|
Abstract:
|
Multiple imputation (MI) becomes a standard approach for handling missing data. To date, software packages do not pool the imputed results from ANOVA (see https://www.youtube.com/watch?v=27NSGTcWaPI). As a special case, this presentation illustrates demands for different methods of result pooling in independent sample t tests. While software offers MI result pooling in either “Equal variances not assumed” or “Equal variances assumed” category, the "across-the-board" approach may not work when imputed data need a mixture of both methods according to Levene’s test findings. SPSS and SAS syntax is applied to well-documented, public data from UCLA to show the fact that some imputed data reject and others retain the hypothesis from a variance homogeneity test. Partition of the result aggregation is critical because it could lead to p< .05 or p>.05 in a t test on the population mean difference. Random seeds are included to facilitate the result replication.
|
