Multiple Imputation - Factor Variable Interactions

[pgr451]

It is advised that i create all interactions terms first, then impute them like regular variables. However, several of the interactions terms i need to create use factor variables (e.g factor * continuous). How do i create the factor interactions terms so that they can be imputed like regular variables.

[Konrad Zdeb]

My understanding is that you can specify interactions using i. and c. prefixes when appropriate. Maybe if you could post corresponding code it would be easier to reflect on your problem. As a side matter, please note that the forum requires users to use their full names.

[Richard Williams]

You just do it the old fashioned way -- you compute the interactions yourself. So, it would be something like

gen feminc = female * income

For more on why you should do things this way (or at least why Paul Allison thinks you should do it this way) see

http://www.stata.com/statalist/archi.../msg00602.html

http://www.stata.com/statalist/archi.../msg00613.html

A good article by Royston which includes a discussion of the "just another variable" approach for interactions is

http://www.ncbi.nlm.nih.gov/pubmed/21225900

[pgr451]

Here are the two variables.

Religious attendance
0 = none
6 = more than once a week

Sexual Abuse history
0 = none
1 = Only before the military
2 = only during ht military
3 = both before and during the military.

[Sarah Edgington]

Neither of those appears to be a continuous variable.
You also appear to be missing some categories for religious attendance. How are people who attend church 1 time per week (or once per month, or once per year) coded?

[pgr451]

Religious attendance is a continuous variables. Its (0=none) through (6=more than once a week).

[Sarah Edgington]

What are the categories between 0 and 6? It sounds to me like it's an ordinal variable. That is, the categories are ordered but the spacing between them is not equal. Treating an ordinal variable as continuous in your model may not be the best idea.