Message-ID: <OFDC6514CA.095BAC24-ON87256D2B.00760E02@apca.gc.ca>
Date: 2003-05-19T21:52:14Z
From: Jesse.Whittington@pc.gc.ca
Subject: multcomp and glm
I have run the following logistic regression model:
options(contrasts=c("contr.treatment", "contr.poly"))
m <- glm(wolf.cross ~ null.cross + feature, family = "binomial")
where:
wolf.cross = likelihood of wolves crossing a linear feature
null.cross = proportion of random paths that crossed a linear feature
feature = CATEGORY of linear feature with 5 levels: high-use road, low-use
road, high-use trail, low-use trail, and railway line
I would like to determine whether wolves are more likely to cross some
features than others and am using csimtest in the package multcomp to do
so.
x <- coef(model)
var.cov <- vcov(model)
df <- model$df.residual
contrast.matrix <- contrMat(rep(2,length(x), type = "Tukey")
post.hoc <- csimtest(estpar = x, df=df, covm = var.cov, cmatrix =
contrast.matrix)
summary(post.hoc)
My questions are:
(1) Have I entered my parameters for the posthoc analysis correctly and
does it matter which categorical variable I use as my reference category?
(2) How do I omit the continuous variable "null.cross" from the list of
multiple comparisons?
I would be most grateful for any assistance you can give me,
Jesse Whittington
Jasper National Park
Box 10,
Jasper, Alberta
Canada
(780) 852-6187