The factor structure and construct validity of the inventory of callous-unemotional traits in Chinese undergraduate students

The current study assesses the factor structure and construct validity of the self-reported Inventory of Callous–Unemotional Traits (ICU) in 637 Chinese community adults (mean age = 25.98, SD = 5.79). A series of theoretical models proposed in previous studies were tested through confirmatory factor analyses. Results indicated that a shortened form that consists of 11 items (ICU-11) to assess callousness and uncaring factors has excellent overall fit. Additionally, correlations with a wide range of external variables demonstrated that this shortened form has similar construct validity compared to the original ICU. In conclusion, our findings suggest that the ICU-11 may be a promising self-report tool that could be a good substitute for the original form to assess callous-uncaring traits in adults.


Introduction
Psychopathic personality is a multifaceted personality disorder comprised of the interpersonal, affective, and behavioral/lifestyle dimensions [1]. There is growing evidence that the affective component of psychopathy, also called callous and unemotional (CU) traits, could define an important subgroup of children and adolescents with severe conduct problems [2]. CU traits are characterized by a lack of concern about performance, lack of guilt and empathy, and a shallow and deficient affect [3]. These traits are believed to be the developmental precursor to adult psychopathy [3][4][5]. Increasing understanding of the developmental progression of CU traits from childhood to adulthood has received increased attention [6,7]. Therefore, the development of a measure that is appropriate to use in various age groups is timely, necessary, and critical. One way to do this is to validate existing youth measurements in adult samples. Until recently, only a few studies have attempted to address this issue [8][9][10][11]. For example, in a sample of 687 college students, Kimonis and colleagues found that a three-factor structure similar to that found in youth fit the data well through principal components analyses. The final model also showed reasonable convergent and discriminant validity. Other instruments that were developed initially for youth (i.e., the Youth Psychopathic Traits Inventory) have also been validated in adult populations [10,11]. Despite those promising results, more validation studies are warranted, and the current study will add to this area of research.
Given the importance of CU traits for understanding antisocial and delinquent youths, there is a need for an efficient, reliable, and valid measure of these traits. The two most widely used measures are the Psychopathy Checklist: Youth Version (PCL: YV) [12] and the Antisocial Process Screening Device (APSD) [13]. The PCL: YV is a 60-90 minute semi-structured interview and has primarily been used in incarcerated samples of adolescents (ages 12 to 18). It is a time-consuming instrument and thus is less appropriate for use in community samples. Furthermore, it contains only a few items that specifically assess CU traits (n = 4).
The APSD is a 20-item rating scale including parent, teacher [13], and self-report [14] versions. However, this scale contains only a few items (n = 6) to assess the CU traits and is limited with regard to the number of response options available (0 = not at all true, 1 = sometimes true, and 2 = definitely true), thus restricting the range of scores on the measure. Furthermore, many studies have indicated that the internal consistency of the CU factor is unacceptable [15,16]. To overcome these limitations, the Inventory of Callous-Unemotional Traits (ICU) was introduced [17]. Each of the four items that loaded consistently on the CU factor of the APSD was expanded with six new items. Specifically, three positively (e.g., "Shows no remorse when he/she has done something wrong") and three negatively worded (e.g., "Easily admits to being wrong") items were developed from each original item, leading to a 24-item scale with equal number of items worded in each direction [18].
In addition, these items are developmentally appropriate for use with older children as well as adults (e.g., "I do not feel remorseful when I do something wrong" or "I do not like to put the time into doing things well"). Most recently, researchers have revealed that the ICU may be a promising measure that has some utility in adults [8,11]. Still, validation of the structure of the ICU in various samples, specifically in non-Western cultures, is warranted. To this end, the aim of the present study was to examine the psychometric properties of the ICU in a Chinese sample of community adults.
However, further examination of the bifactor model in those studies revealed that this model didn't meet the common model fit criteria (i.e., CFI/TLI > .90 and RMSEA < .08, [28]). Table 1 summarizes findings of those studies that tested the factor structure of the ICU. It is worth noting that almost all studies accepted the bifactor model as the best fit model because it was better than the unidimensional model and the intercorrelated three-factor model (without a higher order general factor). However, detailed examination of the models indicated that the model fit was insufficient. For instance, Ciucci et al. (2014) [19] compared four different models of the self-report ICU in a sample of 540 Italian children. Although the bifactor model exhibited the best fit (χ 2 = 442.06, df = 198, χ 2 /df = 2.23, CFI = .87, TLI = .85 and RMSEA = .05), none of the four models reached the minimum fit criteria [28]. Again using the self-report ICU, Feilhauer and colleagues (2012) [21] compared a one-factor model, a three-factor intercorrelated model, a three-factor hierarchical model, and a bifactor model in a mixed adolescent sample. All models failed to fit the data well; the authors therefore extracted five factors through exploratory factor analyses. Only two studies have examined the factor structure of the original ICU in adults and the findings are inconsistent. Byrd et al. (2013) [8]in a community sample of adult males concluded that the three-factor bifactor model was the best, although the model fit indices did not reach the criteria (CFI = .88, TLI = .91, and RMSEA = .10). In a group of college students  [11]conducted an exploratory principal components analysis with varimax rotation, and concluded that a new three-factor model was suitable with 37.6% of variance explained. In sum, the extant literature provides limited support to the bifactor model as representing the underlying dimensionality of the ICU, and much less is known in adults.

The shortened forms of the ICU
Several studies have developed various shortened forms of the ICU after failing to achieve acceptable fit with original items. For example, Hawes and colleagues (2014a) [29]examined the factor structure of the ICU in 250 boys who exhibited significant conduct problems. With the three-factor bifactor model failing to fit their data, a 12-item shortened form was developed through item response theory. This shortened form of the ICU (SF-ICU) consists of two factors: callousness (7 items) and uncaring (5 items), and its scores demonstrated good reliability and discrimination across the continuum of the CU constructs [29]. The total score of the SF-ICU exhibited the expected associations with relevant external measures, including conduct problems (r = .46, p < .01) and social competence (r = -.55, p < .01).
In a sample of male and female children from the community, Gao and Zhang (2016) [35]created two different shortened versions for the child-and parent-report forms of the ICU. Specifically, the child self-report shortened form (ICU-13) consists of 13 items that were divided into two factors: callousness (7 items) and uncaring (6 items). The α coefficients for the ICU-13 total score and the two subfactor scores were acceptable. In addition, the ICU-13 total score and its two subfactor scores exhibited the expected associations with relevant external measures [35]. In sum, shortened forms of the ICU have recently received promising initial support. However, none has examined the validity of shortened forms in adults or in non-Western samples.

The current study
The primary aim of the current study was to examine the factor structure of the original 24-item ICU and the shortened forms of the ICU (i.e., SF-ICU, ICU-10, and ICU-13) in Chinese adults from the community. To this end, a series of confirmative factor analyses (CFA) were conducted to compare these models. The model specifications are present in Table 2.
On the basis of findings from recent studies, we predicted that the bifactor model of the original ICU would provide unacceptable fit to the data. We also would explore which of the shortened version fit our data best, given that no such study has been done in adult populations. Additionally, we aimed to test the construct validity of the best-fitted model by examining whether the total and factor scores were correlated as expected with constructs including (a) alternative measures of psychopathy (i.e., the Levenson Self-Report Psychopathy Scale (LSRP)); (b) aggression (e.g., Reactive-Proactive Aggression Questionnaire); (c) antisocial personality symptoms, and (d) trait measures of empathy and callousness. Based on previous research, we expected that CU traits would be positively related to the LSRP total and subscale scores, in particular the primary psychopathy subscale score [11], reactive and proactive aggression scores [20] [21,23], and the number of antisocial personality symptoms [11]. In addition, we expected that the CU traits would be positively related to the scores on trait measures of callousness, and negatively correlated with empathy [11]. Additionally, the ICU-callousness factor score would be preferentially associated with the trait measures of callousness.

Participants
Two independent samples of participants were recruited from a community college in Guangzhou City, China. Questionnaires were administered to only those who had given informed consent. This study was approved by the Human Subjects Review Committee at Guangzhou University. Participants completed surveys in school during specific class periods lasting approximately 40 minutes. After answering basic demographic questions (e.g., age, sex, race/ethnicity), participants completed the measures described below. All questionnaires were administered in Chinese.

Measures-Both samples
Inventory of Callous-unemotional traits. The Chinese version of the ICU [17] was translated into Chinese and back-translated to English to ensure accuracy. The translators further discussed items with translation differences, until they reached an agreement. The questionnaires were then piloted in a different sample (n = 22) of college students to assess for readability, and no further revision was needed.
Personality diagnostic questionnaire. The Antisocial Personality Disorder (ASPD) subscale of the PDQ-4 [36,37]was used to measure characteristics of ASPD in both samples. The ASPD scale consists of 22 forced-choice items that are rated as either true or false. Items correspond to diagnostic criteria for the ASPD from the Diagnostic and Statistical Manual of Mental Disorders [38] (4th ed., American Psychiatric Association, 2000). Sample items include "I've been in trouble with the law several times," and "Lying comes easily to me and I often do it." The Chinese version of the PDQ-4 has demonstrated moderate internal consistency (α coefficients ranged from .56 to .78) and test-retest reliability (the coefficients ranged from .49 to .80) in college students [37]. In the current sample, the internal consistency was .66.
IPIP-empathy and IPIP-Callousness. An established 10-item Likert-scale questionnaire assessing empathy was drawn from the International Personality Item Pool (IPIP). Sample items include "Suffer from others' sorrows" and "Don't understand people who get emotional". Scales drawn from the IPIP have well-established reliability and validity in the literature, and are freely available (http://ipip.ori.org/). Ten items correspond to the empathy subscale of Jackson Personality Inventory-Revised [39], and the coefficient α was .80 in initial sample. The Chinese version of the IPIP-Empathy was translated in the current study and the coefficient α was .73.
Similarly, an established 7-item Likert-scale questionnaire assessing callousness was drawn from the IPIP. Sample items include "Am not a caring person" and "Can't be bothered with others' needs". The IPIP-Callousness scale has been scrutinized in community (N = 1,269) and patient samples (N = 628), and the coefficient αs were .85 and .83, respectively [40]. The Chinese version of the IPIP-Callousness was translated in the current study, and the internal consistency was .78.

Measures-Sample one only
The Reactive-Proactive Aggression Questionnaire (RPQ). The RPQ[41] is a 23-item self-report questionnaire that distinguishes between proactive and reactive aggression. A total of 12 items assess proactive aggression (e.g., "Hurt others to win a game"), and 11 items assess reactive aggression (e.g., "Reacted angrily when provoked by others"). Items are scored on a three-point scale (0 = never, 1 = sometimes, 2 = often), and scores of relevant items are summated to form measures of reactive or proactive aggression together with an overall score of total aggression. The questionnaire has high internal consistency and good validity [41]. Prior studies in Chinese samples have shown excellent internal consistency and good factorial validity and construct validity [42,43]. In the current study, consistency measures were comparable, with a coefficient α of .90 for the total scale, .80 for the reactive, and .86 for the proactive aggression subscale, respectively.

Measures-Sample two only
The Levenson Self-Report Psychopathy Scale (LSRP). The LSRP[44] is a 26-item selfreport questionnaire that provides a total score of psychopathy and subscale scores for primary and secondary psychopathy, respectively. The Likert-style items have four response options ranging from 1 (strongly disagree) to 4 (strongly agree). Research has indicated that this measure has adequate reliability, with coefficient αs ranging from .63 to .82 for the two subscales [44]. The Chinese version of the LSRP was created and validated in a sample of Chinese inmates [45]. In that study, the original two-factor structure fit the data reasonably well and provided good construct validity. In the current study, the coefficient αs for the total and factor scores were .78, .68 and .76, respectively.
The aggression questionnaire. The AQ [46]is a 29-item questionnaire assessing aggression in three components: a behavioral component represented by the subscales of physical aggression and verbal aggression, an emotional component covered by the anger subscale, and a cognitive component represented by hostility. Items were scored on a 5-point Likert scale from 1 (extremely unlike me) to 5 (extremely like me). The Chinese revision of the AQ has good internal consistency ranging from .60 to .89 and appropriate construct validity [47]. In the present study, the internal consistency was acceptable to good for the four subscales and the total scale, ranging from .60 to .89.

Statistical analyses
To compare the various models of the ICU, a series of CFAs were conducted via Mplus 7.0 [48] using robust weighted least-squares with a mean and variance adjustment (WLSMV) estimator. This method is strongly recommended for data with ordinal items [48]. Following generally accepted practice, we evaluated the fit of each model by examining multiple fit indices [28], including Chi-square, the root-mean-square error of approximation (RMSEA), the Tucker-Lewis index (TLI), and the comparative fit index (CFI). Conventional guidelines suggest that RMSEA values .08 indicate acceptable model fit and .05 indicate good model fit, and CFI, TLI ! .90 indicate adequate model fit [28].
To evaluate the internal consistency of the ICU scores, Cronbach's αs were calculated and coefficients were evaluated as follows: < .60 = insufficient; .60 to .69 = marginal; .70 to .79 = acceptable; .80 to .89 = good; and .90 or higher = excellent [49]. Given that α depends on inter-item correlations and number of items, we also calculated mean inter-item correlations (MIC), which is considered to be a more straightforward indicator of a scale's internal consistency than Cronbach's α and should be at minimum in the range of .15 to .50 to be considered adequate [50].
Finally, zero-order correlations were examined between ICU subscale scores and criterion variables (i.e., LSRP, RPQ, ASPD, AQ, and IPIP empathy and callousness). Additionally, to further evaluate the distinctive/independent contributions of the subscale scores of the ICU, we performed separate regression analyses, using subscale scores as predictors for each criterion variable. Table 3 summarizes the fit indices of these competing models in the whole sample. The original 3-factor model (M1) fit the data inadequately (WLSMV; χ2 = 1149.93, df = 249, p < .001, CFI = .83, TLI = .81, RMSEA = .08). After deleting item 2 and item 10 that were poorly correlated with the total score in previous studies [23] and also in the current sample, the revised model (M2) still showed poor fit. Two modified three-factor bifactor models displayed adequate fit according to the CFI (>.9) and RMSEA (< .08). The first model (M6) was proposed by Waller and colleagues based on the parent report version (WLSMV; χ2 = 592.27, df = 186, p < .001, CFI = .91, TLI = .89, RMSEA = .06), and the other model (M8) was reported by  [29] and its bifactor form (M14) displayed better fit compared to the other shortened models, and all items had statistically significant and moderate-to large-sized factor loadings on their respective factors (λ = .48-.69, ps < .01), with the exception of item 6 (λ = .14, p < .01). Similar findings have been reported by others [30].

Confirmatory factor analysis
After deleting item 6, the modified model (M15) fit the data very similar to M14 (see Table 4). Finally, the other short form models (M16-M18) fit the data inadequately. Thus, the modified model (M15) without item 6 was considered the best-fitting model and used in following analyses. To compare the differences in correlations between original ICU and this best-fitting model (ICU-11) with relevant variables, the original 3-factor model of ICU was used.  The inventory of callous-unemotional traits in Chinese undergraduate students

Internal consistency and intercorrelations
The coefficient αs for all tested models in the current study are present in Table 2. Overall, the coefficient αs for the callousness factor were acceptable (αs >.70), and higher than those for the other two factors. Specifically, the coefficient α for the ICU total score (24 items) was .80 (MIC = .15). The reliability of the original model was .75 (MIC = .21), .68 (MIC = .21), and .66 (MIC = .28) for the callousness, uncaring, and unemotional subscale, respectively. The uncaring factor showed the strongest correlation with the callousness factor (r = .53), followed by the unemotional factor (r = .21). The callousness and unemotional factor showed the weakest correlation (r = .12). Inter-factor correlations after correcting for unreliability in CFA models were .75 for callousness-uncaring, .20 for callousness-unemotional, and .29 for uncaringunemotional. Those correlations suggested that callousness strongly relates to uncaring, whereas the correlations between unemotional and other factors are moderate at most (< .30). The coefficient αs for the ICU-11 total score, callousness, and uncaring were .76 (MIC = .22), .69 (MIC = .27) and .59 (MIC = .22), respectively. The correlation between the two subfactors was .51 at observed variable level; in contrast, the correlation reached .79 at latent variable level.

Construct validity
Descriptive statistics and internal consistency estimates of all measures in the current sample are presented in Table 4.
The zero-order correlations were calculated to examine the associations between ICU-11 total and subfactor scores and external criteria measures (see Table 5). As expected, significant correlations were found between ICU-11 total and LSRP total scores (r = .50, p < .01), as well as IPIP-callousness and empathy scores (r = .63 and r = -.40, ps < .01, respectively). The ICU-11 scores showed much stronger correlations with the LSRP primary than with the secondary psychopathy scores. In addition, ICU-11 scores showed significant correlations with measures of antisocial personality symptoms (i.e., ASPD). The correlations between ICU-11 scores and aggression measures were moderately significant, with the strongest associations emerging between proactive aggression and the callousness factor. Finally, ICU-11 scores were not significantly associated with verbal or physical aggression, although their associations with anger and hostility were significant.
Correlations between the original ICU (24 items) total and factor scores, and external variables were largely consistent with those for the ICU-11 (see Table 5). It is worth noting that the unemotional factor showed weaker or no associations with the external variables, except that it demonstrated stronger associations with scores on IPIP-empathy, verbal aggression, and anger.

Discussion
The developmental progression of the CU traits from childhood and adolescence to adulthood has received increasing attention of late. Therefore, an efficient, reliable, and valid measure of the CU traits covering various ages is of the upmost priority [52]. Although evidence from previous validation studies has shown that the ICU may be a promising measure in adult populations [18,20,21,23,26], these studies were restricted to the samples and findings on the factorial structure of the ICU have been inconsistent. To the authors' knowledge, the current study is the first to compare various models of the ICU and their psychometric properties in a non-Western adult sample. In particular, this is the first study to compare five recently proposed shortened forms of ICU. In general, we found limited evidence to support the original three-factor bifactor model in our sample. Instead, a shortened version with 11 items (e.g., ICU-11) loaded onto two factors (i.e., callousness and uncaring) demonstrated good fit and reasonable construct validity. Finally, our findings also revealed that the ICU-11 exhibits similar correlations with external variables as compared with the original ICU, suggesting that it this shortened form could be used as a reliable and valid measure of CU traits in community adults.
Consistent with most previous studies [8], the bifactor structure fit our data better as compared with the single-factor and the three correlated factor models, although its fit indices were still unacceptable. Other three-factor models with various modifications [33] were also tested but none provided good model fit. Taken together, we may conclude that at least at the item level, limited evidence supports the three-factor bifactor model in non-Western adult populations.
The coefficient α for the unemotional subscale was only .66 in the current sample. Similar issue with the unemotional factor has been reported in Byrd [8,18,23]. In addition, the unemotional factor of the original ICU showed a weaker or negligible association with the majority of the variables except for IPIP-empathy, AQ-verbal aggression, and AQ-anger. Of  The inventory of callous-unemotional traits in Chinese undergraduate students note, Feilhauer et al. (2012) [21] also failed to find significant associations between the unemotional factor and scores on the ASPD and the PCL: YV. Furthermore, in our sample we found null or negative associations between the unemotional factor and aggression measures, a finding seemed unexpected at the first glance. For example, Feilhauer et al. (2012) [21] found that the unemotional factor was positively associated with aggression as assessed by the AQ and RPQ, in a mixed sample aged from 13 to 20 years. Indeed, the correlation coefficients ranged from .18 to .30 (ps < .01). However, a more careful examination of the constituent items of the anger factor revealed substantial overlap with the items in the unemotional factor. For example, the two items of the anger factor, namely "I have trouble controlling my temper" and "Sometimes I fly off the handle for no good reason", capture the expression of anger emotion; meanwhile, the items in the unemotional factor, namely "I express my feelings openly (reversed)" and "I do not show my emotions to others", reflect the concealment of emotion. Therefore, it is not surprising that the anger factor of the Aggression Questionnaire was found to be negatively associated with the unemotional factor score (r = -.29, p < .001). Taken together, more research on the unemotional factor is warranted. Given that the unemotional factor displayed poor reliability and unexpected associations with theoretically related variables [22,29,30], some authors eliminated the unemotional items to develop a shortened form of the ICU. In general, the shortened version proposed by Hawes et al. (2014a) [29]consists of 12 items loaded onto two factors (i.e., callousness and uncaring) and fit our data well. Notably, in line with recent work [30], item 6 (i.e., "Does not show emotions") demonstrated poor factor loading (λ = .14) and was subsequently deleted from the analyses. In fact, item 6 was not included in any of the two-factor models (except [29]).
With regard to internal consistency, the findings of the current study were consistent with most previous studies [8,18,23]. Specifically, the coefficient αs for the callousness subscale in most of the tested models were acceptable, whereas the uncaring factor demonstrated poor internal consistency. Notably, the coefficient α for the uncaring factor in Hawes et al.' model was only .59, which was much lower than findings in other reports [29,30], although the MICs were in a reasonable range (>.15). Such low internal consistency indicates that the items in the uncaring factor need to be further refined in future studies.
The ICU-11 total score exhibited robust associations with other measures of psychopathic features. Specifically, the primary psychopathy factor of the LSRP assesses a callous, manipulative, and self-centered lifestyle, while the secondary psychopathy factor assesses impulsivity and poor behavior controls [44]. As expected, the ICU-11 total score showed stronger correlations with the primary than with the secondary psychopathy scores, which is in line with previous studies in adults [11]. Furthermore, the ICU-11 total score showed stronger correlations with psychopathy scores (i.e., LSRP) than with the number of antisocial personality symptoms (i.e., ASPD; r = .52 vs. r = .27), indicating that these characteristics are related to but distinct from symptoms associated with antisocial personality disorder. Additionally, in line with prior studies [30,31,33], the correlation pattern for the total score of the ICU-11 highly agrees with that for the original ICU, suggesting that this shortened scale keeps sufficient information from its original form.
At the factor level, both callousness and uncaring factor scores correlated significantly with overall scores on the ASPD, LSRP, and proactive aggression, again in line with previous findings [11,26]. Interestingly, when both factors were entered in regression, only callousness was significantly related to these measures. Moreover, the construct validity of the ICU-11 was supported by its associations with measures of empathy and callousness. Again, the pattern of correlations among ICU-11 callousness/uncaring and external measures was similar to that with the original ICU.