Hello, Can someone suggest a strategy of selecting a parsimonious coxph model when there is a cluster() term, a frailty() term and several potentially important covariates? I have read in the R-help archives that AIC can not be used when cluster() is. My full model looks something like: fit <- coxph(Surv(start, stop, fate) ~ x1 + x2 + x3 + x4 + cluster(site) + frailty(year), data) Thank you. Richard -- Richard Chandler Department of Natural Resources Conservation UMass Amherst (413)545-1237