Hi,
I am trying to fit a negative binomial model with a number of parasite tapeworms
as response variable to geographical coordinates
(actually preparing a trend surface before kriging). When I try an anova, I get
warnings:
> glm4.nb<-glm.nb(wb~X4+Y4+I(X4^2)+I(Y4^2))
> anova(glm4.nb)
Analysis of Deviance Table
Model: Negative Binomial(0.0463), link: log
Response: wb
Terms added sequentially (first to last)
Df Deviance Resid. Df Resid. Dev P(>|Chi|)
NULL 344 225.7
X4 1 0.0 343 9578.7 1.0
Y4 1 1695.7 342 7883.1 0.0
I(X4^2) 1 1992.2 341 5890.9 0.0
I(Y4^2) 1 5687.4 340 203.5 0.0
Warning messages:
1: tests made without re-estimating theta in: anova.negbin(glm4.nb)
2: Algorithm did not converge in: method(x = x[, varseq <= i, drop = FALSE],
y = object$y, weights = object$prior.weights,
3: Algorithm did not converge in: method(x = x[, varseq <= i, drop = FALSE],
y = object$y, weights = object$prior.weights,
4: Algorithm did not converge in: method(x = x[, varseq <= i, drop = FALSE],
y = object$y, weights = object$prior.weights,>
Results look like non sense with an intercept deviance smaller than the next
variables... One can see that X4 has a null deviance.
If X4 is removed from the model, Y4 get a null deviance in the model updated
(due to an intercept deviance smaller), and so on...
Actually smaller intercept deviance and null deviance for the first variable is
obtained for every first independant variable,
except when only one is left in the model.
Can somebody tell me what happens?
Thanks in advance,
Patrick Giraudoux
