Re: [Gretl-devel] Lee's bkw.gfn and vif (was Re: (now vif-related) Re: new accessor)
Am 15.03.2019 um 10:51 schrieb Sven Schreiber:
Am 15.03.2019 um 10:48 schrieb Riccardo (Jack) Lucchetti:
>
> Oh, sorry, I had read your message to superficially, sorry. I thought
> you meant "non-linear" as in "non-OLS". Yes, I think we could
enable
> that.
I did indeed mean non-OLS, nls was just an example, as was your probit
example. So sometimes it works and sometimes it doesn't, it's not clear
to me by what criterion. The aim would be to enable that always, except
when there are intrinsic reasons not to do it.
OK, first of all, here's a reproducible crash recipe (careful!):
<hansl>
open kennan
duration Ti 0 Prod
vif
</hansl>
I discovered this when I semi-systematically checked the estimation
commands if they allow a 'vif' afterwards. Here's my list:
* Works already: heckit, logit, quantreg, probit, intreg, lad, poisson,
tobit
* Doesn't work with vif:
- nls
- mle
- garch (presumably also arch)
- midasreg
- gmm
- biprobit
- tsls
- duration (see above)
I'm sure some of these are intrinsically impossible with vif (midasreg?
biprobit?), but others look as if there is no objective reason.
thanks
sven