Measuring the Dark Core of Personality in German: Psychometric Properties, Measurement Invariance, Predictive Validity, and Self-Other Agreement

Abstract

The Dark Factor of Personality (D)—the underlying disposition of aversive traits—has been shown to account for various ethically and socially aversive behaviors. Whereas previous findings support the reliability and validity of the original English item sets suggested to measure D, a thorough psychometric examination of their German translation is still pending. Using data from four different samples (total N > 33,000), this study comprehensively evaluates the German version of the D70, D35, and D16 with respect to (a) their factor structure, (b) measurement invariance across gender, (c) measurement equivalence with the original English item sets, (d) predictive validity for relevant outcomes across a six-month period, and (e) self-observer agreement. Results confirm the bifactor structure of the D70 and single-factor models for the D35 and the D16. Measurement invariance testing shows partial strict invariance across gender and language versions. Furthermore, predictive validity and a moderate degree of self-other agreement are supported. The German version of the D70 and its shorter versions thus allow for a psychometrically sound assessment of D.

Acknowledgments

We wish to thank Isabel Thielmann for the permission to use parts of the data collected for the Prosocial Personality Project.

Disclosure statement

We have no conflict of interest to disclose.

Open Scholarship

This article has earned the Center for Open Science badges for Open Data and Open Materials through Open Practices Disclosure. The data and materials are openly accessible at https://osf.io/zwjms/. To obtain the author's disclosure form, please contact the Editor.

Data deposition

Data, items, analysis scripts, and supplemental material for all studies are provided on the Open Science Framework (https://osf.io/zwjms/). For items and a detailed sample description of Study 2 see https://osf.io/m2abp/.

Notes

1 Note that the term "dark" as used in personality psychology—and thus in the naming of the Dark Factor of Personality—is exclusively meant to refer to the light-versus-darkness symbolism used across cultures to metaphorically describe the normative dimension of social/ethical appropriateness. Nevertheless, to avoid any unintended connotations with this term, we will instead rely on the term "aversive" to refer to traits and behaviors that conflict with widely held social and ethical norms (such as fairness, respect, care, and honesty, see Schroeder et al., 2019).

2 These 12 traits included amoralism-crudelia, amoralism-frustralia, egoism, greed, Machiavellianism, moral disengagement, narcissism, psychological entitlement, psychopathy, sadism, self-centeredness, and spitefulness (see Table S1 on the OSF for a summary of the respective measures and number of items per aversive trait).

3 So far, the available language versions are Arabic, Chinese, Danish, Dutch, Estonian, French, German, Greek, Hindi, Italian, Norwegian, Polish, Portuguese (European and Brasilian), Romanian, Russian, Serbian, and Spanish.

4 Comparing the observed D16 scores suggested that the participants choosing to complete the D16 (M = 2.14, SD = 0.67), D35 (M = 2.12, SD = 0.65), or D70 (M = 2.19, SD = 0.79) were highly comparable in their average level in D (Cohen’s ds ≤ .10).

5 For instance, a power-analysis using semPower (Moshagen & Erdfelder, 2016) indicated that a sample size of n = 10,542 yields a power of 99% to reject a model with even minor misspecifications corresponding to RMSEA ≥ .01 on α = .05.

6 Although the utility of the CFI for the evaluation of absolute model fit has been questioned, the CFI remains useful for the evaluation of comparative model fit because here models are compared with respect to the same reference model (Moshagen & Auerswald, 2018).

7 Analogous to the sample-corrected robust RMSEA (Brosseau-Liard et al., 2012) and the robust CFI (Brosseau-Liard & Savalei, 2014), we calculated a robust version for the Mc according to ${Mc}_{\mathit{rob}}=\exp (-0.5·\max (0, \frac{{c_n·(T}_{\mathit{MLM}}- df)}{N})),$ where c_n is the Satorra-Bentler scaling factor, T_MLM is the Satorra-Bentler chi-square statistic, df are the degrees of freedom of the model, and N is the total sample size.

8 As completing additional items might have affected the psychometric properties of the D16 and D35, we also performed the analyses based on the subsets of participants who only completed the items of the respective set. Except for slightly lower reliability estimates for the D scores in the D16, the results were highly comparable (see Table S4 in the OSF additional materials).

9 In Table S5 on the OSF, we provide norms for manifest D scores in the D16, D35, and D70 separated for gender and age groups.

10 Estimating a CAB-factor based on all ten items of the respective scale led to an improper solution because of a negative residual variance of one item (“Have you ever stolen a car?”). This was arguably due to the extremely low agreement rate for this item (only 6 out of 1,853 participants indicated “yes”). We therefore excluded said item from all models.

11 Readers interested in the predictive validity of the five specific factors are referred to Table S10 on the OSF. In brief, although some specific factors yielded incremental predictive validity beyond D, none of the specific factors was consistently linked to all outcomes.

12 For two participants, other-reports were available from multiple raters, which were averaged for the present analyses.

13 Please note, however, that if the samples completing different language versions exhibit different factor variances, the correlation of the latent D factor to external criteria may differ in size.

Additional information

Funding

This research was supported by grants from the German Research Foundation (DFG) to Benjamin E. Hilbig (grant number HI 1600/1-2) and to the Research Training Group “Statistical Modeling in Psychology” (SMiP; grant number GRK 2277).