Going beyond the Mean: Using Variances to Enhance Understanding of the Impact of Educational Interventions for Multilevel Models

Variance heterogeneity is a common feature of educational data when treatment differences expressed through means are present, and often reflects a treatment by subject interaction with respect to an outcome variable. Identifying variables that account for this interaction can enhance understanding of whom a treatment does and does not benefit in ways that can inform and improve the treatment. Even in the absence of a treatment effect expressed through means studying variance heterogeneity offers insight into a treatment by identifying subject characteristics related to heterogeneity. This study illustrates four methods of modeling variance heterogeneity for data from a study of the impact of an engineering design-based STEM curriculum on student achievement with a focus on multilevel models.