Plattner, Dankwart
> I encounter a behavior I don't understand. I'm running a regression
> (stripped down to its bones in order to better present my
> problem) with xi
> and an if expression:
>
> . xi: reg qa_ek i.r2 if r2>15
> i.r2 _Ir2_15-93 (naturally coded;
> _Ir2_15 omitted)
>
> Source | SS df MS
> Number of obs =
> 85825
> -------------+------------------------------ F(
> 2, 85822) =
> 132.51
> Model | 207587.248 2 103793.624 Prob
> > F =
> 0.0000
> Residual | 67223628.8 85822 783.291333
> R-squared =
> 0.0031
> -------------+------------------------------ Adj
> R-squared =
> 0.0031
> Total | 67431216.0 85824 785.691834 Root
> MSE =
> 27.987
>
> ------------------------------------------------------------
> ----------------
> --
> qa_ek | Coef. Std. Err. t P>|t| [95% Conf.
> Interval]
> -------------+----------------------------------------------
> ----------------
> --
> _Ir2_91 | -4.046854 .2492706 -16.23 0.000 -4.535422
> -3.558285
> _Ir2_92 | (dropped)
> _Ir2_93 | -2.513805 .2658109 -9.46 0.000 -3.034792
> -1.992818
> _cons | 22.55939 .2058561 109.59 0.000 22.15592
> 22.96287
> ------------------------------------------------------------
> ----------------
> --
>
> My questions are,
> - why does STATA use r2==15 as the omitted category when I
> excluded this
> category in the if exp? Is an if exp useless with xi when
> the same var is in
> the if exp and handled by xi?
> - why is r2=92 dropped? Is it because the omitted category
> was chosen
> wrongly (there should be no with the data)?
> Maybe somebody could possibly point me to an explanation
> and a remedy?
On the first, my guess is that you are seeing
a strict division of labour. -xi- has the job
of setting up the interactions. Only when -regress-
is fired up is there attention to any -if- restrictions.
To put it another way, there is code within the -regress-
command to do this, so it would have been pointless
for the Stata programmers to duplicate that code with
(much slower) interpreted
code within -xi-.
On the second, my guess is that this is just a
consequence of the relationships between
your covariates.
I doubt you did anything wrong.
Nick
[email protected]
*
* 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/
