Will the real nonparticipants please stand up?

PowerPoint Presentation

Will the real nonparticipants please stand up?Exploring selection bias and treatment contamination in employment programsNora Wikoff, MSLS, MSWagendaReentry context in New York StateKnowledge base: work, finances, and crimeAims and research questionsResearch findingsPractice and research implicationsOngoing and future research

reducing the madness OF AN INCARCERATION SOCIETY*25% decline in NY prison populationDeclining crime rates in NYCRockefeller drug laws reformCuomo: reduce fiscal burden on the state11 recent, 4 planned prison closingsReinvest in preventive and rehabilitative services*Cuomo, 2014Prison reduction strategiesCouncil on Community Re-Entry and ReintegrationState council to help nonviolent offenders$5 million proposed FY14 investment in programs:Workforce Investment Board: Oneida CountyTASC Case management and reentry servicesStrategies include job training and supported work

Jobs program logic modelServicesLife skillsTransitional jobsJob coachingJob developmentSupportive servicesIncreased EmploymentIncomeSoft skillsWork readinessStabilityIncreasedEmploymentJob retention

Reduced recidivismReduced recidivismAdapted from Redcross et al. (2012)Weak evaluation supportPrograms do not appear to improve work outcomes or reduce crimeExperimental studies: Most jobs programs show modest or null effects Positive effects observed among: Older former prisonersHigh-risk prisonerslabor force participation Before prisonRising unemploymentHigh levels of work instability High job turnover After releaseInitial boost in formal employmentRates decline to pre-prison levels within three years

Labor force nonparticipationLow opportunity cost to crimeLimited job options availableLimited formal work experienceWeak labor market skillsHigh opportunity cost to formal employmentLow hourly wage, reduced leisure timeGarnished wages (e.g., child support, legal debts)Risk detection at workplace

Research questions, Part 1Do respondents exhibit distinct arrest trajectories before entering prison?Do participants differ from nonparticipants along prior arrest trajectories?Do employment programs improve mens post-release employment and recidivism outcomes?Research questions, Part 2Is labor force non-participation associated with increased recidivism risk?Is labor force participation associated with higher quality employment?What factors break the association between employment and reduced offending?Research aimsExamine whether evaluation findings reflectMens selection into employment servicesContamination from participation in similar programsExamine whether effects persist after controlling for Prior criminal recordWork experienceParticipation in programs that offer overlapping contentSerious and violent offender reentry initiative (SVORI)Target Population: Adult male prisoners under 35 years old Convicted of violent or serious drug offensesStates designed services to fit local contextIntent-to-treat designPropensity score weights: SVORI service receiptdata sourcesFBI National Crime Information CenterLifetime adult arrest recordsSpanning state lines and agency reporting systems

SVORI Evaluation baseline interviewsConducted in prisons during month before releaseDemographics, background, criminal history, prison experiences, physical/mental healthDescriptive statistics (N = 1,575)VariableDefinitionN / M% / SDAgeAt release from prison29.6(7.3)Sentence lengthTime served by date of release2.6(2.6)EducationLess than high school diploma63340.2High school diploma22814.5GED45629.0Trade certificate, Some/more college25616.3Racial/ethnic statusAfrican American87255.4Hispanic, Multi-racial, Other18611.8White51532.7EmploymentWorked last 6 months before prison1,04066.1Ever held job for 2/more years49638.1Job terminationFired from one job31626.4Fired from more than one job33928.4Trajectory modelPredictor variables, final model:Age at each arrest: Linear and squared termsIndicator of arrests during 10 years before SVORI termState indicator for prison siteAge at release from prisonLifetime adult arrest record: Natural log transformationOutcome variable:Predicted probability of group membershipPre-svori arrest trajectories

Probability of arrest each year Age at arrestHigh (38.9%)Middle (45.1%)Low (16.0%)Criminal history: Trajectory groups(High) n = 616(Mid.) n = 706(Low) n = 253N / M% (SD)N / M% (SD)N / M% (SD)Age at first arrest***15.4(3.4)16.3(4.5)18.4(8.2)Lifetime arrests***22.7(13.5)11.3(6.5)3.2(2.1)Lifetime convictions***6.2(5.5)5.0(4.5)2.8(3.3)Times in prison***1.9(1.6)1.2.(1.3)0.7(1.1)SVORI term:Age at release***28.8(6.5)30.3(7.3)29.6(8.8)Sentence length***2.0(1.6)2.6(2.5)3.9(4.1)Parole violation12424.516427.84220.2Violent offense***20233.128640.615662.2Property offense*16426.816823.94417.5Drug offense***26843.923433.23915.5Demographics: trajectory groups(High) n = 616(Mid.) n = 706(Low) n = 253N / M% (SD)N / M% (SD)N / M% (SD)Education ***Less than HSD29848.526337.37228.5High school diploma7912.910815.34116.2GED17428.319127.19136.0Trade school/some coll.6310.214420.44919.3Race***African American39464.237653.310240.3Hispanic, multi-racial, other569.18512.04517.8White16426.724534.710641.9Worked before prison***36358.950070.817770.2Key findingsParticipation: trajectory groups(High) n = 616(Mid.) n = 706(Low) n = 253Percentages engaged in each type of employment-focused programParticipation in each type of program49.758.161.3Education programs (e.g., GED, literacy, college classes)40.746.654.2Work readiness or job training programs19.322.524.5Prison job (work release or prison industry)6.38.68.7Participation in more than one program7.39.312.0Propensity score matchingMultilevel logistic regression model Stata xtmelogit SVORI treatment condition, Prison site (state)Individual-level characteristicsMatching techniquesStata psmatch2Radius matching with caliper (.2 SD/ln odds)Common support condition, ties permittedCriminal history: participationUnm.TotalNon.Emp.Educ.Job2691,304658152515102Age at first arrest***16.116.3*16.616.515.617.6Lifetime arrests***2.32.3***2.52.22.22.4Lifetime convictions***5.15.15.34.84.95.4Times in prison***1.31.5***1.71.31.21.5SVORI term:Age at release***28.5**29.8***30.928.528.131.5Sentence length***4.3***2.2***1.92.52.42.6Parole violation15.2***25.327.121.721.628.4Violent offense***53.3***38.636.439.743.130.7Property offense*25.223.824.425.822.719.8Drug offense***26.3**36.336.039.134.538.6Demographics: trajectory groupsUnm.TotalNon.Emp.Educ.Job2691,304658152515102Education *******Less than HSD47.638.737.224.345.026.5High school diploma13.414.719.611.27.613.7GED24.529.926.135.534.838.2Trade school/some coll.14.516.617.128.912.621.6Race******African American51.756.258.153.955.542.2Hispanic, multi, other20.410.08.811.811.89.8White27.933.733.134.232.648.0Worked before prison***56.3***68.167.871.166.284.3Prison site: participationUnm.TotalNon.Emp.Educ.JobState site ***2691,304658152515102Iowa8.111.29.729.610.98.8Indiana4.811.013.45.310.32.0Kansas7.03.85.20.72.91.0Maryland11.816.521.48.610.518.6Missouri14.83.22.61.33.92.9Nevada9.29.38.514.510.72.9Ohio11.83.92.65.35.82.0Oklahoma5.95.74.76.66.26.9Pennsylvania4.87.96.45.98.721.6South Carolina11.423.722.319.726.028.4Washington10.33.83.22.64.14.9State profilesNonparticipants: Indiana, Kansas, MarylandOlder (M = 32.5 vs. 28.6 years old)Higher statewide recidivism rate (45% vs. 34%)Higher lifetime arrestsHigher proportion African AmericanHigher proportion drug offendersLower proportion violent offendersLower proportion property offendersLess likely to have worked before entering prisonDuration model measuresDuration models Indicators of three employment servicesIndicator of multiple service receiptDemographic characteristicsIndicator for work before prison Criminal historyHazard ratios: Time to first arrest

Strengths and limitationsTrajectory modelPossible bias due to varying length of criminal recordsUnobserved heterogeneity Propensity score modelLingering observed heterogeneityUnobserved heterogeneity Limited common overlapDuration modelVariation in quality and quantity of services receivedOfficial data: timing and observation

Part 2: Job quality model

norawikoff.wordpress.comImplications: researchDirection of the work-crime relationshipFactors that contribute to labor force exit Low wages, debts, garnishments, financial strainFactors that increase labor force attachmentImplications: practiceProgram designOffer intensive programs to a select fewUse desistance signals to identify participantsProgram evaluationModel the selection process (not cream-skimming)Service deliveryOther researchSEED for Oklahoma Kids (SEED OK)Test of universal Child Development AccountsExperimental study designBetter Futures Enterprises (Twin Cities, MN)Social enterprise providing subsidized housing and supported employment to homeless menPay-for-success agreements with nonprofits and governmental agenciesfuture research: Pathways to DesistanceYoung serious offendersRigorous study design

AcknowledgementsThis research is supported by a research grant from the National Institute of Justice:

NIJ Graduate Research FellowshipGrant Award #: 2013-IJ-CX-0042Project Period: 11/1/13 7/31/14Contact informationNora WikoffBrown School of Social WorkWashington University in St. Louis

[email protected][c] 314-703-8731 norawikoff.wordpress.com35