Yunxuan Zheng¹ (), Nadia Haddara¹, Dobromir Rahnev¹; ¹Georgia Tech

Metacognition, the ability to evaluate one’s own decision-making process, often manifests as confidence judgments. While numerous computational models have been developed to explain confidence in perceptual decision making, most focus on 2-choice tasks. However, real-world decisions frequently involve multiple alternatives, and how confidence is computed in such contexts remains less understood. To address this gap, we examined confidence judgments across varying numbers of choice options. Two hundred participants performed a perceptual discrimination task on noisy images of hand-written digits from the MNIST dataset. Participants identified the digit from a given set of options and rated their confidence on a 4-point scale. The task included three conditions: a 2-choice condition (discriminating between digits 5 and 6), a 4-choice condition (discriminating among digits 5–8), and an 8-choice condition (discriminating among digits 1–8). Accuracy was approximately matched across conditions by using a 2-up-1-down staircase procedure that continuously adjusted the noise level of the images for each of the three conditions independently (mean ± SD: 0.66 ± 0.02, 0.63 ± 0.02, 0.64 ± 0.02 for 2-, 4-, and 8-choice, respectively). Despite matched accuracy, we found that reaction time (RT) increased with number of choices (mean ± SD: 0.76 ± 0.25, 0.94 ± 0.24, 1.03 ± 0.23; all pairwise p’s < 0.001), an effect consistent with the notion that motor preparation may take longer in the presence of more choice alternatives. Critically, we also found that confidence ratings also increased as choice set size increased (mean ± SD: 1.88 ± 0.51, 2.21 ± 0.54, 2.43 ± 0.52; all pairwise p’s < 0.001). These results demonstrate that confidence can be strongly influenced by the number of available choices and support the notion that understanding confidence requires that we move beyond simple 2-choice tasks.