Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Yi-Yi Chen <chenyiyi.tku@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: missing standard errors in ML estimation |
Date | Tue, 9 Mar 2010 01:31:23 +0800 |
Dear All, I wrote my own log-likelihood function for a fairly complicated model and estimated it using Stata's -ml max- (Stata 10). A simulation study (using generated data) showed that the coefficients are estimated pretty well. However, I noticed that, for each replication, Stata almost always returns missing values for some or all of the coefficients' standard errors. Why is that? Except for this anomaly, there is no other obvious problem. There is no "flat" or "none concave" type of warning messages, nothing. The only thing that I might add is that because the model's error structure is complicated and a Gauss-Hermite algorithm is used in the ML function, I choose a generous convergence criterion (ltol(1e-2) tolerance(1e-2) ) in the estimation. If I tighten it up, the model often has difficulty declaring convergence. If there is any problem in the Hessian matrix, wouldn't it lead to "flat" or "none concave" type of error message? Anyway, what may have caused the problem and how may I work around it? Thank you! Y. Chen * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/