Apologies for cross posting
I have always been taught that the proper way to handle interactions in
a logistic model was quite similar to the way to do it in an OLS model:
Create an interaction term (usually the product of the variables
involved); add that to the model; and procede as usual.
Recently, I came acroos an article claiming that this is incorrect.
The article is in the STATA Journal, and was sent to me by a client, I
don't use STATA. The full cite is
Norton, EC; Wang, H; Ai, C (2004). Computing interaction effects and
standard errors in logit and probit models. The Stata Journal, v 4 no
2, p 103-116.
They claim that one needs an entire new method to deal with interaction
terms in logit and probit models; I didn't completely follow the math
(which is given very briefly), but they propose that the interaction
term and its se involve partial derivative and second order derivatives,
that they need to be compted per observation, and that they can be
significant for some observations and not for others. Writing these in
e-mail is very hard, but if anyone is interested I could try writing
them up and attaching them in a Word document
I am skeptical, since a) This seems wrong intuitively b) If it were
true, why not publish in some very prestigious journal? c) I've never
heard of the authors, nor are any of them in statistics departments
However, none of these reasons are conclusive. If anyone has read this
article, and could confirm my intuition (or confirm the results of the
paper) I would appreciate it
Thanks in advance
Peter
Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
www.peterflom.com
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)
|