Borderline personality disorder traits
We began with a Varimax rotated factor analysis of the 24 items chosen from the 4 DAPP scales of AI, CD, IP, and IA in our entire sample (n = 44,112). As seen in , we were able to recover the four factors quite cleanly. Each item loaded most heavily on the scale to which it was assigned. For nearly all items, the item loaded much more strongly on the scale from which it came than on any of the other three scales.
The Pearson correlations between the four BPD traits in our entire sample (n = 44,112) were as follows: AI-CD +0.47; AI-IP +0.48; AI-IA +0.49; CD-IP +0.61; CD-IA +0.41; and IP-IA +0.35. A factor analysis of these four traits produced a single Eigen value above unity (2.40) accounting for 60.1% of the variance. Loadings on this single factor were as follows: AI +0.79; CD +0.81; IP +0.79; and IA +0.71.
presents, in the MZ and DZ members of our twin sample, the mean and standard deviation of the four BPD traits as well as the observed polychoric correlations in the complete pairs. Our baseline model to which we compare all subsequent models for the twin analyses of these four BPD traits was a common pathway model with additive genetic, common environmental and individual-specific environmental influences, and separate parameter estimates in men and women (model 1; - 2LL 12,694.8, df = 4276). When we constrained estimates to equality across the sexes, the AIC fit index improved substantially (model 2; Δχ2 = 14.7, Δdf = 18, ΔAIC −21.3). In model 3, we set all shared environmental pathways to zero with a large further improvement in fit (Δχ2 = 18.1, Δdf = 23, ΔAIC −27.9). If, in model 4, we instead set all the additive genetic effects to zero, the model fit deteriorated substantially (Δχ2 = 38.3, Δdf = 23, ΔAIC −7.7). We also fitted independent pathway models (details not shown). The best fit independent pathway model also constrained estimates to equality across the sexes and dropped all shared environmental parameters. However, this model did not fit as well as the best-fit common pathway model (Δχ2 = 13.4, Δdf = 16, ΔAIC −18.6).
The mean and standard deviations and polychoric correlations of the four borderline personality disorder traits and five dimensions of the big five inventory in monozygotic and dizygotic (DZ) twin pairs
The results of the best-fitting model 3 are depicted in . The latent liability to BPD traits was highly heritable (a2 = 0.60) with the remaining variance in liability resulting from individual- specific environment (e2 = 0.40). Loadings from the common factor were similar and strong for three of the BPD traits (AI +0.68, CD +0.72 and IP +0.69). The loading for IA was more modest (+0.54).
Fig. 1 Parameter estimates of the best-fit common pathway model for the items selected from four scales of the DAPP: Affective Instability, Cognitive Distortion, Identity Problems, and Insecure Attachment. ‘A’ stands for additive genetic effects (more ...)
From this best-fit model, it was possible to calculate the total heritability (or a2) for each of the four BPD traits and the proportion that resulted from the common factor vs. genetic factors unique to that trait. These results were as follows: AI a2 = 0.42 (67% common /33% unique); CD a2 = 0.51 (63% common /37% unique); IP a2 = 0.53 (55% common /45% unique); and IA a2 = 0.51 (35% common /65% unique).
Borderline personality disorder traits and normative personality disorder dimensions
As seen in controlling for age and sex, in our entire sample (n = 43,799), we predicted the four BPD traits from the ‘big five’ personality dimensions: O, C, E, A, and N. The proportion of variance predicted by this model was as follows: AI – 44.8%, CD – 29.1%, IP – 38.9%, and IA – 20.3%. Interestingly, all four BPD traits were predicted by high levels of N and low levels of C. Indeed, for every trait, N was the strongest single predictor. By contrast, the other big five dimensions had variable relationships with the individual BPD traits. O was positively related to AI and CD, uncorrelated with IP, and negatively related to IA. E was positively related to AI and IA and negatively related to CD and IP. A was positively related to IA and negatively related to CD, IP and, quite weakly, AI.
Prediction of borderline personality disorder traits from the big five normal personality disorder dimensions
We also calculated the correlation between each individual personality dimension and the common factor obtained from the four BPD traits. As seen in , of these four phenotypic correlations, N was by far and away most strongly related to the BPD common factor (+0.63), followed by C (−0.38) and A (−0.29).
presents, in the MZ and DZ members of our twin sample, the mean and standard deviation of the five personality dimensions and the observed polychoric correlation in the complete pairs. In prior analyses, as we saw for the BPD traits, we found no evidence for shared environmental effects with our five personality dimensions (18
). Therefore, to simplify our model fittings to the BPD and OCEAN variables, we examine models containing only additive genetic and individual-specific environmental effects. presents the AIC fit indices for each of these models. For four of the five analyses, the best fit was achieved with model II which constrained parameters to equality between men and women but allowed for genetic and environmental correlations between the personality dimension and both the latent risk to BPD and to each of the four specific BPD traits. Only for C did a simpler model, which contained correlations only between C and the latent risk to BPD, fit better. However, to permit comparisons across models, we present the results for model II for all five personality traits in . Our discussion here will focus on the genetic results, and we will concentrate on genetic correlations ≥ /0.20/.
Akaike’s information criterion (AIC) values for models examining the genetic and environmental correlations between personality dimensions and borderline personality disorder traits
With respect to the pattern of genetic correlations, these five best-fit models can be meaningfully divided into three groups (). Two of the personality dimensions (O and E) had quite modest negative correlations with the latent genetic BPD factor: −0.16 and −0.23 respectively. Interestingly, both had additional positive genetic correlations with AI, while O had a substantial positive correlation with CD and E a negative genetic correlation with IP. Two of the personality dimensions (C and A) had strong negative correlations with the latent genetic BPD factor: −0.66 and −0.57, respectively, with relatively modest additional correlations with specific BPD traits. The pattern with N was unique having not only a moderately strong positive correlation with the latent BPD genetic factor (+0.44) but also rather robust additional positive correlations with AI and especially IP. The direction of the environmental correlation between the personality dimensions and the latent BPD factor tended to mirror those of the genetic correlations. The environmental correlations tended to be lower than the genetic correlations for C, A, and N and higher for O and E.
Genetic and environmental correlations between personality dimensions and the common factor underlying the borderline personality disorder traits and the specific traits themselves derived from the best-fit twin model*