Multiple imputation with restrictions or constraints

[emre]

I have some data in discrete choice format, and I have a question on using PROC MI for discrete choice data. I am using imputation to fill in some missing independent variables as well as the dependent variable, choice. The problem is that pairs of choices need to be mutually exclusive (one row of data is the chosen option, the other is the option not chosen), something that PROC MI does not take into account. Is there a way to enforce this condition? Should I be using a different method of imputing the data? Should I not bother imputing the dependent variable? Any advice is appreciated.

[emre · Accepted solution]

I will answer my own question if it is permitted. It may not be the most elegant solution.

To recap, my data looked like

pid	choice	var	...
1	1	1	...
1	0	1	...
2	1	0	...
2	0	0	...
3	0	1	...
3	1	0	...
...	...	...	...

To successfully use PROC MI, I "rolled" the observations with the same pid into a single row, so that my input would look something like this

pid	choice	var1	var2	...
1	1	1	1	...
2	1	0	0	...
3	2	1	0	...
...	...	...	...	...

for the same data as the first table. After PROC MI, I would split the rows into the format needed to use PROC MDC.

It turns out this was pretty trivial, I was just in a state of tunnel vision!