Helping some and harming others: Homework frequency and tradeoffs in student performance

Abstract

The authors of this article examine the potential for increased student learning and retention through more frequent assignments. They conduct a field experiment that investigates whether student knowledge retention can be improved by increasing the frequency of homework assignments, motivating students to have more exposure to the material, and reducing the incentives for students to procrastinate. They find that the impact of the treatment on student performance varies based on the student’s past academic performance. Students on the lower end of the academic performance distribution benefit from the structure imposed by more frequent assignments and perform better. However, students with high prior academic performance are harmed by the intervention. The final exam scores of high-performing students are lower in courses with higher assignment frequency.

Keywords:

• Achievement
• behavioral economics
• heterogeneous performance
• homework assignments
• self-determination theory
• undergraduate teaching

JEL CODES:

• A10
• A22

Acknowledgment

The authors thank the attendees of the April 2016 University of Kentucky Economics Workshop and the 2016 Conference on Teaching & Research in Economic Education (CTREE) who provided valuable comments. They are also very appreciative of comments from Justin R. Roush, discussant at the 2016 CTREE meetings and the attendees of the 2019 Spanish Economics Association, and they are also very grateful for excellent research assistance from Trang Pham and Jubal Miller.

Notes

1 See, for example, Cooper et al. (1998), De Jong, Westerhof, and Creemers (2000), Gustafsson (2013), Holmes and Croll (1989), Keith and Cool (1992), Kember et al. (1996), Tymms and Fitz-Gibbon (1992), and Walberg (1991).

2 There are a few examples of ambiguous or contrary findings with respect to the positive relationship between homework and achievement. See, for example Dettmers, Trautwein, and Lüdtke (2009), Trautwein (2007), and Maertens and Johnston (1972).

3 These two institutions are both small private residential liberal arts colleges, and the courses were taught during a standard U.S. in-person 15-week semester. IRB approval was granted for both Institution 1 (protocol 107) and Institution 2 (protocol 405) and verified as exempt according to 45CFR46.101(b)(2).

4 Our results are robust to the inclusion of these students.

5 In this experiment, intermediate macroeconomics was taught at Institution 1 and principles of macroeconomics was taught at Institution 2. Therefore, the institutional dummy variable also captures differences related to course content and instructor fixed effects.

6 The independent variables were tested for multicollinearity, and the results show no major concerns. The highest correlation coefficient is −0.79; however, this is between two dummy variables created from the same categorical variable (with a third omitted category). So, we would expect these variables to be somewhat related. The next highest correlation is between Econ Major and Institution 1 at 0.49. This relationship also makes sense because intermediate macroeconomics was taught at Institution 1 and principles of macroeconomics at Institution 2. We would expect more economics majors in the intermediate macroeconomics course. However, again, this correlation is not high enough to be concerning. All remaining correlations were below 0.35.

7 Our measure of GPA is student self-reported. There are concerns with student self-reported measures of student success (Kuncel, Credé, and Thomas 2005), but it is a common control in the literature when the data are not made available by the registrar (see, for example, Ballard and Johnson [2004]; Al-Bahrani, Patel, and Sheridan [2017]; Linask and Monks [2018]).

8 All subsequent regression specifications were also run with the additional control variables class rank (year in college) and number of math courses taken (in high school and college). All results are robust to the inclusion of these explanatory variables. These results are not reported here but are available upon request.

9 We do not control for ACT or SAT scores in the results reported here. Controlling for ACT or SAT scores would result in a loss of 44 and 18 observations, respectively.

10 The GPAs shown in table 5 are selected based on percentiles in the sample.