A tutorial on cue combination and Signal Detection Theory: Using changes in sensitivity to evaluate how observers integrate sensory information

详细信息    查看全文

• 作者：Pete R. Jones ^{p.r.jones@ucl.ac.uk" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Cue combination ; Multisensory integration ; Weighted linear summation ; Signal Detection Theory ; Internal noise
• 刊名：Journal of Mathematical Psychology
• 出版年：2016
• 出版时间：August 2016
• 年：2016
• 卷：73
• 期：Complete
• 页码：117-139
• 全文大小：2354 K

文摘

Many sensory inputs contain multiple sources of information (‘cues’), such as two sounds of different frequencies, or a voice heard in unison with moving lips. Often, each cue provides a separate estimate of the same physical attribute, such as the size or location of an object. An ideal observer can exploit such redundant sensory information to improve the accuracy of their perceptual judgments. For example, if each cue is modeled as an independent, Gaussian, random variable, then combining 00165&_mathId=si28.gif&_user=111111111&_pii=S0022249616300165&_rdoc=1&_issn=00222496&md5=fbfc2e77cd1922709bc684128d454f97" title="Click to view the MathML source">N$N$cues should provide up to a 00165&_mathId=si77.gif&_user=111111111&_pii=S0022249616300165&_rdoc=1&_issn=00222496&md5=0048d0077773537c9c2b321a3380562a" title="Click to view the MathML source">√N$\surd N$ improvement in detection/discrimination sensitivity. Alternatively, a less efficient observer may base their decision on only a subset of the available information, and so gain little or no benefit from having access to multiple sources of information. Here we use Signal Detection Theory to formulate and compare various models of cue-combination, many of which are commonly used to explain empirical data. We alert the reader to the key assumptions inherent in each model, and provide formulas for deriving quantitative predictions. Code is also provided for simulating each model, allowing expected levels of measurement error to be quantified. Based on these results, it is shown that predicted sensitivity often differs surprisingly little between qualitatively distinct models of combination. This means that sensitivity alone is not sufficient for understanding decision efficiency, and the implications of this are discussed.