I don't think you can conclude that standard errors would be smaller
using the pooled data and -lincom- or estimating separate regressions
by group. Here is an example regressing blood pressure on age groups
and race groups:
. webuse nhanes2
. qui xi: regress bpd i.agegrp*i.race sex
. lincom _Irace_2+_IageXrac_2_2
( 1) _Irace_2 + _IageXrac_2_2 = 0
------------------------------------------------------------------------------
bpdiast | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
(1) | 2.886063 .9669886 2.98 0.003 .9905782 4.781548
------------------------------------------------------------------------------
. xi: regress bpd i.race sex if agegrp==2
i.race _Irace_1-3 (naturally coded; _Irace_1 omitted)
Source | SS df MS Number of obs = 1622
-------------+------------------------------ F( 3, 1618) = 30.67
Model | 12325.8838 3 4108.62793 Prob > F = 0.0000
Residual | 216728.047 1618 133.948113 R-squared = 0.0538
-------------+------------------------------ Adj R-squared = 0.0521
Total | 229053.931 1621 141.304091 Root MSE = 11.574
------------------------------------------------------------------------------
bpdiast | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_Irace_2 | 2.892348 .9182858 3.15 0.002 1.091193 4.693502
_Irace_3 | -1.08313 2.070185 -0.52 0.601 -5.143656 2.977396
sex | -5.216742 .5758084 -9.06 0.000 -6.34615 -4.087333
_cons | 87.29417 .9318357 93.68 0.000 85.46644 89.1219
------------------------------------------------------------------------------
In any case, choosing a model because it has smaller standard errors
is a good way to introduce bias (sort of publication bias writ small).
The question you should ask yourself is, Should the covariance matrix
be estimated separately for each group? and the answer must come from
your theory of the data generating process, not your observations of
what gives you more statistically significant results.
On 12/1/05, Raoul C Reulen <[email protected]> wrote:
> Oke, let me clarify this:
> This is the model that I am using:
> > > [omitted]
*
* 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/
