[R] Pairwise comparisons/contrasts from a coxph model?

Kaspar Pflugshaupt pflugshaupt at geobot.umnw.ethz.ch
Tue Jun 20 16:07:47 CEST 2000


this is probably more a statistical question than an R-specific problem, but
I'll risk it.

I've fitted a Cox Proportional hazard model with one factor Treatment (seven
levels) as a predictor variable. The general Null hypothesis (all groups
show the same survival behaviour) is clearly rejected. Now, is there any
(statistically sensible) way of doing pairwise comparisons and/or contrasts
among levels?

In the R online help as well as in the literature I've looked up (I'll look
up more), nothing of the sort is mentioned. Does this mean it's not a
reasonable thing to do, or that it's trivial?  :-)

And, has somebody already done this in R or could give me a hint?

One way I considered was (for all pairs of levels) to exclude all but the
respective two levels from my data and run the model. In the end I would do
a Bonferroni correction on all the p values I receive. Seems to be on the
save side, maybe too much so. Anyway, there ought to be a more elegant way.

Thankful for any hint


Here's the output from my model

> poll.survmodel2
coxph(formula = Surv(Week, Cens) ~ Treatment, data = poll.survdata2)

              coef exp(coef) se(coef)      z       p
TreatmentT1  0.342     1.408    0.256  1.337 0.18000
TreatmentT2  0.102     1.108    0.246  0.416 0.68000
TreatmentT3  0.205     1.227    0.247  0.829 0.41000
TreatmentT4 -0.303     0.739    0.256 -1.184 0.24000
TreatmentT5 -1.066     0.344    0.306 -3.485 0.00049
TreatmentT6 -0.733     0.480    0.281 -2.611 0.00900

Likelihood ratio test=30  on 6 df, p=3.89e-05  n= 210


Kaspar Pflugshaupt
Geobotanisches Institut
Zuerichbergstr. 38
CH-8044 Zuerich

Tel. ++41 1 632 43 19
Fax  ++41 1 632 12 15

mailto:pflugshaupt at geobot.umnw.ethz.ch
privat:pflugshaupt at mails.ch

r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch

More information about the R-help mailing list