Comparisons of Three Strategies for Addressing Missing Data in a Graded Response Model

• In Event: Roundtable Session 117
  In Roundtable Session: Item Response Theory

Abstract

Three strategies for addressing missing data on a summated rating scale analyzed with the graded response model are compared, namely complete case analysis, ordered categorical logistic regression imputation with nose, and multiple imputation. If data are Missing Completely at Random (MCAR), the differences among the three methods are negligible. However, if data are Missing at Random (MAR), both complete case analysis and ordered categorical logistic regression produce biased item and ability parameter estimates. Multiple Imputation is a practical and conservative method of choice, since we never know whether data are MAR or MCAR.

Authors

• Arnond Sakworawich, Fordham University

• Jennifer Hill, New York University

• Charles Lewis, Fordham University