# Designing state-trace experiments to assess the number of latent psychological variables underlying binary choices

- Guy Hawkins,
*University of Newcastle*
- Melissa Prince,
*University of Newcastle*
- Scott Brown,
*University of Newcastle*
- Andrew Heathcote,
*University of Newcastle*

## Abstract

State-trace analysis is a non-parametric method that can identify
the number of latent variables (dimensionality) required to explain the effect of
two or more experimental factors on performance. Heathcote, Brown & Prince
(submitted) recently proposed a Bayes Factor method for estimating the evidence
favoring one or more than one latent variable in a state-trace experiment, known
as Bayesian Ordinal Analysis of State-Traces (BOAST). We report results from a
series of simulations indicating that for larger sample sizes BOAST performs well
in identifying dimensionality for single and multiple latent variable models. A
method of group analysis convenient for smaller sample sizes is presented with
mixed results across experimental designs. We use the simulation results to
provide guidance on designing state-trace experiments to maximize the probability
of correct classification of dimensionality.

Back to Table of Contents