Applying mixed-effects receiver operating characteristic (ROC) curve analysis to diagnostic evaluations of human learning

Date

2001-08

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

This dissertation introduces a novel application of mixed-effects receiver operating characteristic (ROC) curve analysis in the study of individual differences. By applying a mixed-effects ordinal regression model to the rating outcomes of a spelling discrimination study, the underlying sensitivities of good and poor undergraduate spellers to a set of correctly spelled and misspelled words were estimated to test two competing hypotheses about individual differences in spelling. Variation in random slopes was modeled as a function of person-level covariates to identify possible predictors of spelling ability. Results provide strong support for the role of strategy-use in spelling discrimination. After partialling for strategy-use, between-group differences for good and poor spellers were greatest on irregular words, lending support to the hypotheses that good spellers have more well-defined mental lexicons than poor spellers. A significant proportion of unexplained variation in the random slopes remained after modeling the ratings assigned to irregular words, suggesting that spelling discrimination may involve additional strategies not identified in this study. This study showed that discrimination accuracy tasks can be sensitive, valid and reliable measures of underlying ability, attesting to their use in the study of individual differences.

Description

text

Citation