HLM vs. SPSS GLM repeated-measures

#1
HLM 7.0 - Different p-values in HLM vs. SPSS GLM Repeated-measures

My model is a 4-way interaction between 3 dichotomous dummy variables and a continuous variable, with 2 dummy variables at Level 1 (within-Ss) and 1 dummy and the continuous variable at Level 2 (between-Ss). I'm looking for the 2-way interaction at Level 2 to moderate one of the slopes at Level 1, so in SPSS I want a 3-way interaction and in HLM I want cross-level moderation of the appropriate Level 1 slope by the Level 2 interaction term.

When I run the model in SPSS GLM repeated-measures ANOVA with the full interactive model (i.e., interactions between all categorical and the continuous variable) manually specified in the Model box, I get one set of results. When I run the model in HLM, I get a different set of results. The only result SPSS and HLM have in common is the same p-value for the 4-way interaction. All the lower-order p-values are different. And the weird thing is that SPSS GLM is giving me significant results where HLM is not -- isn't HLM supposed to be more powerful than GLM? So I must be doing something wrong somewhere...

I've double-checked my restructuring of Level 1 and 2 datasets for HLM, and they're error-free. My dummy coding on the categorical variables is also error-free, and I centered the continuous variable prior to creating the Level 1 interaction term in SPSS (all prior to moving everything to HLM). I've double-checked my creation of the within-level interactions (2-ways) at Levels 1 and 2, and they're okay. I always toggle the error terms at Level 2 in HLM and fill in all Level 2 equations with the two Level-2 main effects and interaction term, as well as the Level 1 equations with the two Level-1 main effects and interaction term. So I am stumped.

Any ideas on why GLM in SPSS is giving me different (and better) results than HLM? I must be making some kind of error somewhere...

Thanks so much!