Three recommendations based on a comparison of the reliability and validity of the predominant models used in research on the empirical structure of psychopathology

Miriam K. Forbes*, Ashley L. Greene, Holly F. Levin-Aspenson, Ashley L. Watts, Michael Hallquist, Benjamin B. Lahey, Kristian E. Markon, Christopher J. Patrick, Jennifer L. Tackett, Irwin D. Waldman, Aidan G. C. Wright, Avshalom Caspi, Masha Ivanova, Roman Kotov, Douglas B. Samuel, Nicholas R. Eaton, Robert F. Krueger

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

6 Citations (Scopus)
10 Downloads (Pure)

Abstract

The present study compared the primary models used in research on the structure of psychopathology (i.e., correlated factor, higher-order, and bifactor models) in terms of structural validity (model fit and factor reliability), longitudinal measurement invariance, concurrent and prospective predictive validity in relation to important outcomes, and longitudinal consistency in individuals' factor score profiles. Two simpler operationalizations of a general factor of psychopathology were also examined-a single-factor model and a count of diagnoses. Models were estimated based on structured clinical interview diagnoses in two longitudinal waves of nationally representative data from the United States (n = 43,093 and n = 34,653). Models that included narrower factors (fear, distress, and externalizing) were needed to capture the observed multidimensionality of the data. In the correlated factor and higher-order models these narrower factors were reliable, largely invariant over time, had consistent associations with indicators of adaptive functioning, and had moderate stability within individuals over time. By contrast, the fear- and distress-specific factors in the bifactor model did not show good reliability or validity throughout the analyses. Notably, the general factor of psychopathology (p factor) performed similarly well across tests of reliability and validity regardless of whether the higher-order or bifactor model was used; the simplest (single factor) model was also comparable across most tests, with the exception of model fit. Given the limitations of categorical diagnoses, it will be important to repeat these analyses using dimensional measures. We conclude that when aiming to understand the structure and correlates of psychopathology it is important to (a) look beyond model fit indices to choose between different models, (b) examine the reliability of latent variables directly, and (c) be cautious when isolating and interpreting the unique effects of specific psychopathology factors, regardless of which model is used. 

Original languageEnglish
Pages (from-to)297-317
Number of pages21
JournalJournal of Abnormal Psychology
Volume130
Issue number3
DOIs
Publication statusPublished - Apr 2021

Keywords

  • bifactor model
  • hierarchical model
  • latent variable models
  • quantitative psychopathology
  • validity

Fingerprint

Dive into the research topics of 'Three recommendations based on a comparison of the reliability and validity of the predominant models used in research on the empirical structure of psychopathology'. Together they form a unique fingerprint.

Cite this