A Bayesian Perspective on Methodologies for Drawing Causal Inferences in Experimental and Non-Experimental Settings

In recent years, attention in the education community has focused on the need for evidenced-based research, particularly educational policies and interventions that rest on "scientifically based research". The emphasis on scientifically based research in education has led to a corresponding increase in studies designed to provide strong warrants for causal claims. These studies have been conducted relying almost entirely on analytic procedures that are rooted in the frequentist (classical) paradigm of statistics. Over the last decade, however, significant advances have been made in the area of Bayesian statistical inference, owing mostly to computational developments and readily available software (see e.g. Gilks, Richardson, & Spiegelhalter, 1996). With these advances have come important applications of Bayesian methods to problems in the social and behavioral sciences. However, few applications of Bayesian methods to educational problems can be found, with the exception of Bayesian statistical methods for item response theory models in educational measurement (e.g. Fox & Glas, 2001). This talk proposes that Bayesian methods offer an alternative analytic paradigm that can support and enhance scientifically based research in the educational sciences.