Web Appendix for The Rate of Return to the High/Scope Perry Preschool Program James J. Heckmana,1, Seong Hyeok Moona,2, Rodrigo Pintoa,2, Peter A. Savelyeva,2, Adam Yavitza,3 aDepartment of Economics, University of Chicago, 1126 East 59th Street, Chicago, Illinois 60637 Key words: rate of return, cost-benefit analysis, Perry Preschool Program, early childhood intervention programs, deadweight costs JEL Codes: D62, I22, I28. Email addresses: [email protected] (James J. Heckman), [email protected] (Seong Hyeok Moon), [email protected] (Rodrigo Pinto), [email protected] (Peter A. Savelyev), [email protected] (Adam Yavitz) 1Henry Schultz Distinguished Service Professor of Economics at the University of Chicago,ProfessorofScienceandSociety,UniversityCollegeDublin,AlfredCowlesDistin- guished Visiting Professor, Cowles Foundation, Yale University and Senior Fellow, Amer- ican Bar Foundation. 2Ph.D. candidate, Department of Economics, University of Chicago. 3Research Professional at Economic Research Center, University of Chicago. Preprint submitted to Elsevier November 23, 2009 Contents A Background on the Perry Preschool Curriculum 4 B Data Sources 6 C Program Initial Costs 10 D Educational Attainment 15 E Life Cycle Activity and Employment Outcomes 19 F Constructing Comparison Groups to Facilitate Interpolation 24 F.1 National Longitudinal Survey of Youth 1979 (NLSY79) . . . . 24 F.2 Panel Study of Income Dynamics (PSID) . . . . . . . . . . . . 26 F.3 Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 G Imputation and Extrapolation of Earning Profiles 29 G.1 Imputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 G.1.1 Imputation Method 1: Piece-wise Linear Interpolation 29 G.1.2 Imputation Method 2: Cross-section Regression Impu- tation . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 G.1.3 Imputation Method 3: Kernel Matching . . . . . . . . 30 G.1.4 Imputation Method 4: Hause Procedure . . . . . . . . 31 G.1.5 Comparing Imputation Methods . . . . . . . . . . . . . 32 G.2 Extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 G.2.1 Extrapolation 1: CPS Projection . . . . . . . . . . . . 32 G.2.2 Extrapolation 2: PSID Projection . . . . . . . . . . . . 33 G.2.3 Extrapolation 3: Hause Procedure . . . . . . . . . . . . 33 G.2.4 Comparing Extrapolation Methods . . . . . . . . . . . 34 G.3 Estimated Earnings Profiles from the Various Procedures . . . 34 2 H Estimating the Social Cost of Crime 41 H.1 General Framework . . . . . . . . . . . . . . . . . . . . . . . . 45 H.2 Perry Crime Data . . . . . . . . . . . . . . . . . . . . . . . . . 50 H.3 Estimating Victimization Levels . . . . . . . . . . . . . . . . . 53 H.3.1 National Datasets . . . . . . . . . . . . . . . . . . . . . 54 H.3.2 Estimated Victimization Rates & Levels . . . . . . . . 58 H.4 Estimating Unit Costs . . . . . . . . . . . . . . . . . . . . . . 76 H.4.1 Victimization Unit Costs . . . . . . . . . . . . . . . . . 76 H.4.2 CJS Unit Costs . . . . . . . . . . . . . . . . . . . . . . 77 I Use of the Welfare System 83 J Lifetime Profiles and the Sensitivity of Estimates to Alterna- tive Imputations and Extrapolation Approaches and to Al- ternative Assumptions About Deadweight Costs of Taxation 90 K Calculating the Standard Errors for the IRR and Benefit-to- Cost Ratio 101 L Adjusting for the Compromised Randomization 103 References 107 3 A. Background on the Perry Preschool Curriculum Preschool Overview. During each wave of the experiment, the preschool class consisted of 20–25 children, whose ages ranged from 3 to 4. This is true even of the first and last waves, as the first wave admitted 4-year-olds, who only received one year of treatment, and the last wave was taught alongside a group of 3-year-olds, who are not included in our data. Classes were 2-1/2 hours every weekday during the regular school year (mid-October through May). Thepreschoolteachingstaffoffourproducedachild-teacherratioranging from 5 to 6.25 over the course of the program. Teaching positions were filled by public-school teachers who were “certified in elementary, early childhood, and special education,” (Schweinhart et al., 1993, p.32). Home Visits. Home visits lasting 1-1/2 hours were conducted weekly by the preschool teachers. The purpose of these visits was to “involve the mother in the educational process,” and “implement the curriculum in the home,” (Schweinhart et al., 1993, p.32). By way of encouraging the mothers’ par- ticipation, teachers also helped with any other problems arising in the home during the visit. Occasionally, these visits would consist of field trips to stimulating environments such as a zoo. Curriculum. The Perry Preschool curriculum was based on the Piagetian concept of active learning, which is centered around play that is based on problem-solving and guided by open-ended questions. Children are encour- aged to plan, carry out, and then reflect on their own activities. The topics in the curriculum are not based on specific facts or topics, but rather on key 4 experiences related to the development of planning, expression, and under- standing. The key experiences are then organized into ten topical categories, such as “creative representation”, “classification” (recognizing similarities and differences), “number”, and “time.”4 These educational principles are reflected in the types of open-ended questions asked by teachers: for exam- ple,“What happened? How did you make that? Can you show me? Can you help another child?” (Schweinhart et al., 1993, p.33) As the curriculum was developed over the course of the program, its de- tails and application varied from year to year. While the first year involved “thoughtful experimentation” on the part of the teachers, experience with the program and series of seminars during subsequent years led to the devel- opment and systematic application of teaching principles with “an essentially Piagetian theory-base.” During the later years of the program, all activities took place within a structured daily routine intended to help children “to de- velop a sense of responsibility and to enjoy opportunities for independence,” (Schweinhart et al., 1993, pp. 32–33). 4For a full list, see Schweinhart et al. (1993). 5 B. Data Sources After initial program costs, each area of costs or benefits relies on two types of data sources: information on the Perry subjects — from subject in- terviews or public records retrieved by High/Scope — and data from external sources, with the latter used to compute unit costs or transform the Perry records into a more usable form. Table B.1 summarizes the data sources used in this paper. Program Costs (Section 3.1, Appendix C). Program costs are those reported in Barnett (1996). Costs related to program administration (such as screen- ing, test administration, staff, and preschool supplies) are recorded directly fromhistoricalprogramexpenditures. Costsrelatedtoconductingthepreschool in a public school facility were estimated using depreciation costs from the Ypsilanti Public School District. Education (Section 3.2, Appendix D). K-12 records on special education, grade completion, and graduation time (number of years held back) were obtained directly from school districts. GED, college, and vocational educa- tion records were also obtained from public or private institutions indicated during subject interviews. Annual per-pupil K-12 school expenditure come from the annual Digest of Education Statistics (1975–1982), with special ed- ucation cost estimates from Chambers et al. (2004). GED expenditures are taken from Heckman and LaFontaine (2008). Public expenditures on college and university tuition are taken from Washtenaw Community College and Michigan State University. Vocational training unit costs are taken from estimates in Tsang (1997). 6 s e rg nt a u h o n c m d) nd urces Contents/Usage ProgramCosts College,vocationaltrainingSpecialeducation,gradecompletion,graduatioPer-pupilK-12schoolexpenditureGEDexpendituresPer-pupilspecialeducationexpenditurePer-pupilcollegeanduniversityexpenditureVocationaltrainingcosts Employmentspells,earningsEarningsimputationandextrapolationEarningsextrapolationSurvivalrates Juvenile,misdemeanor,andfelonyarrestsandIncarcerationandparolerecordsNationalvictimizationlevels,bycrimetypeNationalarrestlevels,bycrimetypeCriminalvictimizationunitcostsOverallcriminal-justicesystemexpenditures EarningsMarginal&FICAtaxrate Timespentonwelfare,typesofbenefitsExtrapolation:cashassistance/foodstampaExtrapolation:In-kindtransfers udyofIncomeDynamics;(c)CurrentPopulationSurvey;( UniformCrimeReports;(g)CriminalJusticeExpenditurea o St ) aS 82) ns nel y;(f TableB.1:Dat DataSource Barnett(1996) PerryInterviewsPublicSchoolRecordsDigestofEducationStatistics(1975–19HeckmanandLaFontaine(2008)Chambers,Parrish,andHarr(2004)StateandcountyeducationalinstitutioTsang(1997) PerryInterviewsaNLSY79bcPSID,MarchCPSdNVSR(2004) County/state/federalpolicerecordsStatedepartmentofcorrectionseNCVSfUCRCohen(2005)gCJEE PerryInterviewsTaxPolicyCenter(2007) StateandcountryrecordsNLSY79,PSIDhSIPP LongitudinalSurveyofYouth,1979;(b)ThePa eports;(e)NationalCriminalVictimizationSurve urveyofIncomeandProgramParticipation. al R S Category ProgramCosts(3.1,Apx.C) Education(3.2,Apx.D) Employment/Earnings(3.3,Apx.G) Crime(3.4,Apx.H) TaxPayments(3.5) Welfare(3.6,Apx.I) Notes:(a)TheNation NationalVitalStatistics EmploymentSurvey;(h) 7 Employment / Earnings (Section 3.3, Appendix G). Subject interviews pro- vided information on the timing and length of employment spells, as well as earnings during those spells. Data quality issues (such as right-censoring of employmentspellsorincompleteearningsinformation)areresolvedusingone of four different imputation methods, all of which use data from the NLSY79. LifetimeEarningsareextrapolatedpastage40usingtheMarchCurrentPop- ulation Survey (CPS) and the PSID. The National Vital Statistics Reports (2004) are used to factor survival rates into this extrapolation. Crime (Section 3.4, Appendix H). Individual criminal activity comes from public record searches: At the county level, arrests, convictions, and incar- ceration are reported; at the state (Michigan) level, arrests are only reported when they lead to convictions. Data quality for subjects residing outside Michigan is low, as many other states did not cooperate with attempted record searches. State and county correctional departments provided incar- cerationandparolerecords. Arrestlevelsweretransformedintovictimization rates using a ratio of arrest levels from the Uniform Crime Report (UCR) and victimization levels from the National Criminal Victimization Survey (NCVS). Criminal victimization unit costs estimates are taken from Cohen (2005). State expenditures on the criminal justice system (CJS) from the CriminalJusticeExpenditureandEmploymentSurvey(CJEE)arecombined with criminal justice system volume records from the UCR to compute CJS unit costs. Tax Payments (Section 3.5). Subject earnings are taken from the sources indicated in a preceding paragraph. Federal Income tax and FICA tax rates are taken from Tax Policy Center (2007). 8 Welfare(Section 3.6, Appendix I). Social service records from Washtenaw County and the State of Michigan provided a timeline of social service re- ceipt by type of benefit. National longitudinal records were used to extrap- olate amounts past age 40: the NLSY79 and PSID for cash assistance and food stamp programs, and the Survey of Income and Program Participation (SIPP) for in-kind transfers. 9 C. Program Initial Costs This appendix breaks out the costs of various components of education. 10