Neural Information Processing

Psignifit

Psignifit is a free toolbox for psychometric function estimation. The current version—psignifit 4—provides a full Bayesian analysis of a beta-binomial model of the psychometric function, providing a tool for robust inference for the psychometric function even for overdispersed (“noisy") data. The toolbox is available as MATLAB code and in addition we provide a clone of it in python.

Where to find the psignifit 4 software:

You find an introduction and general information about the software packages in their Wiki(s):

These are the direct download links for the packages:

Our paper explaining the beta-binomial model and showing how it improves robustness is published in Vision Research (alternatively see http://www.nip.uni-tuebingen.de/publications/peer-reviewed.html).

Psignifit - History:

As you might have guessed from the version number, there have been earlier versions of psignifit. The earliest version (link: http://www.nip.uni-tuebingen.de/research/software/psignifit-legacy.html), which went up till version 2.5.6. provided a maximum likelihood fitting procedure and estimates for confidence intervals from bootstrapping techniques, based on two papers by Felix Wichmann and Jeremy Hill.

Version 3 (link http://psignifit.sourceforge.net/) was developed by Ingo Fründ, Valentin Hänel and Felix Wichmann at the TU Berlin and provided the complete functionality of the previous 2.5.6 version in python and added a first MCMC sampling based Bayesian inference. (However, it is important to note that the sampling, and thus parameter and confidence interval estimates, are not always reliable in this version! This was one of the motivations to write the latest psignifit 4 version based on numerical integration rather than MCMC sampling!) Both of these legacy versions are still available for download, but their maintenance was stopped and they are no longer fully functional with modern computers and software.

We strongly suggest you use psignifit 4 in the future.