Need help with SPSS loglinear

I'm using SPSS loglinear to test for multicollinearity between 7 discrete predictor variables. Details of my model:
  • N=7265 (each case represents a child)
  • 5 of the 7 variables have dichotomous categories
  • 2 of the variables have 5 categories (1-5)
  • Total cells in the multiway contingency table is 800
  • 3 of the 7 variables measure orphanhood (only the mother is dead, only the father is dead, or both parents are dead; there is independence between these 3 variables)

In the SPSS output K-Way and Higher Order Effects table, 1st and 2nd order effects are significant (p=.000 for both levels); however 3rd and higher order effects (up to the 7th level) are perfectly insignificant (p=1.000 for all levels). This seems too-good-to-be-true.

I've troubleshooted by removing the orphanhood variables from the model, (which seems to impact the results a bit), and have also tried collapsing the 5 categories (in the 2 variables) down to 2 categories (which also seems to impact the results somewhat). I suspect the perfect probabilities may be related to the rare orphanhood cases and/or too many cells in the table, i.e both are ratio of cases to cells issues. But I'm really not sure because it's the first time I've ever used loglinear analysis. Any suggestions for further troubleshooting, or should I simply accept the results as valid?? Thanks!
Last edited: