ORGANIZING AND INTERPRETING THE DATAequity <strong>and</strong> use, in figures <strong>and</strong> graphical forms,as <strong>and</strong> when needed. It is also possible to showprogress over time as the number of individualcommunity projects increases. Updating onlyrequires new information on current conditions,not historical analysis.Using aggregated scores from sub-indicators onsustained or effectively used services (e.g. asstacked bar charts, as shown in Figure 8),managers can rank individual communities intotop <strong>and</strong> bottom performers. Ranking givesmanagers <strong>and</strong> staff insight into the characteristicsof both groups. It helps them to bring communitiestogether for horizontal learning <strong>and</strong> to decidewhere to focus limited resources.Comparing community performance on individual<strong>and</strong> aggregated scores also helps identifyinteresting findings for case studies, especiallywhen combined with qualitative information.In situations where MPA is used for more researchfocusedpurposes, such as investigatingrelationships between factors or for comparingclusters of communities on specific aspects,statistical tests appropriate for ordinalmeasurements can be applied, providedconditions for statistual testing are satisfied. Thismeans ensuring that appropriate sample selectionprocedures have been followed, the correctstatistical test is selected, <strong>and</strong> sufficiently smallmargins of error are chosen to conclude that thedifferences or associations found are significant(that is, not due to chance). 17 For hypothesestesting, a 95 percent level of confidence isconsidered sufficient for most social research <strong>and</strong>monitoring purposes, although MPA data haveoften produced results significant at a higher (99percent) level of confidence, i.e., with only 1percent probability that the tested relationshipfound is not true.Statistical testing for associations helps managersto see whether certain clusters of factors, as wellas individual factors, tend to occur together. MPAdata have been used in this manner to answerquestions such as:●●Are better-sustained services also better usedby all? If not, are non-users a mix of differenttypes of households or are certain groups,such as poor households, systematicallyexcluded?Are higher user payments for operation <strong>and</strong>maintenance associated with greater choiceavailable to users (for service levels or for modesof financing, or choice by more groups in thecommunity), more equitable tariff systems,provision of training on financialmanagement, <strong>and</strong>/or better accountability tousers?To test for significant correlations, non-parametrictests are used because most scales used in theMPA are ordinal in nature. The strength of theassociation is measured through the correlationcoefficient. As the name indicates, a correlationmerely says that the two related aspects “co-relate,”or “occur together.” It does not say whetherincrease in scores for one aspect leads to asimultaneous increase in scores for the relatedaspect. Clear causal relationships are hard toestablish using ordinal data, since suchrelationships may only be tested using parametrictests such as multiple regression analysis or factoranalysis - which are applicable at higher levels ofmeasurement <strong>and</strong> require much larger samplesizes than is usual with MPA applications.17 For a fuller explanation readers can refer to st<strong>and</strong>ard textbooks for behavioral <strong>and</strong> social research such asResearch Methods <strong>and</strong> Anthropology: Qualitative <strong>and</strong> Quantitative Approaches by H. Russel Bernard. 1995.Altamira Press.60
ORGANIZING AND INTERPRETING THE DATASupplementing non-parametric tests of associationwith qualitative explorations of communityperceptions of causality can provide projectmanagers with sufficient insights for action. Tocite an example from the global study, bettergender balance in the village water <strong>and</strong> sanitationcommittees was found to be significantly associatedwith higher overall sustainability scores. Qualitativedata from focus groups explained that whencommunity service management organizations aregender-balanced, more management training cango to women, enabling them to exert greaterinfluence on service management, thus makingthe water service more responsive to their dem<strong>and</strong>s<strong>and</strong> therefore more satisfying to the principal users(who are usually women in the community).Satisfied users value the service <strong>and</strong> therefore tendto pay user fees regularly, thus improving theadequacy of financing, which makes the servicemore sustainable .It can also be investigated if such relationshipscontinue to hold when controlling for otherpossible influencing factors, e.g., the level ofdevelopment in the communities concerned orthe complexity <strong>and</strong> age of the installed systems.Through building a model, it is possible to seewhich factors are most crucial in obtaining thedesired results. However, such analyses can onlybe done when the sample <strong>and</strong> data meet certainrequirements (see any methodological orguidebook on statistical analysis in the socialsciences).Insight through statistical analysis is attractive forprogram managers, policymakers <strong>and</strong>researchers. It requires the presence of asociologist, economist or statistician experiencedin the use of non-parametric statistics at theprogram level. Her or his main functions are tosupervise data entry, analyze frequencies <strong>and</strong>cross-tabulations, decide on the statistical test(s),test the strengths of the associations amongindividual influencing factors, <strong>and</strong> between thosefactors <strong>and</strong> the achieved results in terms of howservices are sustained <strong>and</strong> used.61