Purchase this article with an account.
Luis A. Lesmes, Zhong-Lin Lu, Nina T. Tran, Barbara A. Dosher, Thomas D. Albright; An adaptive method for estimating criterion sensitivity (d') levels in yes/no tasks. Journal of Vision 2006;6(6):1097. doi: 10.1167/6.6.1097.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Compared to those available for forced-choice tasks, there are few adaptive methods for estimating thresholds in Yes/No (YN) tasks. Even existing methods1,2, aimed at estimating thresholds for specific Yes rates (e.g., 50%), are unreliable, due to contamination of Yes rate by response bias. We present a novel adaptive procedure (the “quick YN” or “qYN” method), to estimate thresholds at levels of criterion sensitivity (d'), rather than Yes rate. The method uses Bayesian estimation and a minimum entropy criterion3,4,5 to place stimuli at signal intensities providing the most information about three parameters: (1) the observer's response biasβ; (2) the signal intensity, α, corresponding to d'=1; and (3) the steepness of the d' psychometric function, γ. Using simulations and a psychophysical experiment, we compared threshold estimates obtained in a YN task, using the qYN and the method of constant stimuli (MCS). Simulations showed that, for false-alarm (FA) rates from 10–50%, the qYN needed only 25 trials to provide accurate (bias <.5dB) and precise (sd < 2.5dB) estimates of d'=1. For FA=1%, d' estimates were biased by 2dB. For all FA rates, the number of MCS trials needed to match the qYN's precision was roughly five-fold. Psychophysical results showed that qYN threshold estimates (sd < 2.5dB), deviated little from MCS estimates (<1dB). The qYN exhibits strong advantages over previous methods. Most notably, by measuring (and accounting for) response bias, the qYN estimates thresholds at criterion sensitivities with reduced data collection.
1. Green (1993)
2. Kaernbach (1990)
3. Cobo-Lewis (1996)
4. Kontsevich & Tyler (1999)
5. Lesmes et al (2005)
This PDF is available to Subscribers Only