Dealing With Missing Data by EM in Single-Case Studies

• In Event: Marriott Roundtable Session 29
  In Roundtable Session: Missing Data Across Statistical Models

Abstract

The present study extends findings from Smith, Borckardt and Nash (2012) by examining the performance of EM under various missing data proportions, autocorrelations, phase lengths, and magnitudes of effects in an AB design. Findings suggest that neither missing data proportion nor magnitude of effects substantially impacted on the performance of EM. EM performed acceptably, in terms of relative bias, except when the intervention phase was long and the autocorrelation =.9. EM underestimated the RMSE of estimated effects. RMSEs of the level change estimates were consistently larger than those of baseline slope and slope change estimates. In general, results demonstrated that effects of a lag-1 time series model can be estimated by the least squares approach under a general linear model.

Authors

• Li-Ting Chen, University of Nevada, Reno

• Yanan Feng, McKinsey & Company

• Po-Ju Wu, Indiana University

• Chao-Ying J. Peng, Indiana University