A Comparison of the Usefulness of SSW and SST Structure Coefficients for Identifying Variable Importance in Descriptive Discriminant Analysis Following a Significant MANOVA: Examination of the Two-group Case

• Activity Number: 254
• Type: Contributed
• Date/Time: Tuesday, August 5, 2003 : 12:00 PM to 1:50 PM
• Sponsor: General Methodology
• Author(s): Mercedes Schneider*+
• Companies: Ball State University
• Keywords: descriptive discriminant analysis ; variable importance ; MANOVA post hoc procedures ; structure coefficients

Abstract

There is some question as to when one should rely upon SSW or SST structure coefficients (SCs) in order to determine continuous variable importance when using descriptive discriminant analysis (DDA). One rule is that the SSW SCs are appropriate if the groups come from separate populations with identical covariance matrices and that the SST SCs are appropriate if the groups represent one population. The current study employed Monte Carlo simulation to compare patterns of SSW and SST SCs in a two-group DDA analysis where the two groups were assumed to be from different populations with identical covariance matrices. Potential SSW and SST SC patterns were identified across all levels of continuous variable intercorrelation (R) for a given set of population mean vectors. Comparisons of identified SSW and SST SC patterns revealed minor differences between the two coefficients. Both SSW and SST SCs yielded patterns such that decisions regarding variable importance could be reached. Proportions of coefficients following the identified pattern were calculated; preliminary results indicate that both SSW and SST coefficients are useful depending upon effect size, n and R.