Re: [Gretl-users] bivariate probit (in a loop)
On Tue, 21 Oct 2014, Artur Bala wrote:
Ok, here I'm back with an observation (just in case something may
need to
be fixed):
As Jack recommended I used the "catch" option with the probit command and
the loop didnt' stop anymore at that time...but it did stop yesterday when
applied to new data.
When I look at the detailed estimations into the loop, it seems that
there's no problem in estimating the probit itself and gretl shows up the
results normally. Which means that the error I encountered eventually has
no relation with the perfect prediction symptom we suspected initially.
So troubles should come when retrieving the $yhat series. Consequently, I
tried the following "catch genr series predict_external = $yhat[,1]" and
got the code error "30".
Any hint on that?
This means that the "hat" matrices were not added to the biprobit model,
which in turn means that some error was encountered before reaching that
point (failing to calculate the covariance matrix of the estimates,
perhaps).
There was a buglet in the biprobit code that allowed a model to be saved
without an apparent error condition in this sort of situation, so long as
the first phase of estimation seemed to go OK. That's now fixed in CVS:
the "biprobit" command will flag an error unless the whole process is
completed successfully. Updated snapshots will follow later.
Allin Cottrell