Grades, Gender, and Encouragement: A Regression Discontinuity Analysis

Abstract

The author employs a regression discontinuity design to provide direct evidence on the effects of grades earned in economics principles classes on the decision to major in economics and finds a differential effect for male and female students. Specifically, for female students, receiving an A for a final grade in the first economics class is associated with a meaningful increase in the probability of majoring in economics, even after controlling for the numerical grade earned in the class. This suggests that for female students, the feedback that is embedded in the course letter grade has an encouragement effect on their decision to study economics further. The author finds no evidence of a similar effect for male students.

Keywords:

• economics major
• encouragement
• gender
• grades

JEL codes:

• A22
• D03
• J16

Acknowledgments

The author is grateful to Jennifer Borton, Emily Conover, Wayne Grove, Elizabeth Jensen, David Toomey, Julio Videras, and Stephen Wu for helpful comments. She is especially indebted to Rachel Friedberg and David Weil for their assistance and advice. Shirin Rashid provided valuable research assistance.

Notes

1. In a more dated study, Dweck et al. (1978) provided evidence that female students are more likely to attribute negative feedback to their own low ability than male students.

2. As is the case at most institutions, students can place out of this course with satisfactory performance on Advanced Placement exams, the international baccalaureate higher-level exam in economics, or the British A-level exam.

3. By leaving all sophomores in the sample, we find that some students may have learned their grade for introductory economics after they declared an initial major. However, removing these students from the sample does not change the major conclusions of this research.

4. The absence of pluses or minuses makes the distinction between letter grades sharper and may make the magnitude of the effect larger.

5. Interested readers can find additional advice for implementing a regression discontinuity analysis in the works of A. Nichols (2007a) and G. W. Imbens and Lemieux (2008).

6. Unfortunately, I do not have access to student characteristics other than those used in the analysis to verify this claim directly.

7. Because there was only one female instructor in the sample who taught the course multiple times, it is not possible to separate out the effect of instructor gender from the effect of that particular instructor. However, when instructor gender dummies are included instead of section dummies, the major conclusions of the paper are unchanged. Including a seasonal dummy instead of section dummies does not change the results either, and it is significant in the expected way: Taking the course in the fall semester is associated with an increased probability of majoring in economics.

8. Rask and Bailey (2002) found evidence that undergraduates are more likely to declare majors in fields in which they have been taught by instructors of the same gender or race.

9. In addition, the hypothesis that the coefficient on female*A is equal to the coefficient on male*A is rejected at the 5-percent significance level.

10. Although there may be omitted variables, a limited dependent variable version of the RESET test (based on the work of Pregibon 1980) indicates no specification issues at the significance level of 5 percent.

11. The regression discontinuity routine of Nichols (2007b) was used in this analysis.

12. Because there are many more smaller sections in this sample, rather than including section dummy variables, the estimations in columns 5–7 in table 6 include instructor dummy variables. Although not shown, the instructor dummy variables are insignificant. Because one instructor had few female students, none of whom majored in economics, it is not possible to estimate interactions between instructor and female student in the probit specification.

13. As in the main sample, the hypothesis that the coefficients on male*A and female*A are equal is rejected in this specification at the 5-percent level.