Table 8

Robustness to Selective Migration

APIP effect on SAT/ACT performance of non movers by ethnicity
Sample: All students who attended their 12th grade school for at least four years
	1	2	3	4	5	6
Subsample	Black	Hispanic	White	Black	Hispanic	White
	log above 1100/24 on SAT/ACT	log above 1100/24 on SAT/ACT	log above 1100/24 on SAT/ACT	log above 900/19 on SAT/ACT	log above 900/19 on SAT/ACT	log above 900/19 on SAT/ACT
	Missing values assumed to be 2
First year	–	0.018	0.165	0.04	0.234	0.005
First year	–	[0.047]	[0.055]^**	[0.052]	[0.074]^**	[0.047]
Second year	–	0.12	0.242	−0.12	0.378	0.045
Second year	–	[0.051]^*	[0.084]^*	[0.117]	[0.116]^**	[0.063]
Third year and beyond	–	0.18	0.262	0.123	0.385	0.257
Third year and beyond	–	[0.038]^**	[0.15]^*	[0.092]	[0.204]^*	[0.096]^*
First year	–	0.023	0.124	0.031	0.183	0.009
First year	–	[0.019]	[0.046]^*	[0.04]	[0.058]^**	[0.032]
Second year	–	0.084	0.166	−0.104	0.294	0.043
Second year	–	[0.027]^**	[0.075]^*	[0.075]	[0.089]^**	[0.035]
Third year and beyond	–	0.108	0.221	0.043	0.313	0.222
Third year and beyond	–	[0.023]^**	(0.122)	[0.065]	[0.167]⁺	[0.063]^**
	Missing values assumed to be 0
First year	–	0.006	0.254	0.057	0.342	−0.003
First year	–	[0.113]	[0.084]^**	[0.091]	[0.116]^*	[0.096]
Second year	–	0.197	0.405	−0.156	0.56	0.051
Second year	–	[0.13]	[0.118]^**	[0.227]	[0.182]^**	[0.139]
Third year and beyond	–	0.335	0.351	0.295	0.54	0.334
Third year and beyond	–	[0.120]^*	[0.217]	[0.175]	[0.292]⁺	[0.185]⁺

Robust standard errors in brackets. Standard errors clustered at school district level.
↵+ significant at 10 percent; * significant at 5 percent; ** significant at 1 percent.
Notes: All regressions are based on 384 observations from 57 schools. All regression specifications include school-level controls, school fixed effects, year fixed effects, and propensity score decile by year fixed effects. Due to data masking count fewer than five are coded as missing. Treatment effects are not estimated for above 1100/24 for blacks because there are too many missing values. To provide bounds for those masked observations, missing values are assumed to be four and zero.