Fig. 7
From: The interplay of uncertainty, relevance and learning influences auditory categorization
For an individual participant their category choice was driven by both their sensory uncertainty and their subjective estimates of the probability of distractors, but their accuracy was mostly driven by their sensory uncertainty. (A) Cartoon models of different psychometric curves illustrate the qualitative range of sigmoidicity. Left: high sigmoidicity, right: low sigmoidicity. (B) Each data point corresponds to a single participant; color denotes \({\text{p}}_{\text{distractor}}\) which is the participant’s estimated probability of distractors (see color bar); Black: \({\text{p}}_{\text{distractor}}\) \(\sim\) 0; Green: \({\text{p}}_{\text{distractor}}\) \(\sim\) 1. \({\text{p}}_{\text{distractor}}\) is a fitting parameter in the full Bayesian model and has value 0 in the no-distractor model. The model parameter \({\upsigma }_{\text{sensory}}\) captures participant’s sensory uncertainty. Statistics: effect of \({\text{p}}_{\text{distractor}}\) and \({\upsigma }_{\text{sensory}}\) on sigmoidicity using a two-factor regression model. (C) Participants’ accuracy did not significantly depend on \({\text{p}}_{\text{distractor}}\) but was inversely correlated with their sensory uncertainty. \({\text{p}}_{\text{distractor}}\) is color coded as in (B). Statistics: effect of \({\text{p}}_{\text{distractor}}\) and \({\upsigma }_{\text{sensory}}\) on accuracy using a two-factor regression model. In all panels, data point: one participant; †p < 0.05; *p < 0.01, **p < 0.001, ***p < 0.0001. Error bars in (B) and horizontal error bars in (C): 95% confidence intervals derived from respective model fits to 100 bootstrapped samples. Vertical error bars in (C): standard error of the mean, obtained by assuming that accuracy follows a binomial distribution; vertical error bars aren’t large enough to be distinguishable.
