Comparing visual discrimination and detection: the special status of ‘no’ responses

Making a decision about whether something is there or not (detection) is qualitatively different from making a decision about what is there (discrimination). A key aspect of detection is the asymmetry in the availability of evidence for 'yes' and for 'no' responses. While discrimination requires a comparison between the relative evidence for different options, in a detection setting, evidence is only available for the presence of a stimulus and not for its absence. This means that confidence in the absence of a stimulus cannot rely on the magnitude of "evidence for absence" and must instead be based on other information. In line with this conceptual difference, previous studies identified behavioural dissimilarities between 'yes' and 'no' responses: confidence in 'no' responses is generally lower, and is less predictive of objective accuracy (Kanai, Walsh, & Tseng, 2010; Meuwese, van Loon, Lamme, & Fahrenfort, 2014) not only in visual detection but also in detection-like tasks (such as recognition memory; Higham, Perfect, & Bruno, 2009). Here, we set out to further characterise the distinct cognitive processes involved in reporting stimulus absence compared to reporting stimulus presence or identity. We present the results from two psychophysical experiments in which participants were asked to perform detection and discrimination tasks on the same class of stimuli, in different blocks. Although there was no difference in overall accuracy between the two tasks, we find markedly different behavioural signatures for discrimination and detection responses. Specifically, we replicate findings of a weaker relationship between accuracy and confidence for 'No' responses (lower area under the type 2 ROC) and show that 'No' responses exhibit a weaker negative association between reaction-time and confidence compared to both 'Yes' and discrimination responses. We discuss our observations in the context of possible cognitive models, such as unequal-variance SDT and temporal evidence accumulation models.