Abstract
In simple detection tasks observers presumably use a subjective decision criterion based on their noisy internal response, setting a boundary between “yes” and “no” responses. The variability of this decision criterion cannot be directly assessed in a fixed stimulus paradigm. Here, we used external noise to model internal responses, so that criteria can be unambiguously identified in the stimulus space. This also allowed the direct assessment of the interactions between criteria assigned to different targets in conditions where two detection tasks were mixed in a single testing session (Gorea & Sagi, 2000).
Observers were presented with two luminance flashes, one on each side of fixation, with their amplitudes randomly drawn from either “Signal” or “Noise” Gaussian distributions. Flashes were presented within a red and a green circle whose color marked the two Signal/Noise distribution pairs that differed in their variances and/or means. Observers reported whether the flash enclosed in the circle with delayed offset belonged to the corresponding Signal or Noise distribution (partial report). Signal mean and variance differed across the two tasks by a factor of 1 to 3.5. Observers were informed about the differences between Signal means.
For naïve observers the two criteria measured within the dual task were found to be very close, deviating significantly from the close-to-optimal ones observed with a single target. Observers familiar with Signal Detection Theory and stimulus parameterization did not show such interactions. As the present unidimensional external-noise task allows for the use of a stimulus-intensity level as a decision criterion, expert observers may set their criteria in stimulus space while naïves do so internally, as is the case for all observers when only internal noise is present and suboptimal performance is observed. Our results reject a previous suggestion that in multiple detection tasks observers simply match their false alarms rates across tasks.