Authors: Sejal Mistry-Patel (1Department of Psychological and Brain Sciences, Texas A&M University), Tristin Nyman-Mallis (2Kennedy Krieger Institute and Johns Hopkins University School of Medicine), Jessica M. Dollar (3Departments of Kinesiology and Psychology, University of North Carolina at Greensboro), Jeffrey R. Gagne (4Department of Educational Psychology, Texas A&M University), Rebecca J. Brooker (1Department of Psychological and Brain Sciences, Texas A&M University)
Categories: Article
Source: Developmental psychobiology
Doi: 10.1002/dev.22545
Authors: Sejal Mistry-Patel, Tristin Nyman-Mallis, Jessica M. Dollar, Jeffrey R. Gagne, Rebecca J. Brooker
Temperamental characteristics and emerging cognitive control are meaningful predictors of children’s development of adaptive and maladaptive social behaviors during the preschool period. However, knowledge of the interplay of these pathways, when examined concurrently to highlight their individual contributions, is limited. Using a cross-sectional sample of three-year-old children, we examined parent-reported discrete traits of negative (anger, fear, sadness, shyness) and positive (low- and high-intensity pleasure) temperamental reactivity as predictors of children’s prosociality and physical aggression. Further, we tested whether the effects of discrete temperament were moderated by cognitive control, as indexed by the N2 event-related potential, during a go/no-go task. Analyses focus on a subsample of children with an observable N2 (n = 66). When controlling for other relative temperament traits, several significant main effects emerged. Moreover, at low cognitive control (smaller N2), fear was negatively associated with aggression, while at high cognitive control, sadness was positively associated with aggression. Heightened anger was linked to reduced prosocial behavior when cognitive control was low but linked to greater prosocial behavior when cognitive control was high. The results highlight that discrete temperament traits predicts individual differences in child outcomes but that associations depend on concurrent levels of cognitive control.
Temperament is defined as biologically-based differences in reactivity and self-regulation (Rothbart & Bates, 2006) that underlie individual differences in propensities to experience and express basic emotions (Goldsmith et al., 1987; Shiner et al., 2012). Early temperament characteristics predict the development of children’s socioemotional characteristics including prosociality, or voluntary behaviors that benefit others [empathy, cooperation, sharing] (Eisenberg et al., 2006, 2013), and aggressive behaviors (e.g., Rothbart et al., 1994), or behaviors that lead to bodily harm of others [kicking, punching, hitting] (Crick et al., 2006). Prosocial and aggressive behaviors are critical elements of long-term adaptive (Caprara et al., 2000; Jones et al., 2015) and maladaptive (Brennan et al., 2012; Campbell et al., 2006; Loeber et al., 2013) social and academic outcomes.
Although associations between temperament and socioemotional outcomes are well-known, relatively little research has considered concurrently testing discrete forms of temperamental reactivity as facets of children’s early development, which can reveal the role of each domain above and beyond variance that overlaps with other temperament constructs (e.g., Dollar et al., 2023). Further, even less work has examined how intra-level interactions, between domains of temperamental reactivity and regulation, may promote adaptive or maladaptive outcomes. The absence of this work is problematic because regulatory components of temperament such as cognitive control develop early and play an important role in shaping the associations between temperament and socioemotional outcomes (Henderson, 2010). As such, an absence of work delineating independent and interactive contributions of temperament domains limits our understanding of which pathways are the most salient for children’s development of prosocial and aggressive behaviors.
To address these gaps in the literature, we used two samples of three-year-old children to test direct associations between multiple discrete forms of temperamental reactivity (anger, fear, sadness, low-intensity pleasure, high-intensity pleasure, and shyness) and children’s prosociality and physical aggression. We simultaneously examined interactions between discrete temperament domains and cognitive control, indexed by the N2 event-related potential (ERP), as pathways to children’s prosociality and aggression. Testing multiple pathways in one model allowed us to tease apart associations that overlap in emotional valence and arousal, while concurrently identifying the degree to which pathways are shaped by cognitive control. We focused on the preschool stage given that this sensitive period is marked by rapid neural changes related to emotional and cognitive development (Bell et al., 2019; Hoyniak et al., 2018; Kopp, 1982; Tottenham et al., 2011).
Like classic models defining emotion through domains of valence and arousal (Barrett, 1998; Lang et al., 1993), temperament reflects individual differences in reactivity, including the speed and intensity of children’s expression of the primary emotions in response to environmental cues (Goldsmith et al., 1987; Rothbart & Derryberry, 1981; Shiner et al., 2012). Characterized in this way, temperament-based emotion propensities predict children’s long-term socioemotional development (Brooker et al., 2016; Eisenberg et al., 2015; Sanson et al., 2010). Negative and positive emotionality comprise the broadest domains of child temperament. Negative emotionality, reflecting a tendency to experience more frequent and intense levels of anger/frustration, sadness, fear, and shyness (Putnam et al., 2001) positively predicts impulsive, aggressive behaviors and internalizing problems (Eisenberg et al., 2000, 2009; Lipscomb et al., 2012; Sanson et al., 2004). Children who experience greater levels of negative emotionality may be less equipped to identify their own emotions, leading to a dampened ability to identify and appropriately respond to others’ emotions (Shin et al., 2011). As a result, children high in negative emotionality also frequently experience conflict between managing their distress and the needs of others (Eisenberg et al., 2006) thereby limiting prosocial behaviors.
In contrast, the broad temperamental domain of positive emotionality is linked to higher levels of empathy, prosociality, and peer acceptance (Eisenberg et al., 1996; Spinrad & Eisenberg, 2017; Volbrecht et al., 2007). Children who are prone to positive emotionality experience more frequent instances of high-intensity pleasure, smiling, and laughing (Putnam et al., 2001). Greater displays of positive affect are associated with initiating social interactions and maintaining friendships (Shin et al., 2011). At the same time, more extreme levels of positive emotionality are linked to a heightened approach to novelty and risk-taking (Dollar et al., 2023; Hane et al., 2008), suggesting a nonlinear association between positivity and adaptive outcomes. Risk for problematic outcomes, primarily physically aggressive behaviors, appears most salient when the temperament characteristic of high-intensity positivity co-occurs with low cognitive control and emotion regulation deficits (Degnan et al., 2011; Kochanska et al., 2000; Putnam & Stifter, 2005).
Notably, the overall pattern of association between positivity and child outcomes, specifically under what conditions seemingly overlapping constructs predict different outcomes, is not entirely clear. One explanation for this ambiguity may be the tendency to collapse discrete facets of temperament into broader dimensions. For example, low-intensity pleasure is frequently collapsed with high-intensity pleasure to broadly represent temperamental propensities for positive emotion. However, low-intensity pleasure appears to be distinct from high-intensity pleasure, for example, in its stronger relation with cognitive control processes (Rothbart et al., 2001). In this way, combining specific temperament characteristics into broad constructs like positive and negative emotionality ignores the unique properties that distinguish each discrete emotion (e.g., anger vs fear) that may be useful for further specifying pathways of children’s social development (Buss & Goldsmith, 1998; Campos et al., 1989; Goldsmith et al., 1987; Goldsmith & Campos, 1982; Rothbart & Bates, 2006; Shiner et al., 2012). When discrete temperamental domains are discussed in the literature, work most frequently highlights propensities for anger, fear, sadness, pleasure, or shyness.
Anger is believed to reflect persistence when a goal has been blocked or interrupted (Campos et al., 1989; Izard, 1991). Anger-prone children, relative to children who express less anger, are at a significantly greater risk of developing aggressive behavior problems and show fewer basic skills in many aspects of social competence [i.e., communication and interpretation of social information] (Brooker et al., 2014; Dollar & Calkins, 2019; Eisenberg et al., 2009; Nwadinobi & Gagne, 2020). Children prone to anger also experience challenges in developing social skills that facilitate peer relationships, often becoming too emotionally aroused to interpret and respond to social cues accurately (Dollar & Calkins, 2019). Specifically, observed anger in kindergarteners positively predicts later oppositional defiance and conduct problems (Hernández et al., 2015; Nozadi et al., 2015). These children are often rejected by their peers, which likely exacerbates negative social experiences and limits positive prosocial opportunities (Dougherty, 2006; Hernández et al., 2015; Valiente et al., 2012). Indeed, anger proneness as early as child age 2 is associated with more physical aggression (e.g., violence, defiance) and less prosociality (e.g., empathy, helping) at school entry (Sirois et al., 2019). This accumulation of evidence suggests that greater levels of anger predict more physical aggression and less prosocial behavior across childhood.
Fear is thought to surface when individual safety is perceived to be threatened (Izard, 1991). Fearfulness in early childhood is linked to fewer delinquent and aggressive behaviors (Colder et al., 2002; Penela et al., 2012; Rothbart & Bates, 2006) and is positively associated with social reticence (Buss, 2011; Duchesne et al., 2010; Gartstein et al., 2012; Rubin et al., 2009). The positive link between early fear and social withdrawal is particularly salient for children who experience intense fear in non-threatening situations (Brooker et al., 2016; Buss, 2011; Buss & McDoniel, 2016). As fear is linked to less social engagement (Penela et al., 2012; Walker et al., 2015), greater levels of fear are likely to limit opportunities for prosocial behavior. However, it is important to note that there is limited evidence on the direct influence of fear on prosociality as fear is commonly collapsed in broader dimensions of negative emotionality.
Sadness is defined as a lowered mood associated with experiences of loss or disappointment and may reflect “giving up” in response to a blocked goal (Izard, 1991). Elevated sadness is associated with early manifestations of difficulties in aggressive behaviors and social withdrawal (Gartstein et al., 2012), which is believed to impede prosocial engagement (Eisenberg et al., 2001). However, relations between sadness, aggression, and prosociality may vary according to the degree to which children are well-regulated (Posner & Rothbart, 2000) as well as expressions of empathy and sympathy (Rothbart et al., 1994). Specifically, children prone to sadness may show a predisposition to empathy, which is a central motivator to engage in prosocial behaviors (Eisenberg, 2006). Similarly, higher levels of sympathy are linked to more direct (e.g., hugging) and indirect (e.g., getting their mother’s attention for assistance) prosocial behaviors (Edwards et al., 2015), speaking to the utility of investigating potentially different roles of sadness in the development of aggression and prosociality.
Pleasure is enjoyment derived from varying degrees of stimulation-seeking (Gartstein & Rothbart, 2003). As previously noted, low- and high-intensity pleasure are viewed as constructs of positive emotional reactivity (Rothbart et al., 1994) but are linked to distinct developmental outcomes. Children who derive more high-intensity pleasure, or greater enjoyment from high stimulating activities, are at a greater risk for socioemotional problems (Gartstein et al., 2012; Oldehinkel et al., 2004), including difficulties with peers (Kozlova et al., 2020) and aggression (Dollar et al., 2023). In contrast, low-intensity pleasure, which involves greater enjoyment in less stimulating activities, protects against the development of behavioral problems (Gartstein et al., 2012; Kozlova et al., 2020). Children who find pleasure in low-stimulating situations have a larger capacity to self-comfort at an earlier age (Aureli et al., 2015), show greater attentional skills (Slobodskaya et al., 2020), and have higher levels of cognitive control (Shin et al., 2023). These associations may manifest as protection from emotional dysfunction into middle childhood (Dollar et al., 2023) and enhancements in prosocial behaviors (Eisenberg & Fabes, 2006).
Shyness is characterized by inhibited behavior and preoccupations with negative self-evaluations self in response to social interactions (Melchior & Cheek, 1990, though see also Schmidt & Poole, 2019). Shyness appears to be a particularly important pathway for the development of maladaptive child outcomes, such as poor school adjustment, peer rejection, avoidance of social behavior, and anxious symptoms (Coplan & Arbeau, 2008; Hassan & Schmidt, 2022; Karevold et al., 2011; Rubin et al., 2009). Shy children may be especially challenged by novel social contexts, showing reticence to interact with others (Eisenberg & Spinrad, 2014) and thereby reducing opportunities to engage in aggressive and prosocial behaviors (Beier et al., 2017; Endedijk et al., 2015). Other work has noted an absence of a direct link between shyness and prosociality (Grady & Hastings, 2018) and suggested that childhood shyness is may be socially adaptive in certain contexts (Poole & Schmidt, 2019).
It is unclear whether discrete temperamental traits remain predictors of children’s socioemotional development when investigated within the context of one another. This ambiguity exists because the predictive value of each association may vary due to shared valence and arousal among discrete emotions, which are not accounted for in correlations or when collapsing discrete emotions into broader dimensions. For instance, although heightened anger and high-intensity pleasure predict more physical aggression, it is unclear whether greater high-intensity pleasure would incrementally predict more aggression above and beyond that predicted by anger. Testing multiple discrete traits in one model offers the advantage of dissecting the specific pathways that are the most meaningful for children’s social development beyond broader facets of temperament, thereby increasing the potential precision and impact of targeted interventions.
Cognitive control is a self-regulatory component of temperament that includes the ability to inhibit a prepotent response to execute a subdominant response (Rothbart et al., 2003). From a psychobiological perspective, cognitive control reflects the modulation of arousal, particularly the downregulation of negative emotion (Rothbart et al., 2004). In children, higher cognitive control is correlated with enhanced emotion regulation (Hudson & Jacques, 2014) and positive developmental outcomes (Eisenberg et al., 2007). As cognitive control emerges later in life than do propensities for discrete emotions (Izard et al., 2011) and given the overlap of cognitive and affective systems (Bell & Wolfe, 2004), cognitive control may interact with earlier-developing domains of temperament to predict individual differences in children’s socioemotional outcomes.
Beginning in early childhood, cognitive control can be indexed at the level of neural activity using scalp-recorded event-related potentials (ERPs). The N2 is a negative waveform that is visible in children approximately 400 ms after the presentation of a stimulus that requires activation (e.g., button press) and inhibition (e.g., withholding a button press). As the N2 is larger to stimuli that cue inhibition (Nieuwenhuis et al., 2003), functional interpretations of the N2 include the denotation of response inhibition (Hoyniak & Petersen, 2019), executive function (Espinet et al., 2012), emotion regulation (Lewis et al., 2006), response conflict (Groom & Cragg, 2015), and attention (Buss et al., 2011). Interpretations of N2 may also vary across tasks (Lo, 2018; Todd et al., 2008); for example, a larger N2 following a negative mood induction may reflect the recruitment of emotional regulatory processes (Lewis et al., 2006) while enhanced N2 in a non-emotional go/no-go task likely indexes broader cognitive control mechanisms (Lamm et al., 2014). Recognizing these differences, we define N2 as an index of cognitive control in the current study.
In both children and adults, the N2 is linked primarily to activity in medial-frontal areas of the brain, though it is common to see posterior N2 activity in young children (Ciesielski et al., 2004; Jonkman, 2006), likely due to the relative immaturity of the prefrontal cortex (Durston et al., 2002). The sharpening of cognitive control skills from early to middle childhood is reflected by changes in N2 amplitude (decreasing amplitudes), distribution (fewer brain regions recruited during response inhibition), and latency (faster N2 responses; Best & Miller, 2010; Durston et al., 2002; Hoyniak, 2017; Hwang et al., 2010; Tamm et al., 2002). Despite these developmental changes in N2, we conceptualize N2 to have equivalent functional significance across childhood.
Meta-analytic work with young children links a smaller N2 (i.e., more positive) to a greater risk for impulsive and aggressive behaviors (Hoyniak & Petersen, 2019). Moreover, children diagnosed with Attention Deficit/Hyperactivity Disorder (Chen et al., 2021; Yong-Liang et al., 2000) show smaller N2 amplitudes, suggesting diagnosis-related impairments in cognitive control networks associated with response inhibition. In contrast, a larger N2 (i.e., more negative) is linked to a stronger ability to curb behaviors, denoting better response inhibition and self-regulatory capacities (Lewis et al., 2006).
However, underscoring the importance of considering other temperament characteristics, enhanced N2 may be maladaptive for children with hyper-monitoring tendencies (Hoyniak & Petersen, 2019). Shy children with a large N2 are more likely than children with a small N2 to exhibit social-emotional maladjustment (Henderson, 2010). Moreover, links between fearfulness and anxious behaviors are stronger at greater levels of N2 activation (Lahat et al., 2014; Lamm et al., 2014; Thai et al., 2016), highlighting an elevated risk for poor social outcomes at high, or potentially exaggerated, levels of cognitive control.
An imbalance of between levels of approach (e.g., high-intensity pleasure) and self-regulatory processes (e.g., cognitive control) is a critical predictor for the subsequent emergence of aggressive and disruptive problems (Jonas & Kochanska, 2018), as well as social reticence (Hoyniak & Petersen, 2019). Children may be more susceptible to behavioral problems and psychopathology when levels of cognitive control are either too low or too high, as opposed to moderate (Murray & Kochanska, 2002). This body of work underscores the nuanced contributions of cognitive control for children’s psychosocial adjustment (Buzzell et al., 2020; Choe, 2021) and highlights that the role of early cognitive control should be interpreted in the context of individual differences in emotion reactivity.
Using a cross-sectional sample of three-year-old children, we examined the associations between discrete domains of temperamental reactivity, within the context of one another to control for overlapping properties of valence and arousal, and children’s aggressive and prosocial behaviors. Moreover, we tested whether cognitive control, indexed by N2, moderated these associations.
Based on patterns of findings in the broader literature (Buss et al., 2011; Henderson, 2010; Lahat et al., 2014; Lamm et al., 2014; Thai et al., 2016), we hypothesized that anger and high-intensity pleasure would be positively associated with aggression and negatively associated with prosociality; we predicted that this association would be especially visible at low cognitive control (i.e., smaller/more positive N2). We also hypothesized that greater sadness would predict more aggression, visible at low cognitive control, but also more prosociality, visible at high cognitive control (i.e., larger/more negative N2). We hypothesized that greater low-intensity pleasure, fear, and shyness would predict less physical aggression and more prosociality, and these associations would be strongest at high cognitive control.
The participants for the present study were drawn from two cross-sectional samples of three-year-old children who completed the same cognitive control task in our laboratory.
Participants for Sample 1 were drawn from a study examining self-control and school readiness in preschoolers. Community families were recruited from the sister cities of College Station and Bryan, Texas via emails on a university campus; fliers shared with families through local daycares, preschools, children’s programs, gyms, and churches; and advertisements on social media platforms. Preschoolers (n = 69, 58% female) were approximately 3 years of age at the laboratory visit (Mage = 3.10, SDage = .09, Range = 2.89 – 3.31).
Families were eligible to participate if their child was within one month of their third birthday, typically developing, and English-speaking. Eligible families were invited to complete a series of online questionnaires about themselves, their partner, and their child’s behaviors. Immediately after completing the online survey, families were scheduled for a laboratory visit; 58 of the 69 families who completed the online surveys participated in the laboratory portion of the study. Of the eleven families that did not participate in the laboratory portion, 5 were unreachable, 4 could not participate because the laboratory was in the process of moving, and 2 declined. Upon arrival at the laboratory, preschoolers engaged in a series of behavioral and cognitive assessments while physiological data were collected. Families received a 50 for the laboratory visit.
Participants for Sample 2 were drawn from a larger, longitudinal study of emotional development in early childhood. Community families were recruited from the Bozeman, Montana area through mailings based on local birth records, fliers, newspaper and media advertisements, in person recruitment, and via word-of-mouth. Families were eligible to participate if their child was three and a half years of age, typically developing, and had no history of neurological impairment in their immediate family. Preschoolers participated in the initial laboratory visit near their 3.5-year birthday (n = 108; 57% female; Mage = 3.59, SD = .15, Range = 3.13 – 3.99). Parents, and co-parents when applicable, independently completed questionnaire packets before the laboratory visit. Upon arrival at the laboratory, preschoolers engaged in a series of behavioral and cognitive episodes while physiological data were recorded. Families received 30 for the laboratory visit.
Across both samples, most mothers (91.3%) and fathers (94.0%) identified as White; 1.1% of mothers and 1.2% of fathers identified as Black or African American; 3.3% of mothers and 2.4% of fathers identified as Asian; 1.1% of mothers identified as American Indian or Alaska Native; 1.1% of mothers identified as other; 2.7% of mothers and 2.4% of fathers identified as biracial. About eight percent (7.7%) of mothers and three percent (2.9%) of fathers identified as Hispanic or Latino.
Of those parents who reported gross annual income, 6.6% reported earning less than 20,001–30,001–40,001–50,001–60,001–70,001–80,001–90,001 or greater.
Given study hypotheses, the current report focuses on parent-reported temperamental reactivity and social behaviors and electroencephalogram (EEG) data collected during a modified go/no-go paradigm. Although behavioral observations of temperament were part of both laboratory assessments, parent-report measures were used for the current project given our interest in discrete dimensions of temperament, which are historically attributed to parental report rather than observations (Goldsmith & Gagne, 2012; Zentner & Bates, 2008) and because not all temperament domains were targeted in the observational assessments.
Primary caregivers reported their child’s affective behaviors during the previous six months via the Children’s Behavior Questionnaire (CBQ; Rothbart et al., 2001) in Sample 1, while mothers and fathers completed the CBQ Short Form (CBQ-SF; Putnam & Rothbart, 2006) in Sample 2. The CBQ is a 195-item measure and the CBQ-SF is a 94-item measure, derived from the full measure, that asks caregivers to rate, on a 7-point Likert scale (0 = Never, 7 = Always) the degree to which certain behaviors are characteristic of their child. Items were averaged to create Anger/Frustration (e.g., gets angry when told to go to bed), Fear (e.g., afraid of the dark), Low-Intensity Pleasure (e.g., likes being sung to), High-Intensity Pleasure (e.g., likes rough and rowdy games), Sadness (e.g., upset when relatives are getting to leave after a visit), and Shyness (e.g., shy around people they have known for a long time) scales. In Sample 1, all scales demonstrated adequate reliability (Anger/Frustration α = .80, Fear α = .48, Low-Intensity Pleasure α = .73, High-Intensity Pleasure α = .79, Sadness α = .46, Shyness α = .92).
In Sample 2, scales between mothers and fathers were then mean-composited. Parent ratings were significantly correlated and reliability was acceptable for all scales (Anger [r (69) = .36, p < .01, α = .80], Fear [r (69) = .45, p < .001, α = .70], Low-Intensity Pleasure [r (69) = .47, p < .001, α = .80], High-Intensity Pleasure [r (69) = .39, p < .001, α = .74], Sadness [r (69) = .31, p < .01, α = .63], and Shyness [r (69) = .73, p < .001, α = .90]). Raw scale values were transformed into z-scores for analyses given the differences in the number of scale items in CBQ and CBQ-SF, though scales from both questionnaires demonstrate equivalence (Putnam et al., 2014).
In Sample 1, primary caregivers completed the Children’s Social Behaviors-Parent Report (CSB-P; Crick et al., 2006). The CSB-P is a 13-item measure that asked caregivers to rate, on a 5-point Likert scale (1 = Never True, 5 = Always True) the degree to which physically aggressive, relationally aggressive, and prosocial behaviors were characteristic of their child. Analyses focused on scales comprising physical aggression (e.g., pushes other kids) and prosocial behavior scales (e.g., kind to other kids). Four items were averaged per scale and each scale demonstrated good reliability (α = .82, α = .76, respectively).
In Sample 2, families completed the parent version of the MacArthur Health and Behavior Questionnaire (HBQ-P; Armstrong & Goldstein, 2003), an assessment designed to collect information about children’s behaviors between ages 4 and 8. The HBQ-P is a multi-domain, 175–item instrument that asked parents to rate, on a 0–2 scale (0 = never true, 2 = often or very true) the degree to which behaviors were characteristic of their child in the last 6 months. Analyses focused on overt hostility (e.g., gets in many fights) and prosocial behavior (e.g., considerate of other’s feelings) scales, comprising averages of 4 and 20 items, respectfully. Scales between mothers and fathers were mean-composited, given acceptable internal consistency and correlations of parent ratings (overt hostility [r (65) = .22, p = .08, α = .61] and prosocial behavior [r (66) = .51, p < .001, α = .91]). Although CSB-P and HBQ-P are conceptually similar measures for assessing aggressive and prosocial behavior, scales included different metrics and numbers of items. To increase comparability, we transformed raw scale values to z-scores for analyses.
In both samples, children completed an identical modified go/no-go task during their visit to the laboratory (Brooker, 2018). Prior to the start of the task, an experimenter provided instructions on how to play the game using laminated pictures of asteroids and spaceships. Children were instructed to press the button in front of them when they saw an asteroid (go stimulus) but to “wait” if they saw a picture of a spaceship (no-go stimulus). Once children could demonstrate that they understood the game, they completed two computerized practice blocks of 10 trials each. Stimuli were presented vertically and at the center of a 23 in. computer screen using Presentation stimulus delivery software (Neurobehavioral Systems; Berkley, CA). Feedback was only provided during the practice blocks.
The task included two experimental blocks of 40 trials for a maximum of 80 trials per child. Trials were pseudorandomized such that roughly 60% of trials were go trials. Each trial was initiated with a 200 ms fixation cross at the center of the monitor, followed by the presentation of the stimulus for 1200 ms. A fixation cross was then presented again for 300–800 ms prior to the subsequent trial. To equate task difficulty across participants, an automated procedure decreased stimulus presentation time by 50 ms following two consecutive correct responses and increased presentation time by 50 ms following two consecutive incorrect responses. Children were reminded to respond quickly and were rewarded with a sticker after the completion of each block.
In Sample 1, EEG data were collected using an ActiChamp amplifier system and ActiCaps (Brain Products, Gliching Germany). EEG was recorded through 32-channel cap-based electrodes arranged according to the 10–20 labeling system. Data were sampled at a rate of 500 Hz and were referenced to Fp1 during recording. In Sample 2, EEG data were collected using a BioSemi Active 2 recording system (Cortech Solutions, Wilmington, NC). Continuous EEG were recorded through a 64-channel cap using Ag-AgCl-tipped electrodes arranged according to the 10–20 labeling system. Passive electrodes were placed at the outer canthi of the left and right eye to identify eye movements and on the mastoids for later re-referencing. Data were sampled at a rate of 2048 Hz. Data were referenced to the Common Mode Sense and Driven Right Leg electrodes during recording.
Offline, all electrodes were re-referenced to the whole head average, high-pass filtered at 0.1 Hz and low-pass filtered at 20 Hz. Correct and incorrect trials were segmented (−200 to 800 ms) and baseline corrected for 200 ms prior to the response. Artifacts were identified using an automated procedure when one of the following criteria were a difference of 150 μV within 200 ms, a voltage step of more than 75 μV between data points, amplitudes below .05 μV within a 50 ms period, or activity that exceeded +100 μV or −100 μV. ERPs were averaged within conditions (go and no-go) for each participant and were exported for principal components analysis (PCA).
For each sample, a temporospatial PCA was conducted to empirically identify temporally and spatially distinct components consistent with N2 (Dien, 2012) using the ERP PCA Toolkit, version 2.83 (Dien, 2010). Samples were analyzed separately given a small but significant age difference between participants (t(164) = 22.42, p < .001) and known changes in N2 timing and localization with development (Buss et al., 2011; Hoyniak, 2017; Jonkman et al., 2007). First, a temporal PCA was used for the individual averages of trial type (go and no-go) across 32 electrodes that overlapped across both samples. Electrode sites were identified as variables while participants, trial types, and temporal components were identified as observations. A Promax rotation was used to rotate the simple structure in the temporal domain (Dien et al., 2007; Dien, 2010).
To identify temporal components, we used a parallel analysis (Horn, 1965) to compare the scree of the present data against the scree of a completely random data set (Cattell, 1966); this yielded 13 temporal components that accounted for a greater proportion of the variance than those generated by the random data set. Then, a separate spatial PCA was applied for the 13 temporal components (Dien et al., 2007; Dien, 2010). Electrode sites were identified as variables while participants, trial types, and temporal components were identified as observations. Infomax rotation was used to rotate the components to independence in the spatial domain (Dien, 2010). A parallel test via scree plot was used to identify spatial components, yielding five spatial components. Results from the temporospatial PCA revealed 13 temporal and 5 spatial components for each temporal component, resulting in 66 total component combinations with the grand average. Only those temporal components that accounted for more than .5% of the total variance were included in further analysis, yielding 22 components (Dien, 2010; Kayser & Tenke, 2003). Twenty-two component combinations were statistically evaluated using an Analysis of Variance (ANOVA) to identify components that significantly differentiated go and no-go trial type activity; 4 components were significant at p < .05.
One of the four significant components, peaking from 382–384 ms post-stimulus (p < .001), was visually inspected according to a priori expectations about the temporal and spatial components of the N2 in Brain Vision Analyzer (BVA; Brain Products, Gliching Germany). Waveforms showed a negative deflection and a topographic map reflecting negative amplitudes at Pz during no-go trials, compared to go-trials, between 360 to 520 ms. We also examined the grand average waveforms and topographic maps of nearby parietal sites, P3 and P4, to determine whether we would see a traditional N2 effect (more negative no-go trial amplitudes relative to go trial amplitudes). Indeed, we found N2 negativity at the lateralized posterior sites, confirming that a multi-site cluster of P3, Pz, and P4 would best represent N2 observed in Sample 1. Mean amplitudes from P3, Pz, and P4 sites were averaged in the 360 and 520 ms time window and exported for analysis.
The same steps used for PCA in Sample 1 were applied to Sample 2. We used a parallel analysis to compare the scree of the present data against the scree of a completely random data set, yielding 20 temporal components that accounted for a greater proportion of the variance than those generated by the random data set. Then, a separate spatial PCA was applied for the 20 temporal components. Electrode sites were identified as variables while participants, trial types, and temporal components were identified as observations. Infomax rotation was used to rotate the components to independence in the spatial domain. A parallel test via scree plot was used to identify spatial components, yielding 6 spatial components. Results from the temporospatial PCA revealed 20 temporal and 6 spatial components for each temporal component, resulting in 121 total component combinations with the grand average. Only those temporal components that accounted for more than .5% of the total variance were included in further analysis, yielding 19 components. Nineteen component combinations were statistically evaluated using an ANOVA to identify components that significantly differentiated go and no-go trial type activity; 6 components were significant at p < .05.
None of the significant components corresponded with a priori expectations of the temporal and spatial components of the N2. However, a component peaking from 370.8 ms post-stimulus (p = .06) was visually inspected in BVA. Waveforms showed a negative deflection and a topographic map reflected negative activity at Cz during no-go trials, compared to go-trials, between 320 to 480 ms. We also examined the waveforms and topographic maps of nearby central sites, C3 and C4, to determine whether we would see a traditional N2 effect. Indeed, we found N2 negativity at the lateralized central sites, confirming that a multi-site cluster of C3, Cz, and C4 would best represent N2 observed in Sample 2. Mean amplitudes from C3, Cz, and C4 sites were averaged in the 320 to 480 ms time window and exported for analysis.
In both samples, N2 latencies occurred later and N2 amplitudes were more diffuse than is typical in studies with adults, consistent with prior work noting variability in samples of young children (Hoyniak, 2017). Thus, we focused on the mean amplitude, rather than the peak amplitude, for each defined area. Residual N2 difference scores were created by regressing activity during no-go trials on activity during go trials and saving the standardized residual scores (Hanna et al., 2020; Meyer et al., 2017). Calculated in this manner, more negative numbers (i.e., more negative amplitude) reflect a larger N2 and more cognitive control.
In Sample 1, there was no missing questionnaire data except for 2 families that chose not to report their household income. The most common reason for missing EEG data was nonparticipation in the laboratory visit (n = 11). Of families that did participate in the laboratory visit, N2 was missing from 19 children (9 = visit ended early due to distress, 6 = excessive artifact, 4 = technical issues). In Sample 2, there were missing data for 19 children (10 = visit ended early due to distress, 6 = excessive artifact, 2 = technical error, 1 = unknown). Eleven families did not return their questionnaire packets.
Tests of patterns of missingness suggested that data were missing at random (MAR). Patterns of missingness were unrelated to study variables (ps > .05) with four exceptions. Children who were missing N2 showed less high-intensity pleasure (t[165] = −2.74, p < .01), sadness (t[165] = −2.00, p < .05) and prosocial behavior (t[164] = −2.14, p < .05) but more physical aggression (t[164] = 2.97, p < .01) relative to children who were not missing N2. One child did not provide data for the measures used in the current analyses and so was excluded from the sample, resulting in a full sample of 176 preschoolers.
A path analysis was conducted to test the degree to which children’s social behaviors (i.e., prosociality and aggression) were independently predicted by domains of temperamental reactivity (i.e., anger, fear, low-intensity pleasure, high-intensity pleasure, sadness, shyness) and cognitive control (i.e., N2). To do this, social behaviors were simultaneously regressed onto domains of temperament and cognitive control. Given our interest in cognitive control as a moderator, we also regressed social behaviors onto the interaction between each domain of temperamental reactivity and cognitive control. Correlations between cognitive control and discrete temperament traits were fixed to zero, but the remaining variables were allowed to correlate. Child sex was controlled for in the analyses by regressing aggression onto child sex. Models were conducted using full-information maximum-likelihood estimation with robust standard errors in Mplus (Version 8.5; Muthén & Muthén, 2017) to account for mild nonnormality in physical aggression and use all available information.
Recognizing that the full sample did not show an average N2 effect by visual inspection (Figure 1) and that variability in N2 between preschoolers may reflect differences in response inhibition capacities that are still maturing (Hoyniak, 2017), we ran analyses focusing on only preschoolers who showed a pattern of neural activity consistent with the traditional N2 effect (n = 66). Importantly, doing this allows us to ask questions about the role of cognitive control in a subsample that demonstrates at least preliminary evidence of the presence of N2 but does not try to address this question in a sample where no N2 exists. However, for transparency in reporting, results from the full sample can be found in the supplementary material.
Preliminary analyses of the full sample were performed using SPSS Version 29 (see supplementary material). Results from independent sample t-tests, with the full sample, indicated that males were rated as higher in physical aggression than females (t(164) = 3.58, p = .03, d = .35, Mmales = .20, Mfemales= −.15), but all remaining t-values were nonsignificant (ps > .05).
Descriptive statistics for the N2 subsample are found on Table 1. Partial correlations controlling for biological sex of child indicated anger was positively associated with sadness, physical aggression, and cognitive control (Table 2). Fear was positively linked to cognitive control. Low-intensity pleasure was positively associated with high-intensity pleasure and prosociality while high-intensity pleasure was positively linked to sadness and prosociality but negatively associated with cognitive control. Sadness was positively related to shyness and physical aggression.
Results are shown in Table 3. Greater high-intensity pleasure (p = .01) and sadness (p = .00), along with less cognitive control (more positive N2; p = .01), predicted more aggression. Anger (p = .12), fear (p = .24), low-intensity pleasure (p = .98), and shyness (p = .07) were unrelated to aggression.
Interactions between cognitive control and anger (p = .90), low-intensity pleasure (p = .59), high-intensity pleasure (p = .13), and shyness (p = .20) predicting aggression were not significant. However, cognitive control significantly interacted with fear (p = .02) and sadness (p = .03) to predict aggression. These interactions were probed by re-centering cognitive control at low (+1 SD; more positive N2*)* and high (−1 SD; more negative N2) levels. Less fear was associated with more aggression at low (β = −.47, SE (β) = .14, p = .00), but not high (β = .02, SE (β) = .16, p = .81) cognitive control [Figure 3] . Additionally, sadness was unrelated to aggression at low cognitive control (β = .19, SE (β) = .15, p = .20) but positively predicted aggression at high cognitive control (β = .58, SE (β) = .11, p < .001).
Higher levels of low-intensity pleasure predicted more prosociality (p < .001; Table 3). Anger (p = .70), fear (p = .91), high-intensity pleasure (p = .45), sadness (p = .33), shyness (p = .25), and cognitive control (p = .20) were unrelated to prosociality. Interactions between cognitive control and fear (p = .40), low-intensity pleasure (p = .21), high-intensity pleasure (p = .21), sadness (p = .36), and shyness (p = .99) predicting prosociality were not significant.
As the interaction between cognitive control and anger predicting prosociality was significant at the trend level (p = .06), this interaction was probed at low (+1 SD) and high (−1 SD) levels of cognitive control. At both low (β = −.30, SE (β) = .17, p = .08) and high (β = .41, SE (β) = .23, p = .07) cognitive control, the associations were nonsignificant. However, recognizing the presence of a moderate-to-large effect and the opposing direction of beta values at differing levels of cognitive control, this interaction was probed at lower (−1.5 SD) and higher (+1.5 SD) levels of cognitive control. At lower cognitive control, greater anger was associated with less prosociality (β = −.43, SE (β) = .16, p = .00). However, at higher cognitive control, greater anger was associated with more prosociality (β = .50, SE (β) = .19, p = .01). A simplified visualization of the model is depicted in Figure 2 and plots of simple slopes are presented in Figure 3.
We tested discrete temperamental traits as predictors of children’s prosocial and aggressive behaviors and investigated whether these pathways were moderated by levels of cognitive control, given its nuanced role in early self-regulation. By testing all pathways linking temperamental reactivity to socioemotional outcomes in one model, we were able to test which pathways had the largest independent effects on children’s adaptive (i.e., prosocial) and maladaptive (i.e., aggressive) outcomes. Our results support theoretical and empirical work grounded in investigating discrete emotion (Campos et al., 1989; Gartstein et al., 2012; Goldsmith & Campos, 1982; Stifter & Dollar, 2016) and cognitive-affective processes (Bell et al., 2019; Hoyniak & Petersen, 2019; Rothbart & Bates, 2006) for understanding children’s developmental outcomes. As hypothesized, children’s levels of cognitive control moderated the associations between discrete temperamental reactivity and socioemotional outcomes.
Anger, relative to other discrete forms of temperamental reactivity, appears to play a distinct role in children’s early social development (Dollar et al., 2018; Dollar & Calkins, 2019), which might begin to explain why interactions among anger and cognitive control predicted prosociality above and beyond other pathways. When children had lower cognitive control, denoted by smaller N2 amplitudes, more anger was linked to less prosociality. This association is not surprising given that children who frequently and intensely display anger are at a disadvantage for developing healthy social skills and building quality peer relationships (Dollar et al., 2018; Rubin et al., 2011). In this manner, deficits in cognitive control and regulation of anger likely impede the suppression of inappropriate social responses that would be necessary for the development of positive prosocial behaviors.
In contrast, at higher cognitive control, denoted by more negative N2 amplitudes, more anger was associated with greater prosocial behaviors. Although heightened anger is typically linked to risk for maladaptive behavioral problems, cognitive control appears to be particularly beneficial when children experience high levels of negative emotional reactivity (Suurland et al., 2016; Wilson et al., 2021). Greater cognitive control may aid children in curbing displays of distress, despite experiencing intense levels of negativity (Putnam et al., 2008; von Hippel & Gonsalkorale, 2005) and protect against the development of social problems for angry-prone children (Davis et al., 2019; Hawes et al., 2016; Kim & Deater-Deckard, 2011). In this way, cognitive control comprises a facet of a broader capacity for self-regulation that allow down-regulation of anger reactivity and potentially provide a foundation for developing prosocial skills.
Our results complement existing literature by suggesting that heightened anger may facilitate healthy goal-directed behavior, prosociality, but only when paired with the capacity to sufficiently inhibit a dominant response. From a functionalist perspective, anger is an approach-related emotion that supports achieving a difficult or blocked goal (Campos et al., 1989). In the preschool period, anger-prone children may be adept at recognizing goal blockage. High levels of cognitive control may leverage approach tendencies associated with anger (Harmon-Jones, 2003) to maintain persistence toward a healthy resolve of the situation. Indeed, anger has been linked to approach related positive emotions (Harmon-Jones et al., 2011) that may be particularly useful in social contexts (Posner & Rothbart, 2000). For example, if a child’s goal is to make friends, higher cognitive control may help anger-prone children to funnel approach tendencies into situationally appropriate behavior, ultimately promoting social cooperation with peers (Wilson et al., 2021). Future work will be needed to further delineate adaptive and maladaptive functions of anger within the context of developing cognitive control. However, this finding is consistent with the idea that approach-oriented emotions may be most concerning when coupled with a lack of control and may be purposeful when no deficits in control exist.
Cognitive control also moderated associations between fear and aggression such that fear was negatively associated with aggression at low levels of cognitive control. Given links between fear and propensities for avoidance (Rothbart, 1989), greater levels of fear may lead to very low levels of aggression in children who tend not to engage with peers and may avoid social difficulties or confrontation (Fox & Pine, 2012; Rothbart et al., 1994). High levels of cognitive control may temper this association, resulting in the nonsignificant effect observed in our data and potentially leading to adaptive aggressive behaviors (Heilbron & Prinstein, 2008) allowing children to engage and persist even in the face of difficult social interactions. Adaptive aggression has gone relatively unstudied in fearful children but may a unique pathway to social competence in young children.
Likewise, it was somewhat surprising that sadness was positively associated with aggression at high levels of control. Again, it may be the case that high levels of control allow for a “break” from the behaviors that are typically associated with discrete emotions to allow for task or goal persistence. Functionalist perspectives on emotion characterize sadness as reflecting one’s giving up when a goal is blocked (Campos et al., 1994). As such a suppression of this tendency at high levels of cognitive control may allow for goal persistence, even if that persistence is viewed as aggressive. On the other hand, previous work that has demonstrated positive associations between sadness and aggressive/externalizing problems when children experience when experiencing loneliness or peer rejection (Eisenberg et al., 2005; Gartstein et al., 2012), highlighting the possibility that other moderators exist in the complex pathways linking discrete emotions to child outcomes.
In line with our predictions, lower cognitive control, indexed by smaller/more positive N2 amplitudes, was associated with more physical aggression. A recent meta-analysis provides evidence that smaller N2 amplitudes are predictive of more aggressive behaviors in early childhood (Hoyniak & Petersen, 2019), likely reflecting deficits in higher-order cognitive processes crucial for adaptive self-regulation. Moreover, greater levels of sadness predicted more aggression regardless of levels of cognitive control. Although sadness generally precedes social withdrawal, heightened sadness is associated with the development of aggressive problems in the preschool period (Eisenberg et al., 2001; Gartstein et al., 2012). Children who experience frequent and intense sadness have more difficulty in social situations and may engage in socially inappropriate behaviors when their needs are not met (Denham et al., 2003). In general, children who have difficulty managing their negative emotions are more likely to experience peer victimization (Sugimura & Rudolph, 2012), which may lead to feelings of isolation that are expressed through physical aggression.
Although not the primary focus of our investigation, it is worth noting that a traditional N2 effect was observed in approximately 51% of the preschoolers who provided usable EEG data. Moreover, N2 topography and latency differed somewhat across samples. N2 in Sample 1 was observed parietally and later in latency, while N2 in sample 2 was observed centrally and earlier in latency. Neural mechanisms underlying cognitive control are notably diffuse in early childhood and change in both morphology (becoming less parietal) and latency (occurring earlier) over time (Bunge et al., 2002; Ciesielski et al., 2004; Jonkman et al., 2007; Lewis et al., 2006). Such patterns are not isolated to N2 (Jonkman et al., 2007; Luna & Sweeney, 2004). Given that children in Sample 1 were slightly younger than children in Sample 2, differences in N2 across samples is likely attributable to sample-related differences in the maturation of the neural regions involved in the generation of N2. Furthermore, N2 latency decreases across age, reflecting improvements in brain regions that support successful response inhibition (Cragg et al., 2009; Hoyniak, 2017). Nonetheless, studies at this age can uniquely capitalize on individual differences, as children between 39 and 41 months of age can successfully inhibit a response while this ability is absent in children from 30 to 39 months, indicating how rapidly neural networks that support cognitive control processes unfold (Jones et al., 2003). However, developmental differences in N2 may not be purely age-related; for instance, N2 has also been elicited as early as 2.5 years of age (Cragg et al., 2009; Hoyniak, 2017) but not at age 6 (Davis et al., 2003).
Additionally, greater low-intensity pleasure was associated with more prosocial behavior, while greater high-intensity pleasure was associated with more aggression. These results support the importance of examining dimensions of temperamental positivity separately. Differences in links between pleasure and social behaviors may be explained by the motivation underlying the expression of positive affect. Low-intensity pleasure may offer a form of positivity that is protective of emotional and physical well-being, while high-intensity pleasure is linked to heightened approach-driven emotions and is associated with risky behaviors (Dollar et al., 2023; Putnam, 2015). Low-intensity pleasure is a significant contributor of stable regulatory and attentional processes (Putnam et al., 2008), better emotional understanding (LaBounty et al., 2017), and prosocial behaviors relating to helping, caring, and kindness (Kozlova et al., 2020; Slobodskaya et al., 2020). Although more low-intensity pleasure is often positively correlated with higher-order cognitive processes (Rothbart et al., 2001), low-intensity pleasure and N2 were not associated with each other in the N2 subsample, suggesting that N2 may capture mechanisms of cognitive control that are less sensitive to the environmental details (Hoyniak et al., 2018). Moreover, links between low-intensity pleasure and N2 may emerge in samples of children with higher variability in cognitive control processes (see Supplementary Material). These results highlight the heterogeneity of positive reactivity for early childhood outcomes, suggesting that low-intensity pleasure may scaffold adaptive behaviors while high-intensity pleasure may increase risk for maladaptive functioning.
Although anger and aggression were positively correlated at the bivariate level, anger did not significantly predict aggression when controlling for other forms of temperamental reactivity. This adds to the growing literature of mixed findings regarding the role of anger in early childhood. Some studies have found an absence of a main effect of anger on early behavioral problems (Brooker et al., 2014; He et al., 2010) while others have reported a significant association (Hernández et al., 2015; Nwadinobi & Gagne, 2020). One likely explanation for our nonsignificant association is that anger was also positively correlated with another discrete domain of temperament, sadness, which was partialled out in our analytic model. This suggests that, during this developmental period, anger is not predictive of aggression above and beyond other discrete forms of temperamental reactivity.
Moreover, cognitive control did not moderate associations between anger and aggression. We expected that low cognitive control would amplify a positive link between anger and aggressive behavior, as observed in previous work (Eisenberg et al., 2009; Suurland et al., 2016), but our findings did not support this hypothesis. Factors such as high life stress (Brooker et al., 2014), socioeconomic disadvantage (Kim & Kochanska, 2021), emotional dysregulation (Garofalo & Velotti, 2017), and age (Gartstein et al., 2012; Suurland et al., 2016) appear to be important moderators for the association between childhood anger and the development of aggression. For example, the negative association between cognitive control and temperamental negative emotionality is more evident in predicting aggressive problems in older preschoolers (4 to 5 years) relative to younger preschoolers (Gartstein et al., 2012; Suurland et al., 2016). Nevertheless, aggressive behaviors in three-year-old children are fairly common and longitudinal designs that assess how multiple factors influence the links between early anger and aggression will be an important avenue for future research.
Although the current study is poised to add specificity to our understanding of pathways to children’s adaptive and maladaptive social behaviors, it is not without limitations. We used a cross-sectional design to maximize the number of children at age 3, a developmental period when cognitive control increasingly evolves. This prohibits the exploration of causal influences or conclusions about development over time. It also limits our ability to directly speak to why some of our pathways of children’s social outcomes were less salient relative to other developmental pathways. Future work may identify whether the maturation of cognitive control processes from the preschool to school period influence the links between discrete temperamental reactivity and social behaviors. That is, the stability of fine-grained temperamental traits may vary, or be relatively indifferent to, increases in cognitive control development. These fluctuations may be critical to identifying when and how cognitive-affective systems act jointly for the growth of social behaviors.
Second, while accounted for statistically, we have patterns of systematic missingness related to N2 that may be attributable to the challenges of collecting EEG data in children. Specifically, children who refused EEG capping evidenced more parent-reported problematic behaviors. As such, a need for more work on N2 in children with propensities toward problem behaviors is needed.
Third, as behavioral data were collected from parent-reported questionnaires, there may be shared method variance across our temperament and social behavior measures. Although we aggregated data from two parents where possible, we recognize that our results may differ if we used a combination of observed and self-report measures to assess children’s behaviors (Olino et al., 2013).
Some attention is also due to statistical limitations. For example, the reliability for our fear and sadness scales from CBQ in Sample 1 was lower than expected (α < .60), suggesting a possible need for caution for drawing conclusions about these temperamental traits. In addition, relative to the full sample, parameter estimates in the N2 subsample may be less precise due its smaller sample size. Despite these statistical considerations, the N2 subset, albeit smaller, allows use to deliberately steer clear of probing questions about cognitive control where the N2 effect is not observed.
Finally, while our community families represented a full range of sociodemographic, racial, and ethnic diversity, participants were predominantly White, non-Hispanic, and from moderately high-income households. Future work with more diverse samples, along with different cultural groups, is needed to enhance the generalizability of our findings.
Consistent with theory and prior research, our study offers evidence of novel and unique contributions to preschoolers’ preschooler’s adaptive and maladaptive development from discrete temperamental traits and cognitive control processes. The study findings demonstrate that specific forms of temperament are differentially associated with risk or protection at varying degrees of cognitive control, highlighting the importance of examining lower-order temperamental reactivity. This research adds to the growing literature on how cognitive control contributes to the self-regulation of emotion and the development of early social functioning. This work, along with future investigations, will be essential for precisely delineating affective-cognitive pathways that can be targeted in identification and prevention efforts.