<strong>The</strong> Impact <strong>of</strong> Government-SponsoredTra<strong>in</strong><strong>in</strong>g Programs on the Labor Market Transitions <strong>of</strong> Disadvantaged Men<strong>The</strong> Impact <strong>of</strong> Government-Sponsored Tra<strong>in</strong><strong>in</strong>g Programs on the Labor Market Transitions <strong>of</strong> Disadvantaged Men 349recipients that experienced a spell at any time between 1987 <strong>and</strong> 1993 <strong>in</strong> the prov<strong>in</strong>ce <strong>of</strong>Québec, Canada. To be <strong>in</strong>cluded <strong>in</strong> the sample, <strong>in</strong>dividuals had to be aged 18 or 19 at anytime dur<strong>in</strong>g that period <strong>and</strong> to have less than a high-school degree. Sample stratification isused to avoid over-parameterization <strong>of</strong> the statistical model that would result if too manyexogenous variables had to be controlled for.By merg<strong>in</strong>g various adm<strong>in</strong>istrative data files we can recreate complete <strong>in</strong>dividuals’ historieson the labour market back to age 16, the legal school-leav<strong>in</strong>g age <strong>in</strong> Canada. Consequently,each <strong>in</strong>dividual <strong>in</strong> our sample is necessarily observed <strong>in</strong> the OLF state at the beg<strong>in</strong>n<strong>in</strong>g <strong>of</strong>his history. This sampl<strong>in</strong>g scheme thus removes the necessity to control for stock samplebiases <strong>and</strong> has the additional benefit <strong>of</strong> provid<strong>in</strong>g rich transition patterns over a relativelylong sample frame.<strong>The</strong> econometric model is built on cont<strong>in</strong>uous labour market transitions processes <strong>and</strong>allows entry rates <strong>in</strong>to each state to depend on observed <strong>and</strong> unobserved heterogeneitycomponents. Heterogeneity terms can be dest<strong>in</strong>ation-specific, orig<strong>in</strong>-specific or both. In allcases, correlation across heterogeneity terms is allowed. We further <strong>in</strong>vestigate the sensitivity<strong>of</strong> the parameter estimates to various distributions <strong>of</strong> the heterogeneity components. Whenparametric distribution functions are used, the model is estimated by Simulated MaximumLikelihood (SML).<strong>The</strong> rema<strong>in</strong>der <strong>of</strong> the chapter is organized as follows. Section 2 provides a detailed description<strong>of</strong> the data. Section 3 discusses the econometric model <strong>and</strong> the various statistical assumptionregard<strong>in</strong>g the distributions <strong>of</strong> the heterogeneity terms. Section 4 reports our empiricalf<strong>in</strong>d<strong>in</strong>gs. Section 5 concludes the chapter.2. Data description<strong>The</strong> basic data used for this study are drawn from the caseload records <strong>of</strong> Québec’s M<strong>in</strong>istèrede la Solidarité sociale. <strong>The</strong> files conta<strong>in</strong> <strong>in</strong>formation on all <strong>in</strong>dividuals who have receivedwelfare benefits at some time between January 1987 <strong>and</strong> December 1993. In particular, thestart dates <strong>and</strong> end dates <strong>of</strong> each welfare <strong>and</strong> welfare tra<strong>in</strong><strong>in</strong>g spells are recorded <strong>in</strong> the files.<strong>The</strong> welfare program conta<strong>in</strong>s special provisions for those who are <strong>in</strong>disposed for work due tomental or physical impediments. <strong>The</strong>se <strong>in</strong>dividuals are not <strong>in</strong>cluded <strong>in</strong> the sample. Thus thef<strong>in</strong>al sample comprises only <strong>in</strong>dividuals who have no h<strong>and</strong>icap or only a m<strong>in</strong>or, <strong>in</strong>termediate,or temporary physical h<strong>and</strong>icap. Furthermore, they are fit to work.<strong>The</strong> welfare adm<strong>in</strong>istrative files conta<strong>in</strong> no <strong>in</strong>formation on employment or unemploymentspells. Our sample was thus l<strong>in</strong>ked to the Status Vector files (SV) <strong>and</strong> the Record <strong>of</strong>Employment (ROE) files, both under the aegis <strong>of</strong> Human Resources Development Canada.<strong>The</strong>se files conta<strong>in</strong> very detailed weekly <strong>in</strong>formation on <strong>in</strong>sured unemployment spells <strong>and</strong>employment spells, respectively. <strong>The</strong> start dates <strong>and</strong> end dates <strong>of</strong> each spell are recorded<strong>in</strong> these files. Similar <strong>in</strong>formation is available with respect to tra<strong>in</strong><strong>in</strong>g spells adm<strong>in</strong>isteredunder the Unemployment Insurance (UI) program. Merg<strong>in</strong>g all three adm<strong>in</strong>istrative filesallows us to def<strong>in</strong>e seven different states on the labour market. Aside from the welfare,
50 Advances <strong>in</strong> Econometrics - <strong>The</strong>ory <strong>and</strong> Applications4 Will-be-set-by-IN-TECHunemployment <strong>and</strong> employment states, we can identify two separate welfare tra<strong>in</strong><strong>in</strong>g states<strong>and</strong> one unemployment tra<strong>in</strong><strong>in</strong>g state. 6<strong>The</strong> focus <strong>of</strong> this chapter is on poorly educated young men. Thus to be <strong>in</strong>cluded <strong>in</strong> the sample,an <strong>in</strong>dividual had to be either 18 or 19 years <strong>of</strong> age at any time between 1987 <strong>and</strong> 1993 <strong>and</strong> havecompleted less than 11 years <strong>of</strong> school<strong>in</strong>g over the sample period. A high-school degree <strong>in</strong>Québec usually entails at least 12 years <strong>of</strong> school<strong>in</strong>g. In pr<strong>in</strong>ciple, then, none <strong>of</strong> the <strong>in</strong>dividuals<strong>in</strong> our sample has earned a high-school diploma. With these selection criteria the f<strong>in</strong>al sampleconta<strong>in</strong>s 3068 <strong>in</strong>dividuals.<strong>The</strong> upper panel <strong>of</strong> Table 1 provides summary statistics for <strong>in</strong>dividuals who have notparticipated <strong>in</strong> a tra<strong>in</strong><strong>in</strong>g program. <strong>The</strong> lower panel presents similar statistics for programparticipants. In the latter case, the mean durations <strong>in</strong> either employment, unemployment orwelfare are calculated both before <strong>and</strong> after tra<strong>in</strong><strong>in</strong>g. An exam<strong>in</strong>ation <strong>of</strong> the table revealsthat the two groups are very similar <strong>in</strong> terms <strong>of</strong> their observable characteristics; <strong>The</strong>y bothhave the same average age <strong>and</strong> nearly identical school<strong>in</strong>g levels. Yet, there are significantdifferences <strong>in</strong> their respective labour market experiences. For <strong>in</strong>stance, non-tra<strong>in</strong>ees havelonger spells <strong>in</strong> each <strong>of</strong> the three states reported <strong>in</strong> the table. On the whole, the proportion <strong>of</strong>time non-tra<strong>in</strong>ees spend employed is slightly larger than that <strong>of</strong> tra<strong>in</strong>ees prior to tra<strong>in</strong><strong>in</strong>g. Onthe other h<strong>and</strong>, once they have had tra<strong>in</strong><strong>in</strong>g, the proportion <strong>of</strong> time tra<strong>in</strong>ees spend employedbecomes larger than that <strong>of</strong> non-tra<strong>in</strong>ees. This <strong>in</strong>crease stems from the fact that the averageemployment duration decreases proportionately less that the average duration <strong>of</strong> welfare <strong>and</strong>unemployment spells. Taken at face value, this would suggest tra<strong>in</strong><strong>in</strong>g programs benefitsomewhat to welfare recipients.Recall that only <strong>in</strong>dividuals who experienced a welfare spell between 1987 <strong>and</strong> 1993 <strong>and</strong> whowere aged 18 or 19 dur<strong>in</strong>g that period are <strong>in</strong>cluded <strong>in</strong> the sample. Those who are 18 or 19 years<strong>of</strong> age <strong>in</strong> January 1987 may have already been on the labour market for 2–3 years at most. Inorder to recreate their complete labour market histories as <strong>of</strong> the age <strong>of</strong> 16, it is necessary <strong>in</strong>some cases to go back as early as January 1984. 7 <strong>The</strong> start date <strong>and</strong> end date <strong>of</strong> each spell isused to create <strong>in</strong>dividual histories on the labour market. Overlaps between states are frequent<strong>and</strong> are not necessarily the result <strong>of</strong> cod<strong>in</strong>g errors. It may well be, for example, that a welfarespell <strong>and</strong> a work spell overlap. Program designs do not forbid this. In pr<strong>in</strong>ciple, such overlapscould be redef<strong>in</strong>ed as a separate state. Given the number <strong>of</strong> possible states, it is simply not6 <strong>The</strong> welfare files conta<strong>in</strong> <strong>in</strong>formation dat<strong>in</strong>g back to 1979 <strong>and</strong> end<strong>in</strong>g <strong>in</strong> December 1993. <strong>The</strong> SV filesconta<strong>in</strong>s <strong>in</strong>formation beg<strong>in</strong>n<strong>in</strong>g <strong>in</strong> January 1987 <strong>and</strong> end<strong>in</strong>g <strong>in</strong> December 1996. F<strong>in</strong>ally, <strong>The</strong> ROEfiles conta<strong>in</strong> <strong>in</strong>formation rang<strong>in</strong>g from January 1975 to December 1996. <strong>The</strong> analysis focuses on the1987–1993 period due to data limitations.7 Data concern<strong>in</strong>g unemployment spells are available only as <strong>of</strong> January 1987. Consequently, a smallproportion <strong>of</strong> unemployment spells occurr<strong>in</strong>g prior to 1987 may be wrongly coded as out <strong>of</strong> the labourforce (OLF). Two factors lead us to believe that the proportion <strong>of</strong> such spells is likely <strong>in</strong>significant. First,the large majority <strong>of</strong> <strong>in</strong>dividuals who were 18 or 19 years <strong>of</strong> age <strong>in</strong> the years 1990 <strong>and</strong> beyond where<strong>in</strong> the OLF, the employment or the welfare states between 16 <strong>and</strong> 19. Second, <strong>of</strong> those <strong>in</strong>dividuals, themajority who had an employment spell would not have qualified for UI benefits given the eligibilityrules that prevailed between 1984 <strong>and</strong> 1987.A similar problem arises with respect to employment spells. Indeed, spells that were ongo<strong>in</strong>g <strong>in</strong>December 1993 will not show up <strong>in</strong> the ROE files until they are term<strong>in</strong>ated. To avoid misclassify<strong>in</strong>gthese spells as OLF, the ROE files are searched as late as December 1996. Given the average length<strong>of</strong> employment spells reported <strong>in</strong> Table 1, it is very unlikely that many employment spells that wereongo<strong>in</strong>g <strong>in</strong> December 1993 will still be ongo<strong>in</strong>g as late as December 1996, <strong>and</strong> thus wrongly classifiedas OLF.