Research

Broadly speaking, my research is based on using mathematical models to draw inferences from behavioral data about unobserved (typically unobservable) states in people. I have used a wide variety of statistical techniques to draw such inferences, including, but not limited to, multilevel (i.e., “mixed effect”) generalized linear models, machine learning classifiers (e.g., support vector machines, naive Bayes classifiers, linear and quadratic discriminant analysis), unidimensional and multidimensional signal detection theory, principal components analysis, and confirmatory factor analysis.

Much of my recent work has focused on developing and using multidimensional models of perception and response selection to draw inferences about perceptual evidence and response bias in the identification of speech sound masked by multi-talker babble.

Here is the Open Science Framework page for a project in which 11 listeners identified 10 tokens from each of 20 talkers (10 male, 10 female) of each of four consonants (t, d, s, z). The OSF page has trial-level identification-confusion data, Python scripts for using PyMC 2 to fit five multilevel GRT models with different structures, Python scripts for visualizing and comparing fitted models, and some fitted models stored in .hdf5 files. A pre-print of the (under review) paper describing all of this in plenty of detail is also hosted there.

More soon… (3 November 2017)