Vision thresholds from psychometric analyses: alternatives to probit analysis

Abstract

Vision scientists and clinicians regularly obtain measures of vision and determine a person's vision threshold by choosing one of numerous methods of analysis. Analytical methods differ in their presumptions about the measures, in their complexity and in the ease of obtaining the threshold estimate. In light of these differences among so-called methods of psychometric analysis, the question is, "Do various analytical methods provide essentially equivalent vision threshold estimates or are there consequential differences that require consideration?" Measures of visual acuity, motion processing, and texture processing were obtained from 20 subjects. Each set of measures was analyzed by five psychometric methods: Logit Analysis, Normit Analysis, and linear regression of z-score transformed, logit transformed and untransformed probabilities vs. stimulus strength. The resulting thresholds were compared to the threshold obtained from Probit Analysis, which was used as a reference or "gold standard." Thresholds from the procedures were remarkably similar to those from Probit Analysis. In addition, examination of the speed of the procedures revealed that Probit Analysis was up to 10 times slower than some of the others. Considering the similarity of thresholds, the speed of computation, and the ease of implementation, Logit and Normit Analyses especially provide effective alternatives to the current gold standard, Probit Analysis, for the estimation of psychometric thresholds. In addition, z-score, logit, and linear regressions also produced unbiased threshold estimates under many circumstances, but the latter method should be applied with some caution.