How to Fit a Model to Different Subsets of Data

Full information maximum likelihood was operationalized early on by Paul Allison in the 80s. The idea of not imputing data is very attractive to me, and I think it's a useful starting point. If you have X and Z and then you estimate a model that has a multiplicative interaction, that actually just goes into in part defining the likelihood of the model. And you don't have that comes up because the XZ and the XZ product are another variable that goes into whether or not they're weighted for probability.