Table 4

Heterogeneous Treatment Effect on Test Scores at Post-intervention (T2) and in the One-Year Follow-up (T3) across High/Low Preschool Quality, Baseline Skills, and Parent Education

Post-intervention (T2)Follow-up (T3)
Summary ScoreMathematicsExecutive FunctioningLanguageSummary ScoreMathematicsExecutive FunctioningLanguage
Panel A: Preschool Center Quality
Treat × High−0.065−0.113−0.019−0.038−0.186*−0.153−0.030−0.264
Wild P (treat)0.1420.1700.2610.7910.0270.0250.3300.034
Wild P (T × H)0.5450.5220.8830.7360.1020.2460.7960.132
Adj. R20.6050.4330.4900.5260.5190.3570.3850.489
Mean low−0.171−0.152−0.157−0.105−0.123−0.098−0.126−0.078
Mean high0.1740.1520.1610.1080.1240.0840.1390.081
Panel B: Baseline Skills
Treat × High0.1070.0490.0260.1480.0230.0350.0880.068
Wild P (treat)0.5030.4010.2610.4810.2820.0880.9330.908
Wild P (T × H)0.3950.7410.7950.2880.8210.7840.5110.608
Adj. R20.6160.4440.4920.5340.5270.3660.3830.491
Mean low−0.637−0.496−0.553−0.547−0.584−0.442−0.477−0.469
Mean high0.6250.4870.5430.5500.5860.4410.4790.465
Panel C: Parent Education
Treat × High0.0600.0100.163−0.0440.039−0.0420.1080.036
Wild P (treat)0.4000.2400.7590.6920.3940.0380.9020.799
Wild P (T × H)0.6180.9380.2060.6470.7870.7560.5060.791
Adj. R20.6010.4320.4860.5190.5060.3580.3720.475
Mean low−0.200−0.192−0.179−0.117−0.192−0.174−0.120−0.176
Mean high0.2220.2030.1950.1470.2270.1890.1510.215
  • Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Each column in each panel presents regression coefficients with standard errors in parentheses, using ordinary least squares. The model specification is in line with Model 1 in Table 3. We add controls for gender, birth month and parental characteristics (education, earnings, and indicator for non-Western country of birth), baseline test scores, indicators for late consent and not having participated in the T1 assessment, and randomization block, with all regressions clustered on randomization block. Preschool center quality is measured as the preschool center fixed effect (center average covariate-adjusted test score). High/low center quality is split at median value. Parental education is measured as the average of mother’s and father’s number of years of education. 12 children with no information on parental education (balanced across treatment status) are excluded from the sample. High/low parental education and high/low baseline skills are split at median value. Mean low and mean high are means of outcome variable in relevant subsamples. We have utilized the boottest package in Stata to do a few-cluster-correction of the p-value, reported in the table as Wild P (treat) and Wild P (T×H).