Re: st: Difference between using dummy variables and the '==1' command

[Maarten Buis <maartenlbuis@gmail.com>]

On Wed, Aug 24, 2011 at 4:58 PM, lreine ycenna wrote:
> I compared the regression results based on two different comamnds,
> namely those derived from dummy variables and the 'if female==1'
> command and found that at times these results can be quite different,
> though they can also be very similar in some cases. For example,
> comparing results from two separate regressions - (1) 'regress A B
> AxFemale BxFemale' (from which I then add up the coefficients for A
> and AxFemale), and (2) 'regress A B if female==1'.  Normally, at least
> I expect, the coeffcients for A and female should be very similar
> under these two commands, but I am unable to explain why sometimes
> these coeffcients (e.g. A and Female) are very different depending on
> the method used (e.g. 0.6 v.s -0.02).  I am wondering whether this is
> normal? Can anyone explain to me whether there is any significant
> methodological difference between these two comamnds?

If you do this in a linear regression and you added interaction terms
with all the variables than the only difference is the residual
variance, which is constrained to be equal across groups in the dummy
variable model (homoscedasticity) and is allowed to differ between
groups when estimating separate models. So you would, as you said, not
expect major differences between these models. Interpretation of those
main effects can be tricky, so I guess that the most likely
explanation is that you just misinterpreted them.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------

*
*   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/