Abstract
Perceptual accuracy can improve by combining multiple people’s judgments into a group decision. Assuming that internal observer responses are Gaussian and Independent (GI), a classic prediction by Signal Detection Theory (GI-SDT) is that the maximum possible group-d’ is equal to the square root of the sum of individual’s squared d-primes (Sorkin & Dai, 1994). When observers scrutinize images limited by external noise (e.g., medical, satellite, aerial imagery), judgments are typically correlated across observers, and the benefits from optimal collective integration are lower than that predicted by GI-SDT (group-d’observed / group-d’GI-SDT < 1). Exp1 demonstrates the typical reduced benefits of collective integration (i.e., d’ratio < 1) using a single location, noise limited, yes/no task. Fifteen observers responded on an 8-point rating scale whether a Gaussian luminance signal was embedded in white noise at a known location (signal present 50%). Group d-primes obtained using an optimal linear combination model were significantly larger than individual d-primes, but significantly lower (p < .01; bootstrap resampling) than that predicted by GI-SDT (d’ratio = .86 for groups of 2; d’ratio = .79 for groups of 3). In contrast, Exp2 demonstrates that visual search tasks can lead to large increases in the benefits of collective integration. Fifteen observers searched (2 second time limit) for a target (target present 50%) in a random and unknown location within a white noise field (28 x 22.5 degrees). Observers responded using an 8-point rating scale about the presence of the target. Group d-primes obtained for the search task achieved benefits comparable or larger than that predicted by GI-SDT (d’ratio = 1.25 for groups of 2; d’ratio = 1.09 for groups of 3). Conclusion: Search with spatial uncertainty in noise limited tasks results in collective integration benefits that are significantly larger than single location perceptual judgments, and can approach or exceed the predictions of GI-SDT.
Meeting abstract presented at VSS 2015