How can we determine with certainty whether a stimulus is conscious or not? This is still a matter of debate (Persaud, McLeod & Cowey, 2007; Schurger & Sher, 2008). Early work simply relied on the introspection and verbal report of participants. In the interests of greater rigor, the last thirty years of research in cognitive psychology have been dominated by the search for objective criteria of consciousness, based on signal detection theory. According to this perspective, a stimulus is said to be non-conscious or subliminal (below threshold) only if performance remains at chance level in a direct detection or classification task (zero d-prime criterion). However, such an objective definition raises a number of difficulties. Firstly, it overestimates conscious perception, since it is not uncommon for a participant to perform better than chance while denying having perceived the stimuli. Secondly, performance may be better than chance in some tasks, but not in others, raising the question of which tasks count towards a definition of consciousness, and which are simply under the influence of non-conscious operations. Thirdly, the proposed approach poses the statistical difficulty of accepting the null hypothesis (that performance should not deviate significantly from chance). In reality, the value of the d-prime never goes quite down to zero, and its statistical significance then depends solely on the number of trials dedicated to its measurement, which robs it of much of its objective character.
It is for reasons such as these that new proposals have recently emerged. Some suggest a return to purely subjective reporting, based for example on quantitative evaluation of visibility (Sergent & Dehaene, 2004). Others call for metacognitive or second-order judgments such as performance betting (do you want to bet that your forced-choice answer was correct? Persaud et al., 2007).