Abstract
In single-stimulus categorization and yes-no signal-detection tasks with asymmetric payoffs or priors, human observers often display conservatism: The likelihood ratio at criterion (β) is closer to one than the optimal criterion maximizing expected gain (assuming an underlying Gaussian model). Are observers similarly conservative in a typical 2-alternative, forced-choice (2AFC) discrimination task? Methods: Six subjects viewed brief (200 ms) displays containing two noise patches (diameter: 3 deg) to the left and right of fixation (eccentricity: 8 deg). A vertical Gabor patch (3 cycle/deg) was added to one of the noise patches. Subjects indicated which noise patch contained the Gabor. Eye movements were monitored; trials were excluded if subjects moved their gaze more than 1 deg away from the fixation mark. A pilot experiment with symmetric payoffs was used to estimate Gabor contrasts corresponding to d′ values of 0.5, 1 and 2. In the main experiment, incorrect responses resulted in a 100-point penalty. Correctly detecting a target on the left resulted in a 100-point gain. Rewards for correctly identifying targets on the right varied between blocks (40, 65, 100, 160 or 250 points). Subjects completed 15 blocks (5 payoff schedules × 3 contrast levels) of 200 trials. Results: Values of d′ and β were estimated by maximum likelihood for each subject. A linear regression of optimal vs. estimated β yielded a slope less than one (i.e., conservatism) for nearly every subject and d′ condition. Pooling data across subjects and conditions, a similar linear regression yields a slope significantly less than one (pConclusion: Observers are conservative in setting their criterion in a 2AFC discrimination task.
NIH EY08266 to MSL and NIH EY017988 to MP.