Dear all,
I want to model the self-stated propensity to vote for parties of the
extreme right in seven European countries by fitting a linear regression
model.
The dependent variable is assumed to be affected by gender, age,
education and church-involvement. Theory suggests that the level of the
dependent variable as well as the effect of church-involvement are
country-specific. Therefore I included country-dummies as well as
interactions between country-dummies and church-involvement. The result
is this:
.xi:regress xrprob female age educat i.country*churchat ,robust
Robust
xrprob Coef. Std. Err. t
(stuff deleted)
female -.2281458 .0545067 -4.19
_Icountry_2 2.746436 .7298551 3.76
(stuff deleted)
churchatt .1102863 .1467994 0.75
_IcounXch -.6626895 .2569513 -2.58
(stuff deleted)
However, my observations are not fully independent since respondents
from the same country are prone to be affected by country-specific
factors not in the model. I therefore tried
. xi:regress xrprob female age educat i.country*churchat ,robust
cluster(country)
While the standard error for gender is a bit bigger now, the standard
errors for the country-dummies, the interaction and even for the
church-involvement variable are much, much smaller:
Robust
xrprob Coef. Std. Err. t
(stuff deleted)
female -.2281458 .0619557 -3.68
_country_2 2.746436 .020654 132.97
(stuff deleted)
churchatt .1102863 .0095317 11.57
(stuff deleted)
_IvcounXch -.6626895 .0091083 -72.76
Which standard errors should I trust? And more in general: Is it a
stupid idea to use the cluster()-Option in conjunction with variables
that indicate cluster-membership?
Thanks,
Kai
*
* 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/
