Multinomial-Processing-Tree (MPT) Modeling: Basic Methods and Recent Advances

This repository provides data and materials for a workshop on multinomial processing tree (MPT) models given by Edgar Erdfelder, Daniel W. Heck, and Franziska Meissner at various occasions.

Abstract

Multinomial processing tree (MPT) models provide a powerful tool to disentangle different cognitive processes contributing to the same observable responses. In many fields of psychology (e.g., memory, decision making, and social cognition), MPT models have been successfully used to test hypotheses concerning specific processes via separate model parameters. So far, however, applications of this model class were restricted to discrete data such as response frequencies. Furthermore, these response frequencies were usually pooled across participants which prohibits applications of MPT models to hypotheses involving individual differences and also involves the risk of standard errors being calculated incorrectly. Recent research, however, resulted in sophisticated improvements and extensions that will broaden the potential applications of this model class.

In our two-day workshop, we provide a systematic and application-oriented overview of the basics and the most recent developments in MPT modeling. Using the software multiTree and other tools, participants will practice how to develop their own MPT models for categorical data, how to assess model fit, how to test hypotheses on model parameters and assess the statistical power of these tests, how to check for identifiability of different model variants, and how to select the best model from a set of candidate models that provides optimal balance between goodness-of-fit and model simplicity. Furthermore, workshop participants will learn about recent advancements. One of them is hierarchical Bayesian modeling, an approach that provides researchers with all the advantages of parameter estimation at the participant level while achieving a higher reliability by assuming a group structure of the parameters on the aggregate level. The second important development is the extension of MPT models to capture categorical and continuous data jointly, enabling applications to a considerably larger set of paradigms and dependent variables (e.g., latency-based or process-tracing measures). Participants will exercise the application of both of these advancements with user-friendly software (TreeBUGS, gpt).

Preliminary Knowledge

The workshop will be largely application-oriented. Nevertheless, we will provide a solid conceptual and statistical treatment. Thus, basic knowledge about MPT models is useful but not necessary for participation. A solid background in statistics is sufficient (Master's level). Note that this workshop does not solely address substantive researchers who want to learn about recent MPT developments. We explicitly invite applications from researchers who have been interested in MPT modeling but did not know how to start.