Addressing Missing Data in Educational Evaluation: Predictive Mean Matching Imputation for Test Score Data

• In Event: Missing Data, Multicollinearity, and Variable Selection in Regression

Abstract

Multiple imputation is widely-used to address missing data in quantitative research, particularly with education data. One limitation is the imputation of implausible values that fall outside of the acceptable range for a given continuous variable. With education data, this can be problematic when test scores generated through imputation are beyond the test scale, potentially biasing analyses. Predictive mean matching (PMM) has emerged in recent years as a potential solution to preserving the distribution of imputed variables. Few studies in education have used PMM, however, and its utility has not been thoroughly investigated. Using multiple diagnostic tests, we compare PMM with other traditional approaches to imputation using educational data from a public school district implementing an integrated student support intervention.

Authors

• Noman Khanani, WestEd

• Jordan Lawson, Boston College