The likelihood ratio test is more reliable when one model is nested in the other. This true for your case. AIC/SBC are usually used when two models are in a hiearchical structure. Please also note that any decision made made based on AIC/SBC scores are very subjective since no sampling distribution can be used to make a "rigorous" decision. Regarding the magnitutes between the loglikelihood and AIC/SBC, I would say the author must used a modified version in coxme() since several different modified AIC/SBC scores are running in practice.
My suggestion would be to use LR test for your case: For the integrated likelihhod: LL.small.model = - 467.3549 (including lifedxm) LL.large.model = - 471.3333 (excluding lifedxm) DF.diff = 3 - 1 = 2 LR: -2*(- 471.3333 + 467.3549) = 7.9568 p-value: 1-pchisq(7.9568,2) = 0.01871556 For the penalized likelihhod: LPL.small.model = -435.2096 (including lifedxm) LPL.large.model = -436.0478 (excluding lifedxm) DF.diff = 3 - 1 = 2 PLR: -2*(- 436.0478 + 435.2096 ) = 1.6764 p-value: 1-pchisq(1.6764,2) = 0.4324883 Two different likehood methods produce different results, which one you should use depends on which likelihood makes more sense to you (or which likehood is better). HTH -- View this message in context: http://r.789695.n4.nabble.com/Question-regarding-significance-of-a-covariate-in-a-coxme-survival-model-tp2313880p2399114.html Sent from the R help mailing list archive at Nabble.com. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
