[R] Cumulative Link Models
Katrin Ronnenberg
ronnenberg at gmx.com
Sun Jul 1 20:42:05 CEST 2012
Dear R community,
I'm trying to analyze a model with an ordinal response variable.
I wonder if clm()s (Cumulative Link Models) are appropriate in my case.
The study compares parasite infestation of porpoises in 1995 and 2009. The degree of infestation is a rank (mild to severe, as ordered factor). In some parasite species clm gives meaningful results. But in one case I started to wonder. In the first study (1995) zero animals were infested by this parasite. In 2009 most animals were, some of which severely. I have two more factors, thus I was happy finding the package - ordinal.
Using a simple Mann-Whitney-U and Chi-square test I find a strong sigificant year term. Also lm, assuming normal distribution, finds significant year (and age) term. But in clm I find no significance. Does clm have a problem with one factor level being zero? I have compared the log-likelihood of the link functions and the cauchit gave the highest results. Here is the r code:
model<-clm(factor(Parasite,ordered = is.ordered(0:3))~
Study*Age*Sex,link=c("cauchit"))
summary(model)
Thanks in advance,
Katrin
Institute of terrestrial and aquatic wildlife
Hannover-Büsum
Germany
More information about the R-help
mailing list