Appendix Table 3.

Effect of excluding “missing data” values on mean treatment effects for African Americans reported in Krueger and Zhu (2004a) using Peterson and Howell or Krueger and Zhu definition of African American

	Using Mathematica weights		Using our inverse p-score weights
	Our replication, K&Z	Our estimates	Our estimates	Our estimates
	0s included	0s excluded	0s included	0s excluded
Panel A: Effects for Peterson & Howell sample of African Americans, math NPRs
Year 1	4.54*** (1.53)	3.65** (1.59)	2.64 (1.42)	2.37 (1.46)
Year 2	2.59 (2.06)	3.03 (2.09)	2.01 (1.67)	2.02 (1.73)
Year 3	4.00** (1.86)	3.38* (1.92)	4.22** (1.73)	3.94** (1.78)
Panel B: Effects for Krueger & Zhu sample of African Americans, math NPRs
Year 1	3.18** (1.53)	2.21 (1.57)	1.41 (1.38)	1.31 (1.41)
Year 2	1.33 (1.97)	1.74 (2.01)	0.48 (1.65)	0.43 (1.72)
Year 3	2.83 (1.76)	2.32 (1.81)	2.33 (1.66)	2.10 (1.70)

Notes: Table reports original results from panel 3 of Table 3B of Krueger and Zhu (2004a), our replication of these results, and then shows the impact of excluding the 0 percentile values (which are invalid percentiles), as well as the impact of using inverse propensity score weights as an alternative to the non-response adjusted weights in the data. Dependent variable is the math (Panel A) or reading (Panel B) National Percentile Ranking scores from the Iowa Test of Basic Skills in the spring for years 1–3 of voucher distribution. Estimates in columns 1 and 2 use the MPR provided weights, while those in 3 and 4 use inverse propensity score weights. The 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. 95% CIs are obtained by bootstrapping families with replacement. Data from the New York City School Choice Scholarships Program evaluation conducted by Mathematica Policy Research.