During the last two decades, Bayesian methods have become a popular alternative to their frequentist counterparts. They do not rely on asymptotic theory and enable researchers to estimate complex models that would not be feasible under the frequentist paradigm. Moreover, by using informative prior distributions, Bayesian methods allow researchers to incorporate available background information. This inherent cumulative characteristic of Bayesian methods is by far their greatest advantage but, at the same time, their most criticized aspect.
A large body of research documents the benefits of Bayesian methods for educational and psychological research, the development and refinement of Markov chain Monte Carlo sampling algorithms, and the strengths and weaknesses of a variety of prior distributions for parameters in numerous statistical models. Most of this research, unfortunately, focuses on the technical benefits and the use of non-informative prior distributions. Popular software packages such as Mplus, SPSS, and Blavaan also offer default Bayesian estimation with non-informative priors. Informative prior distributions, representing the conceptual benefit of Bayesian methods, are used only tentatively. To achieve the full potential of Bayesian methods, however, it is necessary to move beyond using Bayes as ‘just another estimator’ and beyond non-informative prior distributions. Building confidence in the use of informative prior distributions requires providing answers to the following questions: How does one specify proper informative prior distributions? When is background knowledge eligible to be included? How does one account for between-study heterogeneity in background knowledge? How does one elicit and quantify background knowledge?
Thus, this Research Topic aims at providing an overview of our current understanding of the benefits, specification, and use of informative prior distributions in psychological research. Consequently, we welcome the following kinds of contributions focusing on a variety of statistical approaches/models:
(1) Reviews and discussions of conceptual and theoretical considerations when using Bayesian estimation methods with informative prior distributions, for instance, the benefits and potential caveats of Bayesian updating for cumulative science, small sample research, and replication.
(2) Investigations, applications and illustrations of novel approaches to elicit and quantify background knowledge for the specification of informative prior distributions for substantial model parameters (e.g., regression weights).
(3) Investigations, applications and illustrations of the use of weakly informative prior distributions for technical model parameters (e.g., variance components), especially their use in conjunction with informative prior distributions.
Except for the conceptual/theoretical category, all contributions are required to include detailed annotated code as supplementary material.
During the last two decades, Bayesian methods have become a popular alternative to their frequentist counterparts. They do not rely on asymptotic theory and enable researchers to estimate complex models that would not be feasible under the frequentist paradigm. Moreover, by using informative prior distributions, Bayesian methods allow researchers to incorporate available background information. This inherent cumulative characteristic of Bayesian methods is by far their greatest advantage but, at the same time, their most criticized aspect.
A large body of research documents the benefits of Bayesian methods for educational and psychological research, the development and refinement of Markov chain Monte Carlo sampling algorithms, and the strengths and weaknesses of a variety of prior distributions for parameters in numerous statistical models. Most of this research, unfortunately, focuses on the technical benefits and the use of non-informative prior distributions. Popular software packages such as Mplus, SPSS, and Blavaan also offer default Bayesian estimation with non-informative priors. Informative prior distributions, representing the conceptual benefit of Bayesian methods, are used only tentatively. To achieve the full potential of Bayesian methods, however, it is necessary to move beyond using Bayes as ‘just another estimator’ and beyond non-informative prior distributions. Building confidence in the use of informative prior distributions requires providing answers to the following questions: How does one specify proper informative prior distributions? When is background knowledge eligible to be included? How does one account for between-study heterogeneity in background knowledge? How does one elicit and quantify background knowledge?
Thus, this Research Topic aims at providing an overview of our current understanding of the benefits, specification, and use of informative prior distributions in psychological research. Consequently, we welcome the following kinds of contributions focusing on a variety of statistical approaches/models:
(1) Reviews and discussions of conceptual and theoretical considerations when using Bayesian estimation methods with informative prior distributions, for instance, the benefits and potential caveats of Bayesian updating for cumulative science, small sample research, and replication.
(2) Investigations, applications and illustrations of novel approaches to elicit and quantify background knowledge for the specification of informative prior distributions for substantial model parameters (e.g., regression weights).
(3) Investigations, applications and illustrations of the use of weakly informative prior distributions for technical model parameters (e.g., variance components), especially their use in conjunction with informative prior distributions.
Except for the conceptual/theoretical category, all contributions are required to include detailed annotated code as supplementary material.