Suryadipta Roy--
If you use the 1500 observations, or 15 years of data on 100
countries, and you use the option cluster(country) on your regression
command, then you take account of dependence of the errors within
country over time, which allows the "effective sample size" to be some
number between 100 and 1500. In the case where there is really no new
information across years, the cluster() option should make the results
look more like the -collapse-d version in terms of statistical
significance of coefficients. The cluster() option is advisable for
the 1500 obs case even if you have a lot of within-country variation
because it is likely you have some kind of serial correlation.
Example:
webuse grunfeld, clear
egen mean_inv=mean(inv), by(comp)
egen mean_k=mean(k), by(comp)
reg mean_k mean_inv, nohe
reg mean_k mean_inv, nohe cluster(comp)
collapse k inv, by(comp)
reg kstock inv, nohe
*
* 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/
