Abstract
Structural models of psychopathology have emerged as an alternative to traditional categorical approaches. The bifactor model, which incorporates a general p-factor, has become the preferred structure. The p-factor is claimed to represent a substantive construct or property of the system; however, recent evidence suggests that it may be without substantive meaning. If a universal substantive p, and associated specific factors, is to be developed they not only must be applicable and consistent between populations but also must be applicable and consistent within subgroups of a population. This consistency needs to include not only factor loadings but also factor correlates. We used a simulated data approach to explore the applicability and consistency of four popular models of psychopathology to a range of heterogeneous subgroups and examined the consistency of their neurocognitive correlates. We found that only eight out of sixty-three subgroups fitted any of the models with all significant loadings, no negative loadings, no non-positive-definite identification issues, and no negative variance. All eight of these subgroups fit the correlated factors model, none fit the original bifactor model, four subgroups fit the revised bifactor model, and one subgroup fit the single-factor model. Correlates of the factors also varied substantially between the subgroups fitted to the same model. We discuss the implications of the findings, including the implications for the development of universal substantive factors of psychopathology.