Is the p-factor an artifact of evaluative consistency bias? A multitrait-multimethod examination

• In Event: Substantive Developmental Research Questions Addressed with Applied Measurement Methods

Abstract

In child and adolescent psychopathology, much research has focused on two higher-order factors, externalizing and internalizing. Based on the fact that externalizing and internalizing dimensions are generally moderately to highly correlated, various researchers have proposed that a theoretical dimension referred to as the p-factor explains their covariation and gives rise to psychopathology broadly construed. More often than not, the p-factor is presumed to reflect a substantive mechanism (e.g., emotion dysregulation, negative emotionality, disordered thought). Little research has considered potential artefactual, or methodological, explanations of the p-factor. To this end, we drew inspiration from the p-factor’s conceptual cousin, the general factor of personality. Not too long ago, the general factor of personality was determined to arise from evaluative consistency bias in single-reporter designs. That is, the tendency to rate oneself overly negatively or positively caused excess covariation among personality dimensions that spuriously produced a general factor. All told, when personality is modeled in a multitrait-multimethod framework with three or more raters on the same target participant, empirical support for the general factor of personality essentially vanishes. The present work extended this concept to psychopathology and modeled multi-informant psychopathology data on youth by means of multitrait-multimethod models. We used 3 cross-sectional and longitudinal samples where at least three individuals (e.g., self, parent, teacher; N raters per study ranged from 3 to 5) provided ratings on a child or adolescent’s psychopathology as assessed by the Child Behavior Checklist and its progeny. Samples included the Healthy Brain Network (N=303), the Adolescent Brain and Cognitive Development Study (N=2119), and the NICHD Study of Early Child Care and Youth Development (N=714). Like in other research, we conceptualized support for the p-factor as the magnitude of the correlation between externalizing and internalizing dimensions. In single-reporter models, externalizing and internalizing dimensions were moderately to highly correlated across samples (rs ranged from .52 to .90) depending on the rater. In multitrait multimethod models, the externalizing-internalizing correlation became attenuated, although the two dimensions remained moderately correlated (rs ranged from .42 to .74). The extent to which the externalizing-internalizing correlation diminished in multitrait-multimethod models, compared with single-reporter models, differed by sample and by whether items or subscales were modeled. Taken together, empirical support for the p-factor may be stronger in single-reporter designs, although it is clear that the p-factor does not vanish in multi-informant models. Our findings suggest that the p-factor in any given single-reporter model may reflect tendencies towards evaluative consistency bias, among other things. Ultimately, the p-factor is likely not a monolith but an amalgam of substantive and methodologic variance that cannot be disentangled in single-reporter designs.

Authors

• Ashley Lauren Watts, University of Missouri - Columbia
  Presenting Author

• Isabella Palumbo, Georgia State University
  Non-Presenting Author

• Bridget Makol, University of Maryland, College Park
  Non-Presenting Author

• Colin DeYoung, University of Minnesota
  Non-Presenting Author

• Andres De Los Reyes, University of Maryland - College Park
  Non-Presenting Author