Clustering Variable I have a set of 10 variables (v1..v10) which are continuous. I have another two control variables ctrl11 and ctrl12 which of course are categorical. ctrl11 can take any values from 1 to 50 and ctrl12 can take any value from 1 to 200. 
I am trying to shrink ctrl11 and ctrl12 to about 5-10 categories rather than the current 50 and 200 respectively. The end goal is to use the shrunk version of one of these two in a logit regression along with v1..v10, ie a total of 11 DVs. So my idea is to see how the variables v1..v10 cluster. 
Is this strategy sensible?
How can I find which control variable they cluster along so that I shrink that variable and discard the other for use in the regression? 
 A: As an aside, there is no "of course" that control variables have to be categorical. Control variables can take any form - dichotomous, categorical, ordinal, continuous....
Next, your general goal of collapsing some categories is almost certainly worthwhile - even if your N is large enough to include all those levels, it will be hard to evaluate a model with all of them included. But then your final sentences seem to switch the goal to using only one of the two variables - that seems unnecessary; it is certainly reasonable to have more than one control variable (provided they aren't colinear and that you have large enough N).
It would really help to know what these variables are. But, generally, I would collapse categories based on substantive knowledge, not statistical considerations. For example, if one control variable was "state" (as in which of the United States) I might collapse into region. I would also look at the frequencies of each category - categories with small frequencies would be higher on my priority list for combining. Sometimes, a few categories have to be combined into an "other" category. 
