The supermatrix technique: A simple framework for hypothesis testing with missing data

Abstract

We present a new paradigm that allows simplified testing of multiparameter hypotheses in the presence of incomplete data. The proposed technique is a straight-forward procedure that combines the benefits of two powerful data analytic tools: multiple imputation and nested-model χ2 difference testing. A Monte Carlo simulation study was conducted to assess the performance of the proposed technique. Full information maximum likelihood (FIML) and single regression imputation were included as comparison conditions against which the performance of the suggested technique was judged. The imputation-based conditions demonstrated much higher convergence rates than the FIML conditions. Δχ2 statistics derived from the proposed technique were more accurate than such statistics derived from both the FIML conditions and the regression imputation conditions. Limitations of the current work and suggestions for future directions are also addressed.