Re: Again the requrest: SIMULATION TO INTERPRET AND PRESENT LOGIT RESULTS without formatting - hopefully looks better

[David Hoaglin <dchoaglin@gmail.com>]

Dear Alex,

The reason for the .01 in line 7 is explained in Appendix 1 of the
Zelner (not Zellner) paper.  For the particular example, the code is
stepping the variable from .02 to .19.

I have not studied the paper closely enough to see whether Zelner is
being careful about the  fact that changing one variable in the model
while holding others constant may produce predictions that are not
supported by any data.  The wider confidence intervals in one of the
figures can warn about such extrapolation.

It should be possible for you to explore the surface of predicted
probabilities from your model, as a function of the two variables and
their interaction.

I would like to see a discussion of the "marginal effect" (in footnote
1 of the paper) in connection with the fact that the coefficient of a
predictor variable in a regression model tells how the dependent
variable changes (on average) in response to a change (of 1 unit) in
that predictor variable, adjusting for the contributions of the other
predictors in the model.  In general, it is not correct to interpret
that coefficient with language that talks about holding those other
variables constant; that is simply not how multiple regression and
similar methods work.

Regards,

David Hoaglin
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/