Ourprimaryinterestisonthelong-termeffectsofthecivilianeconomyonearnings,wealth,andretirementdecision.However,recallthattheliteraturesuggeststhatindividualsmaybeabletotimetheirentranceintothelabormarketthroughobtainingincreasedlevelsofeducationandnotethatTable1Asuggeststhatveteranswhoenterthecivilianlabormarketinperiodsofhighunemploymentareslightlymorelikelytocompletecollegeandaresomewhatlesslikelytobemarriedorliveincoupledhouseholds.Therefore,westartbyinvestigatingtheeffectoftheunemploymentrateatthetimeofexitfromactiveserviceoneducationattainmentandmaritalstatus.Weinvestigatetheeffectsontheseoutcomesbecausetheyareimportantdeterminantsofearnings,wealthvariablesandretirementdecisionsinthelong-run.Forthecaseofeducationattainmentandthatofmaritalstatus,weestimatethefollowingmultinomiallogitmodel:
\begin{equation}
Pr\left(Y_{i}=k\right)=\frac{e^{\beta_{UR,k}UR_{i}+\beta_{X,k}X_{i}}}{\sum_{j=1}^{J}e^{\beta_{UR,j}UR_{i}+\beta_{X,j}X_{i}}}\nonumber \\
\end{equation}
()
Thesubscript\(i\)indexespeople(here,veterans).Foreachveteran,weusedthefirstobservationavailableintheHRS.Thevariable\(Y_{i}\)denotestheoutcomeofinterest(educationalattainment,maritalstatus),and\(k\)isoneofthe\(J\)potentialcategories.Thevariable\(\text{\ U}R_{i}\)istheprevailingunemploymentrateatthetimeofleavingactiveservice.Thevector\(X_{i}\)areothercontrolsincludingage,gender,race,Hispanicethnicity,ageatenlistment,whetherenlistedduringaconflict,unemploymentrateatenlistment,andwhethertherespondenthasadisabilityrelatedtoactiveservice.Themodel(andallmodelsthatweestimate)alsoincludesindicatorsofyear.Inaddition,wealsoestimatethelogisticmodeldescribedinequation(2),forwhethertheveteranlivesinacoupledhousehold(usingagainonlythefirstavailableobservationperindividual).
\begin{equation}
Pr\left(\text{Coupled}_{i}=1\right)=\frac{e^{\theta_{\text{UR}}UR_{i}+\theta_{X}X_{i}}}{1+e^{\theta_{\text{UR}}UR_{i}+\theta_{X}X_{i}}}\nonumber \\
\end{equation}
Ineachcase,wealsoestimateaversionofthemodelthatutilizesa2-stageleastsquares(2SLS)methodologydiscussedabove.Theestimationresults(marginaleffects)areshowninTable2.ConsistentwiththedescriptivestatisticsinTable1A,wefindthatveteranswhoreenteredcivilianlifeduringperiodsofhighunemploymentaremorelikelytohaveobtainedacollegeeducation.Ourestimationsindicatethat1percentage-pointincreaseintheunemploymentrate(intheyearofreenteringthecivilianlife)increasestheprobabilityofhavingacollegedegree(orhighereducation)by1.4percentagepoints;our(preferred)2SLSresultsindicateasubstantiveeffectsizeof3.0percentagepoints(roughly15percent).Althoughwecannottestitdirectlyinourdata(becausewedonothaveyearsofenrollmentandgraduationfromcollege),alikelyexplanationforthisfindingisthatveteranswhoexitactiveserviceinperiodsofhighunemploymentaremorelikelythanotherstoenrollincollege;veteransduringtheerascoveredintheHRSgenerallyhadaccesstoeducationbenefitsuponleaveingactiveduty.AsimilarbehaviorwasfoundforindividualsgraduatingfromhighschoolinperiodofhighunemploymentbyArkes(2003)andbyHershbein(2012).
Table2:RegressionResults:ProbaabilityofCompletingCollege;ProbabilityofBeingMarried:Effectof1ppincreaseinunemploymentrateattimeofleavingactiveservice