Re: [Gretl-users] Multinomial Logit - getting the probabilties as predictions?
On Fri, 9 Jul 2010, Pindar wrote:
when performing a multinomial logit estimation Gretl gives as
fitted values a certain category for each id, which is nice...
Sorry, not as nice as it could be: there was a bug which biased
the fitted value towards prediction of the base case. This is now
fixed in CVS and the gretl snapshots for Windows and OS X. Thanks
for directing our attention to this area.
To make amends for this error ;-) I have now added an accessor
$mnlprobs. After estimation of a multinomial logit model this gets
a matrix of the outcome probabilities (rows = observations,
columns = outcomes).
logit ... --multinomial
matrix P = $mnlprobs
If you prefer, the columns can be extracted as series, as in
series P2 = $mnlprobs[,2]
No doubt this will get added to a GUI menu at some point, but
right now it's a command-line thing only.
Allin Cottrell