Vanessa Torman ( Federal University of Rio Grande do Sul): Bayesian Analysis of the Log-Binomial Model: a Comparison with the Frequentist Approach for the Estimation of Relative Risk

The log-binomial model is considered one of the best options for estimating relative risk (RR). With the frequentist approach, however, the usual programs often fail to converge. Chu e Cole (Epidemiology. 2010; 21:855-862) proposed a Bayesian estimator through MCMC, and explored by simulation properties of the model coefficients estimators but not of RR. In addition, they only considered continuous predictor. In this paper, we added the evaluation of RR, three point estimators (mean, median and mode) and two interval estimators (equal tail and HPD). We simulate two models: the first with one categorical predictor and the second with one categorical predictor and one continuous. We conclude that the mode point estimator and the equal-tail interval estimator are preferred.

Keywords: log-binomial model; relative risk; MCMC; simulation

Joint work with Suzi Alves Camey