Confidence judgments can, and do, integrate multiple sources of information. As many others in the field, we wonder: How are confidence judgments formed? What do they represent? Which sources of information do they integrate? How does this integration happen?

We often work on these projects together with Nathan Faivre and Michael Pereira.

How does prior information contribute to confidence?

Bayesian models tell us that decisions result from the integration of priors and likelihoods. But, do we take prior information into account when judging confidence? Do we weight them equally for confidence as we do for decisions?

As we describe in this tweeprint and the paper below, we found that priors are in general suboptimally weighted and that, perhaps surprisingly, this is more strongly the case in discrimination decisions than in the confidence ratings that follow them. It's as if after we make decisions, we reeevaluate the information that was available from priors, and adjust confidence accordingly.

Motor effects on confidence judgements

Confidence ratings often correlate (negatively) with reaction times. This, intuitively, makes sense: we're probably more sure when we respond quickly, and less sure when we have to think about an answer. We asked whether this is just a correlational effect or a causal one.
As far as our results show, it's just a correlational effect: when we used a "no-report" paradigm to remove, and bypass, first-order responses, there were no changes in either confidence or metacognitive sensitivity.

Confidence on multisensory stimuli

One common question in metacognition research is whether monitoring is domain-specific (each modality has its own inbult monitoring mechanism) or domain-general (a single mechanism can monitor any modality). Here we focussed only on the monitoring of perceptual domains (visual, auditory, tactile). We found that perceptual confidence seems rather supramodal: First, we found correlations between M-ratios across unisensory and multisensory conditions. Second, we found that the computational models that best the data considered explained confidence in multisensory perception as the result of the integration of two sensory signals, not as two independent confidence estimates that are then integrated. And finally, we found that a correlate of confidence in the EEG signal was independent of the sensory modality.
The degree to which metacognition is domain-general or domain-specific turned out to be more complicated (and interesting) than anticipated, so this is just a little glimpse onto a small corner of the entire landscape of metacognitive tasks. There's loads more to learn on this.