I have created 5 piecewise models for answering the same question for 5 age groups and the whole population in TYPE=COMPLEX. Then, I constrained each model to have equal slopes for comparison. Now I want to compare the models using both the 1 df chi-square difference test and the logliklihood. I used the formulas on http://www.statmodel.com/chidiff.shtml and computed them using the scaling correction factors. Yet my results differ for the chi-square difference test and the logliklihood. Shouldn't they yield the same answer? What could explain the different findings?
In large samples, the conclusion should be the same using either chi-square or the loglikelihood. If you want more information, send the relevant information and your license number to email@example.com.