The Need for Bayesian Hierarchical Models in Cognition: Assessing Subliminal Priming

• Activity Number: 230
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jeff Rouder* and Richard Morey
• Companies: University of Missouri and Cardiff University
• Keywords: Bayesian modeling ; Subliminal Perception ; Item-response model ; Bayes factor ; Hierarchical models ; Mixture models

Abstract

Subliminal priming refers to the phenomena where stimuli that are too faint to be seen nonetheless affect subsequent behavior. Establishing subliminal priming is difficult because one must establish that the stimuli cannot be seen, which is a null result. Traditionally, researchers have used low-power designs to establish this null result, an approach that is unpersuasive. Instead, we use a hierarchical Bayesian nonlinear model where the stimulus gives rise to a a latent sensation. If the sensation is above threshold, then detection occurs at an above-chance probability dependent on the strength of the sensation. If the latent sensation is below threshold, then detection occurs at chance rates. A fully Bayesian analysis is provided, and the posterior probability that the latent sensation is above threshold may be computed for each participant in each of several experimental conditions. Based on this approach, we may test whether indeed subliminal priming exists, and positive results from a recent semantic task are reported.