|
|
If you run the nominal LRA procedure with, say, 4 dv levels, one would expect the regression coefficients for the covariates to be the same as from 3 separate LRA's using the same ref group (e.g., if the ref group is 4 in the nominal LRA, then 1 vs 4, 2 vs 4 and 3 vs 4). However, this does not seem to be the case. Altho the coefficients are fairly close and the SEs similar, there is much more difference than rounding errors. Note that my analyses are based on complete data so there is no discrepancy in sample sizes among the runs. Anyone have any reasonable ideas why things don't work out as expected?
_______________________________________
C. Mitchell Dayton, Professor & Chair
Measurement, Statistics & Evaluation
University of Maryland
301-405-3626
www.education.umd.edu/EDMS/
|
Locked
