Do We Hold Males and Females to the Same Standard? A Measurement Invariance Study on the Psychopathy Checklist-Revised

Abstract

Psychopathy in females has been understudied. Extant data on gender comparisons using the predominant measure of assessment in clinical practice, the Psychopathy Checklist Revised (PCL-R), points to a potential lack of measurement invariance (MI). If indeed the instrument does not perform equally (well) in both genders, straightforward comparison of psychopathy scores in males and females is unwarranted. Using a sample of female and male forensic patients (N = 110 and N = 147 respectively), we formally tested for MI in a structural equation modeling framework. We found that the PCL-R in its current form does not attain full MI. Four items showed threshold-biases and particularly Factor 2 (the Social Deviance Factor) is gender biased. Based on our findings, it seems reasonable to expect that specific scoring adjustments might go a long way in bringing about more equivalent assessment of psychopathic features in men and women. Only then can we begin to meaningfully compare the genders on the prevalence, structure, and external correlates of psychopathy.

Disclosure statement

We have no conflicts of interests to disclose.

Public significance and data availability statement

This study shows that the most widely used and validated instrument for the assessment of psychopathy, the PCL-R, needs modification to adequately assess psychopathy in females. Adequate assessment is a prerequisite for consistent and informative research on psychopathy in women, its societal consequences and role in judicial systems, as well as the development of effective interventions.

The data that support the findings of this study are available on request from the corresponding author. It is up to the author to determine whether a request is reasonable. The data are not publicly available due to privacy restrictions.

Notes

1 Note that we treat the item scores explicitly as ordinal (which is equivalent to using the graded response model as Bolt et al., 2004, did.). As a result, similarly as in the graded response model, the factor models considered here do not include intercept parameters, but category thresholds (as there are three answer categories, each item contains two thresholds). In addition, the models considered do not contain residual variances. However, given appropriate identification restrictions (see Millsap & Yun-Tein, 2004) both the intercepts and the residual variances can be estimated and tested for invariance. Here however, we fixed the residual variances and intercepts to be equal across groups in all models, and we only freed specific residual variance parameters or intercepts parameters if this is indicated by the modeling results (i.e., poor model fit and/or large modification indices for the residual variances). Note that if the thresholds and factor loadings are shown to be invariant, it can be concluded that the intercepts and residual variances are also invariant.

2 As the second-order factor loadings of first-order Facets 1 and 2 and the second-order factor loadings of Facets 3 and 4 are constrained to be the same, see above, we only had to fix 2 parameters, i.e., the two loadings, and we had to free 2 parameters, i.e., the second-order factor variances

3 As suggested by a reviewer, we verified these results in the Cooke and Michie (2001) second-order Three-Factor model. Similar to our third-order Four-Facet model, we found item 2 to violate threshold invariance in the Cooke and Michie model. The other items that we found to be non-invariant in the third-order Four-Facet model, items 10, 11, and 17, are not part of the Three-Factor model. The results are available upon request.

4 In the Cooke and Michie (2001) second-order Three-Factor model (see previous footnote), Factor 3 was found to be unbiased with respect to the second-order factor. However, note that the second-order General Psychopathy Factor in the Cooke and Michie model is statistically and substantively different from our second-order Social Deviance Factor in the third-order Four-Facet model.