Reexamining the Growth Mindset Treatment Effect: An Illustration of Why Distributional Assumption Matters

• In Event: Advanced Techniques in Multilevel Modeling and Treatment Effect Analysis

Abstract

Despite the ordinal and censoring nature of GPA scores, they are frequently modeled with the normality assumption. One natural choice for modeling ordinal variables is the generalized ordinal logistic regression that relaxes the parallel trends assumption. However, this model requires more degrees of freedom and can be unnecessarily exhaustive. To re-examine the treatment effect of the growth mindset intervention on student post-treatment mathematics GPA, which has a zero-inflated distribution, scores were rounded down to integers. A two-level hurdle model was applied on these integer values to account for variation across schools and a significant treatment effect was found for students on the cusp of failure, reducing their probability of failing or dropping out of the class.

Authors

• Jiachen Liu, Michigan State University

• William H. Schmidt, Michigan State University

• David Yeager, University of Texas at Austin