Table 2.
Balance in Xs and test scores at baseline across treatment and control groups, our inverse propensity-score weights
| Control mean | T-C difference | SE | P-value | |
|---|---|---|---|---|
| Child characteristics | ||||
| Male | 0.496 | 0.001 | 0.019 | 0.951 |
| African-American | 0.438 | −0.002 | 0.019 | 0.932 |
| Hispanic | 0.489 | −0.0001 | 0.019 | 0.994 |
| Labeled gifted | 0.109 | −0.0007 | 0.012 | 0.958 |
| Labeled special education | 0.110 | 0.0007 | 0.012 | 0.958 |
| Speaks English at home | 0.764 | −0.003 | 0.017 | 0.854 |
| Mother/family characteristics | ||||
| Mother’s years of schooling | 12.977 | −0.006 | 0.069 | 0.932 |
| Mother works full time | 0.215 | 0.0002 | 0.017 | 0.989 |
| Mother born in the US (not PR) | 0.600 | 0.001 | 0.020 | 0.960 |
| Family gets some welfare | 0.779 | 0.001 | 0.016 | 0.947 |
| Mother in same house 1 year ago | 0.914 | 0.0005 | 0.011 | 0.965 |
| Mother is Catholic | 0.539 | −0.00031 | 0.020 | 0.988 |
| Probability income ≤ $15, 000 | 0.502 | −0.001 | 0.025 | 0.958 |
| Baseline test scores (no K scores) | ||||
| Math score | 19.79 | 0.012 | 1.004 | 0.990 |
| Reading score | 25.57 | −0.111 | 1.067 | 0.917 |
Notes: Table reports treatment control differences for baseline demographics and baseline test scores, treating the invalid 99 scores as missing. Column 1 reports the control group mean, column 2 the T-C difference, column 3 the SE on this difference, and column 4 the p-value. The inverse p-score weights are 1/p̂ for treatment observations and 1/(1 − p̂) for control observations, where p̂ is generated from a logistic regression of treatment status on baseline demographics, sample design variables, and baseline test scores. SEs clustered by family. Mother born in the US denotes born in one of the 50 states and Washington DC< and not Puerto Rico, and family welfare use denotes use of Food Stamps, AFDC/public assistance, Social Security, or Medicaid. The probability that income is less than or equal to $15,000 is reported in the table, the specifications (following others) control for the natural log of the midpoint of income ranges. A small number of observations are missing demographics. Data from the New York City School Choice Scholarships Program evaluation conducted by Mathematica Policy Research.