Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Poolability Test - Fixed Effect Estimation, Panel Robust Errors or a Roy-Zellner Test


From   Carla Müller <csm53@cam.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   st: Poolability Test - Fixed Effect Estimation, Panel Robust Errors or a Roy-Zellner Test
Date   Sun, 23 Feb 2014 11:01:55 +0000

Dear all,

I have a panel of 33 countries over 7 periods (7*3 year period averages). I am trying to perform a test for poolability.

My restricted model is a fixed effect model with homogeneous slopes, my unrestricted model is a fixed effect model with slopes on the variable of interest varying across countries (maintaining the assumption that the coefficients on all control variables are constant across countries).

Following Vaona (2008) ("A quick trick to perform a Roy-Zellner test for poolability") I have created 33 interaction terms between the countries and the variable of interest and am able to run regressions with these 33 interaction terms.

My question concerns the testing procedure. Having read previous Stata posts on this topic (see e.g.: http://hsphsun3.harvard.edu/cgi-bin/lwgate/STATALIST/archives/statalist.0303/Subject/article-391.html) I remain unsure about which of the following approaches is valid:

(1) Run the unrestricted model including all interaction terms using a fixed effect estimator, then use an F-test to test the equality of all coefficients.

(2) Same procedure as in (1) but using panel robust (clustered) errors (or alternatively bootstrapped errors) (having tested for autocorrelation, heteroscedasticity and spatial correlation I am using these robust errors throughout the whole paper). Problem: Using cluster robust errors reduces the degrees of freedom and as a result I cannot test the equality of all interaction terms simultaneously (while I expected the degrees of freedom to be reduced to the number of clusters-1, Stata in fact allows me to test only 14 coefficients at a time (all additional constraints are dropped))

(3) Perform a Roy-Zellner test instead of a Chow test given the presence of non-spherical errors (e.g. "Econometrics", Baltagi 2008 ) - how would this be different from (2) which is also robust to non-spherical errors?
Problem: How do I perform a Roy-Zellner test in this case?

Thank you for any help and advice,

Best regards,

Carla Müller

Emmanuel College
Cambridge University




*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index