Data Imputation Algorithms for Mixed Variable Types: An Application in Large-Scale Educational Assessment

• In Event: 16.057 - Roundtable Session 5
  In Roundtable Session: 16.057-4 - Missing Data and Investigations of Statistical Assumptions

Abstract

Missing data is a major issue in working on large scale assessment programs in educational research. In addition, such assessments frequently include variables with a wide variety of distributions; i.e. continuous, categorical, counts. This variation in data distributions makes preferred methods for handling missing data, such as multiple imputation, less than optimal. The current simulation study investigates two methods for imputing missing data, random forest imputation and multivariate imputation using chained equations, that are gaining attention in genetics research, but have not yet become widely used in education. Results reveal that random forest imputation worked particularly well under a variety of conditions. Implications of these results and recommendations for practice are discussed.

Authors

• William Holmes Finch, Ball State University

• Maria E. Hernandez Finch, Ball State University

• Melissa Singh, Ball State University