Table 6

Head Start Impact for Representative Eligible Children, Participants, and Siblings Using Reweighting

FFEReweighted ATE, Target=Diff. b/w FFE and Participant ATE
Garces, Thomas, and Currie/DemingExpand Sample/ReplicateHS EligibleParticipantsSiblings
Panel A: Some College (PSID)
Head Start0.281**
(0.108)
0.120**
(0.053)
0.052
(0.064)
0.021
(0.059)
0.064
(0.061)
0.099**
(0.032)
Y mean in target0.5560.3870.4370.556
Panel B: Economic Sufficiency Index, Age 30 (PSID)
Head Start−0.023
(0.102)
−0.071
(0.101)
−0.040
(0.099)
0.021
(0.113)
0.017
(0.066)
Y mean in target0.213−0.198−0.4850.213
Panel C: High School Graduation (CNLSY)
Head Start0.086***
(0.031)
0.085***
(0.030)
0.033
(0.042)
0.048
(0.037)
0.020
(0.044)
0.037*
(0.023)
Y mean in target0.7760.7340.7660.776
Panel D: Idle (CNLSY)
Head Start−0.071*
(0.038)
−0.072*
(0.037)
−0.061
(0.045)
−0.055
(0.042)
−0.067
(0.050)
−0.017
(0.026)
Y mean in target0.1970.2210.2010.197
Panel E: Learning Disability (CNLSY)
Head Start−0.059***
(0.020)
−0.059***
(0.021)
−0.031
(0.026)
−0.042*
(0.022)
−0.040
(0.026)
0.017
(0.015)
Y mean in target0.0510.0550.0410.051
Panel F: Poor Health (CNLSY)
Head Start−0.070***
(0.026)
−0.069***
(0.026)
−0.063*
(0.037)
−0.067**
(0.034)
−0.050
(0.038)
−0.003
(0.020)
Y mean in target0.1030.0980.0740.103
  • Notes: Column 1 of this table shows the FFE estimated impacts of Head Start for whites from Garces, Thomas, and Currie (2002) or for the whole sample from Deming (2009). Column 2 shows the FFE estimate using our expanded sample for PSID outcomes and using our replication sample for CNLSY outcomes. The outcomes in Panels A and B are taken from the PSID white sample, and the outcomes in Panels C to F are taken from the CNLSY sample. Columns 3–5 present reweighted estimates of the effect of Head Start for three target populations (shown in the column header) using the post-regression reweighting procedure, in which we multiply group-level estimates of the impact of Head Start by the representative weight for the target population of interest. Column 6 presents the difference in the estimate in Column 2 (FFE) and Column 4 (reweighted for participants). Sample size is N = 2,986 for the expanded sample, and 1,036 for Garces, Thomas, and Currie (2002). Standard errors obtained by bootstrapping.

  • * p < 0.10,

  • ** p < 0.05,

  • *** p < 0.01.