Abstract
Signal Detection Theory assumes no sensory threshold, that is, an internal response monotonically increasing with stimulus strength, available for decision. While a wealth of empirical observations unfailingly sustained this principle, the debate on the existence of a sensory threshold persists. An educated SDT intuition is that an identification response of a non-detected stimulus (Miss) should yield above chance performance. A high threshold theory predicts chance performance. Using a single-presentation double-task paradigm, consisting of independent detection (Yes/No) and identification (2AFC) reports, data of three highly trained observers show that orientation (±45°) as well as position (±2° eccentricity) discrimination performance for a missed Gabor patch (5 cpd) is very close to chance for criteria as high as 1.3 noise units (σ) above mean noise level (d′ up to 2.5). To compare the data with SDT predictions, we make the standard assumption that observers are monitoring two independent neuronal populations corresponding to the two possible targets in a given task. A ‘Yes’ report in the detection task occurs when a response from any of the two populations crosses a criterion level; an identification report corresponds to the identity of the population producing the largest response (‘labelled line’). Receiver Operating Characteristics of these populations were estimated using rating experiments, with the experimental results conforming to earlier reports (σ = 1 + r/4, with r the signal evoked mean internal response; Green & Swets, 1966; Graham, 1989). SDT predictions using these ROCs match observers' identification rate for Hit trials (r2=0.9) but exceed by far the measured identification rate for Miss trials (r2=0.04). While the rating data support the availability of a continuous internal response, the detection/identification data point to the fact that decision criteria in a Yes/No task are more akin to a high threshold inasmuch as they are considered to be boundaries for ‘invisible’ events.