pyWitness 1.0: A Python eyewitness identification analysis toolkit

Abstract

pyWitness is a python toolkit for recognition memory experiments, with a focus on eyewitness identification (ID) data analysis and model fitting. The current practice is for researchers to use different statistical packages to analyze a single dataset. pyWitness streamlines the process of the data analysis. In addition to conducting key data analyses (e.g., receiver operating characteristic analysis, confidence accuracy characteristic analysis), statistical comparisons, signal-detection- based model fits, simulated data generation, and power

Application of artificial intelligence to eyewitness identification

“…To compare the discriminability of the participants and FRS, we conducted confidence-based (e.g., Grounlund et al, 2014) and similarity value-based ROC analyses, respectively. A ROC plots correct ID rate and false ID rate pairs for every level of confidence or similarity, cumulating as confidence decreases or similarity increases (e.g., Gronlund et al, 2014 ; Mickes et al, 2023 ). The further the points bow toward the upper left corner, the better the ability to discriminate innocent from guilty suspects.…”

“…The data and scripts are available at OSF ( https://osf.io/6tfuj/?view_only=c0ea0e5d02b34a529e1366f8daac62da ). The analyses were conducted in pyWitness ( https://lmickes.github.io/pyWitness/ ; Mickes et al, 2023 ).…”

Eyewitness identifications based on biased or unbiased line‐up instructions after a realistic and violent hostage simulation

“…In TP line‐ups we also included random intercepts for targets and participants and in TA line‐ups for participants (in TA line‐ups there were no targets) to control for any variance due to these factors. We included a calculation of the confidence accuracy relationship, in accordance with Mickes (2015) and we have plotted receiver operating characteristic (ROC) curves by confidence for the line‐up instructions and analysed the discriminability of both using the pyWitness software (L. Mickes et al., 2022).…”

Comments on the use of mathematical models in eyewitness identification research

“…By adjusting the parameters of the predictive model, the fit between observed and expected outcomes may be minimized and the parameters of the latent variables estimated. While model fitting generally requires a deep understanding of mathematical and/or computational modelling, this approach has become much more accessible to eyewitness researchers through the recent release of new and comprehensive open-source software, such as pyWitness for Python (Mickes, Seale-Carlisle, Chen, & Boogert, 2022), which provides for the fitting of empirical data to SDT models of eyewitness identification, data simulation, and power analysis, in addition to the production and analysis of ROC curves. The following sections provide a brief commentary of these various approaches, focusing on how SDT models inform the analytical frameworks for the measurement of eyewitness identification performance, and emphasize the richness inherent in the data.…”

“…However, this is not surprising due to the difficulty in making shifts towards theory-based data analysis, especially where those theories are being actively developed. Although the recent introduction of new open-source software, such as pyWitness (Mickes et al, 2022) is promising, it should be noted that the accessibility of these analytical tools, particularly to those researchers who rely on proprietary software such as SPSS, is probably rather low. Indeed, without access to tools that provide a simple graphic user interface, such as a Shiny app or other interactive web-based programs, researchers are unlikely to adopt any new methodology, simply because it would require too great an investment in retraining.…”