And did you actually look at the fitted values? I got 22 ones. For a substantial part of your x1-x2 space there are no failures. The warning is telling you that the fitted probabilities are so close to one as to be unreliable. The largest is 1-exp-20!
On Thu, 6 Nov 2003, L Z wrote: > Not real data. It was gererated randomly. The original codes are the following: > > par(mfrow=c(2,1)) > n <- 500 > > ######################### > #DATA GENERATING PROCESS# > ######################### > x1 <- rnorm(n,0,1) > x2 <- rchisq(n,df=3,ncp=0)-3 > sigma <- 1 > u1 <- rnorm(n,0,sigma) > ylatent1 <-x1+x2+u1 > y1 <- (ylatent1 >=0) # create the binary indicator > ####################### > #THE Probit Estimation# > ####################### > probit<-glm(y1~x1+x2-1, family=binomial(link=probit)) > bp<-probit$coef[2]/probit$coef[1] > bp; > I also tried family=quasibinomial. There seems no error message. But the result is > different from what I got from Gauss. For u1 belongs to another distribution (not > normal), the difference is even larger. I used the same data for the comparison. > > Thanks a lot! > > Steve Sullivan <[EMAIL PROTECTED]> wrote: > Is this simulated or actual data? > > STS > > Steven Sullivan, Ph.D. > Senior Associate > The QED Group, LLC > 1250 Eye St. NW, Suite 802 > Washington, DC 20005 > [EMAIL PROTECTED] > 202.898.1910.x15 (v) > 202.898.0887 (f) > 202.421.8161 (m) > > > -----Original Message----- > From: L Z [mailto:[EMAIL PROTECTED] > Sent: Wednesday, November 05, 2003 12:10 PM > To: [EMAIL PROTECTED] > Subject: [R] for help about R > > just want to ask the following > > > question: > > > > probit<-glm(y1~x1+x2-1, > > > family=binomial(link=probit)) > > > Warning message: > > > fitted probabilities numerically 0 or 1 occurred > in: > > > glm.fit(x = X, y = Y, > > > weights = weights, start = start, etastart = > > > etastart, > > > why does that happen? > > > > > --------------------------------- > > > [[alternative HTML version deleted]] > > ______________________________________________ > [EMAIL PROTECTED] mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help > > -- Brian D. Ripley, [EMAIL PROTECTED] Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595 ______________________________________________ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help
