Move "big problem" warning into matching routines

[markmfredrickson]

Right now we warn people when they create big distance matrices. The problem is that this line

match_on(treatment ~ x1 + x2, data = mydata) + exactMatch(treatment ~ mypairs, data = mydata)

would give a waring even if the exact match pairs up treated and control units (i.e., the call to match_on can't know it is about to get added to an exact match).

By moving to the matching stage, the user would get the warning after all components of the match have been combined. On the other hand, it might be helpful to inform him/her earlier. Thoughts?

[benthestatistician]

wouldn't this hypothetical user's computer be happier if she instead did

match_on(treatment ~ x1 + x2, data = mydata, 
                  within=exactMatch(treatment ~ mypairs, data = mydata) 
                )

or the equivalent? I suspect so, and also that we should be steering users away from building big distances. I think that's why we chose to put warnings in the distance building functions.

Perhaps warnings in themselves aren't quite sufficient to make the point. Maybe they'd be better if they pointed to a short vignette on the topic of managing problem size...