Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | statalist <statalistrw@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: degrees of freedom correction: Frisch Waugh |
Date | Fri, 22 Oct 2010 08:18:10 -0400 |
I am estimating a linear model with a large number of fixed effects (e.g. worker/firm, student/teacher, etc...). One way to do this would be to use a two-way fixed effects estimator (e.g. a2reg, felsdvreg, etc...), but I am not really interested in the estimates of the fixed effects per se. An alternative would be to estimate the model in stages, using a wage equation as an example: xtreg wage, fe i(worker_id) predict wage_res1, e xtreg experience, fe i(worker_id) predict experience_res1, e xtreg wage_res1, fe i(firm_id) predict wage_res2, e xtreg experience_res1, fe i(firm_id) predict experience_res2, e reg wage_res2 experience_res2 This should produce the exact same result as "xtreg wage experience i.firm_id, fe i(worker_id)" with the exception that the standard errors will be slightly smaller in the former example due to a degrees of freedom issue. My question is as follows: is there an easy way to correct the standard errors from the two+ stage model to mimic the one stage model? This obviously will not make a huge difference in large samples, but I want to know if such a solution exists. * * 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/