wagering (e.g., charitable,casino, pari-mutuel); this would have the effect ofreducing the assumed 1.5% prevalence rate.sents those who would require or demand treatmentservices. This strategy is sensitive to thecentral goals of an objective-driven populationsurvey: to obtain the optimal information aboutrespondents who will qualify for (i.e., meet diagnosticcriteria), and make use of, gamblingtreatment services. Unlike the traditional epidemiologicalapproach to conducting householdsurveys that simply yields an estimate of disorderedgambling among people in the generalpopulation, this research strategy does not simplyidentify rates of people who meet some diagnosticscreen, and then assume that this groupis in need of treatment. Traditional epidemiologicalsurveys disregard the essential treatmentplanning aspect that serves as the centerpiece ofa genuine treatment needs assessment project.Conventional gambling prevalence studies stillhave not addressed the issues associated witheach of the following: (1) whether people actuallywould enter treatment for a gambling disorderif it were available, (2) identifying the obstaclesto entering treatment, for those whowould be eligible to receive such clinical services,or (3) identifying the rates of natural recoveryamong disordered gamblers (i.e., thosewho will not enter treatment but still satisfytheir need to change; e.g., Cunningham, Sobell,Sobell, Kapur, 1995; Sobell, Cunningham, &Sobell, 1996).An optimal allocation sampling strategy is directlyresponsive to social indicators (e.g., bankruptcy,loan defaults, gambling hotline calls,treatment utilization) that point to the severity ofgambling-related problems and suggest the needfor treatment services. This methodology beginswith samples proportional to the populationin each sub-state area and then adjusts the numberof respondents to be sampled as a functionof the estimated prevalence of disordered gambling.Scientists should sample from regionsthat have higher rates of gambling disorder moreheavily than areas where there are indications offewer gambling problems. This approach isquite different from traditional epidemiologicalsurvey strategies that generate simple prevalencerates for gambling problems because thisapproach “purposively” samples the major attributeof interest (i.e., respondents who areHoward J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gamblinglikely to make use of gambling treatment services).62Respondent Selection MethodInvestigators should select respondents forprevalence studies randomly. For example,once a household or “cluster” is contacted in atelephone survey, we recommend that researchteams select respondents randomly by using thenearest birthdate method. Research protocolsshould specify the respondent selection methodin detail. Protocols also should specify how thesurvey will address non-English-speaking respondents.Connection Attempts, Call-Back Procedures,and Conversion of RefusalsWe suggest that survey protocols specify aminimum of eight connection attempts to initiallycontact a respondent and ten call-backs torespondents once they have been contacted. Aspecial effort should be made to call during“off” hours when problem gamblers are morelikely to be home. Previous survey researchdemonstrates that as many as 35% of householdsinitially refuse to participate in telephonesurveys. We also suggest that research protocolsspecify that at least 15% of the initial refuserswill be converted to active study participants.Throughout the survey, the actual numberof calls depends upon the contractor’s successin achieving an acceptable response rate.This criterion for determining an acceptable responserate is determined by the sampling parametersand the need to draw a scientific sample.Sample response rates of less than 50% areunscientific and offer little to our understandingof disordered gambling. As we suggest in thefollowing section, investigators, funding agencies,and administrators, must establish responserates of 70% as a minimum for satisfactoryprevalence research.62 For additional detailed information about optimalallocation procedures, see McAuliffe et al. (1995).117Response and Completion RatesGiven the considerations introduced inthe power analysis and sampling sectionabove, it is important to remember thatpeople with gambling disorders representa very small proportion of the general populationbut compose a major portion of the consumersof state-funded treatment services—where these exist. A well-designed samplingprocedure will increase the number of disorderedgamblers who are surveyed. However,this increase will produce an absolute number ofrespondents that is large enough to support confidentdata analyses and interpretation only ifthe total survey sample is sufficiently large andthe response rate is adequate. Therefore, wesuggest that, for general population studies,5,000 surveys be completed. In this metaanalysis,we examined 106 prevalence estimates,of which 50 were adult general populationstudies; of these adult general populationstudies, the average sample size was only 1,581.This sample size—while acceptable for managingmeasurement error—is insufficient fortreatment planning purposes or understandingthe attributes of disordered gamblers. For example,an inadequate sampling design forced therecently released Connecticut population survey(WEFA Group, ICR Survey Research Group,Lesieur, & Thompson, 1997) study to concludethat, “...there were only 12 individuals in thetelephone survey of the general population whoindicated probable pathological gambling on alifetime basis. To generalize social costs fromsuch a small number of individuals is not possible”(p. 10). “...only 12 persons on a lifetimebasis, 6 of them on a current basis” (p. 6-14).“The present study was unable to produce generalizabledata on the social costs of pathologicalgambling or the demographic profile ofpathological gamblers. A survey of at least6,000 adults is required in order to obtain reliableinformation on these topics” (p. 11). Carefulplanning will help investigators avoid spendingvaluable resources and then being forced toconclude that they could not deliver critical informationabout the specific topic under investigation.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling118To achieve the suggested goal of 5,000 completedsurveys, investigators should plan to contactat least 7,143 households. This representsan approximate response rate of 70% and shouldbe considered as minimally acceptable for scientificanalysis. We suggest that every investigatorrequire at least a 70% response rate fromtheir survey data collection team. While thisrate applies to the overall study response rate,this rate of completion also should be achievedin all sub-state areas (i.e., sampling strata)(McAuliffe et al., 1995). In areas where respondentsare difficult to contact, the number ofcall-backs should be increased (e.g., to 15) untilthe target completion rate—as determined bythe optimal allocation sampling design andpower analysis—is obtained.Calculating Response and CompletionRatesThis meta-analysis identified a variety of problemsassociated with the calculation of responserates for general population telephone surveys.Frankel (1982) describes the most widely acceptedstandard for calculating response rates.The general formula for calculating a responserate is as follows:Response rate =# of respondents participating in the studytotal # of respondents eligible to participate in the studyIn this meta-analysis, we observed a very commoncalculation problem: many investigatorsused an improper denominator; that is, they improperlycalculated the total number of peopleeligible to participate in their studies. For example,investigators often inflated their responserates by deleting randomly selected householdsthat failed to answer the phone from the denominatorof their response ratio. This methodis incorrect. All eligible random numbers reflectitems that must be included in the denominator.When a household fails to answer, theonly way investigators can increase the responserate is to increase the number of call-backs andvary the time of day these call-backs are made.Calculation of response rates for multi-stagesampling protocols can be simplified with theuse of the following guideline: when there aremultiple stages to a survey protocol (e.g., selectionof sites, then selection of respondents fromthe selected sites), investigators should calculatea completion rate for each stage. Investigatorscan then calculate the final response rate for thestudy by multiplying these completion rates.Investigators who are conducting householdtelephone surveys can consider the selection ofhouseholds to be the first stage of sampling andthe selection of respondents from these householdsto be the second stage of sampling. Usingthis model, the ratio of calls answered to eligiblehouseholds represents the site completion rate.The ratio of participating respondents to eligiblehouseholds contacted represents the respondentcompletion rate. The response rate for the entirestudy is calculated by multiplying the sitecompletion rate by the respondent completionrate (Frankel, 1982).To illustrate these calculations, readers shouldconsider the following hypothetical set of randomlyselected household telephone numbersthat correspond to a particular jurisdiction. Thisset was selected in accordance with the objectivesof a household survey and a power analysisthat considered the comparisons of interest. 63As the discussion above describes, the responserate for this illustrative project is calculatedbased upon two different levels of analysis: (1)the household (site) completion rate; and (2) thewithin-site respondent completion rate. Imaginea sample of 1,000 telephone numbers representinghouseholds with the following rates of participation:800 homes answered; 100 homes didnot answer; 100 telephone numbers were disconnectedor not valid for the purposes of thestudy (e.g., these numbers represented fax linesor business organizations); the 100 invalid sites63 This particular example reflects a survey protocolin which one respondent per household is surveyed.Other protocols could require multiple respondents tobe sampled in each household (e.g., in parent-childsurveys).Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling119were eliminated from the set of eligible homesand were replaced with 100 new randomly selectedtelephone numbers from the appropriatepool of numbers. Of these 100 new households,70 answered and 30 did not answer. Thus, thefirst-stage (i.e., site) completion rate was calculatedas follows:Site completion rate =(800 + 70) participating sites1,000 total eligible sites= 0.87For the second stage, 700 respondents completedthe survey, 100 eligible participants refusedto participate, and 70 agreed to participatebut were unable to complete the survey accordingto the established methodology or were unableto re-schedule the survey within the timelimits of the study. Thus, the second-stage (i.e.,within-site respondent completion rate) was calculatedas follows:Respondent completion rate =700 participating respondents870 eligible respondents within participating sites = 0.80Finally, as a result of these operations, the dataset in this hypothetical prevalence study representsan overall response rate of (0.87)*(0.80) =0.70, or 70%.Although we have illustrated the calculation ofthe response rate using an example of a householdtelephone survey, this method applies toother sampling designs as well, including theselection of respondents from treatment programsand schools.Interviewer Selection and TrainingDisordered gambling survey protocolsshould include a section that specifieshow the interviewers who conduct thesurvey will be trained and monitored.In addition, details about pre-testing are essentialto the development of a quality survey protocol—evenif it rests in large part on instrumentspreviously used in field studies. Finally,the interviewer selection and training part of theresearch protocol should include a considerationof how issues of ethnicity and knowledge ofgambling and addiction treatment modalitieswill be addressed during the interviewer trainingactivities.Data Coding and EntryResearch protocols should specify howdata will be coded and entered. Thisinformation is vital to data integrity.Computer-Aided Telephone Interviews(CATI) result in data that is coded and enteredin a single step. The quality of data should becarefully monitored. We suggest that a researchmonitor observe a minimum of 10% of the interviews,selected at random, complete a codedinterview, and compare it to the CATI interviewto check for accuracy. The comparison betweenthe monitored interviews and the CATI datashould yield an error rate of less than 1%. Finally,provisions must be made that assure confidentialityand anonymity of respondent data.Data AnalysisThe survey protocol should anticipate howinvestigators will analyze the obtaineddata. A variety of descriptive and multivariatetechniques will provide stateplanners with a range of indicators that will assiststatewide and sub-state planning activitiesso long as an adequate sample of qualified respondentswas obtained during the data collectionprocess (i.e., optimal allocation acrossstatewide treatment planning regions). Whensmaller samples are available, scientists shouldconsider non-parametric statistics as the tools ofchoice. Research protocols should not simplylist a variety of statistical instruments as the analyticchoices. Data sets should be diagnosed andevaluated for the proper statistical “fit.” Investigatorsshould provide a rationale for using specificstatistical devices and be certain that theirobtained data set matches the requirements ofthese statistics.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling120Applying the ResultsTo ensure that a survey protocol includesthe essential sampling and design elementsnecessary to meet each investigator'sunique data needs, we suggest thatthe proposed survey protocol consider how thestudy results will be applied once the analyzeddata is obtained. What are the critical treatmentplanning decisions that face health planners?Who will be involved in the development ofnew gambling treatment programs? Which jurisdictions(i.e., level of analysis) will use thesurvey results for treatment planning? Each ofthese and other key questions will assist healthplanners to make key sampling decisions thatwill assure adequate respondent samples in keysub-state or respondent attribute strata.State and Sub-state Approaches forTreatment PlanningWithin any state, prevalence data for treatmentplanning will be most useful if it is categorizedby county, city, or town, depending upon clinicaland treatment planning considerations.Linking the survey results with state and substatesocial indicator data is particularly important.These goals should be specified in the protocolso that appropriate sampling strategies canbe devised to assure researchers that an adequategroup of respondents will be contacted ineach sampling stratum. More information aboutestimating sample size can be found in the previoussections on power analysis and optimalallocation.SummaryThis brief guide provided only the mostbasic architecture necessary for investigatorsto develop quality research protocolsfor the conduct of disordered gamblingprevalence research. In this guide, weemphasized the use of prevalence research fortreatment planning or resource allocation. Administrators,policy makers, and legislative bodiescan use this guide during the development ofrequests for proposals and contract negotiationswith potential research vendors. In addition,this guide can help non-scientist consumers ofresearch better understand the basic elements ofresearch design for estimating the prevalence ofdisordered gambling.ReferencesAmerican Psychiatric Association. (1994).DSM-IV: Diagnostic and statistical manualof mental disorders. (Fourth ed.). Washington,D.C.: American Psychiatric Association.Bland, R. C., Newman, S. C., Orn, H., & Stebelsky,G. (1993). Epidemiology of pathologicalgambling in Edmonton. Canadian Journal ofPsychiatry, 38, 108-112.Boyd, J. H., Burke, J. D., Gruenberg, E., Holzer,C., E., Rae, D., S., George, L. K., Darno, M.,Stoltzman, R., McEvoy, L., & Nestadt, G.(1984). Exclusion criteria of DSM-III: a studyof co-occurrence of hierarchy-free syndromes.Archives of General Psychiatry,41(October), 983-989.Cohen, J. (1988). Statistical power analyses forthe behavioral sciences. (second ed.). Hillsdale,New Jersey: Lawrence Erlbaum.Cunningham, J. A., Sobell, L. C., Sobell, M. B.,& Kapur, G. (1995). Resolution from alcoholproblems with and without treatment: reasonsfor change. Journal of Substance Abuse, 7(3),365-72.Fleiss, J. L. (1981). Statistical methods for ratesand proportions. (Second ed.). New York:John Wiley & Sons.Frankel, J. R. (1982). On the definition of responserates: A special report of the CASROtask force on completion rates. Port Jefferson,NY: The Council of American SurveyResearch Organizations.Frankel, M. (1983). Sampling theory. In P. H.Rossi, J. D. Wright, & A. B. Anderson (Eds.),Handbook of Survey Research (pp. 21-67).New York: Academic Press, Inc.Kish, L. (1965). Survey sampling. New York:John Wiley & Sons, Inc.Lesieur, H. R., & Blume, S. B. (1987). TheSouth Oaks gambling screen (SOGS): A newinstrument for the identification of pathologicalgamblers. American Journal of Psychiatry,144(9), 1184-1188.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt


Prevalence of Disordered Gambling121McAuliffe, W. E., LaBrie, R. A., Mulvaney, N.,Shaffer, H. J., Geller, S., Fournier, E. A., Levine,E. B., Wang, Q., Wortman, S. M., &Miller, K. A. (1995). Assessment of substancedependence treatment needs: A telephonesurvey manual and questionnaire (RevisedEd.). Cambridge, MA: National TechnicalCenter on Substance Abuse Needs Assessment.Rose, G., & Barker, D. J. P. (1978). What is acase? Dichotomy or continuum? BritishMedical Journal, 2, 873-874.Shaw, G. B. (1911). The doctor’s dilemma:with preface on doctors. New York: Brentano’s.Sobell, L. C., Cunningham, J. A., & Sobell, M.B. (1996). Recovery from alcohol problemswith and without treatment: prevalence intwo population surveys. American Journal ofPublic Health, 86(7), 966-72.Sudman, S. (1983). Applied sampling. In P. H.Rossi, J. D. Wright, & A. B. Anderson (Eds.),Handbook of Survey Research (pp. 145-194).New York: Academic Press.Vaillant, G. E., & Schnurr, P. (1988). What is acase? A 45-year study of psychiatric impairmentwithin a college sample selected formental health. Archives of General Psychiatry,45(4), 313-319.WEFA Group, ICR Survey Research Group,Lesieur, H., & Thompson, W. (1997). A studyconcerning the effects of legalized gamblingon the citizens of the state of Connecticut :State of Connecticut Department of RevenueServices, Division of Special Revenue.Howard J. Shaffer, Matthew N. Hall, & Joni Vander Bilt

More magazines by this user
Similar magazines