Bayesian analysis of recognition memory: The case of the list-length effect

Abstract

Recognition memory experiments are an important source of empirical constraints for theories of memory. Unfortunately, standard methods for analyzing recognition memory data have problems that are often severe enough to prevent clear answers being obtained. A key example is whether longer lists lead to poorer recognition performance. The presence or absence of such a list-length effect is a critical test of competing item- and context-noise based theories of interference and bares on whether recognition involves “recall-like” components as dual process theories would contend. However, the issue has remained unresolved, in part, because of the weaknesses of the standard analysis. In this paper, we develop a Bayesian method of analysis and apply it to new data on the list-length effect. The analysis allows us to find positive evidence in favor of a null list-length effect as predicted by context noise models. The data also illustrate the importance of the contextual reinstatement process on recognition performance and show how previous work demonstrating a list-length effect may have been contaminated by reinstatement confounds. By contrasting our new method against the standard approach we highlight the advantages of the Bayesian framework when inferring the values of psychologically meaningful variables, and in choosing between models representing different theoretical assumptions about memory.