Table 6

Robustness Checks

Dependent variable: Overall HSEE test score	Percent of teachers of		Percent of teachers with university degree	Percent of teachers with informal training	Average years of teaching
Dependent variable: Overall HSEE test score	Rank II	Rank III/IV	Percent of teachers with university degree	Percent of teachers with informal training	Average years of teaching
Base result (from Table 5)
(1) Coefficient estimates 95 percent confidence interval	0.76^*** [0.43 – 1.10]	1.10^*** [0.61 – 1.58]	0.06 [−0.09 – 0.20]	−0.45^*** [−0.70 – −0.19]	−4.22^*** [−6.30 – −2.14]
Robustness to omitted school characteristics
(2) Controls for school characteristics	0.74^***	1.04^***	−0.02	−0.54^***	−4.37^***
Robustness to randomization and student transfers
(3) Controls for individual characteristics	0.77^***	1.23^***	0.05	−0.37^***	−4.86^***
Adding transfer students
(4) 2 SLS	0.86^***	1.04^***	0.07	−0.28^*	−4.24^***
Overidentification test for 2SLS: Sargan statistic = 4.5, p-value = 0.21
Robustness to missing HSEE test scores
(5) Controls for missing scores	0.51^***	0.68^**	0.27^***	−0.29^**	−2.86^**
Alternative model specifications
(6) Weighted regression	0.89^***	1.99^***	−0.11	−0.61^**	−6.83^***
(7) All controls	0.64^***	0.87^***	0.14	−0.33^**	−3.86^***

95 percent confidence intervals in brackets;
↵* Significant at 10 percent;
↵** significant at 5 percent;
↵*** significant at 1 percent.
All estimations include student gender and primary test score, parents’ education and income, primary school dummy variables, and selection channel fixed effects. In addition:
- (2) Includes ratio of the number of libraries to 100 students, teacher-student ratio, and nonrandomized students’ parent income.
- (3) Includes parents having a relative in the school, parental ideal for child final education level (6 levels), the number of Type A schools in application sequence, the average scores in 1999 of the first three choices, and the average percentage of teachers of Rank II and higher in the first three choices.
- (5) Includes percentage of students with missing HSEE score and its square, relative score on last semester exam of students with missing score and those with observed scores.
- (7) Includes all controls used in (2), (3), and (5).
(4) Instrumenting teacher characteristics with those of the schools in which the student should be enrolled had he not transferred.
(6) With weights inversely proportional to the estimated probabilities of not having a missing score.