Examining the Signal-Detection Account of Visual Working Memory

The study of visual working memory has long centered on debates between Signal Detection Theory (SDT) and discrete-slots models. A notable limitation of these debates is the strong reliance on parametric assumptions and selective-influence manipulations that rarely receive direct test. Here we take a different approach by examining whether visual working-memory judgments satisfy the structural constraints implied by a random-scale representation—a general latent-variable framework from which both SDT and discrete-slots models can be derived. In Experiments 1a and 1b, multiple-alternative forced-choice judgments conformed to these constraints, allowing the reconstruction of single-item ROC functions without response-bias manipulations. The reconstructed ROC functions were curved and asymmetric, contradicting the linear predictions of discrete-slots accounts. Experiment 2 provided a complementary failure case, showing that random-scale constraints break down precisely when their theoretical conditions are not met. Experiment 3 extended the findings to item-location bindings in change detection, again yielding an asymmetric ROC function that neither equal-variance SDT nor discrete-slots models can accommodate. Together, the results show that visual working memory judgments respect the general assumptions underlying SDT while undermining the core commitments of discrete-slots theories.