Of particular significance to emotional competence is emotion knowledge, which comprises a broad range of skills necessary to recognize emotion expressions and to interpret social contexts and behavioral cues (Denham, 1998). Children adept in these skills are better able to regulate their own emotions (Barrett et al., 2001; Eisenberg et al., 2005; Miller et al., 2006; Hudson and Jacques, 2014). They are also far more socially competent. For instance, preschoolers with high emotion knowledge are rated as more likeable by peers, receive more positive nominations from teachers, and are more likely to succeed in school (Izard et al., 2001; Denham et al., 2003, 2012, 2015; Trentacosta and Izard, 2007). Moreover, poor emotion knowledge has consistently been shown to predict internalizing behaviors (e.g., anxiety, depression, and social isolation) and externalizing behaviors (e.g., aggression, hyperactivity, and conduct problems; for review, see Trentacosta and Fine, 2010), both of which come with their own set of negative implications for social competence (Bongers et al., 2008; Bierman et al., 2015; Heinze et al., 2015) and long-term psychological wellbeing (Farrington, 1989; Moffitt, 1993; Betz, 1995; Liu et al., 2011; Thompson et al., 2011). Given these far-reaching consequences, it is imperative to improve upon current understanding of the factors which contribute to poor emotion knowledge and, in turn, decreased emotional and social competence.

Behavioral withdrawal is one factor which may impede upon the development of emotion knowledge, as it often includes decreased environmental interaction, which is vital for normative social, emotional, and cognitive development (Rubin et al., 1998). In regard to emotional development, withdrawn children may learn less information about emotions due to disengagement from or avoidance of social situations. More precisely, decreased social interaction may translate into delays in developing relations between facial expressions, situational activators, and emotion-eliciting experiences (Schultz et al., 2001; Trentacosta and Fine, 2010). The quality of such interactions may also be compromised (McClure and Nowicki, 2001). For instance, social deficits (e.g., gaze aversion) may decrease a child’s ability to gather emotion-related cues from their environment (McClure and Nowicki, 2001). Taken together, these effects of withdrawn behavior may have profound consequences for emotion knowledge.

In particular, withdrawn behavior may impact emotion knowledge by disrupting normative language development. With fewer peer interactions, withdrawn children may miss experiences necessary for reaching language milestones (Guedeney et al., 2016). This is a reasonable assertion given that social interaction is considered by many to be central to language development (Kuhl, 2007; Bruce and Hansson, 2011). Decreased peer interactions, then, may provide one explanation for the commonly observed negative association between childhood withdrawal and both receptive and expressive language (Paul and Kellogg, 1997; Cohen and Mendez, 2009; Spere et al., 2009; Smith-Watts et al., 2014; Curby et al., 2015). This delayed language development may, in turn, have a negative influence on the development of emotion knowledge. According to Conceptual Act Theory (CAT; Barrett, 2006), language not only scaffolds emotion concept knowledge but also allows individuals to later use such concept knowledge to make sense of sensory perceptions in specific contexts (Lindquist et al., 2015). Evidence of a positive association between language and emotion knowledge (Cutting and Dunn, 1999; Schultz et al., 2001; Miller et al., 2006) supports this emotion knowledge-language link proposed by CAT. Cumulatively, these associative ties between withdrawn behavior, language, and emotion knowledge suggest that language deficits would be a likely mediator for poorer emotion knowledge in withdrawn children.

If withdrawn children are more prone to difficulties in understanding the emotional context, they may also struggle to regulate emotions in an adaptive, situationally responsive way (Thompson, 2001; Miller et al., 2006). For instance, they may show affect which is “context-inappropriate (CI; Locke et al., 2009),” such as responding with anger in positive, rewarding situations. Previous studies have demonstrated that children who display CI anger do have worse emotion knowledge than their peers (Locke et al., 2015; Locke and Lang, 2016). Perhaps then, a tendency to behaviorally withdraw would be associated with CI anger responses. In such cases (i.e., where children show both behavioral withdrawal and CI anger), withdrawal may negatively impact emotion knowledge through deficits in language (as we generally suspect for withdrawn children). Alternatively, children who show CI anger may both withdraw and have poorer emotion knowledge because they are not motivated to engage with affective cues in their environment.

This latter explanation may prove viable given that children who show CI anger have decreased hypothalamic-pituitary-adrenal (HPA) axis activity (as indicated by their lower basal morning cortisol levels; Locke et al., 2009), and low cortisol (or hypocortisolism) may serve to impede arousal and engagement with one’s environment (Adam and Gunnar, 2001; Gunnar and Vazquez, 2001). Hypocortisolism has also been implicated in self-regulatory behaviors and executive functions (Blair et al., 2005) that may impact emotion knowledge (Schultz et al., 2001; Denham et al., 2012), above and beyond language ability (von Salisch et al., 2015). Thus, the limited attentional and behavioral control abilities inherent to poor self-regulation may make it difficult for children to learn about emotions through social relationships (Schultz et al., 2001). Considering CI anger’s associations with low cortisol (Locke et al., 2009), attention difficulties (Locke and Lang, 2016), and externalizing behavior (Locke et al., 2015, 2017), poor executive function is likely present in these children. Thus, even with adequate language skills, withdrawn children showing CI anger may still have poorer emotion knowledge than their agemates.

Taken together, these studies suggest that there may be multiple pathways leading to poor emotion knowledge in early childhood. They also highlight various existent gaps in the literature, which upon exploration may serve to further elucidate understanding of this critical component of emotional competence. Regarding the relationship between withdrawn behavior and language, many studies have evaluated the effect of language impairments/deficits on withdrawn behavior, but only a few (e.g., Redmond and Rice, 1998) have considered if withdrawn behavior may impact language, particularly when this construct is broadened to include forms of withdrawal not related to fearfulness/avoidance. Furthermore, to our knowledge, the ability of language to mediate the effect of withdrawn behavior on emotion knowledge has not been evaluated. Secondly, though CI anger has been demonstrated to predict emotion knowledge (Locke et al., 2015; Locke and Lang, 2016), its relationship to withdrawn behavior has not been considered. Finally, exploration of how CI anger impacts the proposed mediation effect is warranted, as other factors (e.g., poor executive function and self-regulation; Blair et al., 2005; Willcutt et al., 2005) may be more relevant than language ability for proper acquisition of emotion knowledge in withdrawn children who show CI anger.

The objective of the present cross-sectional study was to assess how withdrawn behavior, receptive language, and CI anger interplayed to influence emotion knowledge in a non-clinical, community-based sample of preschoolers. Our primary aim was to determine if withdrawn behavior predicted emotion knowledge and if language acted as a mediator in this effect. We expected children’s receptive language would mediate the association between withdrawn behavior and situation EK (but not recognition EK; hypothesis one), as the former requires higher-level cognitive abilities (Denham and Couchoud, 1990; Widen and Russell, 2010; Bassett et al., 2012) and should, thus, be more vulnerable to decreased peer interaction and language delays (Kuhl, 2007; Bruce and Hansson, 2011; Guedeney et al., 2016). Our secondary aim was to assess associations among CI anger and both withdrawn and externalizing behavior. This aim allowed us to attempt replication of the positive association between externalizing behavior and CI anger noted in previous studies (Locke et al., 2015, 2017; Locke and Lang, 2016); it also allowed us to explore novel associations between CI anger and withdrawn behavior. Here, we expected that children who expressed greater CI anger would show more externalizing behavior (hypothesis two). We also expected such children to be more withdrawn than other children (hypothesis three), in part, perhaps because of their inherent hypoarousal and the resultant disengagement with their environment.

Finally, a tertiary aim of this study was to test whether CI anger moderated the relationship between language and emotion knowledge and, further, if our predicted mediation effect of language on the relationship between withdrawn behavior and emotion knowledge would be moderated by CI anger. We expected CI anger would moderate the relationship between receptive language and situational emotion knowledge (hypothesis four). If factors other than language are more pertinent to emotion knowledge development in children who show CI anger, then increasing levels of CI anger may diminish the ability of language to predict situational emotion knowledge. Finally, we expected CI anger would moderate the mediating effect of language on the association between withdrawn behavior and emotion knowledge (hypothesis five). If language ability has less impact on emotion knowledge, then language may not mediate proper emotion knowledge acquisition for withdrawn children showing CI anger. Such children may instead have more salient cognitive deficits (e.g., poor executive function) and affective deficits (e.g., disengagement) that hinder acquisition of emotion knowledge. Consequently, we expected that children who show CI anger may possess attributes which can operate independently of language to impair a child’s ability to understand contextual cues surrounding emotion. For a conceptual model of the proposed moderated mediation analyses based on Hayes (2018), see Figure 1.

Participants were 74 4 and 5 year-old preschoolers (M = 55 months, SD = 6 months; 49% male) from the northeast region of the United States. The community-based, non-clinical sample equally recruited from local public and private preschools was reflective of local demographics (66.7% Caucasian, 6.7% African-American, 1.7% Asian, 3.3% other race, 21.7% multiracial; 16.3% Hispanic/Latino) and economically diverse (annual income: M = $41,190, SD = $37,768; $3,600 ≤ x ≤ $170,000; 52% <poverty level). In prior research, CI anger was not shown by a majority of children (Locke et al., 2009, 2015; Locke and Lang, 2016). Therefore, a portion of the study sample of 74 children (42%) was recruited from a larger screening sample of 149 children based on parent-reported CI anger (described in greater detail within “Materials and Methods”). Overall, the study sample had lower receptive language (M = 99.88, SD = 13.61) than the larger screening sample (M = 105.86, SD = 13.07), t(140) = 2.67, p = 0.01, but the two samples did not differ on externalizing behavior: t(124) = 0.10, p = 0.92; withdrawn behavior: t(124) = 0.18, p = 0.86; or CI anger: t(127) = 0.38, p = 0.71. Parents provided written informed consent, and all procedures were approved by the University Institutional Review Board.

The subsample of 74 children participated in a larger study of emotion, which entailed additional child assessments completed across multiple days at their preschool. Relevant to this study, children were individually interviewed with age-appropriate measures that assessed emotion knowledge and receptive language. Each of these interviews took approximately 20 min to administer. On separate days prior to the emotion knowledge and receptive language interviews, children’s teachers and parents also completed questionnaires to assess child functioning. Families were compensated $40.00, and teachers were compensated $10.00 for their participation.

When we tested the alternative model, withdrawn behavior did not mediate the association between receptive language and situational EK (indirect b = 0.002, SE = 0.004, 95% CI [−0.004, 0.01]). However, although receptive language was significantly associated with lower withdrawn behavior (b = −0.12, SE = 0.05, 95% CI [−0.22, −0.03]), lower withdrawn behavior was not significantly associated with greater situational EK (b = −0.02, SE = 0.02, 95% CI [−0.06, 0.03]). The total effect was significant (b = 0.04, SE = 0.01, 95% CI [0.03, 0.06]), and the direct effect of receptive language on situational EK was significant (b = 0.04, SE = 0.01, 95% CI [0.02, 0.06]).

As expected, children who expressed more CI anger were rated by parents as showing more externalizing behavior than other children (see Table 2)—demonstrating support for hypothesis two. Results also supported hypothesis three. That is, children who expressed more CI anger were rated by their parents as showing more withdrawn behavior (see Table 3).

Per hypothesis four, there was a significant interaction between CI anger and receptive language on situational EK (see Table 4), wherein the positive relationship between receptive language and situational EK was steeper at lower levels of CI anger (see Figure 3). Further, hypothesis five was supported with a significant moderated mediation effect (Index = 0.02, SE = 0.01, 95% CI [0.0002, 0.04]). Specifically, there was a significant indirect effect of withdrawn behavior on situational EK via receptive language when CI anger was absent or low but not when CI anger was moderate (see Table 4).

Receptive language was found to fully mediate the effect of withdrawn behavior on situational EK. Specifically, greater withdrawn behavior predicted poorer receptive language, and poorer receptive language, in turn, predicted poorer situational EK. These findings support previous studies, which have demonstrated an association between language and EK (e.g., Miller et al., 2006; Morgan et al., 2010). They also expand upon the extant literature in two key ways. Firstly, present findings suggest a direction of effect which differs from most prior research (e.g., Smith-Watts et al., 2014). Previous studies have proposed that language deficits may contribute to withdrawn behavior, but only a few have suggested that withdrawn behavior may lead to language deficits (e.g., Evans, 2010; Strand et al., 2011). Those that have often studied only avoidant/fearful children (e.g., Strand et al., 2011). Secondly, our study findings outline an explicit mechanism by which withdrawn behavior may impact EK through language. Given that peer interaction is cardinal to language development (Kuhl, 2007; Bruce and Hansson, 2011; Curby et al., 2015), and language development is cardinal to EK (e.g., Dunn et al., 1991; Izard et al., 2001), the proposed mechanism is justifiable. Moreover, alternative models with reverse directionality (i.e., language deficits impacting EK through withdrawn behavior), were unsupported by the present data set.

The positive association between CI anger and externalizing behavior observed in previous studies (Locke et al., 2015, 2017; Locke and Lang, 2016) was replicated. CI anger was also found to be positively associated with withdrawn behavior. Hence, decreased emotional competence associated with CI anger may be a risk factor for co-current behavior problems. Given that co-occurrence of internalizing and externalizing behaviors has been observed in prior studies (Jackson et al., 2000; Verduin and Kendall, 2003), such findings are not unexpected. Furthermore, there may be several plausible explanations for such co-occurrent maladaptive behaviors in children specifically showing CI anger. For instance, CI anger is negatively associated with EK, and poor EK is associated with both externalizing behavior (Dodge et al., 2002; Garner et al., 2008; Trentacosta and Fine, 2010) and withdrawn behavior (Schultz et al., 2001; Trentacosta and Fine, 2010). Additionally, children showing CI anger are more likely to show hypoarousal (Locke et al., 2009), which may simultaneously predict externalizing behavior (e.g., McBurnett et al., 2000) and disengagement with one’s surroundings (Adam and Gunnar, 2001; Gunnar and Vazquez, 2001).

Although children who show CI anger may be more likely to withdraw, it appears that how that translates to their ability to learn situational EK goes beyond language deficits. Results from our moderation analysis indicated that increasing levels of CI anger significantly attenuated receptive language’s ability to predict situational EK. Additionally, when incorporated into our moderated mediation model, moderate levels of CI anger rendered the mediating effect of receptive language non-significant. One possible explanation is that there may be fundamentally different reasons why children with CI anger withdraw and, thus, different mechanisms by which withdrawn behavior predicts decrements to situational EK. Several researchers conceptualize withdrawn behavior as a heterogeneous construct (Asendorpf, 1990; Rubin and Asendorpf, 1993). Asendorpf (1990), specifically, distinguishes between three forms of withdrawn behavior based on co-current approach and avoidance motivations: shyness (high approach, high avoidance), unsociability (low approach, low-to-moderate avoidance), and avoidance (low approach, high avoidance). It may be the case, then, that children with moderate or higher levels of CI anger who display withdrawn behaviors are unsociable (i.e., have a non-fearful preference for solitude; Coplan et al., 2019). Consequently, such children may lack motivation to engage in and learn contextual cues related to emotion regardless of language abilities. Conversely, shy children who develop poor EK may do so because their fearfulness drives them to avoid social interactions that would allow them to actively learn language skills necessary for the development of emotional knowledge.

Findings from the present study are significant as they both illustrate the intimate interconnections which exist between withdrawal, language, and emotion knowledge and elucidate the nature of CI anger, a relatively unexplored form of emotion dysregulation proposed by Locke et al. (2009). Moreover, our proposed conceptual model was supported, suggesting that there may be several developmental pathways leading to decrements in emotion knowledge. Taken together, these findings may have significant implications for interventions aimed at improving emotion knowledge among preschoolers, as they suggest that for some withdrawn children improving language ability may improve emotion knowledge, while for others alternative factors (e.g., hypocortisolism and resultant low arousal, disengagement, and poor executive function) may need to be addressed. However, though this study and the conceptual model proposed within it do serve to advance the literature in key ways, neither are without limitations.

Firstly, the cross-sectional nature of our design prevents any supposition on temporal precedence. However, as noted in Hayes (2018), the use of mediation analyses should not be considered inappropriate for cross-sectional designs, as inferences about causality are not the product of these mathematical tools but, rather, the product of our interpretation. Consequently, though we must caution readers that there may be other plausible theoretical explanations for our findings, we still advocate that this study and the conceptual model presented within it serve to advance the literature. To improve upon this model further, future research should utilize longitudinal designs and explore additional competing models. The latter would be particularly helpful in clarifying some of the intricacies of how behavioral tendencies impact CI anger (and vice versa). For instance, we found CI anger was associated with withdrawn behavior after controlling for age. This may indicate that temperamental factors associated with CI anger are relevant across preschool ages.

A second limitation concerns the generalizability of our findings. Firstly, we used the PPVT as a proxy for general language comprehension. Although this is in line with other research on emotion knowledge (Cutting and Dunn, 1999; Schultz et al., 2001; Miller et al., 2006), additional research using other language measures would be prudent. Given the commonly observed negative association between childhood withdrawal and both receptive and expressive language, expressive language would be a likely candidate for exploration. Additionally, if vocabulary is integral to forming adequate emotion knowledge, as our study suggests, we advise future research investigate populations with less proficiency, such as those who grow up in bilingual, non-English speaking, or low SES homes (Hua and Wang, 2021). For these children, a clear link between language and social-emotional development has been demonstrated (Washington-Nortey et al., 2022). We would also advocate for a more extensive assessment of the association between emotional competence and CI anger in children growing up with elevated cumulative risk factors in the home, including low SES. Although children living in poverty appear to be more at risk for dysregulated emotion (e.g., Evans, 2003; Raver, 2004), this association may be tempered by factors relevant for CI anger. For example, low-income preschoolers may find it more difficult to regulate anger rather than sadness (Garner and Spears, 2000). Therefore, in studies of early risk factors it would be important to assess regulation of various emotions to capture dysregulation, and we would advocate doing so in various emotional contexts. Our measure of withdrawn behavior may also not generalize beyond parental observation. Knowing parents and teachers observe child behavior in different contexts, we garnered both perspectives. However, most teachers did not endorse withdrawn behavior or focused only on extreme cases. This is not entirely surprising, as other studies have observed internalizing behaviors to be particularly underreported by teachers and affected by biases (Splett et al., 2020; Zee and Rudasill, 2021). This may be because teachers are more aware of externalizing behaviors, given that they are generally more disruptive to the classroom. To combat the limitations of other-rater reports, future research should consider measuring withdrawn behavior through behavioral observation. Finally, our study was limited to children at the older preschool ages, prompting future designs to broaden their scope to include earlier preschool ages that would capture unfolding language and emotion knowledge abilities.

A third limitation of our study is lack of specificity. The present study did not differentiate withdrawn behavior subtypes on the basis of approach and avoidance motivations. Consequently, assertions about differing motives for withdrawn behavior and how they may differentially influence situational emotion knowledge remain speculative. Future research should, thus, aim to elucidate the nature of the moderated mediation effect observed in this study by evaluating withdrawn behavior as a multidimensional construct (i.e., by evaluating each type of withdrawal separately within the context of our conceptual model). Future studies could also include specific aspects of withdrawn behavior (e.g., inhibition and introversion) as covariates within analyses of a broader withdrawal construct to determine if certain aspects of withdrawal are not implicated in poor EK and CI anger. However, to this last point, the existing literature supports the notion that even predominantly inhibited, fearful children tend to have poorer EK than their peers. For instance, studies suggest that inhibited children may misinterpret others emotions, particularly in ways that may promote their fearful responses (Reeb-Sutherland et al., 2015), and those who socially observe do not appear to learn about emotion any better from afar (LaBounty et al., 2017).

Another caveat inherent to not conceptualizing withdrawal as a multidimensional construct is that, while ratings on the CBCL withdrawn subscale can broadly identify a child who behaviorally withdraws from their surroundings, and it commonly serves as an indicator of social withdrawal (e.g., Rubin et al., 2013), it may only be mildly related to fearful withdrawal in children who are behaviorally inhibited (Ballespi et al., 2013). In contrast, it may capture lack of engagement for children who are more unemotional (Benesch et al., 2014). Callous-unemotional traits are more likely seen in aggressive children (Kahn et al., 2012). Consequently, when accompanied by low arousal, greater withdrawal on the CBCL may indicate lack of engagement. It is therefore important to consider other factors in addition to behavioral measures of withdrawal. Taken with our past findings on hypoarousal, our findings would suggest that callous-unemotionality would be a likely candidate for clarifying the greater withdrawn behavior seen in children showing CI anger. Additionally, it may be beneficial to further explore biological correlates (e.g., HPA-axis activity, cardiac regulation, prefrontal cortex-amygdala neurocircuitry) of CI anger in order to improve understanding of the physiology underlying withdrawn behavior in children who show CI anger versus children who do not. Other cognitive and affective factors which may serve to inform our observed moderated mediation should also be investigated. Thus, in addition to exploring physiological correlates and specific withdrawn behavior motivations, additional factors which may reasonably lead to poverty in emotion knowledge should be explored. Considering CI anger’s relationship with externalizing behavior and its shared cortisol profile of attenuated HPA-axis activity (Locke et al., 2009), affective and cognitive factors specific to externalizing behavior may be particularly worth studying.

A fourth limitation surrounds our measurement of emotion knowledge. We propose specific affective and social cognitive factors that could be particularly salient for CI anger given our findings. However, there are specific affective tasks that should be more relevant for CI anger than others. For example, during the non-stereotypical version of the AKT puppet task (Denham, 1986) the child’s atypical emotions are enacted during the vignettes, but only common emotional choices are portrayed in the task (e.g., “happy” or “sad” to come to day care). For future studies in this area, it would be worthwhile to have puppets enact uncommon emotions (e.g., “angry” to come to daycare). Further, rather than measuring overall accuracy in emotion knowledge, it may be important to also consider biased emotion knowledge. For children who show CI anger their anger may stem from misinterpreting others’ intentions as hostile in otherwise ambiguous situations (hostile attribution bias; Dodge and Coie, 1987). This tendency to see more hostility in others may have come about in part because they misread others’ emotional expressions as angry, as seen in more aggressive children (Fine et al., 2004; Garner et al., 2008). Longitudinal studies that capture some of these factors could provide some insight on the nature of CI anger and how it relates to affective processing.

Finally, exploration of our study variables (i.e., withdrawal, CI anger, receptive language, and situational emotion knowledge) in pediatric clinical populations at high risk for poor social functioning is warranted. These may include, but are not limited to, children with externalizing disorders [e.g., attention deficit hyperactivity disorder (ADHD), disruptive disorders] and/or internalizing disorders (e.g., social anxiety, depression), as well as those who commonly present a combination of internalizing and externalizing behaviors [e.g., autism spectrum disorder (ASD)]. These children would be important to consider given the possible deficits in language, emotion knowledge, and emotion regulation associated with these disorders (Philip et al., 2010; Jaffee, 2017; Beauchaine and Crowell, 2020; Northrup et al., 2021). Moreover, evaluating such constructs more broadly across disorders may serve to not only inform our understanding of various developmental trajectories which lead to decreased emotional competence, broader affective social competence, and maladaptive behaviors but may also ultimately serve to support the development of transdiagnostic treatment approaches.