Here are some issues worth examining. Initial thought is to examine them from unweighted and weighted file-quality perspective, but those that improve file quality also need to be explored from disclosure-risk perspective.:
Pros/cons of different seed variables or X variables.
Pros/cons of different orderings of variables for synthesis, and figuring out to extent practical the logic of ordering.
Pros/cons of splitting the file - fitting and synthesizing mutually-exclusive subsets and then putting the synthesized subsets together.
More-complex forms of fitting and predicting - methods and functional form, two-step procedures, etc.
Here are some issues worth examining. Initial thought is to examine them from unweighted and weighted file-quality perspective, but those that improve file quality also need to be explored from disclosure-risk perspective.: