Charles Geyer wrote: ... > BTW the particular example given doesn't make clear WHAT question cannot > be answered correctly. Some questions can be without fuss, for example > >> y <- c(0,0,0,0,0,1,1,1,1,1) >> x <- seq(along = y) >> out1 <- glm(y ~ x, family = binomial) > Warning messages: > 1: In glm.fit(x = X, y = Y, weights = weights, start = start, etastart = > etastart, : > algorithm did not converge > 2: In glm.fit(x = X, y = Y, weights = weights, start = start, etastart = > etastart, : > fitted probabilities numerically 0 or 1 occurred >> out0 <- glm(y ~ 1, family = binomial) >> anova(out0, out1, test = "Chisq") > Analysis of Deviance Table > > Model 1: y ~ 1 > Model 2: y ~ x > Resid. Df Resid. Dev Df Deviance P(>|Chi|) > 1 9 13.8629 > 2 8 7.865e-10 1 13.8629 0.0002 > > This P-value (P = 0.0002) is valid, because the MLE does exist for the null > hypothesis. Hence we see that we have to use the model y ~ x for which > the MLE does not exist in the conventional sense.
It may be valid in some senses, but I can't help notice that it is off by a factor of at least 10, since the experiment has only 1024 outcomes, two of which are as extreme as the one observed, and where all outcomes are equally likely under the corresponding y~1 model. -p -- O__ ---- Peter Dalgaard Øster Farimagsgade 5, Entr.B c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - (p.dalga...@biostat.ku.dk) FAX: (+45) 35327907 ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
