Workshop on Multinomial Processing Tree (MPT) Models

This workshop will follow MathPsych 2011 and precede the Meeting of the Cognitive Science Society in Boston.

Venue: Tufts University in Medford, MA 02155

Date: July 19, 2011

Instructors

William Batchelder, Richard Chechile, and Xiangen Hu


Background

The purpose of this workshop is to provide researchers with in detailed, hands-on access to recent mathematical, empirical, computational, and statistical developments concerning the class of Multinomial Processing Tree (MPT) models. MPT models are an increasingly popular class of parametric probability models for categorical data generated in cognitive experiments, and models within this class have been applied in such areas as perception, categorization, decision making, social processes, reasoning, and especially in many experimental paradigms in human memory, including recognition memory, free recall, source monitoring, process dissociation, and prospective memory. A 2009 Special Issue of Zeitschrift für Psychologie was devoted to recent developments in these models, and the lead article by Erdfelder et al. reviews over 100 papers developing versions of these models in 20 different research areas. MPT models are relatively simple and paradigm specific, and as a consequence many of their mathematical and statistical properties have been developed in depth at the class level. Models in the MPT class are specified as latent decision trees, where branches correspond to hypothetical processing sequences that eventuate in a manifest categorical response. The main purpose of using a MPT model is to provide a way to indirectly measure latent cognitive processes, such as memory storage, memory retrieval, logical inference, attention focus, metacognition, guessing biases, and perceptual integration, which otherwise are not directly accessible with any single dependent variable measure.

Workshop Overview

This workshop involves more than six hours of instruction by three researchers with wide experience in the history and development of MPT models in psychology. The course is designed to be informative and practical for researchers from diverse backgrounds. Minimal prerequisites would include a basic course in calculus and some background in probability and statistics. Both graduate students in experimental psychology as well as experienced modelers who are inexperienced with MPT models should learn valuable information from this workshop. The instruction includes discussion of task and model design, model specification, model identifiability, parameter validation, both Bayesian and classical methods for statistical estimation, hierarchical modeling, and model selection. Among the statistical estimation methods covered will be maximum likelihood estimation, random effects models for item and subjects, and population parameter mapping. Importantly, software for the different approaches will be provided and explained.

MPT Models Workshop Details