Timothy Brady¹ (), Yong Hoon Chung², Jamal Williams³, Viola Stoermer²; ¹University of California, San Diego, ²Dartmouth College, ³Yale University

The dual-process signal detection (DPSD) model has been influential in long-term memory research. This model makes the strong claim that when a large number of high-confidence responses occurs in a memory task, this provides direct evidence for an all-or-none recollection process that is distinct from familiarity signals. This model has recently been extended to visual working memory; for example, suggesting that stimuli like real-world objects and colors engage different amounts of 'recollection'. Here, we demonstrate that the DPSD model incorrectly considers trial-by-trial variability in memory to be evidence of recollection. We collected confidence ratings in two color change detection conditions matched for overall performance but differing in variability. In Experiment 1, we compared a fixed condition (set size 3) to a variable condition that mixed set sizes (1-6). Despite matched overall performance, the DPSD model attributed the excess high-confidence responses in the variable condition to 'recollection' (Ro=0.31 vs. Ro<0.01 in fixed, p<0.01). Experiment 2 used a fixed set size (four) but different test probes to induce different amounts of variability. In the fixed condition, we used consistent 40° color wheel separations between targets and foils, while the variable condition used varying foil distances. Again, the DPSD model incorrectly interpreted the excess high-confidence responses in the variable condition as greater 'recollection' (p<0.001). These results challenge the DPSD model's claim to measure 'recollection' through confidence ratings alone, without independent measures like context memory. While multiple processes may contribute to memory match signals, it is likely that people collapse these onto a single evidence axis when making change detection decisions, making them inseparable through confidence reports and ROCs derived from those reports. By demonstrating how task variability shapes confidence patterns, this work provides a more parsimonious framework for interpreting memory decisions across different stimulus types and experimental conditions.

Acknowledgements: NSF BCS 2146988