In article <[EMAIL PROTECTED]>, Ryan Schram <[EMAIL PROTECTED]> wrote:
>I am using R-1.7.0 for a logit analysis, as a part of a graduate seminar >in quantitative methods. The rest of the class is using LIMDEP, which (I >think) uses the Berndt, Hall, Hall, and Hausman (BHHH) algorithm for >doing MLE. The glm package in Splus/R uses iteratively reweighted least >squares (IRLS). What's the difference? Specifically, I think perhaps you, and certainly the respondants to your post, may be a bit confused. IRLS is a computational method for computing the MLE. I'm not familiar with BHHH, but it is presumably another computational method for computing the MLE. If both methods work correctly, you should get very nearly the same answer with either one. Furthermore, finding the MLE for logistic regression is a relatively easy convex optimization problem, for which one would expect the algorithms to work well, except in probematic cases of extreme colinearity or perfectly separable data. To be sure, you could set the number of iterations for IRLS to be larger than the default, which if I recall correctly is a bit small. Radford Neal ---------------------------------------------------------------------------- Radford M. Neal [EMAIL PROTECTED] Dept. of Statistics and Dept. of Computer Science [EMAIL PROTECTED] University of Toronto http://www.cs.utoronto.ca/~radford ---------------------------------------------------------------------------- . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
