st: Robust cluster variance estimator: Clustering on a "lower" level than unobserved effects

[Christopher Parker <christopher.parker101@gmail.com>]

Dear Statalists,

I'm analyzing the development of a liberalization Indicator across 24
countries and 7 network industries (Timeperiod: 1975-2007).
I'm regressing the indicator for each industry in each country on
explanatory variables that vary only on a country level and a
timetrend. Additionally im controlling for country and industry
specific time constant effects via dummy variables. Due to intragroup
and serial correlation I want to make use of the robust cluster
variance estimator. I'm of the opinion that controlling for country
specific effects that have an effect on all industries,  implies the
errors terms of industries within a country being related to each
other. In this case clustering seems like the obvious solution to me.

Here's my problem: Due to the fact that I only have 24 clusters and
already 24 country dummies, I cant calculate the joint F statistic. My
idea for a possible solution is using clusters for each industry in
each country (7*24=168 clusters). Not only would this mean that I can
perform the F-test, as far as I understand it would also improve
asymptotic properties of the variance estimator . The only thing is:
Im unsure if I'm "allowed" to cluster on a "lower" level/ if it causes
any other problems? What is your opinion on this ? Remark: I fould
similar threads like this but they did not directly answer my
question.

Bonus question: If there are unobserved effects for a speficic
industry of a certain country(eg. switzerland telecom), do I have to
worry about them being related to explanatory variables and thus
causing an endogeneity problem? Intuitivley I would say no, as the
explanatory variables only vary on a country level.

I would be thankful for your help!

"Chris"- Christopher Parker
-Master student-
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/