ABSTRACT
Objective
The Inventory of Callous-Unemotional Traits (ICU) is a widely used, comprehensive measure of callous-unemotional (CU) traits. While the ICU total score is used frequently in research, the scale’s factor structure remains highly debated. Inconsistencies in past factor structure research appear to be largely due to the use of small non-representative samples and failure to control for method variance (i.e., item wording direction).
Method
The current study used a multitrait-multimethod (MTMM) confirmatory factor analysis (CFA) approach that considers both trait and method variance to test the factor structure of a 22-item version of the self-report ICU in a multinational community sample of 4,683 adolescents (ages 11–17).
Results
Results showed that a hierarchical four-factor model (i.e., one overarching CU factor, four latent trait factors) that controlled for method variance (i.e., by allowing residuals from positively worded items to covary) provided the best fit (χ2 = 2797.307, df = 160, RMSEA=.059, CFI=.922, TLI=.888, SRMR=.045).
Conclusions
After controlling for method variance, the best-fitting factor structure is consistent with how the ICU was developed and corresponds to the four symptoms of Limited Prosocial Emotions (LPE) specifier in the DSM-5 criteria for Conduct Disorder (CD). In addition, measurement invariance of this factor structure across age (i.e., younger versus older adolescents) and sex was supported. As a result, mean differences in ICU total score across age and sex can be interpreted as reflecting true variations in these traits. Further, we documented that boys generally scored higher than girls on the ICU, and this sex difference was larger in later adolescence.
Acknowledgments
We would like to thank our gracious collaborators who collected and provided the data to make this multinational ICU database possible.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Supplementary material
Supplemental material for this article can be accessed online at https://doi.org/10.1080/15374416.2022.2148531.
Notes
1 Tests of measurement invariance across ICU language were not included in the main analyses due to widely varying sample sizes. However, exploratory analyses were conducted using two different approaches to address sample size variability. The first was to randomly sample 400 youth from each subsample to be of approximately equal weight as the smallest (i.e., the Dutch) subsample. The second approach was to test invariance between the Greek subsample, which was approximately half of the entire sample, to all others. The results of these analyses can be found in Table S5. As noted in this table, consistent support across fit indices was found for configural and weak factorial models, but support was not consistent for strong and strict factorial invariance.