Academic journal article Attention, Perception and Psychophysics

Using PPT to Account for Randomness in Perception

Academic journal article Attention, Perception and Psychophysics

Using PPT to Account for Randomness in Perception

Article excerpt

Published online: 31 May 2012

(C> Psychonomic Society, Inc. 2012

Abstract According to many theories of decision making, of which signal detection theory is the most prominent, randomness is the main factor responsible for imperfect performance. These theories imply that correcting for attenuation due to randomness should result in perfect scores as long as the participants use nonextreme decision criteria. On the basis of a recent advance termed potential performance theory (Trafimow & Rice, Psychological Review 115:447-462, 2008), we performed auditory and visual detection experiments and corrected the scores for attenuation. Most participants in both experiments tended to perform at a less-than-perfect level, even after their scores were corrected. The findings demonstrate that at least one systematic factor influences detection that is not included in signal detection theory.

Keywords Potential performance theory * Signal detection theory * Randomness * Perception

(ProQuest: ... denotes formula omitted.)

Suppose that participants are presented with visual or auditory stimuli in a detection or discrimination task. If the task is sufficiently difficult, participants will make errors. The large class of statistical decision models, of which signal detection theory is the most prominent member (e.g., Swets, Tanner, & Birdsall, 1961; Wald, 1950), features the notion of random noise as the main factor responsible for the lack of perfect responding. Although the present research focuses on signal detection theory, due to its prominence in the literature, the research to be presented is more generally applicable to the notion of randomness as an explanatory factor in perception studies.

Consider a two-interval forced choice (2IFC) detection task in which the participant attempts to determine the interval that contains the signal stimulus. According to signal detection theory (hereafter, SDT), the observer is given a pair of percepts (one for each interval) and generates a response by identifying the interval that contains the largest percept. SDT is based on the assumption that random noise is included in the percepts given to the observer for classification: Distributions of percepts (typically Gaussian) are associated with each stimulus type. Given an unlucky sampling, the percept associated with the signal interval could be less intense than the percept associated with the noise interval, and an error could occiu.

Of course, a variety of challenges to the classic equalvariances Gaussian version of SDT have been put forward over the years. Most notably, empirical evidence has accumulated to support the unequal-variances version of SDT over the classic equal-variances version (Ratcliff, Shue, & Gronlund, 1992; Swets et al., 1979; Wixted, 2007). Typically, the variance of the target (signal) distribution is estimated to be larger than that of the noise distribution (Egan, 1975; Mickes, Wixted, & Wais, 2007; Ratcliff et al., 1992). Representing a more fundamental attack on SDT, Balakrishnan (1999; see also Balakrishnan & MacDonald, 2008) presented evidence suggesting that the decision criterion does not shift under response bias manipulations. Mueller and Weidemann (2008; see also Benjamin, Diaz, & Wee, 2009) attempted to account for these findings by adding random noise to the placement of the decision criterion. Complicating matters, Wickelgren and Norman (1966) demonstrated that there is no theoretical way to distinguish percept noise from criterion noise within the SDT framework, effectively requiring empirical methods to lend support to the criterion variability account.

Each of these refinements to the classical equal-variances SDT model essentially amount to inserting additional variability (randomness) into the model. Because there might be different types of randomness (criterial, perceptual, or others) and different conditions under which different types of randomness matter more or less, and because the variances of the signal and noise distributions may be equal or unequal, it is very difficult to evaluate the basic question of concern: Can the SDT framework provide a frill and accurate account of decision-making behavior? …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.