PsiMLE: A maximum-likelihood estimation approach to estimating psychophysical scaling and variability more reliably, efficiently, and flexibly

A simple and popular psychophysical model—usually described as overlapping Gaussian tuning curves arranged along an ordered internal scale—is capable of accurately describing both human and nonhuman behavioral performance and neural coding in magnitude estimation, production, and reproduction tasks for most psychological dimensions (e.g., time, space, number, or brightness). This model traditionally includes two parameters that determine how a physical stimulus is transformed into a psychological magnitude: (1) an exponent that describes the compression or expansion of the physical signal into the relevant psychological scale ( β ), and (2) an estimate of the amount of inherent variability (often called internal noise ) in the Gaussian activations along the psychological scale ( σ ). To date, linear slopes on log–log plots have traditionally been used to estimate β , and a completely separate method of averaging coefficients of variance has been used to estimate σ . We provide a respectful, yet critical, review of these traditional methods, and offer a tutorial on a maximum-likelihood estimation (MLE) and a Bayesian estimation method for estimating both β and σ [ PsiMLE ( β , σ )], coupled with free software that researchers can use to implement it without a background in MLE or Bayesian statistics (R-PsiMLE). We demonstrate the validity, reliability, efficiency, and flexibility of this method through a series of simulations and behavioral experiments, and find the new method to be superior to the traditional methods in all respects.