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

Area of Science:

• Psychophysics
• Cognitive Neuroscience
• Computational Psychology

Background:

• A widely used psychophysical model employs overlapping Gaussian tuning curves to explain human and nonhuman performance in magnitude estimation tasks across various psychological dimensions.
• This model traditionally estimates two parameters: the exponent (β) for stimulus transformation and internal noise (σ) using separate, less robust methods like log-log plots and coefficient of variance averaging.

Purpose of the Study:

• To critically review traditional methods for estimating psychophysical model parameters (β and σ).
• To introduce and validate a novel maximum-likelihood estimation (MLE) and Bayesian approach (PsiMLE) for simultaneously estimating both β and σ.
• To provide accessible software (R-PsiMLE) for researchers to implement these advanced statistical techniques.

Main Methods:

• A new integrated method combining maximum-likelihood estimation (MLE) and Bayesian inference was developed to estimate both the exponent (β) and internal noise (σ) parameters.
• The R-PsiMLE software package was created to facilitate the application of this new method without requiring specialized statistical expertise.
• Extensive simulations and behavioral experiments were conducted to rigorously test the validity, reliability, efficiency, and flexibility of the proposed estimation technique.

Main Results:

• The novel PsiMLE method, utilizing MLE and Bayesian estimation, demonstrated superior performance compared to traditional approaches for estimating both β and σ.
• Simulations and empirical data confirmed the validity, reliability, efficiency, and flexibility of the new estimation method.
• The R-PsiMLE software successfully enabled researchers to implement the advanced estimation techniques.

Conclusions:

• The integrated MLE and Bayesian estimation method (PsiMLE) offers a more accurate and robust approach to analyzing psychophysical data than traditional techniques.
• The R-PsiMLE software democratizes access to advanced statistical modeling for psychophysical research.
• This work advances the precise quantification of internal representations in sensory and cognitive processes.