Emotion Recognition Deficits among Children with Conduct Problems and Callous-Unemotional Behaviors

Abstract

Deficits in emotion recognition have been associated with psychopathic and callous-unemotional (CU) behaviors among adults, adolescents, and children. However, few previous studies have examined such associations exclusively during early and middle childhood, or demographic differences in emotion recognition that may result from early emotion socialization experiences. The current study used a large, population-stratified, randomly-selected sample of 2^nd grade children living in areas of high rural poverty to examine group differences in emotion recognition among children showing no conduct problems or CU behaviors (typical), conduct problems without CU behaviors (CP-only), and both CP and CU behaviors (CP+CU). Primary caregivers reported on children’s conduct problems and callous-unemotional behaviors at 1^st grade and children completed a computerized facial emotion recognition task at 2^nd grade. Results indicated that CP/CU group differences in emotion recognition accuracy were moderated by child race, with children in the typical group showing better overall accuracy and better recognition of fearful and happy faces among European American children, whereas no group differences were found among African American children. Implications for emotion socialization, etiology of CP and CU behaviors, and future directions for research and treatment are discussed.

During the past decade, callous-unemotional (CU) behaviors, a dimension of adult psychopathy that can be measured in middle (Frick & Viding, 2009; Hawes et al., 2014; Willoughby, Mills-Koonce, Waschbusch, & Gottfredson, & the Family Life Project Investigators, 2015) and early childhood (Hyde et al., 2013; Kimonis et al., 2016; Willoughby, Mills-Koonce, Gottfredson, & Wagner, 2014; Willoughby, Waschbusch, Moore, & Propper, 2011), has been used to account for some of the heterogeneity in child conduct problems (Rowe et al., 2010), and to predict later antisocial behavior and psychopathy (Frick & Viding, 2009). Although moderate conduct problems (e.g., defiance and uncooperativeness) are relatively normative during early childhood, most children do not show high levels of such behavior and they generally show decreasing problems with age (Bongers, Koot, van der Ende, & Verhulst, 2004). However, approximately one third of all children who show clinical levels of conduct problems can be characterized as having high CU behaviors, suggesting that approximately 2–4% of all children meet this condition (Frick & Viding, 2009). Such co-occurrence of clinical levels of conduct problems and CU behaviors is associated with a greater frequency, severity, and persistence of aggressive and antisocial behaviors over time, making this a small but highly significant population for study (see Frick & White, 2008; and Kotler & McMahon, 2005 for a review)

Whereas conduct problems (CP)—which subsume both oppositional defiant and conduct disorder behaviors—refer to angry, defiant, antisocial, aggressive, and norm-violating behaviors among children and adolescents (Kimonis, Frick, & McMahon, 2014; Lorber, 2004), CU behaviors refer to the affective component of psychopathy and are characterized by callousness, a lack of empathy, a lack of guilt, and shallow and/or deficient emotions (Frick, Ray, Thornton, & Kahn, 2014). Given these affective deficits, atypical development of emotion functioning and knowledge, including emotion recognition, may characterize an important pathway toward CU behaviors in early and middle childhood. Throughout childhood, recognizing facial emotion cues is crucial for children’s social competence, in that it helps them to form positive relationships with other (Chronaki, Hadwin, Garner, Maurage, & Sonuga-Barke, 2015; Denham, 1998). Cross-sectional studies suggest that facial emotion recognition abilities develop throughout early childhood and adolescence (Herba, Lanau, Russell, Ecker, & Phillips, 2006), with children and adolescents over 11 years old making fewer errors than younger children (Tonks, Williams, Frampton, Yates, & Slater, 2007). However, previous studies suggest that those with psychopathic and CU behaviors may not fit this normative developmental progression of emotion recognition.

As part of their impaired emotional processing capacities, both adults and adolescents with elevated psychopathic and/or CU behaviors demonstrate deficits in emotion recognition. Numerous studies report associations between psychopathy in adults and adolescents and the recognition of fear (Blair et al., 2004; Del Gaizo & Falkenbach, 2008; Montagne et al., 2005; Blair & Coles, 2000; Blair, Colledge, Murray, & Mitchell, 2001; Blair, Budhani, Colledge, & Scott, 2005; Lemos Vasconcellos, Salvador-Silva, Gauer, & Gauer, 2014; Stevens, Charman, & Blair, 2001; Stanković, Nešić, Obrenović, Stojanović, & Milošević, 2015), sadness (Blair & Coles, 2000; Blair et al., 2001; Dolan & Fullam, 2006; Fairchild, Stobbe, van Goozen, Calder, & Goodyer, 2010; Stevens et al., 2001), surprise (Fairchild, van Goozen, Calder, Stollery, & Goodyer, 2009), and happiness (Bagley et al., 2009). Impaired ability to recognize others’ emotions may contribute to the association between psychopathy and antisocial behavior due to a lack of realization, or attention to, the negative consequences of their actions for others; thus, allowing them to hurt others for their own gain without feeling badly (Blair, 2006).

There are far fewer studies that have examined associations between CU behaviors and emotion recognition deficits in younger children. Kimonis and colleagues (2015) found that preschool-aged children with high CU behaviors showed poorer recognition of angry, fearful, happy, and sad facial expressions than children with low CU behaviors. With slightly older children, Dadds and colleagues used eye-tracking to examine looking behavior and its association with emotion recognition deficits among children with CU behaviors (Dadds, El Masry, Wimalaweera, & Guastella, 2008; Dadds et al., 2011; Dadds et al., 2006). In these studies, among boys and adolescent males (8–17 years old), CU behaviors negatively correlated with fear recognition, and with focus on the eye region of the face. When the participants were instructed to focus on the eye region of the face, the negative association between CU and fear recognition was temporarily attenuated, suggesting that inattention to socially-relevant stimuli may partially account for deficits in emotion recognition. In addition, general deficits in emotion recognition have also been partially supported by the extant research on CU behaviors in childhood (Muñoz, 2009; Sharp, Vanwoerden, Van Baardewijk, Tackett, & Stegge, 2014; Kimonis et al, 2015), indicating that emotion recognition deficits related to CU behaviors in childhood may be generalized to various emotions.

Neurodevelopmental Models of Emotion Recognition Deficits among Children with Elevated CU Behaviors

Amygdala structure and functioning have been strongly associated with emotional functioning and are thought to contribute to emotion recognition abilities (Adolphs, 2010; Phelps & LeDoux, 2005; Wang et al., 2014), including the processing of threat-related stimuli and fear (Fitzgerald, Angstadt, Jelsone, Nathan, & Phan, 2006; Phan, Wager, Taylor, & Liberzon, 2002; Phillips, Drevets, Rauch, & Lane, 2003). However, evidence suggests that the amygdala also affects general recognition of emotions (including happy, sad, angry, and disgusted expressions; Fitzgerald et al., 2006; Yang et al., 2002). Thus, amygdala structural and/or functional abnormalities likely confer deficits in emotion recognition (Bio, Soerode-Souza, Garcia Otaduy, Machado-Vieira, & Moreno, 2013; Holmes et al., 2012; Sebastian et al., 2014).

Neurological dysfunction has been posited as an explanatory mechanism for associations between emotion recognition deficits and CU behaviors (Blair, 2003). Blair’s (2006) Integrated Emotion Systems (IES) model suggests that amygdala dysfunction is associated with specific deficits in attending to aversive stimuli, including recognizing fear and sadness. In a similar, neurodevelopmental model, Dadds and colleagues (2011) propose that certain emotional expressions are recognized by attending to the most salient part of the face, body, or voice (e.g. the eyes for fearful expressions or the mouth for happy expressions) and that amygdala dysfunction may lead to cascading interactions characterized by reduced eye contact with caregivers as early as infancy, leading to emotion recognition deficits that foster the early emergence of CU behaviors in childhood. In both models, antisocial and callous behavior are thought to grow over time, as signals of distress in others (e.g., fear and sadness) that usually inhibit further antisocial and callous behavior are not perceived because of poor emotion recognition. Given the importance of amygdala functioning and attending to socially-relevant facial stimuli for broad emotion functioning, it is possible that both of these developmental mechanisms would confer general, rather than specific, emotion recognition deficits among children with CP and CU behaviors.

Although genetically-mediated differences in neurological functioning are possible, an alternative, but not mutually exclusive possibility, is that experiential processes play an important role in emotion recognition abilities. Environmental processes (e.g., interactions with parents, teachers, and peers throughout childhood) influence children’s developing emotional functioning, including emotion recognition (Izard et al., 2011; Johnson, 1992; Warren & Stifter, 2008). Emotion socialization processes occurring within parent–child interactions during infancy and early childhood—such as parents’ emotion-related beliefs, labeling and teaching of emotions, and own emotion functioning and skills—may be particularly important for emotional functioning, as children learn about emotions explicitly and implicitly through such socialization (Castro, Halberstadt, Lozada, & Craig, 2015; Dunsmore, Her, Halberstadt, & Perez-Rivera, 2009). In addition, these emotion socialization processes may operate differentially as a function of various sociodemographic factors—including but not limited to child gender, race and ethnicity, and family socioeconomic status.

A small extant literature suggests that parents socialize emotions differently based on gender (Brody, 1985; Cunningham, Kliewer, & Garner, 2009; Garner, Robertson, & Smith, 1997; Wong, McElwain, & Halberstadt, 2009), including attending to and talking about different emotions, using more specific and greater variety of emotion words, talking more about relationships and people, and using more social context for emotions with their daughters than with their sons (Cervantes & Callanan, 1998; Chaplin, Cole, & Zahn-Waxler, 2005; Fivush, 2007; van der Pol et al., 2015). Similarly, a handful of studies have noted differences in emotion socialization, along with differential effects on children, across race. For example, punitive and negative emotion parenting behaviors, such as mothers’ minimizing reactions to children’s negative emotions, have been found to be positively associated with adaptive coping among African American girls and negatively associated with aggression among African American boys (Smith & Walden, 2001), whereas they have been positively associated with negative outcomes among EA children (Eisenberg et al., 1999). Additionally, findings using the current study sample suggest that African American mothers (Garrett-Peters et al., 2008) and fathers (Garrett-Peters, Mills-Koonce, Zerwas, Cox, Vernon-Feagans, & The Family Life Project Key Investigators, 2011) may engage in more emotion talk with their infants than European American parents. Further, family sociodemographic risk (including low income) has been negatively associated with maternal emotion-related parenting and various aspects of young children’s emotion functioning (Brophy-Herb, Stansbury, Bocknek, & Horodynski, 2012; Brophy-Herb et al., 2011; Shaffer, Suveg, Thomassin, & Bradbury, 2012); and in the current sample, greater household chaos and greater duration of poverty have been associated with children’s poorer emotion recognition (Raver, Blair, Garrett-Peters, & The Family Life Project Key Investigators, 2015).

Given the differences in emotion socialization experiences and emotion functioning across child sociodemographic factors, it is possible that such factors may confound or constrain the developmental associations between emotion functioning and emerging CP and CU behaviors. To date, no research (of which we are aware) has addressed this issue and, as such, the current investigation includes child race, gender, and family income simultaneously as both potential covariates and moderators in all analyses. Such exploratory analyses are needed as starting points for advancing our knowledge regarding between-group differences and within-group processes underlying these associations, and may provide important targets for intervention for different children at risk for CP and CU behaviors.

Contradictory Findings, Meta-Analytic Evidence, and the Current Study

It is important to note that not all studies report a negative association between emotion recognition abilities and CU behaviors (or psychopathy). In a study comparing adult, male offenders with psychopathic behaviors to non-offenders without psychopathic behaviors, Pham and Phillippot (2010) found no differences in emotion recognition between the two groups. More directly contradicting previously mentioned studies, Bowen, Morgan, Moore, and van Goozen (2014) found that adolescent, male offenders with psychopathic behaviors showed better recognition of fearful faces than did non-offending males. Likewise, Schwenck and colleagues (2014) reported that girls and adolescent females (8–17 years old) with high CP and CU behaviors displayed greater accuracy in recognizing fearful faces compared to girls without CP and CU behaviors. Finally, Woodworth and Willoughby (2008) found that boys and girls (7–13 years old) with high CP and CU behaviors showed a marginally significant trend toward greater accuracy in recognizing fearful faces, as compared to children with CP and low CU behaviors. Despite such contradictory findings, a recent meta-analysis of 26 studies (many of which have been discussed in the current review) by Dawel and colleagues (2012) found that psychopathy was associated with general impairments across the six “basic” emotions (i.e., happiness, sadness, surprise, fear, anger, and disgust; Dawel et al., 2012). Interestingly, this finding remained when fear and sadness were removed from analyses, adding strength to the suggested presence of a general deficit in emotion recognition.

It should also be noted that many previous studies have been hampered by several methodological limitations. First, a majority of studies on this topic have relied on relatively small samples. Although a handful of studies included samples of 100 or more participants (Bagley et al., 2009; Bowen et al., 2014; Dadds et al., 2008; Del Gaizo & Falkenbach; Fairchild et al., 2009), most have relied on samples less than 100 and as few as 30 participants (Jones et al., 2009). Second, almost no previous studies have focused on psychopathy- and CU-related emotion recognition among females. To our knowledge, only two studies have considered CU behaviors and emotion recognition specifically with respect to female children (Fairchild et al., 2010; Schwenck et al., 2014). Third, only a handful of studies have utilized samples with substantial proportions of racial minority participants, which would allow for the examination of differential emotion recognition across race (for exceptions, see Del Gaizo & Falkenbach, 2008; Kimonis, Frick, Fazekas, & Loney, 2006). Finally, few previous studies on this topic focused on developmental periods prior to adolescence (for exceptions, see Dadds et al., 2011; Kimonis et al., 2015; Woodworth & Willoughby, 2008).

The current study attempts to address these issues using a large and diverse sample of elementary-school age children to examine facial emotion recognition abilities among those characterized by high and low levels of CP and CU behaviors. It was hypothesized that typically developing children would demonstrate better overall facial emotion recognition accuracy than children with high levels of CP. Furthermore, it was hypothesized that children with co-occurring CP and CU behaviors would demonstrate the greatest deficits in emotion recognition. These hypotheses were consistent for the prediction of overall emotion accuracy, as well as for the accuracy of identifying specific emotions. Although no specific hypotheses were made with regard to variation in group differences as a function of child gender, child race, or family income, interactions between child CP/CU groups and demographic factors were examined, given that children’s emotional development, emotion recognition abilities, and externalizing behavior have been reported to differentially vary by child gender (Cunningham et al., 2009; Maxim & Nowicki, 2003), race (Garrett-Peters et al., 2008, 2011), and family income (Raver et al., 2015; Shelleby et al., 2014).

Methods

Participants

The current study utilized a birth cohort of children and families living in two areas of high rural poverty in the United States (three counties each in Eastern North Carolina and Central Pennsylvania) recruited to longitudinally investigate child development within the context of rural poverty. We used random stratified sampling of mothers who recently gave birth in hospitals to recruit a representative sample of 1,292 families across a 12-month period from September 2003 through August 2004. Specifically, the sample was recruited to be representative of every baby born to an English-speaking mother living in the counties selected during the year of recruitment, while also oversampling for poverty and race (i.e., African American). The full sample consisted of 549 African American (42.5%) children, 736 European American (57%) children, 7 children of other race (0.5%), 657 girls (50.9%), and 635 boys (49.1%). Further details on recruitment procedures and sample can be found in Willoughby and colleagues (2013) and Garrett-Peters and Mills-Koonce (2013). A subsample of 761 children (315 African American, 446 European American; 385, girls, 376 boys) with complete data for caregiver report of CP and CU behaviors and complete emotion recognition data were used for the current study. Children from the full sample who were excluded from current analyses due to missing data did not significantly differ from included children on any demographic variables measured.

Procedures

When children were in 1^st grade, primary caregivers were asked to report on children’s levels of conduct problems and CU behaviors during a home data collection visit. When children were in 2^nd grade, they completed a computerized emotion recognition task as part of a school-based assessment. Children were briefly pulled out of class to complete the tasks in a room designated for assessment at the schools.

Measures

Conduct Problems and Callous-Unemotional Behaviors.

Levels of CP were rated by caregivers using the Disruptive Behavior Disorder Rating Scale (DBDRS; Pelham, Gnagy, Greenslade, & Milich, 1992), a DSM-IV guided rating scale that includes subscales for assessing conduct problems (including oppositional defiance and conduct disorder behaviors). The validity of the DBDRS has been previously established (Pelletier, Collett, Gimpel, & Crowley, 2006) and the internal consistency for the conduct problem composite for this sample was high (a = .92). Callous-unemotional behaviors were assessed with the Inventory of Callous-Unemotional traits (ICU; Essau, Sasagawa, & Frick, 2006), a series of 24 items on a 4-point Likert scale (“not at all true” to “definitely true”) developed from other highly established clinical assessments (e.g., APSD, PCL-YV). The ICU has been validated among normative and clinical samples of both children (Hawes et al., 2014, Willoughby et al., 2015) and adolescents (Essau et al., 2006). Initial investigations into the factor structure of the ICU suggested a bifactor model, which included a general factor on which all items loaded and three specific subfactors (i.e., callous, uncaring, and unemotional) and demonstrated limited consistency across ages and cultures (Essau et al., 2006; Fanti, Frick, & Georgiou, 2009; Roose, Bijttebier, Claes, & Lilienfeld, 2011). However, more recent work has reported mixed results regarding the factor structure of the ICU, with researchers providing evidence for models ranging between two and five factors (Feilhauer, Cima, & Arntz, 2012; Kahn, Byrd, & Pardini, 2013; Kimonis, Branch, Hagman, Graham, & Miller, 2013). Willoughby and colleagues (2015) recently used confirmatory factor analysis to test the factor structure of the ICU using the current sample and findings indicated that CU behaviors are best represented in the this data using a two-factor model that distinguishes empathic-prosocial (EP) and callous behaviors. The EP and callous factors were shown to be moderately negatively correlated and were primarily delineated by the positively and negatively worded items, which is consistent with published findings in other samples (Hawes et al., 2014; Houghton, Hunter, & Crow, 2013). Examples of items loading on the callous factor include “does not care who s/he hurts to get what s/he wants” and “seems cold and uncaring to others”, whereas examples of items loading on the EP factor include “expresses his/her feelings openly” and “does things to make others feel good”.

Given the factor analytic findings with the current sample, we examined group differences in emotion recognition accuracy based on CP symptoms and both EP and callous behaviors. Following clinical guidelines from the DSM-V, children were designated as having high CP if their primary caregiver reported three or more conduct disorder symptoms and/or four or more oppositional defiant disorder symptoms on the DBDRS (other children were designated as low CP). Children were designated as having low EP behaviors if they scored in the 10^th percentile or below based on caregivers’ reports on the ICU. Children were designated as showing high callous behaviors if they scored in the 90^th percentile or above based on caregivers’ reports on the ICU. Although there are no set guidelines that designate clinical levels of CU behaviors, we were confident that the 10^th/90^th percentile cutoffs used designate children that were showing very high levels of CU behaviors in the clinical range. As such, three groups were created using CP and CU criteria: (1) typical (low CP, high EP and low callousness), (2) CP-only (high CP, high EP and low callousness), and (3) CP+CU (high CP, low EP or high callousness). Thus, children qualified for the CP+CU group if they showed high CP and either low EP or high callous behaviors. Although some children did fit into a CU-only group (low CP, low EP or high callousness; n = 132), this group was excluded from analyses to be consistent with DSM-V criteria for conduct disorder, which includes CU behaviors only as a specifier, rather than an independent disorder in childhood (American Psychiatric Association, 2013).

Emotion Recognition Accuracy.

Emotion recognition accuracy was assessed using a computerized task, the Increasingly Clear Emotions task (ICE; Halberstadt, Leary, Garrett-Peters, Lozada, & Sibley, 2011). To create the ICE task, adults’ facial expressions were video recorded while they shifted from neutral expressions to prototypical emotional expressions. These recordings were then split into 7 still images, with the first image depicting a neutral expression and progressing to the seventh image depicting a fully prototypical expression. The video recordings were of 20 different adults (half female), with equal representation of African Americans and European Americans within gender. Five emotions were represented within gender and ethnicity (anger, sadness, happiness, fear, surprise). Photographs were full head shots; 13 were taken from Cohn-Kanade AU-Coded Facial Expression Database (Kanade, Cohn, & Tian, 2000), and were supplemented by the recruitment of 7 African American actors. All faces were judged by 20 adults (10 European Americans, 10 African Americans balanced across gender) for neutrality and prototypicality. Children were asked to choose via a forced-choice response format to determine which of the five emotions (angry, sad, happy, afraid, surprised) was expressed by the actor and viewed all 20 faces in each round before proceeding to the next level of expressive intensity. All image sequences within a round were randomized and similar in the stage (level of prototypicality) of the emotion expressiveness. The first round in which participants viewed neutral faces was dropped because the facial expressions were, in fact, neutral, and participants were unable to distinguish among the emotions. In an effort to reduce the length of the task and maintain children’s engagement and focus, the next-hardest sequence was removed for children as well, as was the final round (full prototypical expression) because of ceiling effects. Thus, children viewed four rounds (rounds 3, 4, 5, and 6) of the expressions as they became increasingly clear from neutral to prototypical. Children’s overall accuracy (across all rounds and emotions) and individual emotion accuracy were examined in association with CP/CU group membership.

Potential Moderators.

Child gender and child race were reported by primary caregivers when they were recruited at the time of their child’s birth (and confirmed at each home visit). Family income-to-needs ratio (total household income divided by the 2005 federal poverty threshold) was reported by primary caregivers when children were 6, 15, 24, 36, 48, and 58 months old; the mean income-to-needs ratio across these visits was used for the current analyses. Main effects of these variables, as well as their interactions with CP/CU group were examined.

Additional Covariates.

Primary caregivers’ years of education (reported at recruitment) and children’s age when they completed the ICE task were used as covariates in the analyses, but were not of interest as main effects or as moderators.

Analysis Plan

Differences in emotion recognition accuracy were examined as a function of child CP/CU groups using a series of univariate and multivariate ANCOVAs. Analyses proceeded in two steps. The first step used univariate ANCOVA to examine CP/CU group differences in overall emotion recognition accuracy collapsed across all emotions and across all rounds of assessment. Within this step, models were examined first for main effects, and then examined with the inclusion of interactions between CP/CU groups and child gender, child race, and family income-to-needs ratio. Significant interactions were probed by examining CP/CU group differences separately within demographic groups. The second step used MANCOVA to examine CP/CU group differences for each emotion presentation (happiness, surprise, anger, sadness, and fear) collapsed across all rounds of the ICE task. Within this step, models were examined first for main effects, and then examined with the inclusion of interactions between CP/CU groups and child gender, child race, and family income-to-needs ratio. Significant interactions were probed by examining CP/CU group differences separately within demographic groups. Because of the large number of group contrasts, Benjamini-Hochberg corrections were used to reduce false discovery rates. Benjamini-Hochberg correction uses a stepwise, adaptive procedure by controlling for the expected ratio of the number of erroneous rejections of the null hypothesis to the actual number of rejections, which allows for greater power than techniques that control the familywise Type I error rate, such as the Bonferroni procedure (Benjamini & Hochberg, 2000; Thissen, Steinber, & Kuang, 2002).

Results

Descriptive Statistics

Table 1 presents the bivariate correlations, means, and standard deviations among focal study variables and covariates. Overall accuracy on the ICE task was positively associated with EP behaviors and negatively associated with callous behavior and conduct problems. Conduct problems were negatively associated with child EP behaviors and positively associated with callous behaviors, and the latter two were also negatively correlated. Child race was not significantly associated with overall ICE accuracy or conduct problems, but African American (AA) children were rated as having lower EP behaviors and higher callous behaviors than European American (EA) children. Child gender was not associated with conduct problems or callous behaviors, but female children were rated as having more EP behaviors and had higher overall ICE accuracy scores than male children.

Table 1.

Bivariate Correlations, Means, and Standard Deviations for Central Variables and Covariates

Measure	1	2	3	4	5	6	7	8	9
1. Overall ICE accuracy	–
2. Child EP behaviors	.095**	–
3. Child callous behaviors	−.135**	−.417**	–
4. Child CP	−.099**	−.305**	.467**	–
5. Child racea	.057	−.136**	.087**	.024	–
6. Child genderb	.127**	.146**	−.050	−.044	−.019	–
7. Primary caregiver years of education	.085*	.188**	−.219**	−.136**	−.260**	−.001	–
8. Family mean incomec	.067*	.176**	−.166**	−.132**	−.402**	−.020	.626**	–
9. Child age	.155**	.026	−.008	−.023	−.196**	.022	.037	.074*	–
Mean	.55560	1.9261	.3283	.72	.44	.4994	12.6865	1.8313	7.8998
Standard deviation	.085040	.54007	.39090	2.159	.497	.50028	2.03176	1.37027	.27861
N	893	893	893	893	893	893	893	893	893

Note. ICE = Increasingly Clear Emotions; EP = Empathic-Prosocial; CP = Conduct Problems

^a0 = European American, 1 = African American.

^b0 = male, 1 = female.

^cFamily mean income averaged across visits from the time child was 6 to 58 months old.

^*p < .05.

^**p < .01.

CP/CU Group Differences in Overall Emotion Recognition

Model 1 examined the main effects of demographic variables and CP/CU group membership. Significant main effects were found for CP/CU group, F(2, 753) – 4.12, p = .017, child race, F(1, 753) = 13.93, p < .001, and child gender, F(1, 753) = 11.56, p = .001. Pairwise comparisons showed that children in the typical group had better accuracy than children in the CP+CU group (Cohen’s d = 0.44), whereas children in the CP-only group did not differ significant from either the typical or CP+CU groups. African American children had better accuracy than European American children (Cohen’s d = 0.13) and female children had better accuracy than male children (Cohen’s d = 0.11).

In Model 2, the interactions between CP/CU group and demographic variables were examined. Significant interactions were found between CP/CU group and child race, F(2, 747) = 5.29, p = .005, and between CP/CU group and family income-to-needs ratio, F(2, 747) = 3.59, p = .028, whereas no significant interaction was found between CP/CU group and child gender. We also conducted analyses examining same race bias for the facial expressions and whether this was associated with differences among the CP/CU groups. We found that, for all analyses, the race of the faces presented did not significantly moderate the associations among CP/CU group, child race, and emotion recognition accuracy. Thus, child race, rather than race of faces presented, was key in differentiating between EA and AA children. After trimming the non-significant interactions from the model, the CP/CU group x race interaction remained significant, whereas the CP/CU group x income interaction was reduced to marginal significance, F(2, 749) = 2.85, p = .059. Table 2 presents the adjusted CP/CU group means for overall emotion recognition accuracy separately for EA and AA children, and separately for children with family income-to-needs ratios above and below 1 (reflecting families above and below the poverty line, respectively). Probing the CP/CU x race interaction revealed that, among EA children, children in the typical group scored significantly higher on overall emotion recognition accuracy than children in the CP-only (Cohen’s d = 1.01) and CP+CU groups (Cohen’s d = 0.83), whereas the CP-only and CP+CU groups did not differ significantly from one another. Among AA children, however, there were no significant CP/CU group differences in overall emotion recognition accuracy. Probing the CP/CU x income interaction revealed that, among children with family income-to-needs ratios greater than or equal to 1, children in the typical group performed more accurately than children in the CP+CU group (Cohen’s d = 0.95), whereas children in the CP-only group did not differ from children in either the typical or CP+CU groups. Among children with family income-to-needs ratios below 1, no CP/CU group differences were found.

Table 2.

Adjusted CP/CU Group Means Split by Child Race and Family Income-to-Needs Ratio

	Child Race
	European American (n = 446)			African American (n = 315)
Accuracy	Typical (n = 422)	CP-only (n = 4)	CP+CU (n = 20)	Typical (n = 291)	CP-only (n = 8)	CP+CU (n = 16)
Overall	.558 (.004)ab	.463 (.039)a	.485 (.018)b	.564 (.005)	.567 (.031)	.573 (.021)
Fear	.301 (.008)	.178 (.080)	.227 (.036)	.316 (.010)	.355 (.061)	.309 (.043)
Anger	.704 (.008)	.65 (.088)	.646 (.039)	.747 (.010)	.838 (.058)	.791 (.041)
Happiness	.810 (.007)ab	.637 (.068)a	.730 (.030)b	.816 (.009)	.889 (.052)	.853 (.037)
Sadness	.368 (.009)	.323 (.090)	.320 (.040)	.424 (.010)	.355 (.063)	.366 (.044)
Surprise	.604 (.010)	.524 (.099)	.503 (.044)	.517 (.013)	.396 (.081)	.544 (.057)
	Income-to-Needs Ratio
	≥ 1 (n = 555)			< 1 (n = 206)
Accuracy	Typical (n = 528)	CP-only (n = 6)	CP+CU (n = 21)	Typical (n = 185)	CP-only (n = 6)	CP+CU (n = 15)
Overall	.571 (.004)a	.559 (.032)	.490 (.017)a	.537 (.007)	.489 (.035)	.564 (.022)

Note. Means sharing a common subscript were statistically significantly different using Benjamini-Hochberg corrections for false discovery rates. Covariates include family income, child gender, child age at assessment, and primary caregiver education.

CP/CU Group Differences in Recognition for Specific Emotions

Model 3 used a MANCOVA procedure to assess children’s accuracy within facial emotion expressions. Significant multivariate main effects were found for child race, Λ = .94, F(5, 730) = 10.20, p < .001, and child gender, Λ = .98, F(5, 730) = 3.13, p = .008, but not for CP/CU group. For fear recognition, a marginally significant effect was found for child race, F(1,734) = 3.72, p = .054, but no pairwise differences were found. No significant main effect of child gender was found for fear recognition. For anger recognition a significant main effect was found for child race, F(1, 734) = 14.46, p < .001, with AA children performing more accurately than EA children (Cohen’s d = 0.12), but no significant effect of child gender was found. For happiness recognition, a significant main effect was found for child gender, F(1, 734) = 10.57, p = .001, with girls performing more accurately than boys (Cohen’s d = 0.09), but no significant effect of child race was found. For sadness recognition, a significant main effect was found for child race, F(1, 734) = 18.94, p < .001, with AA children performing more accurately than EA children (Cohen’s d = 0.14), but no significant effect of child gender was found. For surprise recognition, a significant main effect was found for child gender, F(1, 734) = 8.27, p = .004, with girls performing more accurately than boys (Cohen’s d = 0.08), but no significant effect of child race was found.

In Model 4, the interactions between CP/CU group and demographic variables were examined. A significant multivariate interaction between CP/CU group and child race was found, Λ = .97, F(10, 1488) = 2.14, p = .005, whereas the interactions between CP/CU group and child gender and CP/CU group and family income were not significant. A significant univariate interaction between CP/CU group and child race was found for happiness recognition, F(2, 747) = 7.50, p = .001. After trimming the non-significant interactions from the model, the CP/CU x child race interaction for happiness recognition remained significant. Table 2 presents the adjusted CP/CU group means for happiness recognition accuracy separately for EA and AA children. Probing the interaction revealed that, among EA children, children in the typical group performed more accurately than children in both the CP-only (Cohen’s d = 1.24) and CP+CU groups (Cohen’s d = 0.58), whereas the CP-only and CP+CU groups did not differ from one another. However, among AA children, no group differences in happiness recognition accuracy were found.

Discussion

The current study is one of the first to examine emotion recognition accuracy among young children with CP and CU behaviors, and the potential effects of child and family moderators of emotion recognition for these children. Although it was hypothesized that CP/CU group differences in emotion recognition would be found among all children, results indicated these group differences among typical children, children with CP-only, and children with CP+CU were moderated by child race. Specifically, CP/CU group differences in overall emotion recognition accuracy were found only among EA children, with children in the typical group showing better accuracy than children in the CP-only and CP+CU groups, but not among AA children. In addition, CP/CU group differences were found for recognition of happy faces among EA children, with children in the typical group showing better accuracy than children in the CP-only and CP+CU groups, but not among AA children. In addition, moderation by family income was found for children’s overall emotion recognition, with children in the typical group performing more accurately than children in the CP+CU group if their families did not experience extreme poverty, whereas no such group differences were found among children in families experiencing extreme poverty. It should be noted, however, that examination of these moderators relied on small group sizes, so the interpretation of these moderation effects should be considered a preliminary step towards better understanding these processes.

Despite the small group sizes, these findings raise important questions regarding the etiologies of CP and dimensional aspects of CU behaviors and their associated features. In particular, they suggest that there may be important differences in socialization experiences during early- and middle-childhood between racial-ethnic groups, and between children experiencing extreme poverty and children living above extreme poverty that may lead some children to develop CP, as well as the co-occurrence of CP and CU behaviors. However, interpretation of these results should be done carefully, given that the results did not strongly differentiate children in the CP-only group from those in the CP+CU group. Although one of these two groups often differed in emotion recognition from the typical group, they often did not differ from one another. Further, the effect sizes for the CP-only group often exceeded those of the CP+CU group when both differed significantly from the typical group. Thus, these results have implications for both CP and CU behaviors, rather than solely CU behaviors.

The Effects of Early Socialization and Family Income on Emotion Recognition

The biopsychosocial perspective suggests that development of the various aspects of emotional and behavioral functioning occurs through interactions and coactions at multiple levels of analysis; including genetic activity, neural and cognitive activity, behavior, and children’s physical, social and cultural environments (Gottlieb, 2007). Thus, there are likely a number of complex processes that contribute to individual differences in both children’s emotion recognition and their CP and CU behaviors. Among African American families, sociocontextual processes of racial/ethnic and emotional socialization may have important impacts on children’s developing emotional functioning, including emotion recognition. Racial/ethnic socialization refers to verbal and nonverbal information communicated to children (especially from parents) about what it means to be part of a given racial/ethnic group (e.g., African American; Boykin & Toms, 1985; Peters & Massey, 1983)—including messages promoting racial or ethnic pride, preparation for and coping with racial bias, and sometimes mistrust of majority groups (Dunbar, Perry, Cavanaugh, & Leerkes, 2015). Such socialization may be a key way in which ethnic minority families buffer against the effects of discrimination for their children (García Coll et al., 1996). Previous research suggests that racial/ethnic socialization is associated with positive socioemotional outcomes, including greater anger control and less fighting among male adolescents, as well as fewer depressive symptoms among female adolescents (Hughes et al., 2006; Stevenson, Reed, Bodison, & Bishop, 1997). However, these positive socioemotional and behavioral outcomes may not only result from racial/ethnic socialization, but also through effective emotional socialization in African American families.

Dunbar and colleagues (2015) posit that emotional and racial/ethnic socialization by parents may be important for emotion-oriented outcomes, as experiences of discrimination can be emotionally arousing and, thus, likely require effective emotion regulation for coping. As such, there may be differences in emotion socialization processes between EA and AA families that may differentially affect emotional functioning, including emotion recognition. Although there has been limited previous research on similarities and differences among racial/ethnic groups regarding emotion socialization, similarities between EA and AA parents’ emotion socialization have been found. For example, maternal empathy has been positively associated with emotional understanding and negatively associated with aggression among both EA and AA children (Morelen, Jacob, Suveg, Jones, & Thomassin, 2013; Smith & Walden, 2001). On the other hand, punitive and negative emotion parenting behaviors, such as mothers’ minimizing reactions to children’s negative emotions, have been found to be positively associated with adaptive coping among AA girls and negatively associated with aggression among AA boys (Smith & Walden, 2001), whereas they have been positively associated with negative outcomes among EA children (Eisenberg et al., 1999). Furthermore, findings using the current study sample suggest that AA mothers (Garrett-Peters et al., 2008) and fathers (Garrett-Peters et al., 2011) may engage in more emotion talk with their infants than EA parents, and that experiences of racial discrimination predict these higher levels of emotion talk (Odum, Garrett-Peters, Vernon-Feagans, & The Family Life Project Key Investigators, 2016). Such differences in emotion socialization between EA and AA parents may partially explain the differences in emotion recognition found in the current study.

Specifically, AA parents’ potential use of more emotion talk and/or potential use of emotion socialization as part of racial/ethnic socialization may have contributed to AA children’s greater emotion recognition accuracy, as compared to EA children. As a result, AA children may have better emotional competencies, even in the presence of clinically-relevant CP and CU symptoms, as compared to EA children. If the majority of AA children (regardless of clinical status) gained competence in emotion recognition through enhanced emotion socialization experiences, then this might explain why differences in emotion recognition were not found among CP/CU groups of AA children. Alternatively, given that AA children generally performed better than EA children in emotion recognition, it is possible that there were ceiling effects for AA children. A more difficult or complex emotion recognition task may elucidate differences in emotion recognition among CP/CU groups of AA children. Nevertheless, these findings of differential associations with emotion recognition across race suggest that there are likely multiple etiological pathways leading children to develop CP and CU behaviors. It appears possible that, for a subgroup of children, emotion recognition deficits may lead to problematic outcomes. However, for other groups, the emergence of CP and CU behaviors may occur in the absence of any emotion recognition deficits whatsoever. Those children may perceive and recognize facial emotion cues as well as, or better than, children without CP and CU behaviors, yet these cues do not deter them from engaging in aggressive behavior. Rather, the ability to more accurately recognize emotions in others may actually support their CP and CU behaviors by allowing them to exploit the emotions of others. For example, although some research has shown emotional deficits among adults with psychopathic behaviors, some have demonstrated greater accuracy among adults high psychopathic behaviors in identifying people vulnerable to victimization than adults low in psychopathic behaviors (Book, Costello, & Camilleri, 2013; Book, Quinsey, & Langford, 2007; Nagler, Reiter, Furtner, & Rauthmann, 2014; Wheeler, Book, & Costello, 2009). Such varying pathways could help to explain some of the disparate findings from previous studies that have found comparable or greater emotion recognition accuracy among children with CU behaviors compared to other children (Bowen et al., 2014; Pham & Phillippot, 2010; Schwenck et al., 2014; Woodworth & Willoughby, 2008).

The current findings also suggest that family income may be an important modifier of the etiology of CP and CU behaviors. For children experiencing extreme poverty, there may be myriad risk factors that promote the development of CP and CU behaviors (Carter et al., 2010; Shelleby et al., 2014). As such, it is possible that deficits in emotion recognition may not play as large of a role in the development of CP and CU behaviors as compared to other biopsychosocial factors. In contrast, among children not experiencing extreme poverty and the correlated constraints associated with such contextual risk, amygdala dysfunction and/or poor emotion socialization leading to emotion recognition deficits may play a more pivotal role in CP and CU development. More research is needed to better understand the multiple pathways through which children from distinct demographic backgrounds develop CP and CU behaviors, as such pathways may or may not involve deficits in emotion recognition. Person-centered approaches may be useful in delineating unique constellations of risk factors, including whether certain patterns are more common in some demographics than others (Jobe-Shields, Andrews, Parra, & Williams, 2015; Obrien, 2005).

Is the Presence of Callous-Unemotional Behaviors Associated with Increased Emotion Recognition Deficits?

As mentioned previously, children in the CP-only and CP+CU groups often did not perform differently on emotion recognition accuracy in the current study. In addition, the effect sizes for the CP-only group sometimes were larger when both groups differed from the typical group. Thus, the current findings may represent an effect of CP on emotion recognition, rather than an effect of CU behaviors. Some previous studies have found that children and adults with psychopathic and/or CU behaviors showed emotion recognition deficits compared to those with CP-only or antisocial behavior without psychopathy (Bagley et al., 2009; Blair et al, 2005; Blair et al., 2001; Blair et al., 2004; Lemos Vasconcellos et al., 2014; Stanković et al., 2015; Stevens et al., 2001). However, it is possible that those showing the highest levels of CP or antisocial behavior demonstrate emotion recognition deficits and this coincides with psychopathic and/or CU behaviors, but is actually related to CP and antisociality. Therefore, future research should attempt to parse these emotion recognition effects to determine whether they truly related to CP, to CU behaviors, or to their co-occurrence. Moreover, many children showing emotion recognition deficits do not show high CP and CU behaviors. Thus, future research should also attempt to elucidate the developmental mechanisms (e.g., epigenetic, neurological, and social processes) that may link emotion recognition deficits and CP and CU behaviors.

Strengths and Limitations

The strengths of the current study include the use of a large, population-stratified, randomly-selected sample; the assessment of children in middle childhood exclusively; the use of a relatively racially diverse sample; and the use of both EA and AA stimuli in the facial emotion recognition task. As was mentioned previously, many previous studies examining associations between emotion recognition and CU behaviors have used relatively small samples, whereas the current study used a large population-based sample, which may have allowed for better detection of effects and for better generalization of findings (although broad generalization should proceed with caution since the sample reflects rural but not urban American communities) In addition, the current study used a cohort sample of children who were all in 1^st and 2^nd grade at the time of the time the relevant measures were collected. Thus, this is one of only two studies, to our knowledge, that has examined emotion recognition and CU behaviors exclusively among young children. This has allowed for better examination of these associations during early- and middle-childhood, as there are no older children in the sample who are likely qualitatively different from younger children due to their developmental stages. Furthermore, the sample was comprised of approximately 57% EA children and 43% AA children, making it more racially diverse than many other previous samples. Stimuli in the emotion recognition task were also comprised of 50% EA actors and 50% AA actors. As such, we were able to examine race as a moderator of the association between CP/CU group membership and emotion recognition accuracy, while ruling out same-race bias as an explanation of the CP/CU group x child race interaction.

The findings from the current study should also be considered in the context of some limitations. First, it should be noted that when groups were split by child race and child gender to probe the interactions, small group sizes for the CP-only, and CP+CU groups resulted (the smallest being n = 4 for EA children in the CP-only group). Given that the moderating effects found were driven by small group sizes, those results should be considered as preliminary. Future studies oversampling children with CP and CU behaviors are needed to replicate the moderating effects. Additionally, the current study examined associations between emotion recognition and CP and CU behaviors cross-sectionally. The data were collected as part of a large longitudinal study aimed at examining several developmental phenomena in the context of rural poverty, rather than specifically aimed at investigating emotion recognition and CP/CU behaviors. Thus, the current investigation utilized data that were not ideally suited for understanding the direction of effects. Examining emotion recognition and CP/CU prospectively throughout early childhood would allow for better assessment of associations between emotion recognition and CP/CU behaviors over time, and for better understanding of how these psychopathologies emerge. Nonetheless, the current study provides a useful extension of previous studies by examining these associations in young children. Finally, the use of a population-based sample could be considered a weakness, given that we examined a psychopathological phenomenon. However, clinical cutoffs were used to denote high levels of CP and a 90^th percentile cutoff was used to denote CU behaviors. Thus, the children in the CP-only and CP+CU groups can certainly be considered as showing clinical levels of problem behavior.

Conclusion

Understanding the development of CP and CU behaviors in early- and middle-childhood is an important pursuit for developmental researchers, as children showing high levels of such behaviors are at risk for negative psychosocial and psychopathological outcomes. Although emotion recognition deficits have been repeatedly associated with CU behaviors, findings from the current study (along with some previous studies), indicate that such deficits may not be a core feature of CU behaviors for all children. That being said, it will be important for future research to examine the mechanisms through which emotion recognition, as well as other aspects of emotional functioning, are associated with CU behaviors and for which children these mechanisms matter. Given the apparent heterogeneity in both CP and CU behaviors, interventions aimed at preventing and/or reducing these behaviors will benefit greatly from better understanding of these developmental mechanisms.