After updating -cmp- from SCC, I am now able to use the cmp command
"predict XXX, lnl" as outlined by David. However, for right censored
data points the command does not yield observation-level log
likelihoods. Even when I try a simple interval regression:
cmp (WTP: low_bid high_bid = X1 X2 X3), ind($cmp_int).
Is there a remedy?
Chris
@Nick Cox: "Chris Rh" is willing to learn from the kind advices of
Master Cox
Am 02.11.2012 09:37, schrieb Chris Rh:
David, thanks for the advice.
Unfortunately, Stata keeps telling me that "option lnl not allowed"
although our model seems pretty standard:
cmp (RISK: risk= X1 X2 X3 Z1 Z2) (WTP: low_bid high_bid = X1 X2 X3),
ind($cmp_oprobit $cmp_int)
This is all the more puzzling because I can estimate the scores
(predict XXX, score), thus the likelihood must be stored somewhere.
Also, I shall mention that you don't list lnl as a valid option for
predict in the --cmp help file. Can you give me any other hint how to
retrieve the observation-level ll?
Thanks,
Chris
Am 02.11.2012 04:15, schrieb David Roodman (droodman@cgdev.org):
You need the observation-level log likelihoods? You can get them
post-estimation by typing "predict XXX, lnl" where XXX is the name of
a new variable.
Sent: Thursday, November 01, 2012 2:33 PM
To: statalist-digest@hsphsun2.harvard.edu
Subject: st: cmp and Vuong test
Dear all,
we have fitted a limited dependent variable model using David
Roodman's cmp command and want to do a Vuong test (Vuong, Ectrica
1989) for model selection. Is there a way to save the individual
likelihoods when running cmp models? If not how can we extract the
individual likelihoods?
