Hi Clive
Thanks for your input. I still have a question. Does
the command collin incorporate strata, psu and
pweight?
I ran the svyreg and collin as you suggested. But
when I ran svyreg for each of the independent
variables to calculate the VIF manually, the VIF
values are slightly different than the output from
collin.
So it looks like collin does not take into account
strata, psu, and pweights. Am I right?
Thanks for your time.
Right. -collin- isn't set up to handle weights, clustering, whatever. I
think you would have to run a series of -svyreg- commands, regressing one X
on all the others, and from that you could compute the VIFs and
tolerances. It sounds like that is what you did. But, unless your life
depends on getting exact figures, I suspect -collin- would be an adequate
approximation. After all, you are mostly doing this for diagnostic
purposes, and my guess is it won't hurt you too much if your diagnostic
statistics are a little bit off. If somebody were so inclined, maybe they
could write a -svyvif- or -svycollin- command. (I don't know if there
would be some quirky issues of stat theory that would have to be addressed;
for example I thought it was sort of weird when I saw -svy- commands
producing F tests when I was expecting chi-square tests, although I
semi-understand that now.)
