How can choice, confidence, and response times be modeled simultaneously? Here, we propose the new dynamical weighted evidence and visibility (dynWEV) model, an extension of the drift-diffusion model of decision-making, to account for choices, reaction times, and confidence simultaneously. The decision process in a binary perceptual task is described as a Wiener process accumulating sensory evidence about the choice options bounded by two constant thresholds. To account for confidence judgments, we assume a period of postdecisional accumulation of sensory evidence and parallel accumulation of information about the reliability of the present stimulus. We examined model fits in two experiments, a motion discrimination task with random dot kinematograms and a postmasked orientation discrimination task. A comparison between the dynWEV model, two-stage dynamical signal detection theory, and several versions of race models of decision-making showed that only dynWEV produced acceptable fits of choices, confidence, and reaction time. This finding suggests that confidence judgments depend not only on choice evidence but also on a parallel estimate of stimulus discriminability and postdecisional accumulation of evidence.

Meta-d’/d’ has become the quasi-gold standard to quantify metacognitive efficiency because it has been assumed that meta-d’/d’ provides control for discrimination performance, discrimination criteria, and confidence criteria even without the assumption of a specific generative model underlying confidence judgments. Here, we investigate the robustness of meta-d’/d’ using simulated data from a variety of different generative models of confidence. For most models of confidence, meta-d’/d’ is affected by discrimination performance, discrimination task criteria, and confidence criteria. The single exception was a generative model of confidence according to which the evidence underlying confidence judgements is sampled independently from the evidence utilized in the choice process from a Gaussian distribution truncated at the discrimination criterion. According to all alternative generative models of confidence examined in the present study, there exist at least some combination of parameters where meta-d’/d’ is affected by discrimination performance, discrimination criteria and confidence criteria. These simulations imply that a difference in meta-d’/d’ between conditions does not necessarily reflect a difference in metacognitive efficiency but might as well be caused by a difference in discrimination performance, discrimination task criterion, or confidence criteria.