Re: st: lincom command

[Maarten Buis <maartenlbuis@gmail.com>]

On Tue, Mar 12, 2013 at 10:54 PM, Ari Samaranayaka wrote:
> Thank you Maarten for the comment. Your idea was my very first try, it was
> given up later for the difficulty of interpreting results. What you are
> saying is to treat A,B,C,D as 4 separate binary variables. Problem with that
> is overlap between those binary variables is high, therefore colinearity.
> For this reason estimated IRR for ethnicity A and B can be quite different
> eventhough overlap of people between ethnicities A and B are quite large. In
> that situation I am unable to interpret (understand) estimates.

You should get exactly the same effects if you correctly specify your
alternative model with -lincom-, so that won't solve anything. Anyhow,
the fact that there is a large overlap between categories does not
mean that their effects should be the same, so I don't see why the
behaviour you report represents a problem. However, a pattern like the
one you report would make me want to look at the effects of those with
only A, only B, and A and B, as I would suspect that the combination A
and B would in such situations represent a separate category in its
own right. But that is an empirical question, so estimate both models
and see if the results make sense.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/