For estimating conditional survival functions, non-parametric estimators canbe preferred to parametric and semi-parametric estimators due to relaxedassumptions that enable robust estimation. Yet, even when misspecified,parametric and semi-parametric estimators can possess better operatingcharacteristics in small sample sizes due to smaller variance thannon-parametric estimators. Fundamentally, this is a bias-variance tradeoffsituation in that the sample size is not large enough to take advantage of thelow bias of non-parametric estimation. Stacked survival models estimate anoptimally weighted combination of models that can span parametric,semi-parametric, and non-parametric models by minimizing prediction error. Anextensive simulation study demonstrates that stacked survival modelsconsistently perform well across a wide range of scenarios by adaptivelybalancing the strengths and weaknesses of individual candidate survival models.In addition, stacked survival models perform as good as, or better than, themodel selected through cross-validation. Lastly, stacked survival models areapplied to a well-known German breast cancer study.
展开▼