The short answer is not bother too much with Hausman tests.
The long answer is:
(1) Re-parameterize your model as a multilevel model;
(2) Decompose your variables into 'fixed' (within) and 'random' (between)
(3) Run it in R (through gretl, if you wish), and then;
(4) Do a joint F test of the equality of coefficients.
In R, it would look something like this:
library(car,lme4) # assuming you have these packages installed in your R
attach(mydata) # R purists like to lecture you about not using -attach()-;
for now, ignore them
(NB: -timevar- should be ordered by way of a time counter starting at 1
*within* each of your -groupvars-.)
If you have run the above routine successfully, it means that you have now
tested for the equality (or not) of your fixed and random effects, thus
obviating the need for any Hausman tests.
I hope that helps and good luck.
On 26 April 2018 at 03:05, 3jlema <3jlema(a)gmail.com> wrote:
I conducted a random effects panel model (90 observations, 15
cross-sectional units, 6 time-series length). The model has only one
dichotomous predictor (coded 0 and 1). May i know why Hausman Test is not
generated by Gretl?
The Breusch-Pagan test is significant (p<.001). I would appreciate any
suggestion on the appropriate model for my data. That is, is it safe to use
random effects or fixed effects? Again, why Hausman Test is not computed by
Gretl-users mailing list
"My colleagues in the social sciences talk a great deal about methodology.
I prefer to call it style." -- Freeman J. Dyson