Re: [Gretl-devel] Presentation of model results
On Fri, 21 Nov 2008, Sven Schreiber wrote:
Apart from the looks, I'm not sure why Akaike should be
preferred over Hannan-Quinn in terms of econometrics. I'm sure
AIC is good for some objectives, but AFAIK it is not consistent
in the sense that it doesn't pick the true model even
asymptotically, in contrast to Hannan-Quinn.
Am I mixing things up here, or are these thoughts irrelevnant
for the issue we're discussing right now? Otherwise I would tend
to either keep all or ditch AIC.
I agree, the AIC is not all that great. But it's the best known
of the information criteria and if we dropped it I guess we would
have people saying, why don't you print the AIC?
Here's another go at this topic. I've put up
http://ricardo.ecn.wfu.edu/~cottrell/dougherty/table.html
showing a few variants for folks to consider.
* Version A is what I posted yesterday. It is closely (too
closely?) based on eviews and does not have a space for the HQC;
also it prints the Durbin-Watson statistic regardless of whether
or not the data are time series (but does not print rho). We
probably don't want this.
* Version B is what Jack suggested yesterday (although it doesn't
fully correspond with his preferences since it omits the HQC). It
prints rho and DW only for time series.
* Version C is what I'm inclined to favour at present.
* Version D is what Jack canvassed in a later patch.
I'll make a case for C, though I'm not totally committed to this.
First, I don't think it's necessary to print the sample size in
this table, since we always show that in the header for model
output. Second, it's obviously a matter of opinion whether it's
"more logical" to have closely related items (e.g. the F-stat and
its p-value) on the same row (as in Jack's B) or next to each
other vertically (as in C). I find that my eye tends to group
things vertically given this sort of presentation, so I have a
slight preference for C in this regard.
Thoughts?
Allin.