On 26 April 2018 at 11:11, Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 26.04.2018 um 06:41 schrieb Clive Nicholas:
The short answer is not bother too much with Hausman tests.
Such an answer is not really helpful, and I don't think this personal
opinion of yours can be regarded as a professional consensus.
*Frankly, I could care less whether such things are "professional
consensus" or not. It is my personal view and I'm happy to give it. If the
OP finds it useful, so much the better. If they prefer to chuck my post
straight in the bin, that's for them. Moreover, my view is supported by a
highly readable paper on the subject by Bell and Jones (2012) "Explaining
Fixed Effects: Random Effects Modelling of Time-Series Cross-Sectional and
Panel Data" (available via Google Scholar), which promotes this approach
over Hausman tests and explains why.*
(1) Re-parameterize your model as a multilevel model;
> (2) Decompose your variables into 'fixed' (within) and 'random'
> (3) Run it in R (through gretl, if you wish), and then;
> (4) Do a joint F test of the equality of coefficients.
now tested for the equality (or not) of your fixed and random effects,
Typically the fixed or random effects are regarded as nuisance parameters
in these contexts. It presume it is not really interesting for the
application whether they are equal. That's one advantage of the Hausman
test which directly targets the coefficients of interest.
That said, your idea certainly has merit. There's nothing wrong with using
one of R's packages (especially with lme4 I've done it myself), but
providing more hints as to how to achieve it in gretl would even be more
welcome on a gretl list.
*I explained to the OP that s/he could run the R code (if s/he so wished)
through -gretl-. You may have missed that bit. I would welcome any effort
to see HLMs introduced in -gretl-, but it's not really my place to ask for
it. If people want it badly enough, they can learn to code it themselves. I
can't, so I won't.*
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