Class Attendance and Performance in Principles of Economics

Abstract

A sample of 347 students, enrolled in principles of economics classes during the period 1997–2001, is used to examine the relation between class attendance and student performance on examinations. Among the questions examined are: Is attendance related to performance, with and without controls for other factors? Do only substantial levels of absence matter? Do low test scores cause more frequent subsequent absences? Do the results change when individual heterogeneity (in addition to controls for differences in SAT and GPA) is considered in the context of random‐effects and fixed‐effects models, using panel data? Can overall attendance be proxied by attendance at six meetings at the end of the semester, and does such a proxy yield the same relation to performance as overall attendance? We also study the factors that appear to contribute to improved classroom attendance.

Key Words:

• Attendance
• university
• performance
• SAT
• GPA

Acknowledgements

The authors thank Sharon Cohn for invaluable help on every aspect of the development of this paper. Thanks also to Yingyot Chiaravutthi for his assistance with data coding and analysis, and McKinley L. Blackburn and two anonymous referees for helpful suggestions on the methodology. The usual caveats apply.

Notes

1. In fact, the coefficient of attendance in Romer's full model is significantly lower (at the 10% level or better, t = 1.42) when prior GPA is added than the respective coefficient when prior GPA is excluded from the model. Equation (6) below was used to compute the t‐statistic.

2. Other studies on college attendance include Addison (1995), Brocato (1989), Buckalew et al. (1986), Chan et al. (1997), Hovell et al. (1979), and Knox and Dotson (1969).

3. ST.SCORE = [SCORE – μ(SCORE)]/σ(SCORE).

4. Jones used absences rather than attendance, but the two measures are completely complementary.

5. Alternatively, we can test whether the coefficients of PCATT76, PCATT68 and PCATTL68 are significantly less than the coefficients of PCATT92 and PCATT84.

6. This issue was raised by both anonymous referees.

7. The idiosyncratic effect is assumed conditionally mean independent of X_it, PCATT_it (percent attendance for individual I during the attendance period t preceding the test) and μ_i. Moreover, E(ε_it ²| X_it, PCATT_it, μ_i) = σ _e ² for all t, with E(ε_it, ε_is| X_it, PCATT_it, μ_i) = 0 for t ≠ s. Unobserved individual heterogeneity is captured in the random effect μ_i. which is assumed conditionally mean independent of the variables X_it and PCATT_it (and has expectation = 0). Finally, the conditional variance of the individual effect is assumed constant (E[μ_i ²| X_it, PCATT_it] = σ_μ ²) and is uncorrelated across individuals (E[μ_i, μ_j | X_it, PCATT_it] = 0, for i ≠ j).

8. Further analysis of the factors influencing attendance is presented in the following.

9. The full regression is not shown here but is available from the first author upon request.

10. The full equations are available from the first author upon request.

11. t = 1.90, which indicates that the difference between the coefficients is significant at least at the 5% level for a one‐tailed test.

12. We also tested for non‐linearity by introducing quadratic, cubic, and log functions instead of the linear function for PCATT. There is no evidence that non‐linear functions yield superior results. In fact, the coefficients of the square and cubic terms are not statistically different from zero. And the R ² values for the linear, quadratic, cubic, and log functions are all nearly equal.

13. Note that we are not comparing coefficients, because in Table 5 the dependent variable is the standardized score, whereas in Table 8 it is the nominal score. The comparison here concerns the effects in the two cases on the nominal scores.

14. See SC Commission on Higher Education (2003) and Cohn, Balch and Bradley (2004).