[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: truncreg and discontinous likelihood

From   Rich Steinberg <[email protected]>
To   [email protected]
Subject   st: truncreg and discontinous likelihood
Date   Wed, 23 Sep 2009 15:27:54 -0400

I've been trying truncreg and no matter what maximization technique I try, I get a few "backups" and then the iteration terminates with the following message:

numerical derivatives are approximate
nearby values are missing
cannot compute an improvement -- discontinuous region encountered

Most everything I tried failed to cure this problem. -- variations in optimization technique, in set of RHVs. But then I clicked an option that I couldn't find documentation on -- "disable use of the Hessian-scaled gradient (nontolerance)" and I got convergence. However, results were flakey and suspect (clustering errors led to significant sign flips, most everything that is significant in any other estimation technique is insignificant here and some things that are insignificant elsewhere are significant here.

I found the reference in documentation to the instability of truncated regression coefficients in some circumstances, and my coauthor is reading about that. My questions for the list are

a) I understand what lack of concavity and backing up is about. But I am baffled by "discontinuity" of the likelihood function -- is there a simple explanation of what could cause it? b) Checking that "disable" box led to convergence, but since I don't understand the algorithmic issue here, can you reassure me that this solution is plausibly the correct one to the problem I was having. c) Are my flakey results inherent in doing truncated regression, or do they suggest that I am doing something else wrong (or, heaven forbid, that stata is doing something wrong). (forgive me if part of this is a double posting. I got indications when I tried to post it the first time yesterday that I failed)

Rich Steinberg
Department of Economics, 516 Cavanaugh Hall
Indianapolis, IN 46202-5140

"Unanswered email since 1955"

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index