Remedial Education

Notes: Each column shows the coefficients θ_h of Equation 2, that is, the estimated treatment effects at different time horizons for each outcome of interest. The row labeled “p-value of equal coeffs.” shows the p-value of a test H₀ : θ₁ = θ₂ = θ₃. Control mean and SD correspond to the number of correct answers (Columns 1–4) or the percent of correct answers (Column 5). All models include cohort, year, and strata fixed effects. Standard errors, shown in brackets, are clustered at the school level (the unit of randomization). Significance: ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: The table compiles different robustness checks for estimated treatment effects at the end of Grade 3. Panel A adds controls for the main specification. Individual-level controls include age, age-squared, gender, socioeconomic, and disability status. School-level controls include class size, school number of classes in Grade 3, rural, and morning school status. Panel B estimates Lee (2009) bounds. Panel C calculates Westfall–Young adjusted p-values to address for multiple hypotheses testing. Online Appendix Tables A.1, A.2, and A.3 present details and the same robustness for beginning and end of Grade 4. Significance: ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Each column shows the coefficients θ_h of Equation 2, that is, the estimated treatment effects at different time horizons for each outcome of interest. The row labeled “p-value of equal coeffs.” shows the p-value of a test H₀ : θ₁ = θ₂ = θ₃. Control mean and SD correspond to the baseline number of correct answers (Columns 1–3) or the proportion of students who repeated at the end of Grade 3 (Column 4). All models include cohort, year, and strata fixed effects. Standard errors, shown in brackets, are clustered at the school level (the unit of randomization). Significance: ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Each column shows the estimates of the coefficients β_h of Equation 3, that is, the estimated dose response at different time horizons for each outcome of interest. Dosage is measured by the number of days in which students attended the tutorial. The actual attendance was instrumented with the randomized treatment indicator variable. The first stage Kleibergen–Paap F-statistic is equal to 1828, and the associated p-value is zero. The row labeled “p-value of equal coeffs.” shows the p-value of a test H₀ : β₁ = β₂ = β₃. Control mean and SD correspond to the number of correct answers (Columns 1–4) or the percent of correct answers (Column 5). All models include cohort, year, and strata fixed effects. Standard errors, shown in brackets, are clustered at the school level (the unit of randomization). Significance: ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Notes: Each column shows an OLS estimate of a model in which the dependent variable is an outcome measured at baseline (measured by the number of correct answers in that subtask), and the independent variables are dichotomous variables indicating that the students belong to Cohort 2 or Cohort 3. The index shown in Column 4 is a simple average of the scores in the three subtasks shown in Columns 1–3. All models include cohort, year, and strata fixed effects. Standard errors, shown in brackets, are clustered at the school level (the unit of randomization). Significance: ^*p < 0.10, ^**p < 0.05, ^***p < 0.01.