Bayesian hypothesis testing for single-subject designs.

Abstract

Researchers using single-subject designs are typically interested in score differences between intervention phases, such as differences in means or trends. If intervention effects are suspected in data, it is desirable to determine how much evidence the data show for an intervention effect. In Bayesian statistics, Bayes factors quantify the evidence in the data for competing hypotheses. We introduce new Bayes factor tests for single-subject data with 2 phases, taking serial dependency into account: a time-series extension of Rouder, Speckman, Sun, Morey, and Iverson's (2009) Jeffreys-Zellner-Siow Bayes factor for mean differences, and a time-series Bayes factor for testing differences in intercepts and slopes. The models we describe are closely related to interrupted time-series models (McDowall, McCleary, Meidinger, & Hay, 1980).