Improving inference for social research and social policy: The Bayesian paradigm

Abstract

This paper reviews briefly some data interpretation difficulties associated with many social research projects, difficulties which damage efforts to synthesize the results from different studies or to use such results in the formulation of social policy. It is argued that these difficulties are traceable in large part to three causes: (1) the circuitous and nonintuitive logic of inference used customarily in social statistical analysis; (2) the lack of agreement (usually latent and implicit) about the substantive and technical premises adopted in the research argument; and (3) the frequent weakness (low diagnosticity) of social research data for distinguishing among alternative models of a phenomenon. The Bayesian paradigm of statistical inference, especially a broadly conceived version of that paradigm, offers a number of advantages for overcoming these difficulties. A principal advantage is that the Bayesian inference approach lends itself especially well to the systematic cumulation of evidence from a series of related studies. Another advantage of the Bayesian perspective is that it deals well with contingency and with uncertain knowledge. In view of these and other advantages, the Bayesian inference paradigm is proposed as a highly promising vehicle through which social researchers and policy makers may work and communicate more effectively, not only with each other, but also among themselves.