Rens van de Schoot
Bayesian Statistics in the Applied Literature: Prevalence, Implementation, and the Impact of Priors
The goal of this presentation is to highlight different facets of the use of Bayesian methods in applied research. Bayesian methods have shown a steady increase in use in the social sciences, but this estimation framework requires researchers to make (sometimes) difficult decisions throughout the model estimation process, like the choice of priors. In this presentation, several main issues surrounding Bayesian statistics will be highlighted: (1) prevalence in the applied literature, (2) implementation and avoiding the misuse of Bayesian methods, and (3) the specification and impact of priors, which incorporate existing knowledge into an analysis.
The first part of the presentation incorporates a systematic review of the prevalence of Bayesian statistics in the applied psychological literature. All papers published in the last 25 years in the field of Psychology (source: Scopus) were obtained and were screened on why the authors used Bayesian statistics, which software was used, which software specifications were applied (e.g., type of priors, convergence checks, sensitivity analysis), and how the results were different from default statistical estimators. Comments about prevalence, implementation, and problems will be made.
The second part of the presentation focuses on how to implement Bayesian methods and, more importantly, how to avoid misusing these methods. Bayesian statistics allow researchers to include existing knowledge into their analysis through priors. This possibility is not only interesting because it promotes cumulative science, but also because it can improve statistical power. However, there are many cases where the use of priors can be dangerous and lead toward questionable research practices. This presentation will describe several points that should be thoroughly checked when specifying priors in a Bayesian analysis.
The part surrounds the issue of using priors, which is gaining popularity in the social sciences. A Bayesian approach has distinct advantages over traditional approaches. However, the accuracy of parameter estimates may heavily depend on the appropriate specification of prior distributions. This presentation focuses on implementing a Bayesian approach with some applications using empirical data. Emphasis will be placed on the impact of different types of prior distributions and the conditions under which one should use informative versus noninformative priors.