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 <ssullivan at qedgroupllc.com> 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
> ssullivan at qedgroupllc.com
> 202.898.1910.x15 (v)
> 202.898.0887 (f)
> 202.421.8161 (m)
>
>
> -----Original Message-----
> From: L Z [mailto:cougar3721 at yahoo.com]
> Sent: Wednesday, November 05, 2003 12:10 PM
> To: r-help at stat.math.ethz.ch
> 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]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>
--
Brian D. Ripley, ripley at stats.ox.ac.uk
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