perhaps you are running into some numerical issues (exping a really big
number for example) and so shrinking the covariate helped with that problem.
if Zelig works, then use Zelig.
On Tue, Mar 24, 2009 at 12:02 AM, Christopher Michael Muller <
muller at fas.harvard.edu> wrote:
Hi all,
Quick optim question:
The paper Roman and I are replicating uses discrete-time event-history
logistic
regression. We've been able to get the right point estimates and standard
errors thus far using Zelig's logit commands, but when we program our own
logit
function, we run into some problems. Specifically, the Zelig and numeric
MLE
estimates are nearly identical when we specify the model with all 16
covariates
but one. Adding the final covariate radically shrinks all parameter
estimates.
The final covariate is on a greater scale (thousands) than the others
(tenths
to tens). If we divide it by 1000 and then run the regression, we get the
right
estimates for the other covariates and (unsurprisingly) an estimate 1000
times
greater than the correct value for the final covariate. My guess is that
this
has something to do with how optim is performing optimization. Does anyone
have
any suggestions?
Many thanks,
chris
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
--
Patrick Lam
Department of Government and Institute for Quantitative Social Science,
Harvard University
http://www.people.fas.harvard.edu/~plam