Bayesian statistics in educational research: a look at the current state of affairs

Abstract

The ability of a scientific discipline to build cumulative knowledge depends on its predominant method of data analysis. A steady accumulation of knowledge requires approaches which allow researchers to consider results from comparable prior research. Bayesian statistics is especially relevant for establishing a cumulative scientific discipline, because the incorporation of background (or prior) knowledge is fundamentally anchored in its basic principles. The aim of the current systematic review is to provide insights into the current state of methodological affairs in educational research, with a focus on Bayesian statistics and the use of prior information. An analysis of publication histories of the 224 educational journals currently listed in the Thomson Reuters Journal Citation Report 2015 indicates that Bayesian statistics is primarily used to solve methodological problems, rather than used to build cumulative knowledge based on a combination of study results with comparable prior research. The utilisation of Bayesian statistics is motivated by its flexibility: models are estimated which would not be estimable with frequentist approaches, thus expanding the methodological repertoire of educational researchers and producing knowledge which otherwise would not have been available. Lastly, the predominant use of noninformative prior distributions indicates that one of the biggest advantages of Bayesian statistics, namely the combination of study results with comparable prior research, remains underutilised in educational research. Practical implications of these findings for educational research are illustrated and discussed.

Bayesian Versus Frequentist Estimation for Structural Equation Models in Small Sample Contexts: A Systematic Review

“…Such statements can create the impression that using Bayesian estimation universally solves small sample problems. Although several textbooks on Bayesian estimation stress the important role of prior distributions when Bayesian estimation is used with small samples (e.g., Gelman et al, 2013, p. 88;Kaplan, 2014, p. 291;McElreath, 2016, p. 31), in practice prior distributions are often not carefully chosen, and most empirical researchers rely on default software settings (see e.g., König & van de Schoot, 2017;McNeish, 2016b;van de Schoot, Schalken, & Olff, 2017;van de Schoot, Winter, et al, 2017). Popular software programs, such as: Mplus (L. K. Muthén & Muthén, 2017); SPSS (IBM Corp., 2017); JASP (JASP team, 2018); or the R package blavaan (Merkle & Rosseel, 2018), offer Bayesian estimation with diffuse default prior distributions.…”

Using Bayesian statistical modelling as a bridge between quantitative and qualitative analyses: illustrated via analysis of an online teaching tool

“…multiple issues such as missing values, non-normality, excess zeros, nested relationships), • they wish to integrate expert knowledge or findings from previous studies in a transparent way, that is also easily accessible to a wide audience-for instance, Humphreys and Jacobs (2015) provide an explicit technical framework for integrating particular kinds of information using a Bayesian Integrated Quantitative & Qualitative framework; • they find it difficult to interpret and communicate the model findings (what do the statistical tests mean for an SEM? ); • when a sequential approach to learning is required, to enable feasible implementation over a longer period of time, which is flexible to changes in both quantitative and qualitative information sources and collaborative arrangements: as described in 'The Bayesian Superintendent' by Meyer (1964) and more generally in education by König and van de Schoot (2017);…”

“…The benefits of a Bayesian approach were recommended in education decades before computation was accessible to general researchers (e.g. Diamond, 1964;Meyer, 1964), yet Bayesian approaches are still emerging in this field (König and van de Schoot, 2017). These benefits are more well-known in other research fields-such as natural language processing (Cohen, 2016), demography (Bijak and Bryant, 2016), ecology (Ellison, 1996(Ellison, , 2004Hilborn and Mangel, 1997;Low Choy, O'Leary, and Mengersen, 2009), genetics (Shoemaker, Painter, and Weir, 1999), psychology and social science (Jackman, 2009;McElreath, 2016;Oldehinkel, 2016).…”

Systematic search of Bayesian statistics in the field of psychotraumatology

“…In recent years there has been increased interest in Bayesian analysis in many disciplines; for example, see the systematic reviews in the fields of educational science (König & van de Schoot, 2017), organizational science (Kruschke, 2010), psychometrics (Rupp, Dey, & Zumbo, 2004), health technology (Spiegelhalter, Myles, Jones, & Abrams, 2000), epidemiology (Rietbergen, 2017), medicine (Ashby, 2006), and psychology (Van de Schoot, Winter, Ryan, Zondervan-Zwijnenburg, & Depaoli, 2017). Also, the use of Bayesian analyses in the field of psychotraumatology was advocated during a meeting of the International Society for Traumatic Stress Studies (ISTSS) global meetings program 1 .…”