Often poorly chosen functional forms show themselves up in patterns on a
plot of residuals versus predicted. Conversely, a satisfactory
functional form shows up, after some informal or formal smooth, as a
flat trace on such plots. Naturally, you should check other diagnostics
too, e.g. variance functions.
I see no reason why this should not apply in broad terms to panel data
too. A disadvantage of this is that it does not produce a P-value or
sharp test decision as you evidently prefer. An advantage is that it is
flexible and allows you to apply scientific as well as statistical
judgement.
Nick
[email protected]
[email protected]
to compare different functional forms such as linear, log log, linlog,
loglin i would like to do a PE-test. I found -logsumm in Stata Bulletin
1992, but this does not fit with Panel data and i am not sure how to
compare others than linear and linlog.
I found the same question in the archive but unfortunatly without
solution.
(http://www.stata.com/statalist/archive/2007-02/msg00570.html)
Did anyone use that test in the meantime and could provide some help how
to conduct it manually?
The test is described for example in Verbeek, Marno:A Guide to
Modern Econometrics, 2004 p.61
*
* 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/
