Multitrait-multimethod CFA and informant discrepancy models

Description

Confirmatory factor models that measure several traits by several methods or informants and separate construct variance from method variance, instead of averaging disagreeing sources.

Campbell and Fiske's multitrait-multimethod matrix measures several traits with several methods to assess convergent and discriminant validity. CFA versions model it directly. The correlated trait-correlated method model often fails (about 75% improper solutions in one review); correlated uniqueness models treat method effects as error covariances; CTC(M-1) takes one method as reference and models others as deviations from it; multilevel models handle interchangeable raters such as several customers rating one product; trifactor models handle multiple informants. Developmental research adds that disagreement between informants (parents, teachers, observers, self) is systematic and predictable from age, context and construct, and that each source adds unique predictive information. A general common-method factor is defensible only with marker variables such as a social-desirability scale. Plan multiple methods at the design stage.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Measure each trait with at least two structurally different methods or informants (for example self-report, observer rating, telemetry).
  2. Choose the model: CTCM, correlated uniqueness or CTC(M-1) for structurally different methods; multilevel or trifactor models for interchangeable raters or multiple informants.
  3. If testing common-method bias with a general factor, include marker variables such as a social-desirability scale.
  4. Fit the model and evaluate global fit (CFI, RMSEA) and local fit, watching for inadmissible estimates.
  5. Interpret trait (convergent) and method variance components and discriminant validity across traits.
  6. Model predictors of informant discrepancy (context, user group, construct) rather than averaging sources.

Sources

  • Principles and Practice of SEM (Kline, 2023)
  • Handbook of Child Psychology Vol 3 (Lerner & Lamb, 2015)
  • Portfolio Entrepreneurs in China: A Mixed Methods Study (Xu, 2021)
  • A Beginner's Guide to Structural Equation Modeling (Schumacker & Lomax, 2004)
  • Latent Variable Models (Loehlin & Beaujean, 2017)

Also known as

  • MTMM analysis
  • Multitrait-multimethod CFA and common-method variance models
  • Method factor models
  • CTC(M-1)
  • Correlated uniqueness model
  • Informant discrepancy modelling
  • Informant discrepancy and multitrait-multimethod modelling of observers, parents and self-report

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics