Multi‐method assessment of irritability and differential linkages to neurophysiological indicators of attention allocation to emotional faces in young children

Abstract

Facilitated attention toward angry stimuli (attention bias) may contribute to anger proneness and temper outbursts exhibited by children with high irritability. However, most studies linking attention bias and irritability rely on behavioral measures with limited precision and no studies have explored these associations in young children. The present study explores irritability-related attention biases toward anger in young children (N = 128; ages 4–7 years) engaged in a dot-probe task with emotional faces, as assessed with event-related brain potential (ERP) indices of early selective attention and multi-method assessment of irritability. Irritability assessed via semi-structured clinical interview predicted larger anterior N1 amplitudes to all faces. In contrast, irritability assessed via a laboratory observation paradigm predicted reduced P1 amplitudes to angry relative to neutral faces. These findings suggest that altered early attentional processing occurs in young children with high irritability; however, the nature of these patterns may vary with methodological features of the irritability assessments. Future investigations using different assessment tools may provide greater clarity regarding the underlying neurocognitive correlates of irritability. Such studies may also contribute to the ongoing debates about how to best define and measure irritability across the developmental spectrum in a manner that is most informative for linkage to neural processes.

1 |. INTRODUCTION

Irritability has been defined as a relative dispositional tendency to respond to blocked goal attainment with anger and frustration and can be conceptualized as a developmental temperamental trait (Wakschlag et al., 2018). When irritability is developmentally atypical and interferes with functioning, it is viewed as a psychiatric symptom that can predict significant impairment and poor outcomes throughout life (Brotman, Kircanski, & Leibenluft, 2017; Copeland, Shanahan, Egger, Angold, & Costello, 2014; Dougherty et al., 2015). In young children, we have empirically derived features of irritability that differentiate normative variation from clinically salient patterns, particularly dysregulated tantrums and anger occurring at high frequencies and in developmentally unexpected contexts (Wakschlag et al., 2012, 2018). Understanding the mechanisms underlying severe and persistent irritability is critical to developing informed approaches for preventing psychopathology and other adverse outcomes. Aberrant reward processing, social information processing deficits, and dysfunctional threat processing are candidate mechanisms underlying pediatric irritability in current models (Brotman, Kircanski, & Leibenluft, 2017; Brotman, Kircanski, Stringaris, Pine, & Leibenluft, 2017; Leibenluft, 2017). The present study focuses on the domain of dysfunctional threat processing, building on prior work linking irritability with facilitated attention toward angry faces, hereafter called “attention biases” (Hommer et al., 2014; Salum et al., 2017). Heightened attention to angry stimuli may facilitate angry feelings and temper tantrums (Brotman, Kircanski, & Leibenluft, 2017; Stoddard, Scelsa, & Hwang, 2019) and contribute to risk for later emotional and behavioral disorders (Fu & Pérez-Edgar, 2019; Morales, Fu, & Perez-Edgar, 2016; Perez-Edgar et al., 2011; Wakschlag et al., 2015; White et al., 2017; Wiggins et al., 2017).

The present study focuses on attention bias, specifically facilitated attention, toward angry stimuli, measured in terms of behavioral performance on a dot-probe task and event-related brain potentials (ERPs) elicited by angry faces (Bar-Haim, Lamy, Pergamin, Bakermans-Kranenburg, & van IJzendoorn, 2007; Cisler & Koster, 2010). This study seeks to address two limitations in the existing literature. First, to our knowledge, although studies of older children have shown that irritability is associated with attention bias toward angry faces (Hommer et al., 2014; Salum et al., 2017), no prior research has explored these associations in early childhood. The identification of these patterns in early childhood would raise the possibility that interventions designed to reduce biased attention toward threatening stimuli in older youth may have therapeutic value at this earlier phase of the clinical sequence (e.g., Attention Bias Modification; Bar-Haim, 2010; Hakamata et al., 2010).

Second, the neural mechanisms underlying attention bias toward anger in irritability are largely unexplored, thus, hindering understanding of the mechanisms that give rise to and maintain irritability as well as potential treatment targets. Although irritability has been related to dysfunction in brain regions associated with threat processing and emotion regulation when viewing emotional faces (i.e., amygdala and prefrontal cortex, respectively; Brotman et al., 2010; Stoddard et al., 2016; Stoddard et al., 2017; Thomas et al., 2012; Thomas et al., 2013; Tseng et al., 2016; Wiggins et al., 2016), only one neuroimaging study has assessed the irritability in relation to attention biases specifically (Kircanski et al., 2018). In a large sample of school-age children and adolescents with a range of psychiatric conditions or no psychiatric history (N = 197; ages 8–18 years), higher levels of irritability were associated with greater activation in neural regions associated with emotional processing, as well as attention and motor responses (i.e., amygdala, insula, ventrolateral prefrontal cortex, caudate, and inferior parietal lobule; Kircanski et al., 2018). Although these activation patterns were observed in trials involving angry stimuli, the slow temporal resolution of fMRI (Huettel, Song, & McCarthy, 2009) prevented this study from isolating neural responses to the face stimuli, specifically. Techniques with high temporal resolution that permit the isolation of neural responses to faces are necessary to isolate the presence of facilitated attention toward angry faces. This is particularly important in studies of differential attention toward threatening (i.e., angry and fearful) stimuli which can reflect the facilitated attention toward threat that is of primary interest in the present study or difficulty disengaging from threatening stimuli (see Cisler & Koster, 2010 for review).

The present ERP study addresses these gaps by exploring the neurophysiological correlates of biased attention toward angry faces on the affective dot-probe task in a large sample (N = 128) of young children (ages 4–7 years) and examining how these neurophysiological indicators relate to irritability assessed by parent report and by a standardized diagnostic observation specifically designed to elicit clinically salient patterns of irritability in young children.

1.1 |. Irritability and attention biases toward anger

On a behavioral level, studies investigating attention biases in irritable youth have used the dot-probe task (Hommer et al., 2014; Kircanski et al., 2018; Salum et al., 2017). During this task, participants view a pair of faces that are quickly replaced by a target probe in the same location as one of the faces (Macleod, Mathews, & Tata, 1986). The participant’s task is to identify the location of the target probe. In a typical paradigm, the face pairs include one angry face and one neutral face. If an individual preferentially orients attention toward the angry face, she/he will be faster to detect a target probe that appears in the same location as that face (i.e., congruent trials) and slower to detect a target probe appearing in the opposite location (i.e., incongruent trials). The difference in response time (RT) on angry congruent and incongruent trials is used to generate a RT-based angry bias score (Bar-Haim et al., 2007).

Attention biases to angry (but not happy) faces have been identified in irritable children and adolescents relative to psychiatrically healthy youth. This pattern has been identified in samples with clinically significant irritability (severe mood dysregulation; Hommer et al., 2014) and replicated in a large community sample (Salum et al., 2017). Notably, the latter study observed this association using continuous irritability scores on the Child Behavior Checklist (CBCL) and when subsets of children with the highest and lowest levels of irritability on the CBCL were compared (Salum et al., 2017). Irritability was not associated with biases toward happy faces in either study, suggesting that irritability is not associated with facilitated attention toward emotional faces more generally. Moreover, in both studies, the irritability—angry bias associations remained significant when the models accounted for other clinical phenomena that are known to be associated with attention bias toward angry faces, such as anxiety (Bar-Haim et al., 2007). Finally, irritability and anxiety were associated with greater attention orienting variability, but not the traditional RT-based measure of attention bias toward angry faces, in a recent neuroimaging study of a diverse clinical sample of school-age children and adolescents whose irritability was derived from a bifactor analysis of symptom ratings (Kircanski et al., 2018). Whereas several studies suggest links between irritability and attention bias toward angry faces in children and adolescents, whether these patterns are observed in early childhood has not been specifically tested. Broader developmental research has shown that preferential attention to emotional stimuli, including angry and fearful faces, emerges during infancy in normative samples (for a recent review see Fu & Perez-Edgar, 2019). In children ages 3–7 years, attention biases to emotional faces on the dot-probe task have been observed in samples of healthy children, as well as in children exposed to trauma, children at risk for depression, and children at temperamental risk for anxiety (Briggs-Gowan, Grasso, et al., 2016; Briggs-Gowan et al., 2015; Cremone, Kurdziel, Fraticelli-Torres, McDermott, & Spencer, 2017; Kujawa et al., 2011; Nozadi et al., 2016; Perez-Edgar et al., 2011; White et al., 2017). Together, these findings indicate that attention biases exist in early childhood and may relate to individual differences in risk for psychopathology.

1.2 |. Role of ERPs in clarifying the neural and behavioral processes involved in biased attention to threat

To date, most studies of attention biases (regardless of age group) have relied on imprecise measures of attention orienting. When attention bias to threat is assessed with the dot-probe task, it is inferred based on RTs to target stimuli that are presented several hundred milliseconds after the threatening face (Gibb, McGeary, & Beevers, 2016). This fails to capture attentional processes that occur between the onset of the threat stimulus and the behavioral response. Furthermore, it is impossible to specify whether the RT-based measures of threat bias reflect biased initial orienting (i.e., where the participant looks first), difficulty disengaging attention from the threatening stimulus, or both (Grafton & MacLeod, 2014). ERPs allow exploration of the temporal unfolding of cognitive processes in response to specific stimuli (Huettel et al., 2009; Luck, 2014). As such, they offer greater sensitivity than RT-based indicators for detecting temporal characteristics of attention bias to threat, including early processing that occurs within 200 milliseconds of stimulus onset (Gibb et al., 2016; Kappenman, Farrens, Luck, & Proudfit, 2014; Kappenman, MacNamara, & Proudfit, 2015; Torrence & Troup, 2018). Early selective attention to faces can be assessed by examining the P1, N1, and N2pc ERP components. These components each index selective spatial attention, serving as a mechanism by which the sensory signals associated with attended stimuli are enhanced relative to those of unattended stimuli (Eimer & Kiss, 2007; Hillyard, Vogel, & Luck, 1998; Luck & Kappenman, 2012). These components are maximal over parietal-occipital scalp sites and differ primarily in their onset. The P1 component peaks approximately 100-130ms, whereas the N1 and N2pc components occur slightly later (100-200ms and 150-250ms, respectively).

Among dot-probe studies of adults, N1 and N2pc, but not P1, amplitudes are often enhanced for emotional relative to neutral stimuli (Bar-Haim, Lamy, & Glickman, 2005; Eldar & Bar-Haim, 2010; Grimshaw, Foster, & Corballis, 2014; Pintzinger, Pfabigan, Pfau, Kryspin-Exner, & Lamm, 2017; Pourtois, Grandjean, Sander, & Vuilleumier, 2004; Santesso et al., 2008; Shah et al., 2018). Increased anxiety is associated with larger P1, N1, and N2pc amplitudes to threatening faces on the dot-probe task, consistent with findings from RT-based measures (Bar-Haim et al., 2005; Fox, Derakshan, & Shoker, 2008; Holmes, Nielsen, & Green, 2008; Mueller et al., 2009; Pintzinger et al., 2017; Reutter, Hewig, Wieser, & Osinsky, 2017; Rossignol, Campanella, Bissot, & Philippot, 2013; c.f., Kappenman et al., 2014). Together, these studies suggest that ERPs can be used to assess individual differences in early attention allocation to threat and their relationship with psychiatric symptoms.

To the best of our knowledge, no dot-probe ERP studies have been conducted in early childhood. Two ERP studies have examined the effects of emotion or anxiety in children and adolescents on the dot-probe task (Bechor et al., 2018; Thai, Taber-Thomas, & Perez-Edgar, 2016). These studies examined the P1 and N1 components described above, as well as the N170 and P2 components which occur later and index face processing and attention resource mobilization, respectively (Carretie, Mercado, Tapia, & Hinojosa, 2001; Eimer & Holmes, 2007; Hinojosa, Mercado, & Carretie, 2015; Luck, 2014; Thai et al., 2016). Unlike adults, emotional faces failed to modulate ERPs in either pediatric sample (Bechor et al., 2018; Thai et al., 2016). Similar to adults, anxiety predicted variations in the amplitudes of these ERPs (Bechor et al., 2018; Thai et al., 2016). One study found that children and adolescents (ages 8–16 years) with clinically significant anxiety exhibited larger P1 amplitudes to angry faces, relative to nonangry faces, and larger N170 amplitudes to all faces (Bechor et al., 2018). P2 amplitudes were also reduced in clinically anxious relative to nonanxious children in that study (Bechor et al., 2018), a finding that also emerged in a normative sample of 9- to 12-year-olds, where higher social anxiety was associated with smaller P2 amplitudes (Thai et al., 2016). Thus, available data suggest that emotional faces presented in the dot-probe may elicit greater N1 and N2pc amplitudes in all participants (although this may not be observed in children) and individual differences in anxiety proneness modulate P1, N1, P2, N170, and N2pc amplitudes to angry or emotional faces.

To date, no research has examined associations between irritability and ERP measures of attention bias on the dot-probe task. To the extent that irritability and anxiety are associated with similar RT-based measures of threat bias, the ERP patterns associated with anxiety might also be related to irritability. Finally, because the P1, N1, P2, N170, and N2pc components measure different aspects of attention, linking irritability to specific ERP components provides more precise information concerning potential neural correlates of irritability. However, in order to attribute measures of attention bias to irritability, specifically, it is important to account for anxiety symptoms that co-occur frequently with irritability (Cornacchio, Crum, Coxe, Pincus, & Corner, 2016; Stoddard et al., 2014; Wakschlag et al., 2012) and have robust associations with the behavioral and neural measures of attention bias of focal interest to the present study (Bar-Haim et al., 2007; Bechor et al., 2018; Morales, Fu, et al., 2016; Thai et al., 2016).

1.3 |. Developmentally sensitive and multi-method assessments of irritability

A major challenge to identifying atypical irritability in early childhood is the fact that irritability, particularly temper tantrums, reflects the normative misbehavior of this age period. Thus, the use of developmentally sensitive measures specifically designed to differentiate clinically salient from normative patterns is of particular import (Wakschlag et al., 2018). Relatedly, multi-method approaches are strongly recommended when assessing young children’s social—emotional functioning (Briggs-Gowan, Godoy, Heberle, & Carter, 2016; Wakschlag et al., 2008b; Wakschlag et al., 2005), yet virtually all studies to date have relied on parent-checklist ratings. Obtaining information from observation as well as parent report is especially important in this developmental period for several reasons. Young children cannot provide reliable self-reports of their own moods, the differentiation of normative variation from clinically salient patterns is nuanced, and there is evidence that parent report and behavioral observations of irritability and related behaviors are differentially related to salient outcomes (Briggs-Gowan, Godoy, et al., 2016; Petitclerc et al., 2015; Wakschlag et al., 2005; Wakschlag, Tolan, & Leventhal, 2010). The present study uses parent report and direct behavioral observation methods that are specifically designed for developmental sensitivity in early childhood to assess the irritability and provide a more comprehensive understanding of the relationship between irritability and attention biases.

1.4 |. Present study

The present study explored associations between irritability and attention biases toward angry faces in young children. Based on the available literature, we hypothesized that higher levels of irritability: (a) would predict greater RT-based attention bias scores in response to angry faces relative to neutral faces, but would not be associated with happy bias scores; and (b) would predict greater P1, N1, and N2pc and smaller P2 amplitudes to angry faces, but would not be significantly associated with happy faces. Associations with irritability were expected to exist after accounting for co-occurring anxiety. Irritability and anxiety were operationalized using symptom scores from a semi-structured diagnostic interview, as well as coded behavior observed during frustrating and anxiety-provoking diagnostic observation paradigms. Finally, (c) given the lack of ERP studies in early childhood, we take the opportunity to also report the ERP amplitudes elicited by angry, happy, and neutral faces to facilitate comparisons with prior studies in older children and adults. We did not expect to find significant differences on ERP components for happy, angry, and neutral faces given the lack of significant differences observed between emotional and neutral stimuli on the dot-probe task in older children and adolescents (Bechor et al., 2018; Thai et al., 2016).

2 |. METHODS

2.1 |. Participants

Study participants were drawn from a diverse set of young children that participated in the Multidimensional Assessment of Preschoolers Study (MAPS; Wakschlag et al., 2015). A sample of MAPS participants recruited in a survey sample participated in an intensive longitudinal follow-up study, involving intensive clinical and neurocognitive assessments including the dot-probe task described in the present study (N = 425). Because the primary goal of the MAPS study was to explore the risk factors for psychopathology, this cohort represented an oversampling of psychopathology risk based on heightened irritability and other disruptive behaviors and violence exposure. For additional information about the MAPS longitudinal cohort please see Briggs-Gowan et al., (2015), Wakschlag et al. (2015), and Data S1.

In the present study, we report on a subset of children from the longitudinal follow-up study who met the following eligibility criteria: (a) met electroencephalographic (EEG) eligibility criteria (4 years of age or older, right-handed, no skin conditions that interfered with electrode application, and no neurodevelopmental delays or conditions); (b) attempted the dot-probe task with EEG recordings; and (c) had usable behavioral and ERP data (accuracy >65%¹ and at least nine trials/trial type). As described in Data S1, 256 children met EEG eligibility criteria. Of those, 178 completed the dot-probe task with EEG recordings. Data from 50 (28.1%) participants were removed due to poor accuracy (≤65%; n = 25) or an insufficient number of artifact-free trials to estimate ERP reliability (<9 trials/trial type; n = 25). Therefore, the final sample consisted of 128 participants age 48.2–85.2 months (see Table 1 for the demographic and clinical characteristics). Behavioral data from this sample have been reported previously as part of a larger sample of children that included individuals who did not meet EEG eligibility criteria or could not tolerate the EEG session (N = 86; Briggs-Gowan, Grasso, et al., 2016; Briggs-Gowan et al., 2015). The means of the subsample reported in the current study are within 1 standard deviation of those reported in this prior work.

TABLE 1.

Demographics and Clinical Characteristics

	Full sample (N = 128)		Symptom-based sample (N = 114)		Observation-based sample (N = 121)
Characteristic	M	SD	M	SD	M	SD
Age (months)	62.07	8.4	62.08	8.3	62.28	8.4
Parent-reported irritability	—	—	1.18	1.5	1.12	1.4
Parent-reported anxiety	—	—	−0.11	0.5	−0.13	0.5
Observed anxiety	—	—	4.20	2.5	4.21	2.5
Observed irritability	—	—	−0.94	1.9	−1.01	1.9
			N	%	N	%
Male	62	48.4%	56	49.1	58	47.9
African American	50	39.1%	46	40.4	46	38.0
Caucasian	33	25.8%	28	24.6	32	26.4
Hispanic	45	35.2%	40	35.1	43	35.5
Income < poverty line	66	51.6%	59	51.8	62	51.2

Note: Although 128 children completed the task with usable behavioral and ERP data, interview and/or observational data were missing for some of these participants. Demographic, anxiety, and irritability data are summarized for the full sample as well as within the subsamples with interview-based symptom and observational data.

A small number of participants were missing data for the symptom-based (n = 14) or observation-based (n = 7) irritability and anxiety measures, respectively. Therefore, the symptom-based analyses were conducted using the 114 children with symptom data and the observation-based analyses were conducted with the 121 children with observational data. The sociodemographic characteristics of each subsample are included in Table 1. Additional details about participant progress throughout the study and a participant flow diagram are included in Data S1.

We compared sociodemographic and behavioral performance variables between children included in the final sample (N = 128) with individuals who met EEG eligibility criteria but were excluded from the analyses because no EEG data could be collected on the dot-probe task or due to poor behavioral or ERP data on the task (N = 128 see Data S1). Details about these analyses can be found in Data S1. Relative to eligible but excluded children (N = 128), the participants in the final sample (N = 128) were similar in sex, poverty status, anxiety, and behavioral measures of angry and happy biases. However, included children contained a lower percentage who identified as African American, were older, had higher non-verbal reasoning scores, and performed more accurately on the task than children who were excluded. Finally, although children in the present sample did not differ from those excluded on observed or parent-reported anxiety or on parent-reported irritability (described below), included children displayed less irritability during the observational paradigm than those who were excluded (p<.001).

In addition, because previously published work on this sample focused on violence exposure (Briggs-Gowan, Grasso, et al., 2016; Briggs-Gowan et al., 2015), we explored whether irritability was related to violence exposure in our sample of participants. Irritability scores did not differ between the children exposed to violence (22%) relative to those who did not (78%).

Institutional review boards approved the study protocols and mothers provided informed consent. Compensation was provided to parents for participation and transportation.

2.2 |. Irritability measures

The present study adopted a multi-method approach to assessing irritability. The first was based on parent reports of the child’s irritability symptoms during a diagnostic interview. The second was based on coded behavior observed during a laboratory task designed to elicit frustration.

2.2.1 |. Symptom-based measures of irritability

Symptom data were abstracted from the Preschool-Age Psychiatric Assessment (PAPA; Egger et al., 2006), a semi-structured diagnostic interview for early childhood that was administered by trained research assistants under the supervision of a master clinician. The PAPA assesses the presence of Diagnostic and Statistical Manual 4th Edition (American Psychiatric Association, 2000) psychiatric disorders and symptoms in preschoolers. Inter-rater reliability was excellent in this subsample (M intraclass correlation = 0.93).

An established scoring system based on the intensity, duration, and frequency of six irritability-related items from the PAPA (Dougherty et al., 2013) was used to generate an overall irritability symptom score for each participant. The six items included the presence of irritable mood; proneness to feel and express anger and resentment; tendencies to feel frustrated in response to minor provocations; and tendencies to display temper outbursts including name-calling, shouting, and stamping in the preceding 3 months (see Dougherty et al., 2013 for details). Irritability scores ranged from 0 to 5 (M = 1.16, SD = 1.6). Higher scores reflect clinically significant symptoms of irritability.

2.2.2 |. Observational measures of irritability

Irritability was also assessed using the Disruptive Behavior Diagnostic Observational Schedule (DB-DOS; Wakschlag et al., 2008b; Wakschlag, Hill, et al., 2008). This is a 50-min coded diagnostic observation paradigm that examines variations in children’s emotional and behavioral regulation across varied demands and interactional contexts and has demonstrated validity, reliability, and incremental utility for clinical prediction (Wakschlag et al., 2008b; Wakschlag, Hill, et al., 2008). The task was designed to efficiently elicit children’s patterns of irritability with an emphasis on differentiation between normative versus atypical patterns in response to tasks that “press” for clinically relevant behaviors. For example, one DB-DOS task involves completing a puzzle that is rigged to elicit frustration. Ratings of intensity, pervasiveness of irritability are coded within an “anger modulation” score. Scores are ordinal reflecting a continuum of clinical concern, that is, 0 = normative behavior; 1 = normative misbehavior; 2 = of concern; and 3 = atypical. Irritability was measured using the total anger modulation score, which is based on six DB-DOS codes reflecting the intensity, predominance, ease of elicitation, speed of escalation, and pace of recovery from irritability including outbursts, and the pervasiveness of irritability across expectable and unexpectable contexts. Higher scores reflect the extent to which children’s irritability was dysregulated and pervasive. Irritability was coded in parent–child and examiner–child interactions, but since the observation-based measure of anxiety (described below) has been validated only for the parent–child interaction, for comparability, we focused on coding from the parent–child interactions. Inter-rater reliability for the irritability composite score in this sample was excellent (M intraclass correlation = 0.91).

2.3 |. Covariates

Each analysis controlled for factors that might predict performance on the attention bias task as well as ERP amplitudes, consistent with our prior work (Briggs-Gowan, Godoy, et al., 2016; Briggs-Gowan et al., 2015; Deveney et al., 2019). These included age (in months), sex, and nonverbal reasoning and anxiety. Nonverbal reasoning was assessed using the Pictures Similarities subscale of the Differential Ability Scales-Second Edition (DAS-II; Elliott, 1983) Nonverbal reasoning scores were included in order to control for children’s developmental capacity to complete the task. For example, children with poorer nonverbal reasoning skills might find it difficult to remember the mapping between target location and button press. Therefore, they might have additional neural recruitment relative to those who found the mapping easy to recall and implement.

Because our primary interest was in associations of attention bias with irritability, we wanted to ensure that our findings were not influenced by co-occurring anxiety which has been independently associated with threat biases and ERP patterns on the dot-probe task and frequently co-occurs with irritability (Bar-Haim et al., 2007; Bechor et al., 2018; Cisler & Koster, 2010; Thai et al., 2016). Therefore, all analyses included anxiety as a covariate. Two separate anxiety measures were calculated to be analogous to the two irritability measures. The first was based on parent-reported anxiety symptoms in the PAPA (Egger et al., 2006). An anxiety symptom composite was calculated as the mean score of social phobia, separation, generalized anxiety disorder (after removing the irritability item), and panic symptoms from the PAPA scoring algorithms. This composite was based on z scores of the symptom criterion counts to account for variable scoring methods across different diagnostic categories. Anxiety symptom scores ranged from −0.66 to 1.73 (M = −0.11, SD = 0.54). The second was an observation-based measure of anxiety assessed with the Anxiety Dimensional Observation Scale (ANX-DOS; Mian, Carter, Pine, Wakschlag, & Briggs-Gowan, 2015). Similar to the DB-BOS, the ANX-DOS paradigm is a coded diagnostic observation paradigm designed to distinguish normative from clinically salient manifestations of children’s fear/anxiety in response to tasks that “press” for clinically relevant behaviors. In the ANX-DOS, children are asked to touch a remote controlled tarantula, put their hand into a “mystery jar,” respond to an unexpected bell sound, play with a realistic spider that is controlled by an experimenter outside the room, and remain with an experimenter after their parent leaves the observation room. The child’s behavior was coded on a 4-point scale ordinal scale analogous to that used in the DB-DOS (0 = no evidence of the behavior; 1 = normative behavior; 2 = of concern; and 3 = atypical). Codes captured a range of indicators of anxiety and fear, including fear arousal, physical avoidance of the press, exaggerated startle, proximity seeking (e.g., moving toward the parent), separation distress, and hypervigilance. In the present study, anxiety was measured using a composite score based on codes capturing physical avoidance, whole body startle, and observable expressions of fear during the entire observational period. Inter-rater reliability of the ANX-DOS in this sample was very good (M intraclass correlation = 0.90).

2.4 |. Dot-probe task

The children completed a standard affective dot-probe task while ERP recordings were obtained. The details of this task have been published previously (Briggs-Gowan et al., 2015). Briefly, on each trial, participants viewed a fixation cross (1500ms), two faces (left and right side of the fixation cross; 500ms), followed by a gold coin (target probe; 500ms) that appeared in the same location as one of the two faces. The child was asked to “catch” as many coins as possible by pressing a button to identify the target location. Children completed this task in four blocks (90 trials/block). After each block, children were told that they had earned gold coins that could be exchanged for a prize at the end of the task. All children earned a prize.

Faces were drawn from the NimStim set of facial expressions (Tottenham et al., 2009) and were presented in the following pairs: angry/neutral; happy/neutral; and neutral/neutral. Face and target locations (left, right) were counterbalanced across trials. Targets appeared behind each face emotion with equal likelihood. Congruent trials are those where the target appears in the same location as the emotional (angry or happy) face. Incongruent trials refer to trials where the target appeared in the opposite location of the emotional face. This task design generated five different trial types: (a) angry congruent (the target appeared in the same location as the angry face; N = 72); (b) angry incongruent (the target appeared in the opposite location as the angry face; N = 72); (c) happy congruent (N = 72); (d) happy incongruent (N = 72); and (e) neutral-neutral (N = 72).

The behavioral data from the task were analyzed according to the established methods and prior work in the larger sample from which the present sample was drawn (Briggs-Gowan, Grasso, et al., 2016; Briggs-Gowan et al., 2015; Eldar, Yankelevitch, Lamy, & Bar-Haim, 2010; Perez-Edgar et al., 2011). Trials with incorrect or missing responses as well as trials with response times falling outside of a 200–7,000ms response window were excluded from RT analyses (M = 34.80; SD = 30.4). Finally, trials with RTs >2.5 standard deviations (SD) from the individual child’s mean RT were also excluded from RT analyses (M = 10.14; SD = 3.1). Mean RT for each of the five trial types was calculated for each participant using the remaining trials. The number of trials included in the mean RT for each trial type ranged from 17 to 72 and the mean number of trials used to calculate each participant’s mean RT was as follows: angry congruent: M = 61.6, SD = 9.1; angry incongruent: M = 61.4, SD = 9.3; happy congruent: M = 61.5, SD = 8.9; happy incongruent: M = 62.4, SD = 9.2; neutral-neutral M = 61.7, SD = 8.9. The mode for each trial type was ≥64.

Bias scores were calculated by subtracting the RT on congruent trials from the RT to incongruent trials separately for happy and angry faces. Positive bias scores indicate attentional biases toward the emotional stimulus. Negative bias scores indicate attentional biases toward the neutral stimulus.

2.5 |. Electroencephalographic data acquisition and analysis

EEG data were acquired using a 32-channel Ag/AgCl Quick cap and a SynAmp RT amplifier (Neuroscan). Electrooculographic (EOG) data were collected using Ag/AgCl electrodes placed bilaterally on the outer canthi and above and below the left eye. Impedances were kept below 10 kΩ-. During data acquisition, data were digitized at 1000 Hz, filtered using a 100 Hz low pass filter and referenced to the right mastoid. Offline, EEG data were re-referenced to averaged mastoids and filtered using a Butterworth zero phase filter (0.1-30 Hz, 12 dB/oct). Vertical and horizontal eye movements were removed using an independent components analysis (Makeig, Jung, Bell, Ghahremani, & Sejnowski, 1997). Data were segmented into 1,100ms epochs (including a 100ms prestimulus baseline) around each face pair (i.e., angry/neutral; happy/neutral; neutral/neutral). Trials with amplitudes ± 100 μV were removed using an automatic artifact rejection procedure. Artifact-free segments were then averaged separately for each face pair. A minimum of nine trials per trial type were required to include in further analyses. The number of trials included in each ERP ranged from 17 to 144 and the mean number of trials used to calculate each participants’ ERP was as follows: angry/neutral: M = 114.22, SD = 24.5; happy/neutral: M = 114.6, SD = 23.8; neutral/neutral: M = 58.3, SD = 13.1. The mode for each trial type was ≥67.

Event-related brain potentials components were identified based on visual inspection of the grand average waveforms (see Figure 1 and Data S1) and the existing ERP selective attention literature in children and adults (Eldar et al., 2010; Mueller et al., 2009; Perez-Edgar, Fox, Cohn, & Kovacs, 2006; Rossignol et al., 2013). In order to balance the high levels of variability that results from lower signal-to-noise ratios in young children (Hoyniak, 2017) without over-valuing the peak value (Luck, 2014), each ERP component was calculated using the mean amplitude ± 25ms around the peak value within each of the following time intervals: P1 (100-160ms); anterior N1 (100-220ms); posterior N1 (120-220ms); and P2 (250-350ms). ERPs beyond the P2 component (e.g., the P3 and the late positive potential [LPP]) were not evaluated due to contamination by the target probe which appeared 500ms after face stimulus onset. Separate ERPs were calculated for each face pair and each component. P1 and P2 components were maximal over occipital sites and were calculated as an average across O1, Oz, and O2. The N1a component was maximal over frontocentral electrodes and was calculated as an average across FC3, FCz, and FC4. The N1p was maximal over parietal sites and was calculated as an average across P3, Pz, and P4. Although several recent studies examine the N2pc component as a reliable and valid measure of selective attention and anxiety-related threat biases in adults (Kappenman et al., 2014, 2015), we could not definitively identify the N2pc component in the present study. This may be due to age-related changes in the N2pc (Couperus & Quirk, 2015). Therefore, exploratory N2pc analyses are included in Data S1 but are not discussed further.

FIGURE 1.

Grand average waveforms elicited by angry, happy, and neutral faces averaged across frontocentral, parietal, and occipital sites (N = 128)

2.6 |. Analytic strategy

Behavioral data from this sample and their associations with anxiety have been reported previously as part of a larger sample of young children (Briggs-Gowan, Grasso, et al., 2016; Briggs-Gowan et al., 2015). Hence, the current analyses focused on the novel ERP measures of selective attention to each face emotion pair, as well as associations between irritability and the behavioral and ERP measures of selective attention to each face emotion pair. Findings from the behavioral data analyses within this study sample are included in Data S1. Correlations between variables are presented in Table 2.

TABLE 2.

Pearson correlations between demographic, irritability, anxiety, behavioral, and ERP variables

Measure	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21
1. Sex (months)	—	−0.09	0.10	−0.13	−0.14	0.07	−0.04	0.04	−0.08	−0.14	−0.15	−0.18*	0.21*	0.12	0.10	0.07	0.02	0.02	−0.20*	0.22*	0.21*
2. Age	—	—	0.45**	−0.26*	−0.24*	0.08	−0.19*	−0.05	0.16*	−0.04	−0.04	−0.05	−0.10	−0.09	−0.09	0.05	0.05	0.08	−0.06	−0.12	−0.08
3. Nonverbal reasoning	—	—	—	−0.02	−0.31**	−0.07	−0.06	0.00	−0.03	−0.02	−0.07	−0.07	−0.11	−0.17	−0.17*	−0.11	−0.08	−0.08	0.07	−0.01	0.03
4. Parent-reported irritability	—	—	—	—	0.11	0.23*	−0.05	−0.16	0.02	0.02	0.02	0.01	−0.14	−0.06	−0.09	−0.21*	−0.18	−0.20*	0.17	0.20*	0.20*
5. Observed irritability	—	—	—	—	—	0.21*	0.05	−0.16	−0.02	−0.15	−0.10	−0.07	−0.02	0.04	0.04	0.02	−0.01	0.01	−0.05	−0.01	−0.05
6. Parent-reported anxiety	—	—	—	—	—	—	0.02	−0.03	−0.10	−0.07	−0.06	−0.07	0.12	0.13	0.08	0.01	−0.02	0.04	0.04	0.07	0.05
7. Observed anxiety	—	—	—	—	—	—	—	0.14	0.12	0.07	0.12	0.12	0.11	0.20*	0.16	0.11	0.17	0.15	0.02	0.03	0.05
8. Angry RT-bias	—	—	—	—	—	—	—	—	−0.08	0.09	0.13	0.11	−0.09	−0.04	−0.06	−0.10	−0.03	−0.08	0.07	0.10	0.06
9. Happy RT-bias	—	—	—	—	—	—	—	—	—	0.18*	0.19*	0.22*	−0.02	0.08	0.01	0.03	0.05	0.12	0.03	0.05	0.06
10. Angry P1	—	—	—	—	—	—	—	—	—	—	0.96**	0.94**	0.02	0.03	0.04	−0.07	−0.07	−0.07	0.58**	0.57**	0.57**
11. Happy P1	—	—	—	—	—	—	—	—	—	—	—	0.94**	0.00	0.09	0.03	−0.08	−0.01	−0.09	0.57**	0.61**	0.57**
12. Neutral P1	—	—	—	—	—	—	—	—	—	—	——	—	0.02	0.06	0.14	−0.05	−0.02	0.03	0.52**	0.54**	0.56**
13. Angry N1p	—	—	—	—	—	—	—	—	—	—	—	—	—	0.87**	0.81**	0.73**	0.63**	0.60**	0.02	−0.03	0.00
14. Happy N1p	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.79**	0.67**	0.74**	0.59**	−0.01	0.03	0.00
15. Neutral N1p	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.62**	0.56**	0.71**	0.00	−0.01	0.11
16. Angry N1a	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.86**	0.82**	−0.22*	−0.25**	−0.19*
17. Happy N1a	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.79**	−0.19*	−0.16	−0.14
18. Neutral N1a	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	−0.17	−0.21*	−0.09
19. Angry P2	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.95**	0.91**
20. Happy P2	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	0.90**
21. Neutral P2	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—	—

^*p<.05.

^**p≤.001.

To test whether selective attention differed among face emotion pairs (angry/neutral, happy/neutral, and neutral/neutral) in the full sample (N = 128), repeated measures ANCOVAs were conducted with Emotion (angry/neutral, happy/neutral, and neutral/neutral) as the within-subjects factor, age, sex, and nonverbal reasoning as covariates, and ERP amplitude to the face emotion pair as the dependent variable. Congruency was not included in these models because the faces were presented prior to the target stimulus. These analyses were conducted using IBM SPSS Statistics v.22. The Greenhouse-Geisser correction was used when analyses violated sphericity assumptions.

To test our primary hypothesis that irritability will be associated with behavioral and ERP measures of selective attention to angry faces, we used PROC MIXED in SAS with a maximum likelihood model and unstructured covariance matrices to predict the RT bias score or ERP amplitude to each of the different face pairs. Within each model, irritability was entered as a continuous predictor and emotion (angry/neutral, happy/neutral, and neutral/neutral)² was included as a categorical repeated measure. An irritability × emotion interaction term was also included in each model to test for unique the associations between irritability and attention bias to angry faces. Age, sex, nonverbal reasoning scores, and anxiety were included as covariates. The covariates were entered simultaneously with the irritability and emotion predictors. The first set of analyses used the symptom-based measures of irritability and anxiety from the PAPA (N = 114). The second set of analyses used the observation-based measures of irritability and anxiety from the DB-DOS and ANX-DOS (N = 121).

As a conservative secondary analysis, for each of the significant findings, we repeated the analyses after excluding outliers as determined by the median absolute deviation (MAD) method (Leys, Ley, Klein, Bernard, & Licata, 2013). We did not adopt this approach initially because the removal of outliers based on statistical values, especially post-hoc, without corroborating evidence indicating that the data are unreliable (e.g., the experimenter observed the participant not paying attention) is not recommended (Jones, 2019). Removing these data may eliminate data from extreme ends of the distribution and may be particularly problematic in studies of clinical and pediatric populations, where variability is greater than in nonclinical and adult populations (Jones, 2019). However, because outliers can impact the study results, we adopted this secondary approach to evaluate whether our findings hold when more conservative approaches are used.

3 |. RESULTS

3.1 |. Attention to emotional faces in young children

Accuracy and RT did not differ across emotion categories or for congruent and incongruent trials (see Data S1 for detailed information about these analyses).

Grand average ERP waveforms elicited by the face stimuli at the three main scalp regions are presented in Figure 1 (see Data S1 for the grand averages at all scalp sites). Means and standard deviations for each component are in Data S1. Consistent with the behavioral data (Data S1) and prior ERP studies in older children (Bechor et al., 2018; Thai et al., 2016), neural responses to angry/neutral, happy/neutral, and neutral/neutral faces did not differ significantly for any of the ERP components: P1 (Emotion F(2,248) = 0.59, p = .55, ${\eta }_{\text{p}}^2$ = 0.005); N1a (Emotion F(2,248) = 0.18, p = .81, ${\eta }_{\text{p}}^2$ = 0.001); N1p (Emotion F(2,248) = 0.27, p = .73, ${\eta }_{\text{p}}^2$ = 0.002); P2 (Emotion F(2,248) = 2.06, p = .14, ${\eta }_{\text{p}}^2$ = 0.02).

3.2 |. Symptom-based measures of irritability and associations with behavioral and neural responses to emotional faces

The symptom-based measure of irritability was unrelated to RT-based measures of attentional bias (p = .40). The irritability × emotion interaction term was also nonsignificant (p = .10).

Higher irritability symptom scores were associated with larger (i.e., more negative) N1a amplitudes (B = −0.69; SE = 0.30; F(1, 108) = 5.38, p = .02), but no irritability × emotion interaction emerged (p = .66) suggesting greater selective spatial attention to all faces with increasing irritability. A similar but nonsignificant pattern for irritability has emerged for the N1p component (B = −0.53; SE = 0.30; F(1, 108) = 3.31, p = .07). Irritability was unrelated to P1 and P2 amplitudes (ps > 0.20). The irritability × emotion interaction term was nonsignificant for the N1p, P1, and P2 components (ps > 0.35).

3.3 |. Observation-based measures of irritability and associations with behavioral and neural responses to emotional faces

Observation-based irritability in the parent context was unrelated to the RT-based measures of attention bias to the happy and angry faces (B = −1.94; SE = 5.84; F(1, 234) = 2.79, p = .10). The irritability × emotion interaction term was also nonsignificant (p = .19).

An interaction between emotion and the observation-based measure of irritability in the parent context emerged for the P1 component (F(2,238) = 4.74, p = .01). To explore this interaction, we examined estimates of the relationship between irritability to angry/neutral face pairs relative to neutral/neutral face pairs. Estimates for the association between irritability and happy/neutral face pairs relative to neutral/neutral face pairs were also generated. Every one unit increase in observed irritability was associated with a reduction in P1 amplitudes to angry/neutral faces (B = −0.31; SE = 0.10; t = −3.05, p = .002) relative to neutral/neutral faces (see Figure 2). The estimate for happy/neutral face pairs was not significant (B = −0.11; SE = 0.10; t = −1.19, p = .24). Observation-based irritability was not related to N1p, N1a, or P2 amplitudes (ps > 0.50). The irritability × emotion interaction term was also nonsignificant for those components (ps > 0.30).

FIGURE 2.

Association between observation-based measures of irritability and P1 amplitude for the different face emotions (n = 121). The mixed model revealed an irritability x emotion interaction. The estimates from the interaction indicate that higher irritability is associated with a larger decrease in P1 amplitude to angry/neutral face pairs relative to neutral/neutral face pairs. The slope for the happy/neutral face pairs did not differ from neutral/neutral face pairs. Note: low and high irritability reflect scores 1 standard deviation (SD) below and above the sample mean, respectively. Graph values have been adjusted for age, developmental level, sex, and anxiety

Following a suggestion made by an anonymous reviewer, we conducted exploratory analyses of observed irritability from the examiner context to investigate how contextual factors may influence the neurocognitive measures associated with irritability.³ Observed irritability during the examiner context was unrelated to any behavioral or ERP component and no interactions between emotion and irritability emerged (ps > 0.10).

3.4 |. Secondary analysis after excluding outliers

As a final analytic step, we adopted a conservative approach and re-ran the analyses that generated the significant irritability-related findings above after excluding outliers based on the MAD method. The N1a and P1 findings were robust following the exclusion of outliers (results available upon request).

4 |. DISCUSSION

The present study capitalized on the temporal sensitivity of ERPs to extend prior behavioral findings in samples of older children (Hommer et al., 2014; Salum et al., 2017). Findings link symptom-based measures of irritability with enhanced selective spatial attention toward faces and observation-based measures of irritability in the parent context with P1 amplitude reductions for angry/neutral relative to neutral/neutral face pairs. Importantly, these effects existed after accounting for co-occurring anxiety, which has been independently associated with both behavioral and ERP markers of selective attention on the dot-probe task (Bar-Haim et al., 2005; Fu & Pérez-Edgar, 2019). These findings join an emerging literature identifying irritability-related neurocognitive mechanisms in young children (Deveney et al., 2019; Dougherty et al., 2018; Grabell et al., 2017; Grabell, Olson, Tardif, Thompson, & Gehring, 2016; Kessel, Dougherty, et al., 2016; Kessel, Meyer, et al., 2016; Li, Grabell, Wakschlag, Huppert, & Perlman, 2016; Perlman et al., 2015; Perlman, Luna, Hein, & Huppert, 2013). Further, these findings point to altered attention toward socioemotional stimuli among children with higher levels of irritability that may contribute to future interpersonal difficulties and/or risk for later emotional disorders. These patterns also highlight the importance of a multi-method developmentally sensitive approach for assessing the behavioral patterns that may have distinct neurocognitive correlates.

4.1 |. Symptom-based measures of irritability and increased attention to emotional faces

The present findings failed to replicate prior data on RT-based angry biases in older irritable children and adolescents (Hommer et al., 2014; Salum et al., 2017; c.f. Kircanski et al., 2018). Nevertheless, findings for the N1 component were consistent with prior behavioral findings. The N1 component indexes the selective attention—an early attention filter by which information is selected for further processing (Vogel & Luck, 2000). Therefore positive associations between parent-reported irritability and N1 amplitudes to all face types suggest that faces elicit greater selective attention in irritable children. This is consistent with neuroimaging data linking elevated irritability with greater activation in attention-related regions on the dot-probe task (Kircanski et al., 2018). Contrary to our hypotheses, parent-reported irritability was not associated with enhanced selective attention for angry stimuli, specifically. This may be due to the fact that our young sample of children (4–7 years) is still developing the ability to accurately label face emotions (Batty & Taylor, 2006; Herba & Phillips, 2004). Enhanced attention to faces among young children with higher irritability may set the stage for future attention biases toward angry faces, specifically, as the children learn to accurately label emotional expressions. That is, as children refine their ability to discriminate between different emotional expressions, those with higher levels of irritability may begin to display the types of selective attention toward angry faces that have been observed in older populations (Hommer et al., 2014; Salum et al., 2017). Longitudinal studies will be necessary to test this hypothesis.

4.2 |. Observation-based measures of irritability and increased attention to emotional faces

Although an angry-specific bias was detected among children displaying higher levels of irritability in the context of a laboratory paradigm completed with a parent, the association was in the opposite direction than predicted. Higher irritability scores were associated with relative P1 amplitude reductions to angry/neutral versus neutral/neutral face pairs. In most dot-probe ERP studies, smaller P1 amplitudes are interpreted to reflect reduced attention to emotional stimuli. Therefore, this finding suggests that higher observed irritability in young children is associated with reduced early selective attention to angry faces relative to neutral faces. This may reflect the developmental and methodological differences. Prior behavioral and neuroimaging work linking irritability with enhanced attention to angry faces (Hommer et al., 2014; Kircanski et al., 2018; Salum et al., 2017) has relied on parent reports of irritability symptoms and has been conducted in older populations than those in the present study. In addition, this work utilized RT and fMRI techniques with lower temporal resolution and did not isolate responses to the faces. The ERP techniques used in the present study index neural responses to the faces, specifically, with high temporal resolution.

The P1 reductions are consistent with an emerging literature linking distress/anxious-misery symptoms—including depression and generalized anxiety disorder, which share genetic and longitudinal associations with irritability (Savage et al., 2015; Stringaris, Cohen, Pine, & Leibenluft, 2009; Stringaris, Zavos, Leibenluft, Maughan, & Eley, 2012)—with reduced physiological responses to threatening stimuli (Lang, McTeague, & Bradley, 2016; McTeague & Lang, 2012; McTeague et al., 2010). It is also consistent with the studies linking clinically significant irritability in school-age children and adolescents with reduced amygdala activation to affective stimuli (e.g., Brotman et al., 2010; Deveney et al., 2013; Wiggins et al., 2016; c.f. Tseng et al., 2019). Given the amygdala’s influence on attention toward threat (Anderson & Phelps, 2001), such amygdala perturbations may result in relative reductions in attention toward angry stimuli among young children with higher irritability. Future work on this topic is necessary given the behavioral and neuroimaging evidence linking clinically significant irritability with facilitated attention toward angry faces in older samples (Hommer et al., 2014; Kircanski et al., 2018; Salum et al., 2017).

4.3 |. Reconciling discrepant association patterns across varied methods of irritability assessment

Researchers continue to debate how to best define and measure irritability, especially across different developmental stages (Avenevoli, Blader, & Leibenluft, 2015; Wakschlag et al., 2012, 2015). Mechanistic studies have relied on a range of assessment techniques including clinical diagnoses, child- and parent-reported symptom ratings, and observational paradigms. While multi-method approaches are strongly encouraged for assessments of social–emotional functioning in young children, many measures have not been specifically designed to assess the irritability in a developmentally sensitive way. In addition, there has been limited research into whether different methods of irritability assessment identify unique neurocognitive mechanisms in other early childhood samples and at older ages.

The present study compared behavioral and ERP patterns associated with two valid and reliable measures of irritability in young children: parent-reported irritability symptoms or observed behavior during a standardized laboratory paradigm (Dougherty et al., 2015, 2013; Petitclerc et al., 2015; Wakschlag et al., 2008a; Wakschlag, Hill, et al., 2008). The two measures did not correlate significantly (r = .11, p = .25) suggesting that they captured different features of irritability that were, in turn, sensitive to different neurocognitive mechanisms. Elevated irritability on the symptom-based measure reflects parent reports of frequent and impairing temper outbursts as well as tendencies to get angry or become easily frustrated over a 3-month period. As such, children with elevated scores on the PAPA may reflect children with more severe and impairing irritability. Children with this profile showed preferential early attention toward faces which is broadly consistent with prior behavioral work in older children whose irritability was also assessed using clinical symptoms (Hommer et al., 2014; Salum et al., 2017). In contrast, elevated irritability on the observational paradigm reflects difficulty regulating irritability during a discrete laboratory challenge (Wakschlag et al., 2008b; Wakschlag, Hill, et al., 2008). Thus the DB-DOS may capture a snapshot of children’s dysregulated irritability which is pervasive across demand contexts, but is not informative about persistence in the same way that parent report of the history of irritability is. In addition, the DB-DOS may assess irritability in “real time” and may be less susceptible to recall biases and parent perceptions of behaviors (Wakschlag, Hill, et al., 2008). This observed profile was associated with reduced early attention to angry faces. To the best of our knowledge, no studies have linked observation-based measures of irritability with measures of selective attention; therefore, future research is necessary to test whether observed irritability is associated with reduced attention in other samples of young children and at older ages (since this has not been tested in studies of older youth).

Exploratory analyses also suggest that the neurocognitive mechanisms associated with irritability may vary according to the context in which irritability is assessed. While observed irritability in the parent context was associated with reduced early attention to angry faces, observed irritability in the examiner context was unrelated to any behavioral or ERP measure. These findings join prior research suggesting that children’s behavior varies across contexts in this observational paradigm (Dirks, De Los Reyes, Briggs-Gowan, Cella, & Wakschlag, 2012; Petitclerc et al., 2015; Wakschlag et al., 2008b; Wakschlag, Hill, et al., 2008). However, replication of these findings is essential because the examiner-context analyses did not include a measure of observed anxiety from the same context which may have impacted the findings in unanticipated ways.

Together, the present data suggest that research into informant and contextual discrepancies are important for understanding the neurocognitive mechanisms associated with irritability. The present data provide preliminary evidence linking more severe and chronic irritability with preferential attention to faces, whereas difficulty modulating irritability during a specific laboratory situation with a parent was associated with reduced attention to angry faces. Further study in independent samples to plumb this methodologic variation will be important for ascertaining its salience and informativeness (including relative utility of the different methods for prediction to older ages, etc.).

The present study focused on facilitated attention toward angry faces; however, attention can be differentially allocated to angry stimuli in other ways including the degree to which individuals avoid and/or disengage attention from angry faces (Cisler & Koster, 2010). Each of these attentional processes has been associated with anxiety (Bunford et al., 2017; Cisler & Koster, 2010) and may have links with irritability as well. Future research should also examine whether irritability is associated with atypical responses to other threatening stimuli such as fearful faces. While angry faces have theoretical importance for irritability (Brotman, Kircanski, Stringaris, et al., 2017; Leibenluft, 2017), associations between irritability and fearful faces would suggest broader disruption in threat circuitry and additional areas of shared dysfunction with anxiety.

4.4 |. Irritability and RT-based measures of attention bias

Contrary to our hypotheses, irritability was not associated with the RT-based measure of attention bias toward anger. This may suggest that irritability in young children is not associated with attention biases toward angry faces. However, this null finding may be explained by the developmental stage of our participants. In the anxiety literature, effect sizes are smaller for studies with children than with adults (for review see Fu & Perez-Edgar, 2019) and more robust in older versus younger children (Dudeney, Sharpe, & Hunt, 2015). The same may be true for irritability and RT-based measures of attention bias. As the literature on irritability and attention bias increases, we will be better able to test this hypothesis.

4.5 |. ERPs did not discriminate between emotional and neutral faces dot-probe in the full sample of young children

None of the ERPs distinguished between the different emotional faces in the full sample (N = 128). This suggests that, as a group, young children’s early attentional resources are drawn equally to emotional and neutral faces. These findings are consistent with other dot-probe ERP studies in children (Bechor et al., 2018; Thai et al., 2016) as well as studies in adults where modulation of these early components by emotional expression is often absent or only present among participants with higher levels of anxiety (e.g., Bar-Haim et al., 2005; Eldar et al., 2010; Pintzinger et al., 2017; Pourtois et al., 2004; Santesso et al., 2008; Shah et al., 2018). In contrast, later ERP components (e.g., P3 and LPP) are reliably modulated by stimulus emotion in both children and adults (e.g., Bondy et al., 2018; Hajcak & Dennis, 2009; Schupp et al., 2004). Unfortunately, we were unable to explore later components due to a combination of the task design (due to the timing of the face and the target stimuli) and age-related delays in the timing of these ERP components (Hajcak & Dennis, 2009; Hoyniak, 2017) which reduced the chance of capturing later components prior to target onset. Further research is necessary to explore both the typical neurodevelopmental patterns in these selective attention components and at what stage they are modulated by emotion. Such information will be critical to future studies attempting to distinguish between typical and atypical ERP markers in children with psychiatric symptoms.

5 |. LIMITATIONS

Although we examined an array of attention-related ERPs, we did not explore all ERP components associated with visual attention and emotional processing. As described above, event timing in the task prevented us from examining the LPP. Some dot-probe studies have explored the C1 component – an early component which indexes activity from primary visual cortex in the occipital lobe (Luck & Kappenman, 2012). Detection of the C1 component requires stimuli to be presented in a single visual field (Luck & Kappenman, 2012) and thus could not be assessed with our paradigm.

Despite the fact that it is challenging to obtain reliable measures of brain activity in very young children, data from the majority of children who attempted the dot-probe task were included in our final analyses. However, there were a few differences between the demographics and clinical characteristics of the included versus excluded children. Relative to excluded children, those with usable data had lower irritability scores during the laboratory observation, but had similar symptom-based irritability and anxiety scores as well as similar anxiety scores during the laboratory observation. In addition, the children in the present study were somewhat older, had higher nonverbal reasoning scores, were less likely to identify as African American, and performed more accurately on the task than children who were excluded (see Data S1). Although the effect sizes for these differences were small to moderate, our findings may underestimate the associations between irritability and behavioral and neural measures of attention biases toward angry faces among children who are younger, have lower nonverbal reasoning scores, exhibit more anger during laboratory observations, and identify as African American. In addition, the majority of our sample fell below the poverty line. Exploratory analyses indicated that poverty status did not explain associations between irritability and the ERP measures of threat bias; however, future research should examine whether the observed associations exist in a more representative sample.

Similarly, future research should also consider controlling for other psychiatric symptoms that have been associated with irritability as well as biased information processing. For example, externalizing symptoms and reactive aggression have been linked with reduced attention toward negative stimuli (Schippell, Vasey, Cravens-Brown, & Bretveld, 2003) as well as facilitated attention toward happy faces (Morales, Perez-Edgar, & Buss, 2016) on the dot-probe task.

6 |. CONCLUSIONS

The present study provides the first evidence suggesting that irritability in very early childhood may be associated with altered selective attention patterns, although the nature of the associations may vary according to whether irritability is measured by parent-reported symptoms or by coded behavior during a laboratory observation. These findings join an increasing literature linking irritability in early childhood with neurocognitive functioning (Deveney et al., 2019; Dougherty et al., 2018; Grabell et al., 2017, 2016; Kessel, Dougherty, et al., 2016; Kessel, Meyer, et al., 2016; Li et al., 2016; Perlman et al., 2015, 2013). Together, these findings may contribute to the development of early markers of clinically salient irritability that will facilitate treatment and prevention efforts. Finally, since this is the first ERP study of the dot-probe task conducted in young children, these findings provide important foundational knowledge about expected patterns that can be used in future investigations.