Additional analyses and robustness checks: 2SLS results using UPE eligibility and new schools per capita to instrument for years of schooling
| Model . | Effect Estimate . | 95 % Confidence Interval . | Wild Bootstrap p Value . |
|---|---|---|---|
| Model 1: Linear Birth-Year Trend | 0.05 | (−0.03 to 0.13) | .51 |
| Model 2: Quadratic Birth-Year Trend | 0.05 | (−0.03 to 0.13) | .49 |
| Model 3: State Birth-Year Trends | 0.12 | (−0.17 to 0.41) | .58 |
| Model 4: Pre-UPE Primary Enrollment Rate Control | 0.05 | (−0.11 to 0.20) | .57 |
| Model 5: Placebo Treatment | 0.02 | (−0.43 to 0.47) | .83 |
| Model 6: Permanent Residents Only | 0.05 | (−0.04 to 0.13) | .55 |
| Model 7: Resident Since 4 Only | 0.05 | (−0.04 to 0.13) | .55 |
| Model 8: Exclude Lagos | 0.04 | (−0.04 to 0.13) | .64 |
| Model 9: Exclude High Migration Statesa | 0.09 | (−0.01 to 0.20) | .75 |
| Model 10: Exclude States With >10% Migrants | 0.09 | (−0.002 to 0.18) | .78 |
| Model 11: 1976 Cohort Cutoff | 0.02 | (−0.21 to 0.24) | .88 |
| Model 12: 1974 Cohort Cutoff, All Respondents | 0.06 | (−0.08 to 0.20) | .73 |
| Model 13: 1976 Cohort Cutoff, All Respondents | 0.004 | (−0.78 to 0.78) | .88 |
| Model 14: Federal Funds as Instrumentb | −0.27 | (−6.90 to 6.40) | .75 |
| Model 15: Pre-UPE Enrollment as Instrumentc | 0.01 | (−0.04 to 0.07) | .64 |
| Model 16: BMI Outcome | 0.34 | (−0.60 to 1.30) | .65 |
| Model . | Effect Estimate . | 95 % Confidence Interval . | Wild Bootstrap p Value . |
|---|---|---|---|
| Model 1: Linear Birth-Year Trend | 0.05 | (−0.03 to 0.13) | .51 |
| Model 2: Quadratic Birth-Year Trend | 0.05 | (−0.03 to 0.13) | .49 |
| Model 3: State Birth-Year Trends | 0.12 | (−0.17 to 0.41) | .58 |
| Model 4: Pre-UPE Primary Enrollment Rate Control | 0.05 | (−0.11 to 0.20) | .57 |
| Model 5: Placebo Treatment | 0.02 | (−0.43 to 0.47) | .83 |
| Model 6: Permanent Residents Only | 0.05 | (−0.04 to 0.13) | .55 |
| Model 7: Resident Since 4 Only | 0.05 | (−0.04 to 0.13) | .55 |
| Model 8: Exclude Lagos | 0.04 | (−0.04 to 0.13) | .64 |
| Model 9: Exclude High Migration Statesa | 0.09 | (−0.01 to 0.20) | .75 |
| Model 10: Exclude States With >10% Migrants | 0.09 | (−0.002 to 0.18) | .78 |
| Model 11: 1976 Cohort Cutoff | 0.02 | (−0.21 to 0.24) | .88 |
| Model 12: 1974 Cohort Cutoff, All Respondents | 0.06 | (−0.08 to 0.20) | .73 |
| Model 13: 1976 Cohort Cutoff, All Respondents | 0.004 | (−0.78 to 0.78) | .88 |
| Model 14: Federal Funds as Instrumentb | −0.27 | (−6.90 to 6.40) | .75 |
| Model 15: Pre-UPE Enrollment as Instrumentc | 0.01 | (−0.04 to 0.07) | .64 |
| Model 16: BMI Outcome | 0.34 | (−0.60 to 1.30) | .65 |
Notes: Similar exclusion restriction assumptions apply for these instruments and my original instrument. Standard errors to calculate 95% confidence intervals are clustered at the state level. Wild cluster bootstrap p values are calculated using procedures described in Cameron et al. (2008). Residuals are repeatedly resampled by cluster to form a pseudo-dependent variable, and the model is estimated for each resampled data set. The p values are the proportion of bootstrapped t statistics that are at least as large as the t value from the original model. All models are estimated using DHS sample weights. See the main text for a full description of each robustness check. Models 1–9 and 12–16 show an effect of an additional year of schooling on the probability of being overweight or obese from two-stage least squares regression models using UPE eligibility × new schools per capita as an IV; Models 10 and 11 interact eligibility with two alternative instruments listed in the left hand column.
Model 9 excludes states with migrant share more than 1 standard deviation higher than the mean.
F statistic = 18.1.
F statistic = 23.5.
Source: Demographic Health Surveys Nigeria (2003, 2008, 2013).