Dear R helpers, I need to estimate a probit model with box constraints placed on several of the model parameters. I have the following two questions:
1) How are the standard errors calclulated in glm (family=binomial(link="probit")? I ran a typical probit model using the glm probit link and the nlminb function with my own coding of the loglikehood, separately. As nlminb does not produce the hessian matrix, I used hessian (numDeriv) to calculate it. However, the standard errors calculated using hessian function are quite different from the ones generated by the glm function, although the parameter estimates are very close. I was wondering what makes this difference in the estmation of standard errors and how this computation is carried out in glm. 2) Does any one know how to estimate a constrained probit model in R (to be specific, I need to restrain the range of three parameters to [-1,1])? Among the optimation functions, so far nlminb and spg work for my problem, but neither produces a hessian matrix. As I mentioned above, if I use hessian funciton and calculate standard errors manually, the standard errors seem not right. Many thanks in advance for your kind help. Maomao [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.