Hello all,
I have two questions about logit model specification.
First, I specify a logit model: p(y=1)=f(x1 x2), but cannot decide
whether to use logged "x1" (i.e., ln(x1)) or not. "x1" has a st dev of
815.
I care about this because "x1" is statistically INsignificant while
the logged "x1" is significant, although the coefficients have the
same sign in both cases. The models with or without logged "x1" are
otherwise almost identical.
I do find that the Log pseudolikelihood WITHOUT logged "x1" is
-1865.9909, which is way lower than the Log pseudolikelihood WITH
logged "x1", -1390.1781. Does this suggest that I should use logged
"x1"?
Second, suppose I also want to examine the interaction effects of x1
and x2 which are correlated. There is no theoretical reason to simply
discard one variable. If I use logged "x1", and decide to center both
ln(x1) and x2 at the mean, can I then have the interaction of the
centered variables? The result is pretty hard to interpret in this
case, isn't it?
I would greatly appreciate any advice about these two questions about
logit models.
Aaron
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
