LAUSR

In their article, Mulder, Horwood, Tyrer, Carter and Joyce aimed at validating the five personality domains proposed by the ICD-11 Personality Disorders Committee. They selected 57 of the 79 personality disorder criteria assessed in the SCID II, which were judged to fit into the domains, and applied them to 606 patients with major depression. To reduce statistical difficulties with low base rate symptoms, they classified the selected symptoms into 23 unequally loaded item parcels—again corresponding to the hypothesized domains. On these item parcels, they applied confirmatory factor analysis (CFA). CFA offers the opportunity to test whether the empirical data of an independent sample fit a predefined factorial structure. The initial fivefactor model corresponding to the five proposed personality domains as defined by the ICD-11 Personality Disorders Committee yielded a comparative fit index CFI of .69 which indicates ‘unacceptable fit’. Several ad-hoc measures (correlated residuals, correlated residuals plus cross-factor loadings) substantially improved the fit, but still did not reach commonly accepted thresholds such as .90 or .95. Accordingly, the authors defined a revised five-factor model. The fit of this revised model was still unacceptable according to widely accepted standards. The specification was again modified in order to improve model fit. With these measures, the repeatedly revised models finally yielded ‘mediocre’ to ‘acceptable’ comparative fit indexes. In sum, the confirmatory factor analysis failed to confirm the five personality domains as proposed by the ICD-11 Personality Disorders Committee. The authors preselected personality disorder criteria from the DSM-5, which correspond to their proposed model. Rather than clearly reporting the unacceptable fit of the data with proposed ICD-11 personality disorder domains, the authors undertook several ad-hoc measures to modify the initial model specification ‘to improve model fit’ and concluded that the ‘proposed ICD-11 personality disorder domains were partially validated’. Given the procedure chosen by the authors and given the results of the CFA, this conclusion looks biassed.