Authors: Luna Rabl, Valerie Kienhöfer, Morten Moshagen, Karin Labek, Roberto Viviani
Categories: Article, Health care, Psychiatric disorders
Source: Scientific Reports
Unlike the cognitions associated with depressive symptoms, little is known about those associated with antisocial personality and with its related traits (“dark traits”). Using the scrambled sentences task, an instrument from depression research, we investigated cognitions such as justifications (external blaming for one’s behavior) and harm to others (based on the notion that some of these individuals enjoy harming or humiliating others) that we hypothesized may be prevalent in those high in antisocial personality traits. Confirming our hypothesis, these cognitions were associated with ratings on different antisocial personality scales and with antisocial and detachment scores in the alternative model of personality disorders of the DSM-5 (AMPD) in three non-clinical samples, but not with depressive symptoms or neuroticism. Cognitions including harm to others were differentially associated with high sadism. These findings empirically characterize classes of cognitions that are shared by individuals with antisocial tendencies, and that differ from those associated with depressive symptoms.
**Subject ** Health care, Psychiatric disorders
Antisocial personality is characterized by maximization of one’s own utility disregarding the consequences for others, or even by placing positive value on damage or harm to others^1,2^. Individual degrees in the tendency to enact antisocial behaviour (and display low levels of social value orientation)^3^constitute manifestations of dark personality^2,4^. A multitude of traits exist within the dark personality, but factor-analytic studies have identified a common core tendency, the dark factor of personality (D), on which the dark personality traits are based, but also add unique content to it^1,2^. Four of these traits have been most prominently studied and are known as the dark tetrad (psychopathy, Machiavellianism, narcissism and sadism)^5,6^.
Little is known about the cognitions about the self and the world that accompany antisocial personality traits^7^. There is considerable evidence that antisocial personalities entertain justificatory cognitions that rationalize their behaviour^2,8^. This suggests that a personality construct such as D integrates personality traits with specific cognitions^3^, but no empirical approach to document these cognitions has been undertaken.
In the clinical literature, cognitions have been extensively investigated in the context of affective disorders, starting from depression^9,10^. Negative cognitions in depression involve a negative view of oneself, the world and the future (negative cognitive triad)^9,11,12^. Further research in this area led to the development of an empirical task to assess the individual tendency to activate negative cognitions, the scrambled sentences task (SST)^13^. Each item of the SST^13,14^ consists of a few words in random order that may be used to form a grammatically correct sentence. The words and the formed sentence give information on the tendency to activate specific cognitions. For example, with a set such as “is the bright future dismal”, the frequency with which the sentence “the future is dismal” rather than “the future is bright” is selected is higher in depressed individuals and in at-risk participants (such as previously ill individuals)^10,13,15,16^. Healthy participants, in contrast, tend to activate positive cognitions^15,17,18^.
In the present work, we used the SST methodology to investigate the existence of specific cognitions associated with scores in dark personality scales, viewed as proxies of D. In this SST, negative alternative sentences represented antisocial schemas, and the positive alternatives represented prosocial schemas. For example, in the sentence Tim wants others to be hurt/supported the word hurt represents an antisocial schema and the word supported represents a prosocial schema. We tested the association between rates of antisocial cognitions, as elicited by this version of the SST, with scores in dark personality scales. Our approach, implemented in a series of three experiments, was to vary the modalities of sample recruitment and the instrument for the assessment of dark personality between our experiments to verify generalizability of previous results (‘external validation’)^19,20^.
We also verified the discriminatory validity of the SST for antisocial cognitions relative to negative cognitions associated with depressive symptoms and neuroticism^17,21^ using the “shattered assumptions” SST. The SST for shattered assumption was developed to assess cognitions that normally occur in post-traumatic conditions and reflect a negative view of the world and the self^21,22^. These cognitions have been shown in previous research to be strongly associated with depressive symptoms and especially with neuroticism^21^ in the five-factor model of personality. In using the shattered assumptions SST, our intent was to adopt an instrument that was well suited to assessing depressive cognitions related to personality traits and that captured beliefs about an unsupportive environment to provide discriminatory validity.
This SST contained cognitions in two justificatory cognitions (such as the belief that people think about themselves first, as documented by previous research)^1,2,8^, and cognitions representing individuals that are physically harmed, exploited, or humiliated. This second set of cognitions is motivated by the low value of the welfare of others in this personality trait^1^ as well as by the attraction that these situations exert on sadistic individuals^1,5^. Data from the SST for depressive symptoms suggest a role of motivational factors in the individual propensity to select optimistic cognitions^17,18,23^. We hypothesized that scores in sadism may be specifically associated with cognitions in the harm domain, as these individuals may be motivated by representations where others are physically harmed or humiliated.
We also investigated the propensity for antisocial cognitions assessed by the SST within the trait-domains model of the ‘alternative model of personality disorders criterion B’ (AMPD) of the DSM-5^24^, a dimensional approach to psychopathology. We hypothesized that antisocial cognitions would be associated with the AMPD trait-domain antagonism (which represents antisocial traits included in D)^25^ and not with negative affectivity (which is closely related to neuroticism from the five-factor model of personality)^26^.
Three experiments were conducted. The first experiment had the primary aim to explore the plausibility of a specific association between recruitment of antisocial cognitions and scores in antisocial personality, as assessed by common rating scales (SD3 and ASP) ^27,28^. To assess individual differences in behaviour, we administered the social value orientation scale (SVO)^29^ and let participants play a simulated dictator game, a formalized economic interaction that captures antisocial behaviour^1^. In the second experiment, we wanted to replicate the key association of antisocial cognitions with dark personality scores of the first experiment and further characterize individuals giving high antisocial SST scores in the AMPD dimensional system. In the third and final experiment we administered the antisocial cognitions SST and D, a scale recently developed to capture the full scope of dark personality^1,2^, and the five ‘themes’ that characterize its variants^30^ to a large community sample. In all experiments, we also analysed how cognitions related to harm to others and justifications for antisocial behaviour were associated with variants of the dark personality.
Positive sentences (i.e., prosocial sentences/sentences without sadistic content in the SST for dark personality, and sentences with optimistic content in the SST for shattered assumptions) were modelled in a repeated measurements logistic regression with subjects and sentences as random factors. Hence in what follows, a negative coefficient indicates a lower rate of positive sentences. For example, a negative coefficient of antisocial personality scores means that individuals with high scores were making more antisocial sentences. In all models, age and gender were controlled for as covariates. Further information on the scale scores, including their distributional properties, can be found in the Supplementary Material.
Participants in the first experiment (N = 55, average age 30.7, 19 females) were recruited online. Efforts were made to avoid recruitment among students (see Supplementary Methods for details on the sample). Correlations between the covariates of the first experiment are provided for descriptive purposes in the Supplementary Material (Table S1). Dark personality subscales were correlated with each other, as did the behavioural data. Only narcissism showed a significant correlation with the dictator game (more money kept for oneself)^29,31–33^ and the SVO (less social value orientation)^29^.
As in the depression^13^ or shattered-assumption^21^ versions of the SST participants were preferentially forming positive sentences (antisocial sentences, z = 7.23, p < 0.001; shattered assumptions sentences, z = 5.79, p < 0.001). When adding age and sex to the model, we found no significant associations with antisocial sentences (z = 0.40, p = 0.691 and z = 0.61, p = 0.543).
We then proceeded to test one of the main hypotheses of this experiment, namely the association between scores in the antisocial sentences and the scores on the dark personality scales. To verify that this association was specific, we included in the same model the predictors dark personality^27,28^ and neuroticism^34^ simultaneously, after adjusting also for age and gender. There was a significant association between rates of antisocial sentences and dark personality total scores (z = −6.07, p < 0.001), while the association with neuroticism was not significant (z = 0.16, p = 0.874), as expected (Fig. 1A and Table S2 in the Supplementary material).
Figure 1 (A,B) Prediction of sentence formation by dark personality and neuroticism scores in the antisocial and in the shattered assumptions SST. The black line represents the fitted rate of positive sentence formation (prosocial in panel A on the left, optimistic in panel B on the right) for individual differences in dark personality (standardized scores). The yellow line the fitted rate for individual differences in neuroticism (standardized scores). The individual points reproduce the density of positive or negative sentence formation for these two scales. In the insets, odds ratios for prosocial sentences for the standardized scores. (C) Odds ratios for prosocial sentences and 95% confidence intervals in a model with all dark personality subscales were fitted simultaneously. One can see that they provide similar additional effects on sentence formation, after reciprocal adjustment. (D) Odd ratios for prosocial sentences and 95% confidence intervals in models of the interaction of dark personality subscales and ‘harm’ content of sentences. The hypothesis was that this interaction would be significant in the interaction with sadism.
According to this model, one standard deviation in dark personality scores resulted in an increase in the expected frequency of antisocial sentences from 14% (in individuals with average scores) to 25%. In contrast, the SST for shattered assumptions showed that the formation of a positive or negative sentence was primarily associated with neuroticism scores (z = −4.08, p < 0.001), while dark personality score gave an association only at trend level (z = −1.94, p = 0.052; Fig. 1B and Table S2). Here, one standard deviation in neuroticism scores resulted in an expected increase in pessimistic sentences from 14% (in individuals with average neuroticism) to 29%. These different association profiles could be confirmed by significant there were more negative sentences in the antisocial than in the shattered assumptions SST in individuals with high dark personality scores (z = 2.66, p = 0.008), while the opposite was the case in individuals with high neuroticism scores (z = –7.07, p < 0.001). In summary, scores in rating scales were primarily associated with negative sentences in the corresponding content type.
The dark personality scores summarize an overall trait that may be related to the common core tendency D identified in previous studies^1,2^. However, the dark personality scales^27,28^ also contain four psychopathy, narcissism, Machiavellianism, and sadism, whose scores were strongly associated in our data (Table S1 in Supplementary Material), consistently with the existence of a common core tendency. Hence, taken individually, they were all similarly associated with the rate of antisocial cognitions elicited by the SST. Also in a model where all these subscores were simultaneously included to estimate variance uniquely explained by each, they gave similar coefficients (Fig. 1C).
The antisocial sentences belonged in two justifications and harm sentences. There was no interaction between the effect of dark personality scores on the rate of antisocial sentences and these two types of sentences (z = −1.42, p = 0.156). However, this interaction was significant in the sadism subscale (z = −2.30, p = 0.022; Fig. 1D). People with higher sadism scores selected more often the negative target (the word representing the antisocial schema in the SST for antisocial cognitions) in the harm sentences.
We then analysed the correlation between the SST and the results of the behavioural tasks (the association of dark personality scales with behaviour is reported in Table S1 in Supplementary Material). In the SST for antisocial cognitions, there was an association with SVO scores (z = 2.40, p = 0.017). Individuals with higher SVO scores, indicating higher social-value orientation, formed more prosocial sentences. In the shattered assumptions SST, in contrast, the association with SVO scores was not significant (z = 1.39, p = 0.166). In the dictator game there was no association between the money allocated to oneself and rates of prosocial sentences (money kept for oneself and antisocial sentences, z = −0.98, p = 0.330).
In this experiment, participants were given no time limit to complete the sentences. Previous studies with the SST for depressive cognitions found no interaction between response times and individual differences in depressive symptoms^23^. When modelling the effect of response time on sentence formation, we found no significant effect nor any significant interaction with personality traits (dark personality in antisocial z = −0.63, p = 0.528; neuroticism in shattered assumptions z = 0.86, p = 0.393). Therefore, response time had no effect on the tendency to produce sentences that were associated with the respective scales in either SST.
The second experiment (N = 98, 65 females, mean age 23.8) had two aims. One was to replicate the association between antisocial sentences and dark personality scales of the first experiment, using a different scale to probe dark personality (SD4)^35,36^. Personality was assessed in the AMPD (personality inventory for DSM (PID) trait-dimensions)^37,38^ expecting, as a generalized replication of the first experiment, to find an association with antisocial but not with negative affectivity trait-domains. We also assessed depressive symptoms with the CES-D scale^39^. In contrast to the first experiments, the sample was predominantly comprised of students rewarded with study credits. The second aim was to characterize individuals with high rates of antisocial cognitions, as elicited by the SST, in the dimensional characterization of personality of the AMPD across all personality domains. Correlations between the covariates of the second experiment are shown in Table S3 in the Supplementary Material.
In the SST data, we found an overall tendency to form positive sentences (z = 8.39, p < 0.001) as in the first experiment. Females had a 9% higher probability of choosing the positive target (the word that represents the more prosocial schema in SST for antisocial cognitions) compared to males (z = 3.15, p = 0.002).
As in the first experiment, total SD4 scores were significant predictors of formation of antisocial sentences in the SST (z = −2.39, p = 0.017). In contrast, depressive symptoms levels (z = –0.94, p = 0.349) or negative affectivity (z = −0.64, p = 0.521) were not associated with sentence formation. These associations are illustrated in (Fig. 2A) and Table S4 in the Supplementary Material.
Figure 2 (A) Logistic regression of prosocial sentence formation on dark personality and neuroticism scores. The fitted rate of positive sentence formation for individual differences in darkP (assessed with the SD4 scale) and antagonism (standardized scores) are shown by the black and the gray lines, respectively. The yellow and red lines show the fitted rates for individual differences in negative affectivity and depressive levels (CES-D, standardized scores). The individual points reproduce the density of positive or negative sentence formation for these two scales. In the insets, odds ratios for prosocial sentences for the standardized scores. (B) Odds ratios of prosocial sentence selection and 95% confidence intervals in model where all SD4 subscales were fitted simultaneously. One can see that in this dataset sadism is the only subscale that explains additional variance in the reduced selection of antisocial sentences. (C) Odd ratios for prosocial sentences and 95% confidence intervals in models of the interaction of SD4 subscales and ‘harm’ content of sentences. In this dataset, harm sentences were more effective in eliciting an association with any subscale. The hypothesis was that this interaction would be significant in the interaction with sadism.
Figure 2A shows that in this dataset there was one individual with very large standardized SD4 scores, and one with very low CES-D scores. Repeating the analysis without these individuals did not change the effect of SD4 (z = −2.22, p = 0.027) and slightly increased the effect of depressive levels (z = −1.66, p = 0.097).
Again, the SST sentences were divided into two subsets, those with justificatory and those representing harming others. As in the first experiment, the interaction between sadism and harm sentences could be replicated here (z = −2.98, p = 0.003; Fig. 2C and Table S4). Participants higher in sadism more often chose the negative target in the harm subset of sentences than in the justifications set.
When characterizing individuals with high antisocial cognition SST scores in the dimensional space of the AMPD (PID trait-dimensions), we expected them to be associated with the trait-dimension of antagonism. Because the PID trait-dimensions tend to be positively associated with each other, and we were interested in the specific contribution of these trait-dimensions, we included all six of them in the model simultaneously. Antagonism was found to predict sentence formation as expected (z = −2.05, p = 0.040). However, also detachment scores were associated with a higher probability of choosing antisocial sentences (z = −3.76, p < 0.001; Fig. 3A; see also Table S4 in the Supplementary Material for the corresponding odds ratios). The effectiveness of detachment in predicting antisocial sentences was not changed by replacing antagonism with the SD4 total score in the model. The effect of detachment remained substantially unaltered after adjusting for depression scores (z = −3.68, p < 0.001) while depressive symptom levels themselves showed no predictive effect (z = 0.18, p = 0.856).
Figure 3 (A) odds ratios for prosocial sentences in model with all PID subscales fitted simultaneously (95% confidence intervals). Only antagonism and detachment were associated with increased rates of antisocial sentences. (B) color-coded heatmap of fitted ratio of prosocial sentences in the subspace spanned by antisocial personality (given by averaged SD4 and antagonism scores on the x axis) and detachment (on the y axis). The map is a smooth 2-dimensional estimate of the rate of prosocial sentences in individuals (shown by individual points in the map). The majority of individuals fell within two standardized scores of antisociality and detachment. Hence, heatmap estimates are most reliable within this range (the ridge at high levels of detachment is caused by one individual) and show a gradient of increasing antisocial sentences from yellow to green-blue.
To visualize the distribution of antisocial sentences in the subspace spanned by detachment and antisocial traits, we fitted splines to the positive sentence formation in a semiparametric logistic regression^40^ adjusting for age and sex and estimating random effects for subjects and sentences as in the main model (see Methods for details). Antisocial tendencies were estimated by averaging the standardized SD4 and PID antagonism scores. The splines fit estimates the probability to select a positive cognition (in blue to yellow colors) as a function of antisocial tendencies and detachment considered jointly (Fig. 3B) giving a map of expected cognitions in the space spanned by these two predictors, using the smoothing properties of splines to interpolate between observations (because extreme scores were scarce, an appropriate use of this map should be limited to the area around the middle ground, where most individuals are located). The colors show gradients of increasing antisocial sentence rates in either directions, high detachment or antisocial traits, which were alone capable of producing lower positive sentences rates. The Figure also shows that in our sample there were no individuals with high antisocial and low detachment scores.
In the third experiment (N = 960, 549 females, mean age 39.52), the core disposition of dark personality was measured by the D scale^1,2^. This experiment was conducted in a much larger community sample than the previous experiments (see Supplementary Methods from details). Besides providing a validation of the antisocial SST with a different instrument, we hoped to obtain evidence on different effects of sentence types (harm and justification) on personality facets and sex that may have suffered from the relatively low power of the previous experiments.
D displays an internal structure of five ‘themes’ (callousness, deceitfulness, narcissistic entitlement, sadism, and vindictiveness) which were derived from the D scale using factor analysis and represent specific characterizations of dark traits beyond the general factor D^30^. Correlations between the covariates of this experiment are shown in Table S5 in the Supplementary Material. Overall, males and lower age showed a significant association with a higher score on D and the five factors/themes of D.
Also in this study, participants showed an overall tendency to choose the positive sentence (z = 11.51, p < 0.001). At average D scores, females were 3% more likely to choose the positive target than males (z = −3.38, p < 0.001), consistently with their lower D scores. Age was also a significant predictor of prosocial sentence selection (z = 4.85, p < 0.001), similarly again to the association of this variable with D scores. At average D scores, the probability of choosing the positive target increased from 84 to 86% when age increased by one standard deviation.
We then tested the key hypothesis of the current study, namely the association between target choice within the SST and D scores. As in previous experiments, a higher score in D was associated with a lower probability to choose the positive target (z = −21.17, p < 0.001). At average D scores, one standard deviation of higher D increased the probability of choosing an antisocial sentence from 16 to 25%. The fit of this model is shown in (Fig. 4A) (continuous black line). The decreasing occurrence of sentence choices on the right of the figure (visible by the density of points representing positive and negative sentences) reflects the fact that, while most participants had a low to medium high D score, there were a few individuals with very high D scores (with a maximum of 4.66). This distribution likely reflects the fact that, in this much larger study, we had the chance to sample D levels that occur infrequently in the community and were not present in the previous experiments. Accordingly, at the high end of D scores range the fitted rate of prosocial sentences reached much lower levels here, as shown in the Figure. At this high end, there were individuals that choose prosocial sentences with a frequency of 30% or less, completely reversing the tendency to form prosocial sentences in the population (this reversal is what we observe also in high depressive symptom levels, which are much more common than high D^23^, and the rates of negative shattered assumptions sentences in high neuroticism individuals of Fig. 1B).
Figure 4 (A) Logistic regression of sentence choice on D personality scores. The continuous line shows the fitted probability to select a prosocial sentence in all sentence types combined; the short and long dashed lines for the harm and justification sentence types separately. The individual points reproduce the density of prosocial or antisocial sentence selection. (B) Odds ratios of prosocial sentences and 95% confidence intervals in model where all D themes were fitted simultaneously. One can see that in this dataset most themes were explaining additional variance in sentence choice, suggesting specificity of content. (C) Odd ratios for prosocial sentences and 95% confidence intervals in models of the interaction of D themes and ‘harm’ content of sentences. In this dataset, harm sentences were more effective in eliciting an association with the sadism theme. In addition, callousness and vindictiveness showed an interaction in the opposite direction, due to them being more effective in eliciting individual difference in prosocial sentence selection in the justification sentences.
It should also be noted, however, that these few high-D participants raise the danger of single observations unduly affecting the fit. We therefore conducted a sensitivity analysis by log-transforming the D scores (to reduce the influence of the skew of the D scores) and excluding all participants with a log score of 1.1 or higher (corresponding to the exclusion of standardized scores of 3 or higher in the Figure). This analysis confirmed that a higher D score significantly decreased the probability to choose the positive target (z = −14.82, p < 0.001).
Both harm and justification sentences were effective in eliciting individual differences in D (effect of D in harm z = −15.76, p < 0.001; in justification z = −20.64, p < 0.001; Fig. 4A, dashed black lines). However, justification sentences were more effective, with a significant interaction sentence type × D (z = 2.73, p = 0.006). As we show below, the effectiveness of the justification sentences differed depending on the personality themes.
Figure 4B and Table S6 in the Supplementary Material shows the effects on sentence selection of all five themes considered simultaneously in the same model. Callousness (z = −4.36, p < 0.001), deceitfulness (z = −4.17, p < 0.001) and sadism (z = −3.04, p = 0.002) remained significant predictors of the negative sentence selection after controlling for the other themes. The interaction with sentence type (harm or justification sentences), however, demonstrated the existence of differences among personality themes in eliciting of negative sentence selection depending on sentence types (Fig. 4C and Table S6). As in the previous experiments, the interaction with sadism was significantly negative (z = −4.05, p < 0.001), showing that individuals higher in sadism were also more likely to choose the negative target in harm sentences than in justification sentences. In contrast, the interactions between sentence type and callousness (z = 2.77, p = 0.006) as well as vindictiveness (z = 3.55, p < 0.001) showed that it was justification sentences that were best predicted from the scores of these two themes. In the justification sentences one standard deviation in higher callousness and deceitfulness scores took the expected rate of negative sentences from 17 to 22% and 18 to 21%, respectively. In the harm sentences, by contrast, the increase was much lower, from 19 to 21% for callousness, and was stable at 20% for deceitfulness. Hence, it appears that the inclusion of the callousness and vindictiveness themes was responsible for the increased effectiveness of justification sentences in eliciting individual differences with D as a whole.
Numerous studies have shown that the SST may be used to provide evidence of the depressive disposition of individuals and their cognition patterns^41^. The aim of the present work was to use the SST to demonstrate the existence of a pattern of cognitions associated with high dark personality scores. In a clinical context, cognitions have been investigated with rating scales such as personality beliefs questionnaire^42^. However, there have been few studies on antisocial personality with this instrument^7,43,44^.
We showed that the rate of negative sentences elicited by this SST version was associated in a non-clinical sample with ratings of antisocial or psychopathic personality (or inversely associated with social value orientation scores) using a variety of instruments. Furthermore, there were little or no associations between these rates and measures of depressive levels, neuroticism, or negative affectivity, which are reliably associated to the SST for depressive disposition^21,23^.
When investigated within the framework of a dimensional classification such as the AMPD, individuals producing a high rate of antisocial cognitions were located in the subspace defined by the trait-domains antagonism and detachment. In contrast, individuals who produce a high rate of negative sentences in the SST for depression have high scores in negative affectivity and detachment^23^. Detachment is known to be associated with depressiveness^45,46^, but the involvement of detachment in our high-scoring individuals could not be explained by residual depressiveness, as it remained virtually unchanged after adjusting for current depressive symptoms levels and negative affectivity. This finding raises the issue of why detachment appears to be associated not only with depressiveness, but also with antisocial cognitions. One possibility is that some of the cognitions that were associated with high dark personality scores may also imply a view of social interactions as unsupportive or frankly hostile, a view that may be consonant with the detachment trait. A related finding is that antisocial cognitions are specifically associated with low empathic concern in personality functioning scales^47^, as one would expect in relationships characterized by hostility. These findings suggest that schemas associated with high levels of detachment are worth exploring further in a clinical context.
The nature of the processes active in the SST and responsible for eliciting these patterns of cognitions is still being debated^23^. The finding that these patterns are the same irrespective of response times raises the possibility that controlled cognitive processes, which require more resources, may be recruited in sentence selection in the SST. However, neuroimaging findings show the involvement of neural substrates that differ from those associated with effortful control, showing instead recruitment of the substrates of preference-based choice^17,18^. Preference-based choice is deliberate but consistent, and it has been shown to be based on sophisticated mechanisms of elaboration of information on the available options^48–52^ and on subjective motivational factors^53^. Mechanisms with the same properties may also be active in schema selection, explaining both the selection of cognitions that are conformant to schemas as well as motivational influences. According to this model of schema selection, negative views of social interactions may be associated with schematic tendencies present in different groups of individuals, but personalities traits related to sadism may form a specific core based on a distinct motivational mechanism.
A limitation of the present study is that it was conducted in subclinical samples. For this reason, the conclusions may not be applicable to clinical psychopathy.
All data collection took place online through web interfaces. Participant were given information about the study in introductory web-pages, had to provide an age of 18 or older, and gave explicit informed consent before being admitted to the study. The consent form specified the right of participants to withdraw from the experiment at any time; no data were retained for these participants. All studies were approved by the Ethical Review Board of the Institute of Psychology of the University of Innsbruck and were performed in accordance with relevant guidelines and regulations. For more information on the sample and study design, the rating scales and the behavioural task, see the Supplementary Materials and Methods.
To assess antisocial cognitions, new items were developed following the pattern of the SST for depressive schemas^13^. The creation of the items was based on existing scales^27,28^ and the existing literature on characteristics and common features of the individual dark personality traits^2,54,55^. We chose this approach to emphasise the strong intercorrelation between the different dark personality traits, but also the assumption that these are different manifestations of a common dark core^1^. We therefore included as common core features D and its associated factors (e.g. dominance, power, lack of perspective taking), but also previously reported traits such as low agreeableness^6^, exploitation^56^, aggression, emotional coldness and dishonesty^6^. In addition to these traits, specific characteristics from individual traits (such as the cynism and the ability to manipulate associated with Machiavellianism, the impulsivity associated with psychopathy, and the grandiosity and vulnerability associated with narcissism)^27,55,57–60^ and items from already existing scales for measuring dark personality (e.g. SD3, SSIS and ASP)^27,28,61^ were included. A total of 44 sentences could be derived from the gathered content. These sentences were divided into two categories based on D^2^: Harm and Justifications. The former encompasses one’s own maximization of utility and the potential or intended harm towards others while doing so (example: little Tim gets a spanking/praise). The second category, on the other hand, includes the accompanying justifications for utility maximization at the expense of others (example: harming others is unavoidable/prohibited). Each target (for detailed explanations see below) in the SST for antisocial cognitions was matched for length and frequency in the German language^62^ and refers to a specific word in the sentence (the anchor)^17^. To prevent comprehension difficulties in the implementation of the SST, exercise tasks were carried out before the actual start of the task.
In experiment 2, the same antisocial cognitions SST was used, except that a time limit of 7.5 s for each SST sentence was set. If the participant didn’t give a response within this limited time, the response was coded as missing.
The task was originally administered verbally and in paper-pencil format^13–15^ and later converted to a computerized form^17^, which was also the form adopted in the present experiment. In each item, it is possible to form two sentences that differ only by the last word of the sentence (the target)^17^. Thus, only five words are needed per correct sentence, one of which is the target. The two possible targets per item conventionally represent a positive/optimistic schema and a negative/pessimistic schema^13,17^.
To test the divergent validity of the newly developed SST for antisocial cognitions and whether the SST can be conducted online, the already validated SST for shattered assumptions^21^ was used. The concept of “shattered assumptions” emerged in the research on cognitions in post-traumatic conditions^22^ and refers to the post-traumatic emergence of negative cognitions representing the world as unreliable and unsupportive, in lieu of the generically pessimistic cognitions of depression. It is important to underscore, however, that past studies suggest that these cognitions, as assessed by the SST, as well the scores of rating scales for shattered assumptions, are associated to the intensity of depressive symptoms in clinically non-depressive individuals^21^. Hence, the shattered assumptions SST may also be considered an assessment of depressive cognitions. We selected the 14 sentences that had worked best in previous studies^21^ and conducted them online to check if previously found results of the SST can be replicated (example: I feel alive/dead inside)^21^.
Statistical analyses were conducted with RStudio (R version 4.2.0). Mixed-effects logistic regression was used (lme4 package, version 1.1–29), controlling for age and sex and modelling subjects and the SST sentences as random effects to account for repeated measurements. This package has stringent checks for convergence of the procedure used to compute the fit, which were passed by all models in the study. Formation of a positive sentence (prosocial or optimistic in the shattered assumptions SST) was coded as ‘success’. A negative coefficient in this context therefore means that the higher the predictor, the lower the probability of forming the positive sentence. Any selected word found to be neither a positive nor negative target word was coded as an error and not included in the analysis.
Plots were created with the package ggplot2, version 3.3.6^63^. The plot of Fig. 3 was created prior to fitting a semiparametric model to the SST data of Experiment 2 with the function brm of the brms package^64^. Formation of a positive sentence was modelled in a mixed-effects logistic regression with subjects and SST sentences as random effects, thin-plate splines to model the surface with coordinates detachment and the mean values of PID antagonism and SD4 (after standardization), and sex as covariate. The package brm implements a Bayesian approach, which is here to model the coefficients of the splines as a random effect, estimating the degree of smoothing from the data through the estimated variance parameter of these coefficients^40^. The fitted 2-D surface of estimated rates of formation of positive sentences was visualized with the conditional_smooths function of the same package and refined with ggplot2 for labels.
In Experiment 1, trials were excluded if participants stated that they did not answer the questions honestly (one participant) or if they stated that they did not conduct the SST correctly (two participants). In Experiment 2, trials were excluded where participants did not finish the survey (37 participants) and participants who did not select the correct target (a word that at the end of the sentence gave a grammatically correct construct) or failed to respond within the 7.5 s time limit more than twelve times (11 participants). In Experiment 3 we were able to recruit 1453 participants. Participants were excluded who took less than 2 s on average to complete the SST (24 participants) if they made more than 12 error or misses (469), resulting in a total sample of 960 participants.
All studies were approved by the Ethical Review Board of the Institute of Psychology of the University of Innsbruck.
This manuscript has been loaded onto a preprint server (https://psyarxiv.com/gz7hp/).
This study was supported by Tiroler Wissenschaftsförderung (TWF, Project Number F.33312/6-2021).
Additional information The sample and study design were registered before data collection. The registration of experiment 2 was made public (DOI: 10.17605/OSF.IO/YFDQ4), the registration of experiment 1 remained private and is available from the corresponding author. We report in this article how we obtained our sample size and all reasons for excluding data, all statistical methods and software used. The experiments were conducted following guidelines for online experiments^65^.
Conceptualization: L. Rabl, R. Viviani, M. Moshagen; Methodology: L. Rabl, R. Viviani, M. Moshagen, V. Kienhöfer; Software: L. Rabl, R. Viviani, M. Moshagen; Formal Analysis: L. Rabl, R. Viviani, K. Labek; Investigation: L. Rabl, R. Viviani, M. Moshagen; Resources: L. Rabl, R. Viviani, K. Labek, M. Moshagen; Data Curation: L. Rabl, R. Viviani, M. Moshagen; Validation: L. Rabl, R. Viviani, K. Labek, M. Moshagen; Writing—Original Draft Preparation: L. Rabl, R. Viviani, K. Labek; Writing—Review & Editing: L. Rabl, R. Viviani, K. Labek, M. Moshagen, V. Kienhöfer; Visualization: L. Rabl, R. Viviani; Supervision: R. Viviani, K. Labek, M. Moshagen; Project Administration: L. Rabl, R. Viviani; Funding Acquisition: L. Rabl.
The datasets generated and/or analysed during the current study are available on request from the corresponding author (luna.rabl@uibk.ac.at) after verifying that further analyses are compatible with the aims stated in the consent form signed by participants. The SST software used is also available from the corresponding author (luna.rabl@uibk.ac.at). The German items of the SST for antisocial cognitions can be downloaded from OSF (DOI https://doi.org/10.17605/OSF.IO/YFDQ4).
The authors declare no competing interests.
The online version contains supplementary material available at 10.1038/s41598-024-69473-6.
The datasets generated and/or analysed during the current study are available on request from the corresponding author (luna.rabl@uibk.ac.at) after verifying that further analyses are compatible with the aims stated in the consent form signed by participants. The SST software used is also available from the corresponding author (luna.rabl@uibk.ac.at). The German items of the SST for antisocial cognitions can be downloaded from OSF (DOI https://doi.org/10.17605/OSF.IO/YFDQ4).