controlled for. After controlling for personal characteristics, the model's fitness (adjusted R2
) increased by 0.07-0.1 for University
A and by 0.16–0.4 for University B, suggesting that personal characteristics add to the explanatory power of the model.
The interaction of the liberal-art track and CEE captures the different effects of CEE liberal-art and science-engineering track
exams on college GPAs. The estimates in Tables 1A and 1B suggest that college GPAs are considerably lower for liberal-art track
examinees than the science-engineering track in both universities. The interaction of the liberal-art track and CEE is significantly
positive for University B, suggesting that the effect of the liberal-art CEE score on college GPAs is stronger. This effect is of the
same direction but not significant for University A, which is likely because of the small number of liberal-art exam takers in the
University-A sample.13
Moreover, several individual characteristics are highly correlated with college performance. The sign of the estimates is similar
for both universities, except that for University A the estimates for first-time exam takers are statistically significant, while for
University B, first-time exam takers are insignificant but the estimate for urban is significant. For both universities, gender is
significantly correlated with college grades, suggesting that females have higher undergraduate GPAs than males throughout the
4 years of study.
Since we only obtained the longitudinal data from University A, we estimated Eq. (1) using data for each entry class from 1995
to 2005 in University A. To save space, we only reported the coefficient estimate for the CEE score and R2 for the regression of the
freshman year GPA. Without the control for provincial dummy variables, R2 and coefficient estimates for CEE both declined from
the 1995 to 2005 class. But after we controlled for provincial fixed effects, the pattern changed. R2 and coefficient estimates first
declined then increased (Fig. 1). After the year 2000, many provinces were given authority to write their own exam papers under
the same general guideline. The decline in the explanatory power of CEE without control for provincial fixed effects could be
because of differences in exam papers between provinces.14