Practice Makes Perfect? Self-testing with External Rewards

Abstract

Create your own luck: more practice results in a better outcome. We evaluate this conventional wisdom for a group of economics and business students enrolled in a university math course (final dataset N = 280). During the semester, students were offered a set of optional, additional exercises (some awarding extra credit, some not) and a web-based linear algebra game. Controlling for a large set of personal characteristics using machine learning methods (e.g., Lasso regression, random forest), most notably ability as measured by high school GPA and motivational variables, we find a statistically significant effect of voluntary practice on exam performance. The estimated size of the causal effect is large in terms of final grade, i.e., practice leads to better grades. A subsequent quantile regression analysis showed that the overall effect is most pronounced for weak students who would be otherwise expected to have a low exam score.