Introduction

Conduct Disorder (CD) is a highly heterogeneous disorder with a wide variation of severity, course, and aetiology [1]. Its incidence has been associated with criminal involvement and social exclusion as a result and also as an accompanying range of other costs to affected individuals and the society at large. These intangible costs include physical and mental health burden to affected adolescents as adults and the cost to society for providing financial provisions for afflicted individuals [2, 3]. The variability of CD manifestations has led to the difficulties in its diagnosis and management. Previous research has shown strong evidence that the affective component of psychopathy, also called callous-unemotional (CU) traits, defines an important subgroup of children and adolescents with severe conduct problems [4]. In specific, this group of young people differ from other antisocial youth on important biological, emotional, cognitive, and social characteristics [5]. CU traits are also important clinically, being significant enough as to be included in the Diagnostic and Statistical Manual of Mental Disorders (DSM‐5) specifier criteria for conduct disorder as ‘with limited prosocial emotions’ [6, 7]. Ever since its inclusion in the diagnostic classification, there has been an increased focus on the best method to assess CU traits [5].

Youths with high scores of CU traits have typical characteristics such as lacking remorse or guilt, having a shallow affect, and a callous disregard for others which predispose them to a tendency to commit crime or violate the law [5, 9]. The increased use of CU traits, separately from other domains of psychopathy, makes it important to ascertain how these traits relate to other personality constructs. These traits are believed to be the precursor to the development of adult psychopathy [8, 9]. In addition, the temperamental features related to CU traits are risk factors for impairments in young people’s conscience development. Extensive empirical research implicates CU traits in the development of severe and persistent antisocial behaviours [10] and having poorer responses to standard interventions for antisocial behaviour [11]. CU traits have emerged from just being a component under the contruct of psychopathy to being examined, itself, as a construct that is worth studying [12]. Thus, studying CU traits could further enhance the current theoretical models that explain the development of severe antisocial behaviour.

Among young people with severe conduct problems, however, it appers that about a quarter (25–30%) of them were found as having heightened level of CU traits [13]. Despite the number, these subgroup of children and adolescents is important to be examined and treated clinically for many significant reasons, which include but not limited to, them showing more severe and more stable conduct problems. In this case, they also show higher level of aggression, their aggression tend to harm the victims more, and their aggression is more goal-oriented–for personal gain or to dominate the other person [14, 15].

Given the importance of assessing youth for their CU traits, especially among those with conduct problems, two widely used scales in research are the Psychopathy Check List- Youth Version (PCL-YV) [16] and the Antisocial Process Screening Device (APSD) [17]. The downside for the PCL-YV, however, is the time needed to complete the measure, which takes around 60–90 minutes by highly trained clinicians. As for the APSD, it is a 20-item scale which can be self-reported by the young person or by their parents or teachers, but only 6 of the items measures CU traits [17], making it hard to confirm the important facets of CU traits. Motivated by this shortcomings from previous measures, Frick developed the Inventory of Callous-Unemotional Traits (ICU) by expanding the CU traits questions on the APSD [18].

One of the most important studies in testing the psychometric properties of the ICU was conducted by Essau et al. on German school-going adolescents (n = 1443) aged from 13 to 18 years old [19]. Exploratory Factor Analysis (EFA) was conducted in order to test for the scale items’ dimensionality. The researchers also conducted a Confirmatory Factor Analysis (CFA) and they found that the data fit a three-factor model best, which are callousness, uncaring, and unemotional. This three-factor structure fits both boys and girls well. The Cronbach’s alphas for the factors were .70 for callousness, .73 for uncaring, and .64 for unemotional. The total internal consistency was .77, which was acceptable. In addition, the study also found CU traits to be corelated with conduct problems and psychosocial problems.

Kimonis et al. [20] investigated the psychometric properties of the ICU on 687 undergraduate students. They conducted exploratory principal component analysis (PCA) to examine the factor structure of the ICU and also to examine the internal consistency of this scale. The scree plot was referred to detemine the which factors to rotate–which resulted in a three or four factor solution. Using varimax rotation procedure, two PCAs were done. However, the four-factor solution resulted in five cross-loaded items, indicating the factor loadings were not interpretable. The three factor loading, on the other hand, was interpretable with uncaring factor accounting for 14.3% of the variance, 13.3% for unemotional factor, and 10.0% for callousness. Item 2 and 10 were removed to improve the internal consistency of the total score. The internal consistency of the factors were .77 for uncaring, .80 for unemotional, and .59 for callousness. These factors were also found to be modestly intercorrelated. The study also found ICU score to be correlated with Coldheartedness and Self-Centered Impulsivity factors of the Psychopathic Personality Inventory. In addition, it also correlates with the score of Levenson’s Self-Report Psychopathy Scale.

Kimonis et al. [21] also conducted another validation study for the ICU among 248 young offenders. CFA was conducted using AMOS plugin on SPSS. Again, the results rendered three factors, which were uncaring, callousness, and unemotional. Similar to their other study conducted on undergraduate students, this study also found items 2 and 10 from the callousness factor showed poor item-total correlation and poor loadings. Thus, the two items were removd. The internal consistency for the factors were between .78 to .84 for uncaring, between .71 to .88 for callousness, and between .45 to .60 for unemotional factor. There were moderate correlaton between the factors. The study, additionally found that ICU correlated with aggression, delinquency, and psychophysiological and self-reported indices of emotional reactivity.

A study conducted on a 455 dutch community sample (self-report, parent report, teacher report, and combination of all three) to validate the ICU which was translated into Dutch [22]. CFA was conducted using LISREL framework. In terms of factor structure, the researchers did a comparison between a single-factor model (all items loaded into one factor), a three-factor model (callousness, uncaring, and unemotional), and a three-bifactor model, with all items loaded as one factor, and also on three distinct factors. The three-bifactor model showed the best fit. The Alpha coefficients for the three factors were between the range of “acceptable” to “good”. The study also found that CU traits were inversely correlated with insensitivity to punishment. Agreeableness and Conscientiousness subscales from the Big Five scale were also negatively correlated with ICU scores.

Pechorro et al. [23] conducted a study among 782 Portuguese community youth (M_age = 15.87 years) using both self-reported full version of the ICU and the short version with 12 items. Results from the confirmatory factor analysis showed that the three-factor structure of the full version of the ICU after removing two items were the best fit. For the short version, one item was deleted, resulting to the best fitting two-factor structure (without unemotional factor) of the scale. Both versions of the ICU showed adequate internal consistency, convergent validity, discriminant validity, and criterion-related validity.

Among the Chinese samples, Zhang et al. [24] tested the factor structure and psychometric properties of the chinese version of the ICU on 613 chinese detained boys. Results from the confirmatory factor analysis showed that the three-factor model with 24 items from the original ICU scale did not show a good fit to the data. However, the two-factor with callousness and uncaring dimensions with only 11-item showed the best fit for the data. Furthermore, the study also compared the two-factor ICU with other instruments for psychopathic traits, including the Antisocial Process Screening Device and the Youth Psychopathic Traits Inventory–which resulted in comparable and significant associations with the relevant criteria of psychopathy, aggression, and empathy.

Another study was conducted by Allen et al. [25] to test the invariance between the British and Chinese samples using the shortened version of ICU (11 item) and the original 24 items ICU. The study included 437 British children aged between 11 to 14 years old from the East of England and 364 Chinese sample between 10 to 13 years old who were attending school in Guandong, China. The internal consistency of the original 24-item ICU and the 11-item ICU in both samples were good for the total and subfactor scores, but not for the unemotional factor. According to a recent meta-analysis review, the unemotional scale has shown low internal consistency and poor construct validity in past studies [26]. Thus, overall, the 11-item ICU with callouness and uncaring subscales was the best fit for both British and Chinese samples.

Materials and methods

Results

Discussion

The concept and investigation of CU traits among adolescents in Malaysia is still relatively new. To the best of the authors’ knowledge, this is the first study with the aim to extensively examine the factor structure of the ICU among the adolescent population in Malaysia. Thus, validating the M-ICU would be an essential steppingstone in further discovering the prevalence of CU traits and other associated factors for aggression and delinquent behaviour among young people in Malaysia.

In comparison to the original ICU scale with the most popular three-factor construct, the initial EFA of the current study also produced a three-factor construct with the removal of two items, namely Q6 ‘I do not show my emotions to others’ and Q10 ‘I do not let my feelings control me.’ These two items described the emotional regulation of adolescents which could still be in the developmental stage, even more so for children with potential conduct issues who are the targeted group for this ICU scale. This has also been shown in other prior studies which stated item 6 has regularly demonstrated poor factor loading and has consistently been removed from the final versions of other validated ICU scales [33, 50].

The remaining items loaded onto three distinct factors with nine, ten and only three items loading into the uncaring, callousness, and unemotional subscale, respectively. Reliability estimates for the M-ICU are consistent with findings from previous studies which indicate that the internal consistency of the measure is generally acceptable. Although past studies, for example, Essau et al. [19] and Roose et al. [22] have found five items to load into the unemotional subscale (items 1, 6, 14, 19, and 22), the present study found that the internal consistency for the unemotional factor which consists of only three items was markedly lower than the other two factors. This finding was supported by multiple previous studies which all showed poor internal consistency of the unemotional factor [9, 22, 41]. The unemotional factor displayed unexpected and negative associations with other theoretically related variables such as with aggression measures, which was evident in past studies. For instance, a recent meta-analysis by Cardinale and Marsh [26] also showed that the pooled association between the unemotional subscale and with total CU scores assessed with other measures was small with r = 0.16. Meanwhile, the association between total ICU scores, uncaring and callousness subscales had medium to large effect sizes for association with psychopathy scores with r value ranging from 0.40 to 0.47.

Houghton and colleagues [50] argued that for children and adolescents, they may have difficulties to appreciate the (affective) emotions of others, and that “they may not have enough experience to be able to attribute these emotions to themselves or others” [42]. From that study, coupled with the findings of the present study, it could be possible that self-reported unemotional items may not contribute to differentiating between children with high or low CU traits. In fact, according to Kimonis et al., the unemotional factor may be tapping into a distinct construct from CU traits and thus needs to be examined in further detail in future studies to determine the relevance to include them as part of future (self-report) versions of the ICU.

In the EFA, the deleted items that would constitute the unemotional construct which are item 1 ‘I express my feelings openly’, item 14 ‘It is easy for others to tell how I am feeling’, and item 19 ‘I am very expressive and emotional’ all reflect the element of concealing emotions (once they are reverse scored). This could indicate that Malaysia’s adolescent population tend to be more reserved in their emotional expression, thus leading them to think they are indeed expressing appropriate level of emotion. In other words, they may not see themselves as being unemotional although they could be scoring higher on the callous and uncaring factors. A closer look at the wording in these unemotional items would also reveal that most of the questions imply the outward expression of emotions rather than experiencing them, which could contribute to this factor having low internal consistency [28].

To date, there have been more recent studies that have eliminated the unemotional factor entirely and produced a shortened version of the ICU (SF-ICU) with the number of items ranging from 10 to 13 items with most of these studies maintaining a bi-factor model. It is noteworthy to mention that the majority of these newer studies were conducted in mostly Asian populations such as China and Japan [24, 25, 31, 33, 34, 51]. This may also indicate that young people from these Asian countries (i.e., China, Japan, and Malaysia) tend to be more reserved when dealing with their emotions which could reflect the more conservative and reserved Asian cultures that are shared between these nations.

Regarding the retaining of item 20, this was done after discussion between the researchers and noting from prior validation studies that item 20 is not commonly seen to be associated with both item 6 and 10 that were deleted initially or to load together with items 1, 14 and 19 from the unemotional construct [52]. It is interesting to note that when this item was subsequently included in the CFA analysis, it showed an acceptable standardised factor loading of 0.58 and 0.56 for the 3-factor and 2-factor models respectively. This implies that this item is clinically important for the complete assessment of CU traits in our population specifically under the callousness construct as ‘not taking the time to do things well’ could be contrived as having a callous attitude to other people especially those with higher authority, such as teachers or parents. Despite past studies on the Asian population mentioned above have opted for the short form version (12 items) of the ICU, with only the callousness and uncaring subscales (e.g., [24, 25]), it may not fully meet the purpose of why the ICU was created–which is to assess for DSM-5 specifier criterion (i.e., “concerned about performance at school, work, or in other important activities”, “shallow and deficient affect”) [53].

Further examination of the factors using CFA, with Model-1 being produced based on the three-factor model in EFA showed unsatisfactory fit analysis. Two items (Q2 and Q22) were subsequently deleted due to low factor loadings and this resulted in Model-2 which marginally improved the fit indices, but still did not reach a satisfactory level. Re-specifation of the model by adding correlated residuals for items 15 and 23, produced the final Model-3 and was deemed satisfactory with all fit indices achieving the acceptable threshold values. Such correlation between items had been employed in previous studies to achieve a better fit for the model indices [31]. While Hawes et al. concluded that the two items 15 and 23 displayed low discrimination parameters based on Item Response Theory (IRT) analysis and were subsequently removed from their final model [32]. Based on the number of modifications required to retain the items in the Malay-version ICU scale, it suggested that the relationship between the items in this translated version may be different than the original version.

In Model-4, the fit indices were not within the acceptable threshold values for CFI, TLI and WRMR. Two problematic items, Q2 and Q22 were removed from the model. The fit indices improved but still not within the acceptable threshold value. Model-5 was re-specified by adding meaningful correlation between two items’ residual (Q23 and Q15) and this resulted in Model-6. This re-specification method was supported by Kenny [54], where he suggested two strategies for re-specification to improve model fitting which both are best done a priori–first, by testing theoretically meaningful ways to complicate or to simplify the model, or second, by using modification indices and standardised residuals in re-specifying the model. The latter was used in this study. Model-6 fit the data well with all the fit indices achieving the recommended threshold value. According to Hair et al., for a scale with more than 12 items, a CFI and TLI of more than 0.92 is acceptable [45]. All items achieved satisfactory factor loading, except for item 5, which was then being removed. Most of the shortened versions of the ICU proposed by other studies have also excluded item 5 from their final model [31, 32, 51]. Based on the fit indices displayed in Model-3 and Model-6, this study has an acceptable fit to both the two-factor and three-factor models.

The final factors in the present study aligned almost similarly with those found by Roose et al. [22] and Essau et al. [30], only for callousness factor, item 22 load into unemotional subscale, while for uncaring factor, item 8 load into callousness factor. Pechorro et al. [23] analysis showed 9 items fitted into the callousness subscale, with the only difference from the present study was item 8 being in this factor rather than the uncaring factor. For the unemotional factor, items 1, 6, 14, 19, and 22 were retained in this factor.

The M-ICU final model is a valid and reliable tool to be used to assess adolescents’ CU traits which can be predictive of future behavioural issues and conduct disorder symptoms. Therefore, this would be beneficial to help screen and for early detection of any children or adolescents with elevated levels of CU traits, and to refer them for appropriate early interventions and psychological support [34]. It is noteworthy that heightened level of CU traits, which is often being underestimated, is usually a risk factor that can appear somewhat early in a person’s life and propel them down the road to be involved in a perpetual cycle of severe or chronic violence and antisocial behaviour. This could eventually lead to a huge burden on the resources of not just to the public health and criminal justice systems, but the economic and social welfare systems of the society at large [55]. The researchers thus would recommend the usage of the self-report M-ICU to measure the level of CU traits, especially among the vulnerable groups, in the hope of preventing potentially devastating consequences of antisocial behaviour. Additionally, the researchers would argue that this version of the M-ICU based on Model 6 (with two factors and 17 items) would be as effective in measuring CU traits as the original 24-item version, with an added advantage of being less time consuming.

Conclusion

The present study, once again, supported that the M-ICU with two- or three-factor models as having good factor structure validity and satisfactory reliability, consistent with previous validation studies conducted on the ICU in different countries and languages. Nevertheless, it would be useful to replicate this study to further examine the item content of the M-ICU and also to see the replicability of this translated scale in different populations with more diverse age groups including younger children in primary school or even young adults (e.g., college and university students). Overall, the researchers would recommend the usage of the two-factor M-ICU scale for similar age group and among adolescents from the community sample based on Model-6 of the CFA, given that it has robust goodness of fit indices as compared to the three-factor ICU model. This was achieved by removing the unreliable items associated with the unemotional subscale. This is in tandem with multiple other studies especially in non-Western populations which have shown the low or non-reliability of the unemotional factor in measuring CU traits and the frequent negative correlations it displays with future conduct problems [14, 26, 29, 30]. This study has provided further evidence of the possibility of measuring CU traits with a modified ICU scale consisting of fewer and more precise items as compared to older versions of the ICU. To conclude, the findings show this scale can be useful in identifying the level of CU traits among school children. Screening these young people using the ICU can potentially help to identify those with heightened level of CU traits, who can then be given early and targeted intervention, which may prevent future conduct problems or antisocial behaviour.

Supporting information