Hi Joshua,
Do you absolutely need to do a random effects model? I assume you have panel data and by the looks of it you are running wage regressions. If so the main problem you have is to properly account for the serial correlation of individuals over time. An alternative to running a random effect model would be to just pool everybody over time (however keep separate equations for the black and whites) and run standard OLS and use the cluster option to get the standard errors. So in your case you would run OLS for each of the back and non-black groups - without the cluster command. Then use suest with the vce(cluster id) command. This will provide you with consistent estimates and appropriate standard errors.
Anyways, I hope this helps..
Jason
________________________________________
From: [email protected] [[email protected]] On Behalf Of Joshua A Shindell [[email protected]]
Sent: Monday, March 23, 2009 1:17 PM
To: [email protected]
Subject: Re: st: AW: RE: cross-model hypothesis test
Martin, thank you very much for the reply, I ran the following code you provided:
webuse nlswork, clear
generate age2 = age^2
generate ttl_exp2 = ttl_exp^2
generate tenure2 = tenure^2
generate byte black = race==2
xtset idcode
xtreg ln_w grade age* ttl_exp* tenure* not_smsa south if black, re
estimates store first
xtreg ln_w grade age* ttl_exp* tenure* not_smsa south if !black, re
estimates store second
suest first second
I get the following error code:
first was estimated with a non-standard vce (conventional)
r(322);
Any additional suggestions?
Thank you very much,
Joshua A. Shindell
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
