Abstract
Despite almost a century of evidence implicating the influence of decision noise in perception, the great majority of research in psychophysics literature follows classical Signal Detection Theory in assigning the locus of internal noise exclusively to representational processes (Fernberger, 1920; Tanner and Swets, 1954). Recently, Rosner and Kochanski (2009) demonstrated that a rating task could disambiguate representational and decision noise components when experiments involved at least three stimulus intensities and four response categories. Moreover, Klauer and Kellen (2012) showed that the invocation of decision noise in rating tasks led to ambiguities in identifying subjects' underlying decision rules. Previously, we presented a novel framework that provides full recovery of both representation and decision noise components from classic signal detection rating experiments by using multiple passes of signal embedded in external noise (Cabrera, Lu, & Dosher, 2011). Here, we extend these results to jointly determine subjects' underlying decision rules and decision noise. In a simulation study, we show that a model of the decision rules correctly matched to those adopted by a simulated observer give improved precision, accuracy, and χ[sup]2[/sup] fits with increasing trials and passes, and that they outperform mismatched decision rules in every metric. In a subsequent experiment, subjects completed sessions with a restricted number of response categories as well as sessions with an extended number of response categories. For the restricted response structure, none of the decision rules gave significantly better fits compared to a zero decision noise model. For the extended response structure, one decision rule outperformed the others and gave significantly better fits than the null model. Finally, we also show that subjects' internal representations of the stimuli do not significantly differ for different response structures. As the number of response categories increase, subjects' decision noise becomes more prevalent while representational noise remains unchanged.
Meeting abstract presented at VSS 2013