Title of Oral/Poster Presentation

Using Mixture Multitrait-Multimethod Analysis to Assess Trait-Method Interactions

Presenter Information

Kaylee LitsonFollow

Class

Article

Department

Psychology

Faculty Mentor

Christian Geiser

Presentation Type

Oral Presentation

Abstract

Multitrait-multimethod (MTMM) analyses are commonly used in psychological measurement to assess convergent and discriminant validity. Convergent validity, as defined by Campbell and Fiske (1959), is the correlation between two or more methods that measure the same trait. Current MTMM models assume that method effects are homogeneous across all levels of a trait and do not account for trait-method interactions. However, it is possible that when two methods measure the same trait, the relationship between the two methods varies across the different trait levels, leading to a trait-method interaction. In the current paper, I present a novel approach to assess trait-method interactions using mixture structural equation modeling (Muthén, 2001). Mixture models are used to identify sources of heterogeneity within a sample that were not previously known to the researcher (Lubke & Muthén, 2005). In the context of mixture MTMM analysis, sources of heterogeneity (i.e., the trait-method interaction) may lead to different levels of convergent validity. To illustrate the new method, a two-class single-trait multimethod mixture model was applied to mother and teacher ratings of hyperactivity in first-grade Spanish children (N = 718). Results from this model revealed two subpopulations of individuals: (1) a small group of children (11%) with high symptom levels (M_Mothers = 2.33; M_Teachers = 3.33), a smaller covariance (cov = .17) between latent rater factors, and a nonsignificant correlation (r = .21); and (2) a larger group of children (89%) with low symptom levels (M_Mothers = 1.06; M_Teachers = 0.50), a larger covariance (cov = .23) between rater factors, and a significant correlation (r = .41) between rater factors. Results found that mother and teacher ratings show less agreement in subgroups with higher levels of hyperactivity, indicating a trait-method interaction effect. These results may have implications for future MTMM validity research. Limitations and future directions are further discussed.

Start Date

4-9-2015 12:00 PM

This document is currently not available here.

Share

COinS
 
Apr 9th, 12:00 PM

Using Mixture Multitrait-Multimethod Analysis to Assess Trait-Method Interactions

Multitrait-multimethod (MTMM) analyses are commonly used in psychological measurement to assess convergent and discriminant validity. Convergent validity, as defined by Campbell and Fiske (1959), is the correlation between two or more methods that measure the same trait. Current MTMM models assume that method effects are homogeneous across all levels of a trait and do not account for trait-method interactions. However, it is possible that when two methods measure the same trait, the relationship between the two methods varies across the different trait levels, leading to a trait-method interaction. In the current paper, I present a novel approach to assess trait-method interactions using mixture structural equation modeling (Muthén, 2001). Mixture models are used to identify sources of heterogeneity within a sample that were not previously known to the researcher (Lubke & Muthén, 2005). In the context of mixture MTMM analysis, sources of heterogeneity (i.e., the trait-method interaction) may lead to different levels of convergent validity. To illustrate the new method, a two-class single-trait multimethod mixture model was applied to mother and teacher ratings of hyperactivity in first-grade Spanish children (N = 718). Results from this model revealed two subpopulations of individuals: (1) a small group of children (11%) with high symptom levels (M_Mothers = 2.33; M_Teachers = 3.33), a smaller covariance (cov = .17) between latent rater factors, and a nonsignificant correlation (r = .21); and (2) a larger group of children (89%) with low symptom levels (M_Mothers = 1.06; M_Teachers = 0.50), a larger covariance (cov = .23) between rater factors, and a significant correlation (r = .41) between rater factors. Results found that mother and teacher ratings show less agreement in subgroups with higher levels of hyperactivity, indicating a trait-method interaction effect. These results may have implications for future MTMM validity research. Limitations and future directions are further discussed.