I've been running logistic regression models in SPSS using the LOGISTIC REGRESSION and GENLIN commands and have found some strange differences in the output. For the same model specification, I'll get the same omnibus test results, variable coefficients and odds ratios (and p-value). The Wald-Chi Square tests of model effects are the same in both procedures for my covariates, for one of my factor main effect terms with two levels, and both of my interaction terms.
However, the model effects tests for my main factors of interest (one has five levels, one has three, one has six) are very different. The GENLIN procedure produces Wald Chi Square values that are smaller than the LOGISTIC REGRESSION procedure, with substantial changes in p-value.
What's the cause of the difference, and is one procedure more acceptable than the other?