Introduction
The study of externalizing psychopathology dates back to the pioneering work of Thomas Achenbach [
2]. Achenbach, who conducted factor analytic studies in child and adolescent psychiatric patients, found support for two major separate factors of symptoms: an internalizing factor, characterized by symptoms of depressive and anxious character and an externalizing factor, characterized by maladaptive and disruptive behaviors and symptoms directed from the individual towards the surrounding environment (e.g., physical and verbal aggression, disobedience, rule breaking and deceitfulness). This classification has, primarily during the last two decades, increasingly been extended into the realm of adult psychopathology [
3,
4]. It has gained empirical support in the form of studies documenting the systematic comorbidity of psychiatric diagnoses capturing problems along the externalizing spectrum (e.g., deviant and rule-breaking behaviors, inattention, hyperactivity, impulsivity, alcohol and illicit substance abuse, see e.g., [
5‐
7]). Indeed, categorical psychiatric diagnoses incorporating these symptoms and behaviors, such as conduct disorder (CD), attention deficit hyperactivity disorder (ADHD), antisocial personality disorder (ASPD), and substance use disorders (SUDs), have been found to co-occur within individuals well above what would have been expected had they been unrelated [
8‐
10]. These different manifestations of the externalizing spectrum are suggested to stem from a common, latent vulnerability that links the co-occurrence of externalizing behaviors to a core of impulse control problems, sometimes described as disinhibition or trait impulsivity [
11‐
13]. The way this latent, highly heritable [
14,
15], vulnerability is manifested is shaped and influenced by developmental and psychosocial processes, such as experiences of childhood maltreatment and peer group socialization processes in adolescence [
11,
16‐
18].
The systematic co-occurrence of externalizing (as well as internalizing) problems and behaviors, along with perceived limitations of the DSM and ICD approach to classification [
19], has led to the development of novel frameworks of classifying psychopathology. These criticism and perceived limitations of the DSM and ICD system have been extensively detailed elsewhere [
20‐
23] and includes the use of categorical diagnoses in a world of dimensional psychopathology, concerns about the reification of diagnostic entities, and the purported failure of these systems to get at the roots of etiological processes of mental disorders, sacrificing validity for reliability. A recent example of an alternative classification framework is the Hierarchical Taxonomy of Psychopathology (HiTOP) [
19]. The HiTOP is a large-scale research project seeking to provide a dimensional alternative to the conventional categorical, top-down psychiatric nosology, established on the basis of expert clinician consensus, and as exemplified by the DSM-5 and ICD-10 [
24,
25]. In the HiTOP model, the externalizing spectrum is situated second from the top in a hierarchical model, alongside the internalizing spectrum, but below an overarching level of General Psychopathology (see Fig.
1 in Conway et al., [
25]).
In forensic psychiatric settings, externalizing problems represent an important treatment target above and beyond the effects of severe mental illnesses. While there are several reliable and well validated measures of major mental disorders such as psychosis, depression and bipolar disorder, the same cannot yet be said for instruments aimed at the comprehensive and dimensional assessment of externalizing problems, although several instruments for specific, narrower, aspects of the externalizing spectrum already exist. From a clinical perspective, externalizing problems and criminogenic needs (e.g., antisocial personality pattern, antisocial associates, substance abuse) may persist well after any severe mental illness has subsided or been successfully treated [
26]. Recent research suggests that externalizing problems and criminogenic needs are in fact the primary risk factors for recidivism among persons with severe mental illness, factors that they share with offenders without severe mental illnesses [
27‐
29]. Being able to accurately assess and treat externalizing problems and criminogenic needs, therefore, is crucial, and could aid in further reducing forensic patients’ length of stay and risk for recidivism.
Despite decades of research devoted to delineating the externalizing spectrum, and despite the emergence of several novel frameworks, there are yet few available instruments that capture the associated traits and behaviors in adult populations in a coherent, unified manner. In fact, the Externalizing Spectrum Inventory (ESI; [
30]), is, to the best of our knowledge, one of the first instrument developed with the sole purpose of assessing different manifestations of externalizing behaviors, although assessment of externalizing psychopathology has been included as a part of broader instruments (e.g., Achenbach, & Rescorla, 2003 [
31]). The ESI offers a comprehensive, dimensional self-report assessment, containing 415 items parsed into 23 fine-grained facets. Its length, however, has proved challenging in practice, resulting in the use of shorter, semi-official versions [
32,
33]. To address this, an official, 160-item version was developed, called the Externalizing Spectrum Inventory-Brief Form (ESI-BF) [
1]. The ESI-BF aims to provide a more efficient assessment of the externalizing spectrum while retaining the same structure as the original full ESI, with very high correlations at the facet scale level to the full form ESI (
rs 0.89 –0.98) and high internal consistencies (α > 0.85) across all but one facet scale [
1]. An additional goal in the development of the ESI-BF was the creation of three short, item-based subscales indexing different manifestations of the externalizing spectrum. Patrick et al. [
1] chose to create these item-based subscales in order to aid research on the externalizing spectrum by allowing researchers to eschew the use of the, sometimes prohibitively extensive, full-length ESI or ESI-BF. The General Disinhibition subfactor (ESI-BF
DIS) taps the core of the externalizing spectrum and contains 20 items reflecting problematic impulsivity, irresponsibility, lack of planful control, and boredom proneness. The Callous-Aggression subfactor (ESI-BF
AGG) contains 19 items reflecting dishonesty, deficient empathy, destructiveness, and relational aggression. Finally, the Substance Abuse subfactor (ESI-BF
SUB) contains 18 items related to recreational and problematic use of alcohol, marijuana, and other substances. Although a bifactor structure of the ESI-BF has emerged as the best fitting model in previous studies [
1,
34,
35], the item-based three-factor model (i.e., consisting of the ESI-BF
DIS, ESI-BF
AGG and ESI-BF
SUB subfactors), nonetheless appears to be the most widely used variant of the ESI-BF in practice (see e.g., [
35‐
40]). Furthermore, the ESI-BF
DIS and ESI-BF
AGG scales are also included in the Triarchic Model of Psychopathy [
41], which has previously been used in incarcerated samples [
42,
43]. Thus, in this article, we focus our assessment of the criterion validity on the item-based ESI-BF three-factor model. The ESI-BF, if found to be valid and reliable in forensic psychiatric settings, could hopefully serve as a measure that will allow us to better assess and, in the end, may help us direct treatment efforts aimed at externalizing problems and thus reduce future risk and recidivism.
Since its introduction, the ESI, the ESI-BF and a previously developed short form, the ESI-100 [
44], has been translated into multiple languages and employed in a variety of populations and contexts, including prisoners [
45], forensic psychiatric and drug rehabilitation patients [
46], and the general population [
44]. The ESI is also recommended for assessment of the externalizing spectrum within the HiTOP framework [
47]. Nevertheless, so far validation studies of the ESI and the ESI-BF have primarily been conducted in undergraduate [
1] and prison samples [
33] and studies examining the ESI-BF outside of the North American context are still rare (for exceptions, see [
34,
46,
48]). To address these knowledge gaps, and in order to provide a first study exploring the validity of the Swedish translation of the ESI-BF, the current study: (1) presents a descriptive overview and assess the reliability of the ESI-BF, (2) examines the structural fit of three previously proposed models of the ESI-BF [
1,
26], and (3) examines how scores on the ESI-BF subfactors relate to early-onset externalizing behaviors (e.g., truancy and bullying), to lifetime aggregate diagnoses of externalizing disorders, as well as to aggressive and antisocial behaviors.
Discussion
This study is one of the first to examine the psychometric properties and criterion validity of the ESI-BF in a forensic psychiatric inpatient sample. Furthermore, it is the first study to our knowledge to attempt to validate the ESI-BF in Swedish context and one of the first few to investigate the ESI-BF outside of the United States. Overall, we found the basic psychometric properties to be satisfactory, while the structural models fared less well. In regard to criterion validity, however, the ESI-BF showed some promise as a measure of externalizing problems in this population.
In relation to our first aim, we found that the basic psychometric properties of the ESI-BF held up well, with α-values for all but four facet scales (Alienation, Fraud, Honesty, Relational Aggression) falling within the “satisfactory” range (α ≥ 0.80) [
70], and with similarly consistently high ω-values; only three facet scales (Alienation, Fraud, Honesty) had ω-values below 0.80. Notably, since ω-values are generally considered to rely on more plausible model assumptions for psychological attributes than α-values [
71], they may be considered a more precise estimate of an instrument’s basic psychometric properties. Average inter-item correlation values fell within the range of values that indicates that the items sample the intended construct in a way that is neither to broad nor too narrow. Four facet scales, however, (Marijuana Use, Marijuana Problems, Drug Problems, Boredom Proneness) stood out with regards to elevated values, indicating that these facet scales may contain redundant items and tap their respective domain to narrowly. Three of them (Marijuana Use, Marijuana Problems, Boredom Proneness) exhibited similarly elevated values in recent work by Soe-Agnie et al. [
34], also carried out in a forensic psychiatric patient sample, possibly highlighting a potential issue with these facet scales in the forensic psychiatric context.
Confirmatory factor analyses showed poor to mediocre fit for all three models, in essence replicating recent findings from Soe-Agnie et al. [
34], carried out in a Dutch forensic psychiatric sample. Nevertheless, the bifactor model for the ESI-BF, first proposed and tested by Patrick et al. [
1], showed the best fit out of the three models examined in terms of relative and absolute fit indices. Although absolute fit for the bifactor model was mediocre in our study, it is worth noting that its relative fit was similar to what has been observed in previous work [
1,
34,
43]. Whether this can be seen as a vindication of the original models is difficult to tell given that bifactor models are known to be prone to overfitting [
72]. It should also be noted that while it is theoretically possible to improve overall goodness-of-fit by post-hoc model modification based on modification indices, this approach remains debated due to issues with poor generalizability, especially in small samples [
73‐
75]. Thus, given the relatively small sample in the current study, we chose not to pursue any post-hoc model modification. However, for the interested reader we provide all modification indices in the Supplemental Material (see supplementary material 6–8).
There are several possible explanations for the mediocre fit of the bifactor model in our study. First and foremost, the small sample size likely impacted the robustness of our analyses. Another reason could stem from a possible heterogeneity in the clinical manifestation of violence (e.g., psychotic, impulsive and organized violence; [
76]) and other externalizing outcomes among our participants. It is an open question if violent and externalizing outcomes in our group was driven mainly by processes stemming from the severe mental illness (e.g., delusions) as opposed to personality traits along the externalizing spectrum, the two processes of which may contribute differentially to the manifestation of violence. The structural models tested in the current study were all originally derived by Patrick et al. [
1] in undergraduate and correctional samples in which, presumably, the rate of severe mental illnesses was relatively low. Thus, it is possible that the high rates of severe mental illnesses and neurodevelopmental disorders that characterize our sample [
49] contributed independently, beyond any externalizing psychopathology, to outcomes such as violence, and therefore in the end may have affected the fit of the models. Similarly, it is likely that the clinical characteristics of our sample differed from those of Soe-Agnie et al. [
34], despite both being based on forensic psychiatric patients. This would be due to differences in the legislations and definitions pertaining to offenders with severe mental illnesses in Sweden and the Netherlands, resulting in more patients with a primary diagnosis of personality disorder as opposed to psychotic disorder in the Netherlands (for some estimates see: [
50,
51,
77,
78].
With regards to the criterion validity measures, several robust correlations emerged. However, before interpreting these findings a significant caveat must be noted. As stated above, the correlated factors model used in this analysis exhibited mediocre model fit. Therefore, even though the observed correlations were robust, the relationship between the correlations and the structural fit of the models that is, how the structure and content of the ESI-BF give rise to these correlations remains unclear, and should be further investigated.
All ESI-BF subfactors were robustly associated with repeated truancy before the age of 18. This finding is in line with the notion that early and repeated truancy is best seen as a marker of a broad externalizing tendency and not only as a marker of low school engagement, although these processes may be reciprocal [
79]. Early and repeated school truancy should thus warrant increased societal attention and intervention not only because of the risk for poor educational outcomes but also because it could portend a future trajectory of increasingly severe externalizing problems [
79,
80].
In relation to the LHA, which to the best of our knowledge has not been studied in conjunction with either the ESI or ESI-BF before, our findings indicate that all three ESI-BF subfactors were positively and robustly associated with the LHA
TOT, LHA
AGG, and LHA
ANTI subscales. Thus, providing some support for the overarching convergent validity of the ESI-BF as the LHA indexes instances of violent, criminal and norm-breaking behavior. The strongest of these associations was for the ESI-BF
DIS subfactor, suggesting further support for the conception that disinhibition is a core aspect of externalizing psychopathology [
12,
36,
81]. Of note is also that the only ESI-BF subfactor that exhibited a robust, positive, association with LHA
SELF was ESI-BF
DIS. This finding may be interpreted in the light of previous studies that have pointed to a link between poor inhibitory control and impulsivity and non-suicidal self-injury [
82,
83]. As the ESI-BF
DIS subfactor appears, to a relatively high degree, conceptually similar to those constructs, the association with LHA
SELF scores would seem to be theoretically buttressed by this literature.
Several specific associations for the ESI-BF
AGG subfactor emerged. It was the only subfactor to be robustly associated with violence towards a caregiver before the age of 18. Previous literature specifically examining the link between callous traits and violence towards caregivers appears scarce (for one recent example see Curtis et al. [
84]). However, one model proposed by Kuay and colleagues [
85] delineates two proposed groups of children who engage in child-to-parent aggression. One group, with children exhibiting a high degree of callous-unemotional traits who present with more proactive aggression and aggression extending beyond the family (denoted generalists). Another group, with low levels of callous-unemotional traits, do not typically display aggression towards other people than their parents and their aggression is primarily of a reactive character (denoted specialist). Interpreting these findings through the lens of this model may lend some support for the existence of Kuay and colleagues’ proposed generalist group and may also parallel the distinction between disinhibited and antagonistic externalizing in the HiTOP model [
47]. A significant methodological limitation must however be accounted for here; our assessment of callous traits was done in adulthood and not in proximity with the violence towards the caregiver. It is thus not possible for us to establish the temporal precedence of these callous traits in the child-to-parent aggressors in our sample.
In line with some previous findings [
33,
35], ESI-BF
AGG was also the subfactor that exhibited weaker associations with alcohol and substance abuse as compared to the other two subfactors. From the standpoint of previous research this appears unsurprising, as the ESI-BF
AGG was constructed to capture core aspects of interpersonal-affective processes of psychopathic personality traits, above and beyond the influence of the ESI-BF
DIS. The ESI-BF
AGG has previously been labeled under the phenotypic concept of “Meanness” in Patrick and colleagues [
86] triarchic conceptualization of psychopathy. Thus, it describes an individual with tendencies towards narcissism, callousness, proactive aggression and an antagonistic and disaffiliated interpersonal style. Such attributes are not necessarily connected to substance and alcohol abuse, which in fact to some degree often are social activities and as such would stand in contrast to the disaffiliated style of the individual with psychopathic traits. These results therefore also lie in line with the current HiTOP conceptualization of the externalizing spectrum which distinguishes between a disinhibited and an antagonistic component of externalizing [
47].
Moreover, scores on the ESI-BF
AGG were negatively associated with having committed sexual crimes. Previous research on the link between callous traits and sexual offences appears to support such a connection [
87‐
90], although exceptions also exist in the literature on juvenile sexual offenders [
91]. Moreover, recent research in a sample of young Swedish offenders found a negative association between aggression and sexual offences [
92] and an older study of Dutch forensic patients also found significantly lower levels of self-reported hostile and aggressive behaviors among patients convicted of sexual offences as opposed to among patients convicted other violent offences [
93]. Our finding may reflect the nature of the particular sexual offenders found in our sample, as the strength of the association between psychopathic traits and sexual crimes has been found to vary considerably with regard to the type of sexual offence (e.g., rape, extra-familial offences or mixed sexual offences; [
90]). It should also be noted here, however, that our number of sexual offenders in our sample is small and that these findings may also reflect legal praxis surrounding mentally disordered offenders in Sweden. Individuals who have committed sexual offenses and who present with personality disorders in the absence of conditions such as psychosis or autism, have over the last decades increasingly been sentenced to prison rather than to FPC and only a small minority of patients within Swedish FPC are now sentenced for a sexual index offence (7% of male patients and 1% of female patients; [
51]). The sexual violence committed by the patients in our sample may therefore have been driven more by the nature and symptoms of their severe mental illness rather than by processes related to externalizing psychopathology, paraphilias or personality disorders.
Lastly, the ESI-BF was able to discern those patients who committed a larger number of crimes and those who began engaging in criminal behaviors at an early age. This finding suggests that externalizing problems are indeed distributed along a spectrum of severity and that this is, as previously described [
94,
95], manifested by an early initiation of, and persistence in, criminal behaviors. Crucially, studies have shown that a significant subgroup (~ 20%) of individuals who later develop a schizophrenia spectrum disorder display early and extensive externalizing behaviors which persist over the life-course and are often captured in childhood by the diagnoses of CD and ODD [
96,
97]. Accurate assessment of externalizing behaviors thus seems to be of high priority for this subgroup of persons with severe mental illnesses. For example, a recent large-scale epidemiological study [
98] found comorbid SUDs in persons with severe mental illnesses to increase the risk not only for violence perpetration but also for violent victimization.
Summary and conclusion
The potential relevance of ESI-BF is suggested by the robust associations between the ESI-BF subfactors and the measures of criterion validity examined in this study, and further by the good reliability and internal-consistency values exhibited by the ESI-BF. The current study constitutes a step towards making the ESI-BF, and the HiTOP taxonomy under which it is now subsumed, relevant for use within forensic psychiatric settings. While this study alone is of course insufficient to establish the validity or the appropriateness of the clinical use of the ESI-BF in forensic psychiatric settings, it adds a piece of the puzzle which was outlined by Simms and colleagues [
99] in which the need for instruments derived from the HiTOP taxonomy to be increasingly tested in clinical and forensic samples is highlighted. In the long term, improved measurement and understanding of the externalizing spectrum may aid in the development of novel treatments within settings such as forensic services, possibly with a broader transdiagnostic approach [
100]. Such transdiagnostic approaches have historically mainly been focused on internalizing aspects of psychopathology but findings relevant to externalizing problems, among both child and adult populations, are beginning to emerge. Examples of potential transdiagnostic mechanisms that have been implicated in these studies and which may serve as future treatment targets include: impulsivity, emotion dysregulation, irritability, anger and anger rumination and impaired emotion recognition [
101‐
105]. Identifying and targeting such potential mechanism that cut across diagnostic categories and intersect with severe mental illness in this group of offenders could then potentially increase treatment efficacy. At least two such factors, anger and impulsivity, have already been highlighted as potentially relevant in relation to schizophrenia and its relationship with violence [
106,
107].
Strengths and limitations
A limitation in the present study was the relatively small sample size. This is common within forensic psychiatric patient samples but, of course, affects the robustness of the statistical analyses and, subsequently, any conclusions drawn from this study must take the small sample size into account. Another limitation relates to the accuracy of self-report measures in research on forensic psychiatric patients, a population with a high prevalence of psychotic disorders where reduced illness insight and impaired cognitive functions are common features [
37]. Concerns may therefore be raised regarding to which degree the participants were willing and/or able to answer the questions of the ESI-BF truthfully and accurately. The context of compulsory FPC could also quite plausibly have affected the way our participants chose to present themselves on the questionnaire, although they were thoroughly informed about the absolute disconnect between their research participation and any decisions about their care. Although we have no formal way to test to which degree or in which direction, if any, such factors may have impacted the results, there is some literature suggesting that concerns of this nature may be unwarranted. Self-report measures in forensic populations may, overall, remain valid and patients with psychotic disorders may nonetheless be able to report accurately on aspect of themselves [
108‐
111].
A major limitation in regard to our analyses of the criterion validity of the ESI-BF was the mediocre fit of the structural model we chose to use for those analyses. Such an analysis is in the light of this fact necessarily speculative and all results must be interpreted in the light of this significant limitation. Nonetheless, as this model has been used frequently in the ESI literature, we deemed it worth including.
Another limitation pertains to the use of cut-offs, both for the evaluation of CFA models and for interpreting Bayesian findings. While it may be argued that using cut-offs is arbitrary [
66], thus prompting careful interpretation, they may nevertheless be heuristically useful.
A further potential limitation lies in the degree to which these findings may generalize to other forensic psychiatric populations. This is because the Swedish legislation regulating compulsory FPC is, with very few exceptions, internationally unique in its design [
112]. This may hamper the generalizability of our results in other jurisdictions. Nonetheless, the group of offenders with severe mental illnesses studied here is of course not a group of persons unique to Sweden but one that would likely be found in similar systems and context in other nations and jurisdictions. This therefore suggest that some degree of generalizability outside of the Swedish FPC context, despite the legal peculiarities of Sweden, is present in the current study.
A strength of our current study is the rich, comprehensive array of criterion validity measures related to historic externalizing behaviors and outcomes for our participants. Given the extensive coverage of these information sources, and the combination of self-report and external sources, we believe that both these factors strengthen the reliability and accuracy of our data. With our access to this data, we could not only examine the structural aspects of the ESI-BF in this population but also gain a better picture of the validity of the ESI-BF and how it relates to “real world” outcomes that are thought to fall within the externalizing spectrum. Another strength, given the small sample size, is Bayesian approach to investigating criterion validity; Bayesian inferences remain valid regardless of sample size, and allows for genuine probabilistic statements [
113].
Clinical implications and future directions
We suggest that this study may aid not only researchers studying externalizing behaviors in forensic populations but that it may also, in the long run, contribute to the clinical assessment of these problems and behaviors in forensic psychiatric settings. Mounting evidence, which conforms with our clinical experience, points to the independent and primary importance of assessing and treating antisocial and externalizing behaviors in the prevention of violence and subsequent recidivism of offenders with severe mental illnesses [
26‐
29,
114].
Furthermore, future research on the ESI-BF in forensic psychiatric context should seek to recruit larger samples in order to ascertain if this affects the model fit for Patrick’s original models [
1], and in lieu of those models achieving acceptable fit further research could seek test the three-factor model that emerged in the work of Soe-Agnie et al. [
34]. In a sample of Dutch forensic psychiatric patients, Soe-Agnie et al. [
34] described a three-dimensional model, obtained via minimum rank factor analysis and exploratory bifactor analysis. The three dimensions were labeled Disinhibition/Alcohol Abuse, Callous Aggression and Drug Abuse. The Disinhibition/Alcohol Abuse factor contained facet scales tapping alcohol misuse and impulsive and irresponsible behaviors while the Drug Abuse factor encompassed facet scales related to drug abuse, physical aggression as well as other forms of antisocial behavior. Lastly, the Callous Aggression factor in Soe-Agnie’s model corresponded to the Callous Aggression factor found by Patrick et al. [
1], with the exception of one facet scale, Physical Aggression [
34].
Finally, since the ESI-BF was developed in a context where severe mental illness was not the primary clinical concern it is important that future investigations among forensic psychiatric patients continue to scrutinize if the original models hold up in these populations where externalizing psychopathology and severe mental illness intersect, or if a new superior structure emerges.