Background: The Gaussian Equal Variance Model (GEVM) from Signal Detection Theory (SDT) is a widely used approach to modeling recognition tasks (Hautus et al., 2019). However, it fails to account for the asymmetric Receiver Operating Characteristics (ROCs) frequently observed in empirical data. As an alternative, the Poisson Model of Signal Detection (PMSD) has been proposed (Egan, 1975; Kaernbach, 1991). This study aimed to evaluate the PMSD's fit and its potential advantages over the GEVM in the Visual Sensory Memory Task (VSMT; Kaernbach et al., 2019).

Methods: Stimulus presentation time and size were systematically varied in the VSMT. Participants provided old-new judgments and confidence ratings, which were used to estimate model parameters via Maximum Likelihood Estimation. Log-Likelihood Ratio Tests were conducted to compare the PMSD, GEVM, and a full parameter model.

Results: The PMSD demonstrated statistically significant superiority over the GEVM in only 2 out of 56 comparisons. However, in 68% of the analyses, the PMSD represented a valid parameter reduction compared to the full parameter model, highlighting its parsimony without significant loss of fit.

Discussion: While systematic evidence for the PMSD's superiority over the GEVM remains limited, the results indicate that the PMSD could offer an interesting and interpretable alternative for modeling recognition tasks. Future research should further explore its applicability across different experimental paradigms and task variations.