[R] survreg 3-way interaction

tdenes at cogpsyphy.hu tdenes at cogpsyphy.hu
Fri Jan 28 12:01:28 CET 2011

Dear All,

I was wondering why survreg (in survival package) can not handle three-way
interactions. I have an AFT (accelerated failure time) model with a
frailty term. If the model is restricted to two-way interactions, survreg
and gamlss.cens (with random() term) give very similar results, except for
the intercept (I do not know why it differs, anyway, if other coefficients
are pretty the same). The aftreg function in the eha package could handle
three-way interactions, but no frailty. Streg in Stata works fine with
three-way interactions and frailty. However, 1) in the two-way interaction
setting, Stata's results differ from those of gamlss and survreg, 2) I
prefer R over Stata.
So it seems that I can use gamlss.cens to analyze my data, but it would be
more comfortable to use a dedicated library. Is there a theoretical reason
that survreg does not provide the possibility of higher-order
interactions, or is merely a programmatic issue?


More information about the R-help mailing list