Association for Behavior Analysis International

As a branch of applied mathematics, information theory provides a set of tools and axioms for describing the patterning and structure of streams of events. Such streams are an indispensable form of data for learning theory. While measurements of disorder, or “entropy,” are now widely familiar to academics and laypeople, these only represent the tip of information theory’s iceberg. In concert with probability theory, its tools make possible the measure of how surprising an event is (contingent on some probabilistic expectation), how interrelated streams of events are, and how costly it is to translate between different sets of expectations. In this tutorial, I will give an overview of a practical toolkit of information-theoretic measures that can be used to make normative predictions about experimental designs. I will also introduce a strategy for how to measure the degree of correspondence between a model and data in a way that allows entirely different classes of statistical model to be compared with one another. Throughout, my emphasis will be on making these calculations practical for working experimentalists, so that these tools can be put to work in service of advancing learning theory.

Graduate students, experimental psychologists, and quantitative behavior analysts