How do loglinear models and logistic regression differ in testing models?
They don’t, really, although the results will LOOK different. • In the loglinear model, we OMIT the terms we are interested in (e.g., all three-way interactions) and see what happens to the likelihood ratio chi-square. If the G2 increases sizably over the G2 that included the terms of interest (e.g., all three-way interactions), we say the model “doesn’t fit” and those terms must be included or “put back” to make the expected and observed counts in the multi-way table coincide within sampling error. Returning these terms to the model, lowers the G2 statistic, hence “small Chi-squares are good.” They mean the model fits. The saturated model and the MODEL SELECTION program will give you a good idea of which terms are necessary to have a model that fits, i.e., produces a small G2. The saturated MODEL SELECTION program is a real data drudger. It will tell you everything about how your variables fit together, and that’s why I recommend running it first. • In logistic regression, on the othe
