[R] test of parallelism for ordinal regression?

[Duncan Mackay]

Hello,
in order to test the parallelism assumption of ordinal regression, would it
be appropriate to compare the likelihoods of a model fit by polr {MASS
package} with a multinomial model fit by multinom {nnet package}?

e.g.
# dataframe "ologit" previously read from
url("http://www.ats.ucla.edu/stat/r/dae/ologit.csv")
require(MASS)
ologit.polr <- polr(ordered(apply) ~ factor(pared) + factor(public) + gpa,
data=ologit, Hess=TRUE)
summary(ologit.polr)

require(nnet)
ologit.mn <- multinom(factor(apply)~factor(pared) + factor(public) + gpa,
data=ologit)
summary(ologit.mn)

# and then obtain likelihoods for comparison with a chi-square test
logLik(ologit.polr)
logLik(ologit.mn)

Does this seem like a reasonable thing to do?

Thanks,
Duncan
____________________________________________________________________________
__
Dr. Duncan Mackay
School of Biological Sciences
Flinders University
GPO Box 2100
Adelaide
S.A.  5001
AUSTRALIA

Phone    61-8-82012627
FAX          61-8-82013015

http://www.scieng.flinders.edu.au/biology/people/academic/mackay_d/index.htm
l