Intro PPP and ABC Method Evidence Mechanisms Conclusions THE HEALTH EFFECTS OF EARLY CHILDHOOD INTERVENTIONS Gabriella Conti University College London, IFS and NBER James J. Heckman, Seong Moon, Rodrigo Pinto University of Chicago DAHR Seminar April 10, 2014 Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 1 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Motivation Motivation I A consolidated body of evidence documents the importance of early life conditions as determinants of health. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Motivation Motivation I A consolidated body of evidence documents the importance of early life conditions as determinants of health. I The bulk of health care costs result from relatively few individuals with multiple chronic conditions. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Motivation Motivation I A consolidated body of evidence documents the importance of early life conditions as determinants of health. I The bulk of health care costs result from relatively few individuals with multiple chronic conditions. I NCDs (noncommunicable diseases) are the main cause of death in both developed and developing countries. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Motivation Motivation Most of current policy interventions focus on managing these conditions or attempting behavior change in adults. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Motivation Motivation But such conditions are — at least in part preventable. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper This Paper: Are Early Childhood Interventions an effective strategy to prevent disease and promote health? Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 3 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper This Paper: Are Early Childhood Interventions an effective strategy to prevent disease and promote health? → We examine the health effects of the two oldest and most widely cited ECIs: Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 3 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper This Paper: Are Early Childhood Interventions an effective strategy to prevent disease and promote health? → We examine the health effects of the two oldest and most widely cited ECIs: I the High/Scope Perry Preschool Project (PPP) Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 3 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper This Paper: Are Early Childhood Interventions an effective strategy to prevent disease and promote health? → We examine the health effects of the two oldest and most widely cited ECIs: I the High/Scope Perry Preschool Project (PPP) I the Carolina Abecedarian Intervention (ABC) Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 3 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I IQ Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I IQ I problem behavior Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I IQ I problem behavior I education Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I IQ problem behavior education income Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I I IQ problem behavior education income employment Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I I I IQ problem behavior education income employment personal relationships Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I I I IQ problem behavior education income employment personal relationships → inputs in a health production function Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I I I IQ problem behavior education income employment personal relationships → inputs in a health production function + the ABC intervention also included a health component Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions This Paper Why health effects of ECIs? I High-quality ECIs targeted to disadvantaged children have been shown to have longlasting effects on a variety of social and economic outcomes (Heckman et al., QE [2010], AER [2013], Hines et al., Science [2013], Nores and Barnett, EcEdRev [2010]): I I I I I I IQ problem behavior education income employment personal relationships → inputs in a health production function + the ABC intervention also included a health component I but the health effects of ECIs have seldom been investigated (Campbell et al., Science, 2014). Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Preview Preview of Results I Evidence I PPP: males in the treated group have significantly lower prevalence of unhealthy behaviors (esp. smoking) in adulthood. I ABC: males in the treated group have significantly lower prevalence of cardiovascular and metabolic risk factors in their mid 30s. I Mechanisms I PPP: significantly lower prevalence of problem behavior in childhood. I ABC: significantly lower prevalence of overweight in childhood. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 5 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Preview Outline of the Talk I The PPP and ABC Interventions I Methods I Results 1 2 Evidence Mechanisms I Conclusions Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 6 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Where: Chapel Hill, North Carolina I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits newsletter sent to parents I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits newsletter sent to parents I How long: 5y (ages 0-5) I How much: $9,604.30 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP and ABC ABC PPP I What: small RCT (double RCT but ignore school-age component for this talk) I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits newsletter sent to parents I How long: 5y (ages 0-5) I How much: $15,388 (2010$) per child/year Design Conti, Heckman, Moon, Pinto More THE HEALTH EFFECTS OF ECIS 7 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Eligibility Eligibility in the ABC and PPP Interventions PPP ABC Requirements Child IQ<85 No apparent biological conditions Eligibility Criteria Cultural Deprivation Scale High-Risk Index (1) Average Parent Education (1) Absent Father (2) No maternal relatives in area (2) Father’s Occupation Rating (3) Household Density (3) Siblings below age-grade equiv. (4) Welfare Payments (5) Father’s work unstable or unskilled (6) Mother’s or father’s IQ ≤85 (7) Sibling IQ≤90 (8) Social agency referral (9) Family sought counseling or help (10) Other special circumstances (11) Mother’s education (12) Father’s education (13) Family income Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 8 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Eligibility PPP and ABC Target Populations Mother’s Age at Birth Father’s Age at age 3.5 Father absent at age 3.5 Neither Parent Employed Family on Welfare Mother’s Education (years) Father’s Education (years) Household size at age 3.5 No. siblings at age 3.5 Conti, Heckman, Moon, Pinto PPP 25.5 32.8 47% 41% 50% 9.4 8.6 6.3 4.3 THE HEALTH EFFECTS OF ECIS ABC 20.3 26.9 69% 55% 43% 10.3 11.0 4.3 1.1 9 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum Curriculum in the ABC and PPP Interventions: Stimulation PPP Activities Teaching Style Record Keeping Conti, Heckman, Moon, Pinto ABC Curriculum Differences “Plan, Do, Review” Learning games (executive function) (Sparling & Lewis) Curriculum Similarities (1) Individualized teaching: draw from Piaget (active learning) and Vygotsky (scaffolding) (2) Progress tracked through logs and notes THE HEALTH EFFECTS OF ECIS 10 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention I Treated children also received free basic pediatric care. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention I Treated children also received free basic pediatric care. I Medical staff on site: 2 pediatricians, a family nurse practitioner and a licensed practical nurse. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention I Treated children also received free basic pediatric care. I Medical staff on site: 2 pediatricians, a family nurse practitioner and a licensed practical nurse. I Well-child care: assessments at ages 2, 4, 6, 9, 12, 18, 24 months; then yearly. Complete physical exam + parents counseling. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention I Treated children also received free basic pediatric care. I Medical staff on site: 2 pediatricians, a family nurse practitioner and a licensed practical nurse. I Well-child care: assessments at ages 2, 4, 6, 9, 12, 18, 24 months; then yearly. Complete physical exam + parents counseling. I Ill-child care: daily surveillance. In case of illness: doctor visit + lab tests + treatment until recovery. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Curriculum The Nutritional Component of the ABC Intervention I Treated children had breakfast, lunch and afternoon snack at the center. I Food was catered in by kitchens approved by the local health department, with nutritionist consulted to plan menus. The Health Care Component of the ABC Intervention I Treated children also received free basic pediatric care. I Medical staff on site: 2 pediatricians, a family nurse practitioner and a licensed practical nurse. I Well-child care: assessments at ages 2, 4, 6, 9, 12, 18, 24 months; then yearly. Complete physical exam + parents counseling. I Ill-child care: daily surveillance. In case of illness: doctor visit + lab tests + treatment until recovery. I Costs of medicines and referrals to hospitals not included. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 11 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Randomized Controlled Trials I Standard model of program evaluation: Yi = Di Yi (1) + (1 − D)Yi (0) I I I I Yi : observed outcome of participant i, i = 1, ..., N. Di : treatment assignment for participant i. Di = 1 if treatment occurs, Di = 0 otherwise. (Yi (0), Yi (1)): potential outcomes for participant i. I Major benefit of a randomized experiment: (Y (1), Y (0)) ⊥⊥ D|X I Our goal is to estimate average treatment effects and to test the null hypothesis of no treatment effect. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 12 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. Randomization Conti, Heckman, Moon, Pinto Compromises Block Permutation THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Conti, Heckman, Moon, Pinto Compromises Block Permutation THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. =⇒ Multiple hypothesis testing correction: stepdown algorithm developed by Romano and Wolf [2005]. MHT Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. =⇒ Multiple hypothesis testing correction: stepdown algorithm developed by Romano and Wolf [2005]. MHT I Non-random Attrition: biased sample. Conti, Heckman, Moon, Pinto Chart THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. =⇒ Multiple hypothesis testing correction: stepdown algorithm developed by Romano and Wolf [2005]. MHT I Non-random Attrition: biased sample. =⇒ Inverse Probability Weighting (IPW). Conti, Heckman, Moon, Pinto Chart IPW THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. =⇒ Multiple hypothesis testing correction: stepdown algorithm developed by Romano and Wolf [2005]. MHT I Non-random Attrition: biased sample. =⇒ Inverse Probability Weighting (IPW). Chart IPW I Contamination of the Control Group: attendance of alternative childcare by control families (ABC only). Figure Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Problems and Solutions I Small Sample: use of classical asymptotic inference questionable. =⇒ Exact (small sample) permutation-based inference. I Compromised Randomization: non-random re-assignment. =⇒ Block permutation following the randomization protocol. Randomization Compromises Block Permutation I Cherry Picking: pre-testing and selective reporting of significant outcomes. =⇒ Multiple hypothesis testing correction: stepdown algorithm developed by Romano and Wolf [2005]. MHT I Non-random Attrition: biased sample. =⇒ Inverse Probability Weighting (IPW). Chart IPW I Contamination of the Control Group: attendance of alternative childcare by control families (ABC only). Figure =⇒ Control for selection into daycare among the controls. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS Contamination 13 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Evidence Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 14 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Health and Healthy Behaviors PPP I Information on healthy behaviors was collected at ages 27 and 40. I Self-reported health, height and weight recorded at age 40. → On-going age-50 follow-up with biomarkers. ABC I Information on healthy behaviors was collected at ages 21 and 30. I Last data collection (mid 30s) was a biomedical sweep. I Information was collected from two sources: I Physical Exam at the doctor’s office. I Lab tests from venous (non-fasting) blood. → First time biomarkers have been collected in a ECI. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 15 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP: Smoking at Ages 27 and 40, Males Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Not a daily smoker at age 27 0.462 0.581 0.080 0.080 Non- or light smoker (≤10/d) at age 27 0.615 0.903 0.004 0.004 No. of cigarettes at age 27 8.744 4.291 0.006 0.007 Smoked <100 cigarettes by age 40 0.444 0.600 0.040 0.040 Not a daily smoker at age 40 0.472 0.633 0.020 0.041 Non- or light smoker (≤10/d) at age 40 0.743 0.929 0.011 0.021 No. of cigarettes at age 40 6.543 3.714 0.036 0.053 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 16 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP: Other Health Behaviors at Age 40, Males Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Regular physical activity past mo. 0.629 0.567 0.480 0.480 Diet change past 15 yrs. 0.229 0.380 0.018 0.061 Always wears a seat belt 0.618 0.800 0.080 0.080 Never got a traffic ticket 0.265 0.534 0.086 0.097 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 17 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP: Physical Health at Age 40, Males Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Self-reported weight (kgs) 88.619 91.459 0.672 0.888 Body Mass Index (BMI) 27.426 28.720 0.856 0.856 No overweight 0.543 0.533 0.472 0.946 No obese 0.743 0.667 0.869 0.892 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 18 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Biomedical Sweep (Mid 30s): Blood Pressure, Males Variable Control Treatment Mean Mean p-val Blk. IPW P. Stepd. Systolic Blood Pressure (mmHg) 143.333 125.789 0.018 0.029 Diastolic Blood Pressure (mmHg) 92.000 78.526 0.024 0.024 Hypertension (S.>140 & D.>90) 0.444 0.105 0.010 0.018 Hypertension (S.>140 or D.>90) 0.556 0.211 0.038 0.038 Notes: (1) The fourth column “Blk. IPW P. p-val” presents the one-sided single hypothesis block permutation p-values associated with the IPW treatment effect estimate. By block permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimate using the following variables: prematurity, a dichotomous indicator for not having an exam for illness or injury in the past two years at age 30, Achenbach DSM Attention-Deficit/Hyperactivity (AD/H) problems scale at age 30, and Achenbach substance abuse scale. (2) The fifth column “Blk. IPW P. Stepd.” presents the multiple hypothesis stepdown p-values associated with (2). Blocks are delineated by horizontal lines. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 19 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Biomedical Sweep (Mid 30s): Lab Tests, Males Control Treatment Variable Mean Mean p-val Blk. IPW P. Stepd. HDL Cholesterol (mg/dL) 42.000 53.211 0.066 0.110 Dyslipidemia (HDL<40 mg/dL) 0.417 0.106 0.179 0.179 Pre-Diabetes (HbA1C≥5.7%) 0.583 0.473 0.426 0.426 Vit.D Deficiency (<20 ng/mL) 0.750 0.368 0.021 0.021 Notes: (1) The fourth column “Blk. IPW P. p-val” presents the one-sided single hypothesis block permutation p-values associated with the IPW treatment effect estimate. By block permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimate using the following variables: prematurity, a dichotomous indicator for not having an exam for illness or injury in the past two years at age 30, Achenbach DSM Attention-Deficit/Hyperactivity (AD/H) problems scale at age 30, and Achenbach substance abuse scale. (2) The fifth column “Blk. IPW P. Stepd.” presents the multiple hypothesis stepdown p-values associated with (2). Blocks are delineated by horizontal lines. Abbreviations: HDL=High-Density Lipoprotein. HbA1C: Glycosylated Hemoglobin. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 20 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Biomedical Sweep (Mid 30s): Multiple Risk Factors, Males Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Obesity & Hypertension 0.500 0.111 0.016 0.016 Severe Obesity & Hypertension 0.375 0.000 0.005 0.012 Hypertension & Dyslipidemia 0.333 0.000 0.006 0.012 Metabolic Syndrome (NCEP Def.) 0.250 0.000 0.007 0.014 Framingham Risk Score 7.043 4.889 0.038 0.038 Notes: Abbreviations: NCEP=National Cholesterol Education Program. Definitions: Obesity = BMI≥30; Severe Obesity = BMI≥35; Hypertension = Systolic bp>140 or Diastolic bp>90; Metabolic Syndrome = (1) central obesity (waist circumference>102cm or 40in), (2) Dyslipidemia (HDL-C<40 mg/dL), (3) High blood pressure (bp≥130/85 mmHg). The Framingham Risk Score is calculated according to the equation derived in Wilson et al. [1998]; it measures the risk of experiencing “total” Coronary Heart Disease (CHD), defined as both stable and unstable angina, myocardial infarction, or CHD death, within the next 10 years, on the basis of the age, gender, total and HDL cholesterol, systolic and diastolic bp, diabetes and smoking status. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 21 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC at Ages 21 and 30: Healthy Behaviors, Males Variable Control Treatment Mean Mean Blk. IPW P. p-val Stepd. Never a Regular Smoker at age 30 0.391 0.357 0.571 0.571 Age of Onset of Regular Smoking 16.893 19.722 0.051 0.143 Cigarettes/day past 30d. at age 30 11.167 8.685 0.199 0.344 Exercise 4 or more d/w at age 21 0.391 0.307 0.858 0.858 # of Fruit Servings/day at age 21 0.826 0.846 0.524 0.745 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. Females Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 22 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Mechanisms Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 23 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions PPP Mechanisms: IQ, Males 105 Treatment 100 95 IQ 90 85 Control 80 75 Treatment 79.2 94.9 95.4 91.5 91.1 88.3 88.4 83.7 Control 77.8 83.1 84.8 85.8 87.7 89.1 89.0 86.0 Entry 4 5 6 7 8 9 Conti, Heckman, Moon, Pinto Age THE HEALTH EFFECTS OF ECIS 10 24 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions .8 0 0 .2 .2 .4 .6 density .6 .4 density .8 1 1 PPP Mechanisms: Externalizing Behavior, Males 2 2.5 3 3.5 (a) Control 4 4.5 5 2 2.5 ® 3 3.5 4 (b) Treatment 4.5 5 ® Source: Heckman, Pinto, and Savelyev (2013). I Externalizing behavior is the main mediator of the effect of the treatment on healthy behaviors. I Experimentally-induced reductions in externalizing behavior explain between 20% and 55% of the treatment effect on smoking. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 25 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: IQ, Males 120 115 110 105 Treatment 100 95 90 85 Control Age 80 6 6.5 8 12 15 21 00.25 25 00.5 5 00.75 75 1 11.5 5 2 3 33.5 5 4 44.55 5 65 7 Treatment 94.4 103.6106.6111.2108.8 97.0 101.7102.1102.1101.2101.7106.0 99.5 99.0 97.6 93.8 96.9 87.2 Control 94.9 102.0105.0106.1 95.3 87.6 89.5 93.9 90.9 93.6 93.6 95.2 93.7 92.6 93.5 93.8 93.2 86.6 Treatment + SD 96.90105.9109.3113.7111.298.89103.5103.6103.7102.6103.2109.1101.3101.099.3695.1998.6888.92 Treatment - SD 91.93101.1103.8108.7106.395.1699.80100.5100.399.74100.0102.897.7496.8195.9092.3295.0985.54 Control + SD 97.50104.1108.6108.597.8489.4691.8296.1693.1795.8096.2599.0396.2895.3096.2496.1295.8388.37 Control - SD 92.2899.96101.4103.792.7385.8287.2791.6288.5391.3691.0291.2991.2089.8990.7091.4790.5584.77 Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 26 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions 0 .1 .2 .3 .4 .5 ABC Mechanisms: Confidence at 24mo., Males 1 2 Treated, M Conti, Heckman, Moon, Pinto 3 Control, M 4 5 Treated, F THE HEALTH EFFECTS OF ECIS Control, F 27 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions 0 .2 .4 .6 ABC Mechanisms: Confidence at 36mo., Males 1 2 Treated, M Conti, Heckman, Moon, Pinto 3 Control, M 4 5 Treated, F THE HEALTH EFFECTS OF ECIS 6 Control, F 28 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions 0 .2 .4 .6 .8 ABC Mechanisms: Confidence at 48mo., Males 2 Treated, M Conti, Heckman, Moon, Pinto 3 Control, M 4 5 Treated, F THE HEALTH EFFECTS OF ECIS 6 Control, F 29 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions 0 .2 .4 .6 ABC Mechanisms: Confidence at 60mo., Males 2 Treated, M Conti, Heckman, Moon, Pinto 3 Control, M 4 5 Treated, F THE HEALTH EFFECTS OF ECIS 6 Control, F 30 / 56 Intro PPP and ABC Method Evidence Mechanisms Control Treatment Abecedarian Intervention HOME Score Items Mean Mean Child regular preventive health care at 6 mo. 0.826 Child regular preventive health care at 18 mo. Child regular preventive health care at 30 mo. Conclusions Blk. IPW P. p-val Stepd. 1.000 0.023 0.088 0.857 1.000 0.008 0.036 0.810 1.000 0.086 0.402 Delay of food gratification at 42 mo. 0.619 0.760 0.175 0.175 Delay of food gratification at 60 mo. 0.476 0.826 0.017 0.017 Males Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 31 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 Birth 5 10 15 BMI Treated 20 25 Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 32 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males .05 .1 Density .15 .2 .25 .3 3 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 33 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 6 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 34 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .2 Density .4 .6 9 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 35 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 .4 12 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 36 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 .4 .5 18 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 37 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .2 Density .4 .6 24 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 38 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .2 Density .4 .6 36 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 39 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 .4 48 Months 12 14 16 18 20 22 BMI Treated Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 40 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 .4 60 Months 12 14 16 18 20 22 BMI Treated Conti, Heckman, Moon, Pinto Control THE HEALTH EFFECTS OF ECIS ® 41 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males 0 .1 Density .2 .3 96 Months 12 14 16 18 BMI Treated 20 22 24 Control ® Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 42 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -1 -.5 z-Score 0 .5 1 1.5 ABC Mechanisms: BMI, Males 0 3 6 9 12 18 24 Age(months) Treatment Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 36 48 Control 43 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -.5 0 z-Score .5 1 ABC Mechanisms: BMI, Males 24 36 48 60 Age(months) Treatment Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 96 Control 44 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -1 -.5 z Score 0 .5 1 ABC Mechanisms: Height, Males 0 3 6 9 12 18 24 36 48 60 Age (months) Treatment Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 96 Control 45 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -1 -.5 z Score 0 .5 1 ABC Mechanisms: Weight, Males 0 3 6 9 12 18 24 36 48 60 Age (months) Treatment Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 96 Control 46 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -1 -.5 z-Score 0 .5 1 ABC Mechanisms: BMI, Males 0 3 6 9 12 18 24 Age(months) Not Obese Mid-30s Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 36 48 Obese Mid-30s 47 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions -1 -.5 z-Score 0 .5 1 ABC Mechanisms: BMI, Males 24 36 48 60 Age(months) Not Obese Mid-30s Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 96 Obese Mid-30s 48 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: Obesity, Males Control Treatment Condit. Variable Mean Mean TE p-val Blk. IPW P. Stepd. At Risk Overweight at 3 months 0.227 0.037 0.206 0.026 0.121 At Risk Overweight at 6 months 0.250 0.080 0.205 0.074 0.182 At Risk Overweight at 9 months 0.412 0.000 0.446 0.004 0.023 At Risk Overweight at 12 months 0.429 0.000 0.408 0.001 0.009 At Risk Overweight at 18 months 0.389 0.000 0.385 0.000 0.004 At Risk Overweight at 24 months 0.333 0.000 0.343 0.001 0.011 At Risk Overweight at 36 months 0.158 0.080 0.094 0.194 0.194 At Risk Overweight at 48 months 0.300 0.167 0.133 0.150 0.235 At Risk Overweight at 60 months 0.300 0.125 0.187 0.058 0.179 At Risk Overweight at 96 months 0.421 0.120 0.286 0.030 0.117 Weight-for-Length ∆ 0-24m (CDC) 0.858 -0.105 1.176 0.058 0.058 Weight-for-Length ∆ 0-24m (WHO) 1.265 0.166 1.397 0.049 0.057 Notes: We use weight-for-length ≥ 85th percentile for being “at-risk overweight” under 24 months, and BMI-for-age ≥ 85th percentile for being overweight for 24 months and older. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 49 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions ABC Mechanisms: BMI, Males I BMI around one year of age is the main mediator of the effect of the treatment on hypertension and obesity in the mid-30s. I Experimentally-induced reductions in BMI in the treatment group explain between 50% and 70% of the treatment effect on hypertension and obesity. I Preliminary results show that the occurrence of hospitalizations in the last 3 years at age 15 explain a large share of the effect of the treatment on multiple risk factors. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 50 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Parental Responses Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 51 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions 10 20 Proportion 30 40 50 PPP Parental Responses: Beliefs, Males −.015 −.01 −.005 0 .005 Belief in Importance of Parenting Control Conti, Heckman, Moon, Pinto .01 .015 Treatment THE HEALTH EFFECTS OF ECIS 52 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions .0004 .0006 Density .0008 .001 .0012 ABC Parental Responses: Attachment, Males 0 200 400 600 Duration (Seconds) Treated Conti, Heckman, Moon, Pinto 800 1000 Control THE HEALTH EFFECTS OF ECIS 53 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Conclusions I Analyzed the health effects of two flagships ECIs: the Perry Preschool Project and the Abecedarian interventions. I PPP: males in the treated group have significantly lower prevalence of unhealthy behaviors (esp. smoking) in adulthood. I ABC: males in the treated group have significantly lower prevalence of cardiovascular and metabolic risk factors in their mid 30s. I Examined the mechanisms through which these interventions operated to promote health and healthy behaviors. I PPP: significantly lower prevalence of problem behavior in childhood. I ABC: significantly lower prevalence of overweight in childhood. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 54 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Conclusions I Analyzed the health effects of two flagships ECIs: the Perry Preschool Project and the Abecedarian interventions. I PPP: males in the treated group have significantly lower prevalence of unhealthy behaviors (esp. smoking) in adulthood. I ABC: males in the treated group have significantly lower prevalence of cardiovascular and metabolic risk factors in their mid 30s. I Examined the mechanisms through which these interventions operated to promote health and healthy behaviors. I PPP: significantly lower prevalence of problem behavior in childhood. I ABC: significantly lower prevalence of overweight in childhood. I Investigated whether these interventions complement, or substitute for, parental investments. I PPP: increased beliefs in the importance of parenting. I ABC: improved attachment between mother and children. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 55 / 56 Intro PPP and ABC Method Evidence Mechanisms Conclusions Conclusions Our results show the potential of early life interventions to prevent disease and promote health. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 56 / 56 Appendix Appendix Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Abecedarian Design Figure 1 : Design of Abecedarian Study PERSISTENT EFFECTS OF CHILDHOOD EDUCATION Figure 2. Design of Abecedarian Project. Source: Ramey et al. (2000) within both the preschool treatment and control groups. One member of each consecutive pair within grounds in primary education, and successful experience in working with at-risk children and families. back Known as Home–School Resource Teachers (HSTs), each group was then assigned randomly to the AbeceConti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Table 1 : Quality and Service Provision in the ABC and PPP Interventions PPP Training Degree Qualifications Salary Child-Teacher Ratio Home Visits Parent Involvement ABC Educator’s Background Regular Training Periodic on-site training Bachelors Graduate degree and paraprofessionals Certificate in Elementary Extensive Experience + Special Education Similar to public teacher Based on public school pay scale +10% for participation Classroom 5:1 to 6:1 3:1 (Nursery); 4:1 (Toddlers); 6:1 (School) Services Offered Biweekly teacher visits Social work visits as needed, 1.5hrs, 1:1 tutoring for legal help, family planning. Monthly group meetings Newsletter sent home back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Table 2 : Randomization in the ABC and PPP Interventions PPP ABC Randomization Protocol For each of the 5 waves: For all eligible 122 children by cohort: (a) Rank by (1) IQ at entry (2) gender and (a) Match in pairs by HRI rating, mother’s IQ, gender and number of (3) SES of the eldest sibling (b) Assign even and odds by rank and group siblings (precise pairs unknown) in pairs (b)Toss coin for T-C status (c) Toss coin for T-C status Compromises of the Randomization Protocol (1) Initial assignments switched to (1) 8 declined participation after learning balance gender and socio-economic status assignment (7 T, 1 C) (non-compliance) (2) After assignment, working moms switched (2) 2 controls assigned to treatment to control group since could not participate at the request of local authorities in home visits and transportation issues + 1 control diagnosed with retardation associated with seizures was dropped (non-random reassignment) back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Small Sample with Compromised Randomization Problem: Small Sample with Compromised Randomization I Induces a correlation between treatment status and unobserved/observed variables. I Counterfactual outcomes are different from conditional outcomes: d (Y (0), Y (1)) 6= (Y |D = 0), (Y |D = 1) I Casts doubts on standard estimations that ignore compromises. I Needs a conditional independence assumption: (Y (1), Y (0)) ⊥⊥ D | X I where X are the variables used in the randomization protocol: I PPP: cohort, gender, child IQ, SES, maternal employment. I ABC: cohort, gender, maternal IQ, High Risk Index, number of siblings. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Small Sample with Compromised Randomization Solution: Permutation Test tailored to the randomization protocol I Use two assumptions: I Conditional exchangeability property: the distribution of D is invariant under valid permutation (b/c of randomization). I Null of no treatment effect: vector of Y independent of the vector of D (to generate permutation test). Theorem Randomization Hypothesis: the joint distribution of outcomes Y and treatment assignments D are invariant under permutations GX of treatment assignments within strata formed by d values of covariates X , that is: (Y , D) = (Y , gD) ∀ g ∈ GX =⇒ a statistic based on assignments D and outcomes Y is distribution-invariant under permutations g ∈ GX . I Permutation test: compare a test statistic computed on the original (unpermuted) data with a distribution of test statistics computed on resamplings of that data according to permutations in GX . I The p-value, is computed as the fraction of resampled data which yields a test statistic greater than that yielded by the original data. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Cherry Picking Solutions: Multiple Hypothesis Testing (MHT) Type-I Error in Multiple-Hypothesis Testing I Goal: control for familywise error rate (FWER) at a given level α: FWER = P {reject any true null hypothesis in a joint test of a set of hypotheses} ≤ α We perform multiple-hypothesis testing using the Stepdown Algorithm: I Starts by targeting the most significant test statistic. I Decides if it is statistically significant or not. I “Steps down” to smaller, less significant test statistics. I Less Conservative than traditional Bonferroni and Holm procedure by incorporating the dependence structure of the test statistics. I Generates as many adjusted p-values as there are hypotheses. I Romano and Wolf [2005] show that stepdown provides strong FWER control (controls FWER regardless of which subset of the null hypotheses are actually true). back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Attrition Problem: Attrition I Outcome Y not observed for some participants. I Y (d) ⊥⊥ D|X holds, but Y (d) ⊥⊥ D|X , A = 1 may not hold. I There could be differential attrition: D ⊥⊥ A|X may not hold. Solution: Inverse Probability Weighting I Key idea: retrieve statistics for the full outcome distribution through reweighing non-missing observations according to their likelihood of compliance. I IPW relies on a matching on observables assumption: Y (d) ⊥⊥ A|(D, X , Z ) I Z are pre-program variables other than X, and variables in the previous period (Robins et al. [1995], Fitzgerald et al. [1998]) I Applying Bayes’ Theorem we get the standard IPW formula for ATE: Y 1[A=1,D=1] Y 1[A=1,D=0] E(Y (1) − Y (0)) = E( Pr(A=1|D=1,X ,Z ) Pr(D=1|X ,Z ) ) −( Pr(A=1|D=0,X ,Z ) Pr(D=0|X ,Z ) ) I We use a logit to estimate the probability of participation, separately for each gender. I The selection of covariates is based on the lowest AIC among models examining all combinations of covariates imbalanced between attriters and non-attriters. Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Attrition d= ATE N X Yi · 1 [Di = 1] · 1 [Ai = 1] · ωi,1 N1 i=1 where ωi,d = 1 p̂i,d 1 Nd − N X Yi · 1 [Di = 0] · 1 [Ai = 1] · ωi,0 i=1 N X 1 [Di = d] · 1 [Ai = 1] p̂j,d j=1 and pi,d = Pr(A = 1|D = d, Xi , Zi ) Pr(D = d|Xi , Zi ) N0 (1) d ∈ {0, 1} d ∈ {0, 1} N where Nd = X 1 [Di = d] · 1 [Ai = 1]; d ∈ {0, 1} i=1 back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Attrition Attrition in the Abecedarian Intervention Wave Baseline 3m 6m 12m 18m 24m 36m 48m 60m 8y 21y 30y 35y (lab) 35y (PE) Treated M Foll. Attr. 29 28 1 25 4 27 2 26 3 27 2 25 4 24 5 26 3 26 3 28 1 27 2 20 7 19 8 Treated F Foll. Attr. 28 23 5 24 4 24 4 22 6 23 5 21 7 22 6 23 5 23 5 25 3 25 3 18 7 18 7 Control M Foll. Attr. 23 22 1 22 1 21 2 20 3 15 8 19 4 20 3 21 2 19 4 23 0 21 2 12 9 9 11 Control F Foll. Attr. 31 26 5 27 4 25 6 27 4 17 14 23 8 26 5 25 6 24 7 28 3 28 3 22 6 22 6 back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Contamination of the Control Group Problem: Contamination of the Control Group 0 .1 Fraction .2 .3 .4 Figure 2 : ABC, Out-of-Home Care Attendance of the Control group 0 20 40 Months in Day−Care, Countrol Group 60 Notes: Total number of months of out-of-home day-care services experienced by the control group over ages 0–5 years old. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Contamination of the Control Group Solution: Linear Model Based on Matching on Observables I Two issues: I selection of out-of-home care I small sample I Two solutions: I conditional independence: Y (0, t) ⊥⊥ T | (X , W ) ∀ t ∈ supp(T ) Y (0, t) is the outcome Y when the out-of-home care choice T is set at t for D = 0 W is a set of pre-program variables different from X t ∈ [0, 60] is the length of daycare I linearity: Yi = τ + βW Wi + βX Xi + βT Ti + i for Di = 0 I So the counterfactual outcome now is Ŷi (0, 0) = τ̂ + β̂W Wi + β̂X Xi . back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 0/8 Appendix Contamination of the Control Group PPP: Smoking at Ages 27 and 40, Females Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Not a daily smoker at age 27 0.409 0.520 0.107 0.283 Non- or light smoker (≤10/d) at age 27 0.818 0.760 0.661 0.661 No. of cigarettes at age 27 7.682 7.600 0.300 0.487 Smoked <100 cigarettes by age 40 0.409 0.458 0.142 0.503 Not a daily smoker at age 40 0.409 0.522 0.127 0.356 Non- or light smoker (≤10/d) at age 40 0.773 0.870 0.385 0.442 No. of cigarettes at age 40 6.818 5.869 0.483 0.483 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 1/8 Appendix Contamination of the Control Group PPP: Other Health Behaviors at Age 40, Females Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Regular physical activity past mo. 0.091 0.375 0.002 0.038 Diet change past 15 yrs. 0.227 0.375 0.287 0.287 Always wears a seat belt 0.818 0.750 0.813 0.813 Never got a traffic ticket 0.409 0.417 0.661 0.856 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 2/8 Appendix Contamination of the Control Group PPP: Physical Health at Age 40, Females Control Treatment Blk. IPW P. Variable Mean Mean p-val Stepd. Self-reported weight (kgs) 72.665 68.668 0.259 0.766 Body Mass Index (BMI) 26.829 26.283 0.886 0.962 No overweight 0.550 0.556 0.365 0.960 No obese 0.800 0.833 0.884 0.884 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, and the median of SES for the 101 eldest siblings only per cohort. Probabilities of IPW are estimated using the following variables: gender, presence of the father in the home at entry, socio-economic status index (household density, mother and father education, child IQ at entry (Binet)), and number of siblings a entry. Fifth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 3/8 Appendix Contamination of the Control Group ABC Biomedical Sweep (Mid 30s): Blood Pressure, Females Variable Control Treatment Condit. Mean Mean TE p-val Blk. IPW P. Stepd. Systolic Blood Pressure (mmHg) 135.636 129.666 2.185 0.300 0.380 Diastolic Blood Pressure (mmHg) 89.227 85.333 1.204 0.446 0.446 Pre-Hypertension (S.>120 & D.>80) 0.727 0.500 0.101 0.222 0.222 Pre-Hypertension (S.>120 or D.>80) 0.909 0.667 0.244 0.042 0.069 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 4/8 Appendix Contamination of the Control Group ABC Biomedical Sweep (Mid 30s): Obesity, Females Control Treatment Condit. Blk. IPW P. Variable Mean Mean TE p-val Stepd. Overweight (BMI ≥25) 0.955 0.889 0.054 0.482 0.690 Obese (BMI ≥30) 0.727 0.666 -0.112 0.790 0.790 Severely Obese (BMI ≥35) 0.364 0.223 0.143 0.354 0.653 Waist-Hip Ratio (WHR) 0.933 0.876 0.053 0.063 0.101 Abdominal Obesity (WHR≥0.85) 0.762 0.563 0.198 0.080 0.080 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 5/8 Appendix Contamination of the Control Group ABC at Ages 21 and 30: Healthy Behaviors, Females Variable Control Treatment Condit. Mean Mean TE p-val Blk. IPW P. Stepd. Never a Regular Smoker at age 30 0.357 0.600 0.221 0.071 0.195 Age of Onset of Regular Smoking 17.861 17.050 -1.544 0.848 0.848 Cigarettes/day past 30d. at age 30 10.294 10.450 -0.939 0.567 0.807 Exercise 4 or more d/w at age 21 0.071 0.320 0.287 0.004 0.008 # of Fruit Servings/day at age 21 0.286 0.800 0.518 0.004 0.005 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 6/8 Appendix Contamination of the Control Group ABC Age 30: Health Care, Males Control Treatment Condit. Variable Mean Mean TE p-val Blk. IPW P. Stepd. Health Insurance Coverage 0.476 0.704 0.226 0.039 0.039 Buys Health Insurance 0.333 0.650 0.248 0.035 0.080 Hospital or Doctor Care When Sick 0.524 0.815 0.265 0.037 0.068 Notes: Fourth column presents the one-sided single hypothesis constrained permutation p-value based on the IPW t-statistic associated with the difference in means between treatment groups. By constrained permutation we mean that permutations are done within strata defined by the pre-program variables used in the randomization protocol: gender, cohort indicator, mother IQ at entry, number of siblings and high risk index. Probabilities of IPW are estimated using a selection of covariates based on the lowest Akaike Information Criterion (AIC) among models examining all combinations of covariates that present statistically significant imbalance between attriters and non-attriters. Sixth column presents the stepdown p-values associated with (5). Blocks are delineated by horizontal lines. back Conti, Heckman, Moon, Pinto THE HEALTH EFFECTS OF ECIS 7/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small double RCT I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits (day-care treatment) home visits (school-age treatment) newsletter sent to parents I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits (day-care treatment) home visits (school-age treatment) newsletter sent to parents I How long: 5y (ages 0-5), 3y (ages 5-8) Design More THE HEALTH EFFECTS OF ECIS 8/8 Appendix Contamination of the Control Group ABC PPP I What: small RCT I Day-care (age 3–4): n=123: 58 T, 65 C, 5 cohorts Child-Teacher Ratio: 5:1 to 6:1 I Where: Ypsilanti, Michigan I When: 1963–1967, children born 1958–1962 I Who: all African-American children I How: 30 w/y, 12.5 h/w (2.5 h/d) 3 h/w (1.5 h/d) teacher home visits 1/mo parents group meetings I How long: 2y (ages 3-4) for Cohorts 1-4; 1y for Cohort 0 I How much: $9,604.30 (2010$) per child/year Conti, Heckman, Moon, Pinto I What: small double RCT I Day-care (age 0–5): n=111: 57 T, 54 C, 4 cohorts Child-Teacher Ratio: 3:1 (Nursery); 4:1 (Toddlers) I School-age (age 5–8): n=95: 49 T, 47 C Child-Teacher Ratio: 6:1 I Where: Chapel Hill, North Carolina I When: children born 1972–1977 I Who: 98% African-American children I How: 50 w/y, 40 h/w (8 h/d) social work visits (day-care treatment) home visits (school-age treatment) newsletter sent to parents I How long: 5y (ages 0-5), 3y (ages 5-8) I How much: $15,388 (2010$) per child/year Design More THE HEALTH EFFECTS OF ECIS 8/8