The Adult Attachment Interview (AAI) is a widely used measure in developmental science that assesses adults’ current states of mind regarding early attachment-related experiences with their primary caregivers. The standard system for coding the AAI recommends classifying individuals categorically as having an autonomous, dismissing, preoccupied, or unresolved attachment state of mind. However, previous factor and taxometric analyses suggest that: (a) adults’ attachment states of mind are captured by two weakly correlated factors reflecting adults’ dismissing and preoccupied states of mind and (b) individual differences on these factors are continuously rather than categorically distributed. The current study revisited these suggestions about the latent structure of AAI scales by leveraging individual participant data from 40 studies (N = 3,218), with a particular focus on the controversial observation from prior factor analytic work that indicators of preoccupied states of mind and indicators of unresolved states of mind about loss and trauma loaded on a common factor. Confirmatory factor analyses indicated that: (a) a 2-factor model with weakly correlated dismissing and preoccupied factors and (b) a 3-factor model that further distinguished unresolved from preoccupied states of mind were both compatible with the data. The preoccupied and unresolved factors in the 3-factor model were highly correlated. Taxometric analyses suggested that individual differences in dismissing, preoccupied, and unresolved states of mind were more consistent with a continuous than a categorical model. The importance of additional tests of predictive validity of the various models is emphasized.

Taxometric procedures, model-based clustering and latent variable mixture modeling (LVMM) are statistical methods that use the inter-relationships of observed symptoms or questionnaire items to investigate empirically whether the underlying psychiatric or psychological construct is dimensional or categorical. In this review we show why the results of such an investigation depend on the characteristics of the observed symptoms (e.g. symptom prevalence in the sample) and of the sample (e.g. clinical, population sample). Furthermore, the three methods differ with respect to their assumptions and therefore require different types of a priori knowledge about the observed symptoms and their inter-relationships. We argue that the choice of method should optimally match and make use of the existing knowledge about the data that are analyzed.

Taxometric research methods were developed by Paul Meehl and colleagues to distinguish between categorical and dimensional models of latent variables. We have conducted a comprehensive review of published taxometric research that included 177 articles, 311 distinct findings and a combined sample of 533 377 participants. Multilevel logistic regression analyses have examined the methodological and substantive variables associated with taxonic (categorical) findings. Although 38.9% of findings were taxonic, these findings were much less frequent in more recent and methodologically stronger studies, and in those reporting comparative fit indices based on simulated comparison data. When these and other possible confounds were statistically controlled, the true prevalence of taxonic findings was estimated at 14%. The domains of normal personality, mood disorders, anxiety disorders, eating disorders, externalizing disorders, and personality disorders (PDs) other than schizotypal yielded little persuasive evidence of taxa. Promising but still not definitive evidence of psychological taxa was confined to the domains of schizotypy, substance use disorders and autism. This review indicates that most latent variables of interest to psychiatrists and personality and clinical psychologists are dimensional, and that many influential taxonic findings of early taxometric research are likely to be spurious.