Bayesian Survival Analysis in STAN for Improved Measuring of Uncertainty in Parameter Estimates
Riko Kelter
Abstract
Survival analysis is an important analytic method in the social and medical sciences. Also known under the name time-to-event analysis, this method provides parameter estimation and model fitting commonly conducted via maximum-likelihood. Bayesian survival analysis offers multiple advantages over the frequentist approach for measurement practitioners, however, computational difficulties have mitigated interest in Bayesian survival models. This paper shows that Bayesian survival models can be fitted in a straightforward manner via the probabilistic programming language Stan, which offers full Bayesian inference through Hamiltonian Monte Carlo algorithms. Illustrations show the benefits for measurement practitioners in the social and medical sciences.