Cutting our losses: The effects of a loss-aversion strategy on student learning gains

Abstract

In this randomized controlled trial with controls for student characteristics, the authors used a loss-aversion strategy to test whether students achieve greater learning gains from combining research-based instructional strategies with loss aversion. Students in the control group began the class with no grades and built their course average by completing assignments and exams. The treatment group began with a score of 100 percent on each assignment and viewed the deductions on scores from errors and missed work. The results indicated that students in the treatment group experienced gains in learning that were, on average, 5 to 13 percentage points larger than those of the control group. These learning gains were concentrated among students with low SAT math scores.

Keywords:

• Learning gains
• loss aversion
• randomized controlled trial
• research-based instructional strategies

JEL CODE:

• A22

Notes

1 For examples of how the strategies could be implemented in an economics course, see Cosgrove and Olitsky (2018, 2020).

2 While we can only speculate on students’ reasons for not completing graded work, we acknowledge that all assignments are low stakes because they are designed to facilitate learning rather than penalize errors. Perhaps students view the stakes to be too low to be worth their time. Moreover, in a course designed with RBIS, acceptance of late work is not an option because answers and explanations are posted as soon as the due date passes to ensure timely formative feedback.

3 The sample is 31 percent female and 69 percent male students. This result is consistent with Levitt et al.’s (2016) finding that boys were more responsive to incentives than girls.

4 It is not clear from the study when (if) students were made aware of the experiment and consent sought. The percentage of students by gender is not given.

5 Students took an average of 15 to 20 minutes to attempt the pretest because most questions were foreign to most of them.

6 Entry quizzes test knowledge of definitions and basic concepts. The videos were created by the instructor, a full professor with more than 15 years of experience teaching principles of microeconomics, using Camtasia for screen and audio capture.

7 Interleaving, or asking questions requiring knowledge or application of previously taught concepts or methods after time has passed and some forgetting has occurred, is an important RBIS. Interleaved questions also were included in entry quizzes and post-class quizzes.

8 While it is possible that students from one group talked with students from the other group and discussed the difference in appearance of their grades on upcoming assignments, no evidence of student awareness existed during the study. Anecdotally, when students were informed of the experiment on the final exam day, they displayed surprise and conversed with each other asking how they viewed their scores.

9 For this analysis, we use the percentage of quizzes and assignments completed; however, we conducted the analysis presented here for a number of measures of classwork participation and performance (score). These measures include separate analyses of performance on the quizzes (both pre-class and post-class), exit page performance, and exit page participation. For all of these measures, the results were qualitatively the same. For this reason and for simplicity, we confine our attention only to the completion percentage of all quizzes and assignments. The full results are available upon request.

10 We also examine the relationship between gender and the effectiveness of loss aversion. In Apostolova-Mihaylova et al. (2015), the authors report positive effects of loss aversion on males. To check this, we ran additional regressions in which we interacted the male variable with the treatment, period, and the treatment x period variables and found no significant effect of gender on the effectiveness of treatment. Because these results are not directly related to the present study, we omit them, but they are available upon request.

11 As a robustness check, we also conduct this analysis for the first and fourth quartile of SAT math scores. This reduces our sample size, but the results are very similar to the comparison between the lowest 50 percent and the top 50 percent of SAT math scores. For the first exam, neither the lowest quartile nor the top quartile has a significant effect of treatment. For the second exam, treated students in the lowest quartile of SAT math scores experienced learning gains that were 17.7 percentage points larger than those in the control group. For the top quartile of SAT math scores, the effect of treatment is insignificant. Likewise for the pretest/final exam comparison, treated students in the lowest quartile of SAT math scores experienced learning gains that were 11.9 percentage points higher than control students in the lowest quartile of SAT math scores. The treatment effect is insignificant for students in the top quartile. The full results are available upon request.

12 We did not survey students to determine their expected grades but think it is reasonable to assume that generally higher-achieving students expect to earn higher grades.