Table 5

Robustness Checks

	District Performance		STEP Program		Negative Shock		Small School
	Below	Above	STEP	Other	Shock	None	Smallest	Large
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
0–10th percentile in previous year	4.155***	5.572***	5.871***	4.313***	−18.558	2.940***	3.817*	4.329***
	(1.207)	(1.636)	(1.509)	(1.305)	(13.720)	(0.482)	(2.263)	(0.953)
10–20th percentile in previous year	1.833***	2.770***	2.764***	1.942**	−3.379	0.917***	2.068	1.722***
	(0.679)	(0.856)	(0.702)	(0.763)	(33.090)	(0.344)	(1.453)	(0.551)
Diff-diff	Yes	Yes	Yes	Yes	No	No	Yes	Yes
Fixed effects	No	No	No	No	Yes	Yes	No	No
Observations	41,096	36,635	25,395	52,336	932	56,463	15,108	62,623
R²	0.580	0.642	0.669	0.650	0.864	0.815	0.656	0.664

Notes: Each coefficient refers to the decile of within-district performance rank, compared to the middle six deciles. The outcome variable is average school performance, which can take values of 0–250). In Column 1 the sample is restricted to the bottom half of districts, in terms of a district’s average school performance on the previous year exam; in Column 2 the sample is restricted to the top half. In Column 3 the sample is restricted to districts where the STEP remedial education training took place; in Column 4 it is restricted to districts where it did not take place. In Column 5, the sample is restricted to schools that dropped 30 percentiles in its national rank between year t and t – 1. Schools in Column 5 did not experience such test score declines. In Columns 5 and 6 we do not difference out the baseline relationship between rank and performance, since we do not have data for performance in 2010 so do not know which schools in 2011 experienced a large drop. In Column 7, the sample is restricted to smallest quintile of schools—measured in the number of test-takers in 2012. Column 8 is the complement of Column 7. Standard errors are clustered at the district level.