New approaches to subliminal perception: Exploring the unconscious perception of masked primes with Bayesian regression models and General Recognition Theory (GRT)

One of the central debates in experimental psychology concerns the extent to which information outside conscious awareness can influence behavior. This study aims to assess whether two novel approaches to unconscious perception, a regression-based Bayesian generative model and Sensitivity vs. Awareness curves derived from General Recognition Theory (GRT), produce convergent results when examining the visual processing of briefly presented masked stimuli. While both investigate “unconscious processing,” they differ not only in how they define “unconscious” (chance-level discrimination vs. absence of subjective awareness) and “processing” (priming vs. detection), but also in their underlying mathematical frameworks. To investigate this, we examined the grouping of local elements into global shapes using two prime-mask stimulus onset asynchronies (SOA): a shorter baseline SOA of 40 ms (Experiment 1) and a longer SOA of 67 ms (Experiment 2). Both methods provided evidence for the unconscious processing of the primes’ global shape in Experiment 2. However, while GRT analyses also supported unconscious processing at the shorter 40 ms SOA, the Bayesian generative models produced strong evidence against it. We show that discrepancies between the regression-based and GRT-based approaches may arise not only from their operational definitions, but also their underlying mathematical frameworks and validity. Preliminary SvA simulations indicate limited robustness of the GRT-based model, warranting caution prior to further validation. More broadly, while both approaches offer refined tools for studying unconscious perception, persistent methodological challenges remain, underscoring the need for clearer definitions of unconscious processing, converging measures of awareness, and more thorough analyses of method validity.