Developmental Patterns of Emotion Regulation in Toddlerhood: Examining Predictors of Change and Long-Term Resilience

Abstract

Emotion regulation is critical for optimal functioning across a wide range of domains and may be even more important for individuals in high-risk environments. While evidence suggests that childhood is generally a period of emotion regulation growth and development, research is needed to examine factors that may contribute to deviations from a typical trajectory. In a prospective study of 1,905 children, latent class growth analysis (LCGA) was used to identify trajectory groups of emotion regulation across toddlerhood (age 14–36 months), examine predictors of those trajectory groups from child temperament, parenting behaviors, and environmental risk, and explore predictions of resilience in 5^th grade from the identified groups. LGCA supported a three-class model, with a Stable Incline group, a Decline group, and a Catch-Up group. Child negative emotionality, positive and negative parenting, and environmental risk predicted group membership. These trajectory groups in toddlerhood were predictive of child resilient functioning in the 5^th grade. Our findings highlight the importance of utilizing developmental models of emotion regulation and provide implications for prevention and early intervention services to enhance emotion regulation development in early childhood.

The development of emotion regulation, or “behaviors, skills, and strategies, whether conscious or unconscious, automatic or effortful, that serve to modulate, inhibit, and enhance emotional experiences and expressions,” (Calkins & Hill, 2007, p. 229) is a critical milestone of childhood (Cicchetti et al., 1991). As social, academic, and occupational demands increase with age, it is advantageous to approach these new demands with a toolbox of reliable, flexible, and reflexive strategies for modulating emotional reactions. Further, emotion regulation may be even more critical for children growing up in high-risk environments (e.g., poverty, neighborhood violence, and residential instability; Raver, 2004), as emotion regulation has been identified as a predictor of resilience under adverse conditions (Cicchetti, 2013). Thus, characterizing individual differences in emotion regulation development during early childhood and identifying key child- and family-level predictors and long-term outcomes of these early developmental trajectories is a public health and scientific priority.

Development of emotion regulation during toddlerhood

Toddlerhood is a particularly unique developmental period during which linguistic, attentional, and motor abilities rapidly emerge. These emerging domains, as well as the development of control over them (e.g., attentional control), results in toddlers’ unfolding capacity for self-regulation. (Bridgett et al., 2015; Eisenberg et al., 2004; Kochanska et al., 2001; Rueda et al., 2004). Self-regulation is a broad and overarching term referring to the general ability to control one’s inner states (e.g., attention, thoughts, emotions) and behaviors to achieve one’s goals (Vohs & Baumeister, 2016) Within the broader domain of self-regulation, emotional self-regulation (i.e., emotion regulation) refers to the specific ability to control emotional experiences and expressions (Calkins & Hill, 2007). During the toddler period, a range of emotion regulation skills begin to appear (Calkins, 2007), including increased implementation of adaptive regulatory behavioral strategies such as distraction away from the stressor and calm bids for information about the stressor (Bendezú et al., 2018). A strong foundation in emotion regulation beginning in toddlerhood has been identified as a key predictor of successful navigation of upcoming challenges (Blair, 2002; Calkins, 2007; Diamond & Aspinwall, 2003; Feldman, 2015; Kopp, 1982). Indeed, emotion regulation during toddlerhood has been prospectively associated with mental and physical health outcomes, including reduced risk for externalizing behaviors (Blandon et al., 2010), impulsivity (Graziano, Keane, et al., 2010), and pediatric obesity (Graziano, Calkins, et al., 2010). Because of the enduring effects of early emotion regulation, recent efforts have been directed to examination of development within this domain over time.

Importantly, although measures of emotion regulation are generally correlated across the first five years of life, the magnitude of these associations has been relatively small (Feldman, 2009), which reflects the malleability of regulatory processes during this early developmental period and a need to examine changes in emotion regulation across time. In line with the emergence of the attentional, linguistic, and motor abilities during toddlerhood that facilitate emotion regulation, prospective, longitudinal studies have found that, on average, emotion regulation increases over early childhood (Blandon et al., 2008) and emotion dysregulation generally decreases (Fabes et al., 2002; Noroña et al., 2017). However, there are likely many individual differences in patterns of emotion regulation development, with some toddlers advancing in their emotion regulation skills at different rates than others. To date, relatively few studies have investigated individual differences in trajectories of emotion regulation development starting from toddlerhood, an essential step to differentiating early developmental pathways.

Person-centered analyses have been valuable in other fields (e.g., pediatrics) in identifying developmental stages during which trajectories are forming and prevention and early intervention efforts may be most effective (Ziyab et al., 2014). Emerging studies employing person-centered approaches (e.g., latent class growth analysis) that model within-person changes are beginning to detect emotion regulation trajectory groups across a range of developmental periods. These examinations have included parent-report of emotion regulation difficulties in toddlerhood (age 1–4; Jusiene, Breidokiene, & Pakalniskiene, 2015), observed emotion regulation strategy use during early childhood (Bendezú et al., 2018; Supplee et al., 2011) observed emotion regulation and behavior regulation in middle childhood (age 3–7; Montroy, Bowles, Skibbe, McClelland, & Morrison, 2016), and self-report of intentional self-regulation in adolescence (grade 5–11; Bowers et al., 2011). Across these prior studies, person-centered approaches resulted in identification of trajectory groups including an at-risk group (e.g., pronounced delay or decline in regulation) and a more optimal developmental group (e.g., elevated regulation). Some studies also identified a third group that starts similar to the at-risk group and improves over time (Bendezú et al., 2018; Jusiene et al., 2015). It is important to point out that these prior studies have been inconsistent in their measurement of emotion regulation (parent report vs. observed child behavior) and in their identified trajectory classes. Thus, the present study aims to apply a person-centered approach to analyze developmental trajectories of emotion regulation in toddlers, using an observed measure of regulatory behaviors. We also extend existing findings by testing whether empirical class membership is associated with child- and family-level factors and predicts long-term school-age outcomes.

Internal and external correlates of early emotion regulation

Given the malleability of emotion regulation during early childhood, a large body of work has examined correlates of and influences on this critical domain. Two broad factors have been associated with the early development of emotion regulation: factors internal to the child and factors external to the child (for the seminal review, see Calkins, 1994). Within the child, an internal factor that has been widely investigated in relation to emotion regulation is early temperament. Negative emotionality (i.e., an individual’s reactivity and intensity of emotional expressions) is one aspect of temperament that has been consistently associated with difficulties with emotion regulation (Calkins & Johnson, 1998; Kim & Kochanska, 2012; Thomas, Letourneau, Campbell, Tomfohr-Madsen, & Giesbrecht, 2017; Zalewski, Lengua, Wilson, Trancik, & Bazinet, 2011). A limitation of this area of research is that the majority of work has been cross-sectional and thus unable to examine how temperament is related to change in emotion regulation over time. The current study directly addresses this gap and determines whether early negative emotionality is associated with profiles of emotion regulation development over toddlerhood.

A number of external factors have been consistently linked to the development of emotion regulation, with parenting behaviors being a particularly key predictor in early childhood (for reviews, see Morris, Criss, Silk, & Houltberg, 2017; Morris et al., 2007). Positive parenting behaviors, such as sensitivity (Feldman, 2007; Fogel, 1993; Halligan et al., 2013; Leerkes et al., 2009) and expression of positive affect (Cumberland-Li, Eisenberg, Champion, Gershoff, & Fabes, 2003; Eisenberg et al., 2001) facilitate adaptive emotion regulation development during childhood, whereas the absence hinders it (NICHD Early Child Care Research Network, 2004). Negative parenting behaviors, such as intrusiveness (Cabrera et al., 2007; Graziano, Keane, et al., 2010; Stevenson & Crnic, 2013) and parental expression of negative emotions (Morris et al., 2017) increases a child’s risk for difficulties with emotion regulation. A critical limitation of this prior work is that parenting behaviors have been associated with emotion regulation outcomes at an outcome time point. Because emotion regulation develops rapidly over early childhood, it is important to test whether parenting is associated with trajectories of emotion regulation over time.

Beyond the parenting environment, a particularly salient aspect of the broader family environment that affects general child socioemotional functioning is socioeconomic status (Brooks-Gunn et al., 1997; Evans & English, 2002; Raver, 2004). In particular, a large body of evidence has documented associations between poverty and early difficulties with self-regulation, a broad measure of variability in the capacity to direct and focus actions or attention in the service of goals (Flouri et al., 2014; Lengua et al., 2007; McCoy & Raver, 2014). Part of the reason low-income has such a tenacious effect on child development is that poverty is linked with a range of co-occurring, persisting family-level stressors, such as teenage motherhood, single parenthood, parental unemployment, and lower parental educational attainment (Bocknek et al., 2009; Riva Crugnola et al., 2014). Early exposure to chronic environmental stressors affect the development of the brain, particularly prefrontal cortex-corticolimbic connectivity, and stress-response systems, such as the HPA-axis, which in turn impair emotion regulation (Blair, 2010). For example, exposure to these cumulative poverty-related risk factors predicts externalizing behavior problems in childhood (Ackerman et al., 1999) and adolescence (Li et al., 2007). Examination of whether these relations translate to specifically to emotion regulation (vs. self-regulation more broadly) is an important next step.

Emotion regulation as a protective factor

In a promising line of research directed at resilience, emotion regulation, and self-regulation more broadly, have been found to be protective factors for children in high-risk environments, with evidence coming from studies of poverty (Ingoldsby et al., 1999), maltreatment (Cicchetti et al., 1993), and overall environmental risk (Lengua, 2002). In one such study, self-regulation moderated the association between an environmental risk index and positive adjustment (i.e., social competence, self-worth, life satisfaction), with children low in self-regulation evidencing more vulnerability to risk (Lengua, 2002). The critical next step is to identify predictors of emotion regulation in children experiencing adversity, and to replicate findings for emotion regulation having long-term implications in this population.

Limitations of previous studies

A primary limitation of current work is little simultaneous examination of early influences on emotion regulation, such that many studies have examined temperament, positive parenting, negative parenting, or environmental risk, not a simultaneous examination that allows for comparison across predictors. In addition, while a wealth of studies has documented the link between emotion regulation at one time point and a subsequent outcome (Graziano et al., 2007; Romens & Pollak, 2012; Ursache et al., 2013; Zeman et al., 2002), work is needed to characterize the longitudinal patterns of emotion regulation across toddlerhood and examine how these patterns predict later functional outcomes. Given the developmental nature of emotion regulation in the early years of life, potentially when these trajectory groups are first identifiable and most sensitive to change, interrogation of whether one’s emotion regulation trajectory predicts long-term outcomes is critical.

The present study

The primary goals of the current study were to directly address these gaps and employ a person-centered approach to answer the following questions: (1) Are there subgroups of children who exhibit more rapid increases or even decreases in emotion regulation in toddlerhood? (2) If so, do established early internal and external predictors of emotion regulation (e.g., parenting, temperament) predict subgroup membership? (3) Lastly, do the trajectory groups have any predictive validity for functional outcomes later on in development?

We advanced these aims in a sample of children enrolled in the Early Head Start Research and Evaluation (EHSRE) project (United States Department of Health and Human Services, Administration for Children and Families, 1996–2010). This study builds upon a well-established literature of sources of individual differences in childhood emotion regulation while adding a developmental lens to our understanding of early emotion regulation development and its long-term effects. Findings have the potential to identify risk factors for a poor developmental profile of emotion regulation in early childhood, as well as protective factors that portend an optimal profile associated with long-term resilient functioning.

Method

Participants

Participants were 1,905 children and families involved in the EHSRE project, funded by the Administration on Children, Youth, and Families and conducted by Mathematica Policy Research (MPR) between 1996 and 2010. MPR obtained IRB approval for all study procedures. The EHSRE dataset is public use, made freely available by Child Care & Early Education Research Connections (United States Department of Health and Human Services. Administration for Children and Families, 2011a). Low-income families with pregnant mothers or children up to 12 months old at the time of enrollment were recruited for participation. Participants were randomly assigned to the program group (received Early Head Start; n = 1513) or the control group (received community services as usual, n = 1488) at the site level. Analyses conducted by MPR revealed that, as expected, baseline characteristics of the two groups were comparable (United States Department of Health and Human Services. Administration for Children and Families, 2011b). The full sample was comprised of 3,001 children and their families, and the present sample was reduced to participants who had data for at least one key study variable. Families included in the present sample had overall higher income than families excluded from the present sample (p = .002), though included families were predominantly of low-income status (i.e., mean income was at 63% of the poverty line).

The EHSRE study was implemented across 17 sites in the United States, representing diverse program models, racial or ethnic makeup, urban-rural location, program auspice, and program experience in serving infants and toddlers. Three phases comprised the data collection: (1) birth-to-three, (2) pre-kindergarten follow-up, and (3) elementary school (5^th grade) follow-up. The birth-to-three phase consisted of assessments at child ages 14, 24, and 36 months. The present study utilized data from the first and third phases.

Approximately half (49.2%) of the present sample of target children was female. In terms of ethnicity and race, 39.1% were non-Hispanic white, 33.5% black or African American, 23.0% of Hispanic origin, and 4.4% from other racial or ethnic groups. Most respondents spoke English as their primary language, but 20.2% did not. Among those who did not speak English as their primary language, 85.0% were Spanish-speaking. The recruited sample is high-risk, as, at baseline, approximately half of the parents involved in the study were unemployed, 72.9% were single parents, 51.8% were receiving welfare, and only 55.4% had graduated from high school or obtained a GED. Lastly, 38.3% of the children were born to teenage mothers.

Procedures

At each visit, parents were interviewed about services, family and child health, child development, and family functioning. In addition, field interviewers recorded information from their observations of child behavior and home environments. Direct child assessments included standardized assessment batteries of cognitive and academic functioning and videotaped semi-structured parent-child interactions. The present study utilized data obtained from the parent interviews, observations of child behavior and home environments, and direct child assessments.

Measures

Temperament.

The Emotionality, Activity, Sociability, and Impulsivity (EASI; Buss & Plomin, 1984) is a 40-item questionnaire that evaluates respondents based on four temperament factors (i.e., emotionality, activity, sociability, and impulsivity). It was administered during the 14-month parent interview. Parents were asked how accurately behaviors or personality traits characterized their child. The 5-item emotionality subscale was used in the present study to represent negative emotionality; it consists of items such as “often fusses and cries,” “reacts intensely when upset,” and “tends to be somewhat emotional.” Item response scales range from 1 (not very typical of your child) to 5 (very typical of your child). Each child’s emotionality score was an average of the five item scores.

Emotion regulation.

The Bayley Behavior Rating Scale (BRS; Bayley, 1993) is a 30-item examiner-report scale that rates children’s relevant behaviors and measures orientation or engagement, emotional regulation, and motor quality. The examiner who administered the Bayley Scales of Infant Development (BSID) completes the BRS immediately following administration, based on observations made during testing. The examiner assesses the child’s behavior by scoring BRS items on a 5-point scale, with 5 indicating more adaptive behavior (for example, less frustration and more cooperation). The present used the BRS Emotional Regulation subscale (7 items), which assesses the child’s ability to transition between tasks and test materials, expression of negative affect, persistence in attempting to complete tasks, and frustration with tasks during the assessment. Each child’s Emotional Regulation subscale score is the average of the 7 items. Based on the Bayley-II validation sample, score reliability is moderate to high, with internal consistency coefficients ranging from .73 to .90. For children 24 months of age, test-retest reliability coefficients for the three factor scale scores range from .61 to .71 (Bayley, 1993). In the EHSRE sample, Cronbach’s alpha internal consistency coefficient was .92. The BRS, along with the BSID, was administered at child ages 14, 24, and 36 months.

Parent behavior.

Parent behavior was measured during a parent-child semi-structured play task at child age 14 months. The parent and child were given three bags of toys, and parents were asked to play with the toys in sequence. This play task was adapted for the EHSRE study from the Three Bag coding scales used in the NICHD Study of Early Child Care (NICHD Early Child Care Research Network, 1999). The play task was videotaped, and child and parent behaviors were coded by child development researchers, masked to the program status of each child. Aspects of parent’s behavior with the child were rated on a seven-point scale (“1” indicating a very low incidence of the behavior and “7” indicating a very high incidence of the behavior). Coders reached 85% agreement or higher with a master coder before coding interactions independently. Fifteen percent of all tapes assigned thereafter were used to check coders’ ongoing reliability. Coder intraclass correlations for domains of parent behavior (e.g., positive and negative behaviors) ranged from .68 to .76 (percent agreement within 1 point ranged from 87% to 96%)

Positive parenting.

The positive parenting composite is comprised of coder ratings of parental sensitivity, cognitive stimulation, and positive regard during play with the child. The resultant score for each parent’s positive parenting was an average score of each of these components. Behavior consistent with positive parenting includes acknowledgement of the child’s affect, vocalizations, and activity, facilitating the child’s play, taking advantage of the activities and toys to facilitate learning, development, and achievement, praising the child, expressing affection, and showing clear enjoyment of the child. The positive parenting composite in the full EHSRE sample was determined to have acceptable reliability; Cronbach’s alphas published in the EHSRE Codebook by race were as follows: Latina (n = 431) = .78, African American (n = 573) = .83, White (n = 693) = .83 (United States Department of Health and Human Services, n.d.)

Negative parenting.

The negative parenting scale measures the parent’s expression of discontent with, anger toward, disapproval of, or rejection of the child. High scores on negative regard indicate that the parent used a disapproving or negative tone, showed frustration, anger, physical roughness, or harshness toward the child, threatened the child for failing at a task or not playing the way the parent desired, or belittled the child. While data are not available regarding the internal consistency of this negative parenting scale within the EHSRE sample, observer ratings of punitive interactions overall had acceptable reliability (Cronbach’s alpha = .78).

Family environment risk.

A family environment risk index score was computed for each child from the five following established risks, collected at study intake: (1) birth mother under 20 years of age when child was born; (2) caregiver not employed, in school, or training; (3) caregiver neither married nor cohabitating; (4) caregiver receiving welfare; (5) and caregiver with no high school diploma or GED. Each of these was coded 0 (absent) or 1 (present), and they were summed for the family environmental risk index. Researchers have advocated looking at the cumulative effects of risk factors as opposed to studying singular risk factors, as (a) the effect of one isolated risk factor compared to multiple is minimal, and (b) a significant proportion of children experience a cluster of risk factors (Evans et al., 2013). These particular risks were selected because they include a range of co-occurring, persisting family-level stressors associated with poverty (Bocknek, Brophy-Herb, & Banerjee, 2009; Riva Crugnola, Ierardi, Gazzotti, & Albizzati, 2014). Previous work has demonstrated that cumulative risk factors relate with externalizing behavior problems in childhood (Ackerman, Schoff, Levinson, Youngstrom, & Izard, 1999) and adolescence (Li, Nussbaum, & Richards, 2007).

Long-term resilience.

The cumulative resilience index is based on recent recommendations in the resilience field to measure resilience in a multi-domain, multi-dimensional fashion (Cicchetti, 2013). As resilient functioning necessitates a threat to adaptation and development, the threat in the present ESHRE sample is poverty - an established chronic stressor with effects on child development (Horning & Rouse, 2002; Miller et al., 2011, 2015). The resilience index is based on the sum of 16 outcomes from multiple domains including physical health, academic and school performance, cognitive functioning, and socio-emotional functioning. Data were collected from direct assessments of child functioning as well as caregivers’ reports of child functioning when participants were in the 5^th grade. Each measure was scored from 0 to 1, with 1 indicating competent functioning. Cronbach’s alpha for the resilience composite was .7, which is considered acceptable (Santos, 1999). See Table 1 for descriptions of how measures were coded and used in the resilience index.

Table 1.

Measures included in the long-term resilience index

Domain	Measure	Instrument	Description	Scoring
Physical	Healthy weight	Body Measurement Index (BMI)	Direct measurement of child’s weight	1 if BMI below 95th percentile
	No chronic illness	--	Parent report	1 if no chronic illness
	Good general health	--	Parent report	1 if good general health
Academics	Consistent attendance	--	Parent report	1 if 0–2 absences during the school year .5 if 3 absences during the school year
	No Individualized Education Plan (IEP)	--	Parent report	1 if no IEP
	No grade retention	--	Parent report	1 if no retention
	Math achievement	Early Childhood Longitudinal Study-Kindergarten Cohort (ECLS-K; Tourangeau et al., 2015)	Assessment of child’s number sense, properties, and operations, measurement, geometry and spatial sense, data analysis, statistics, and probability, and patterns, algebra, and functions	1 if standard score above 16th percentile
	Reading achievement	ECLS-K (Tourangeau et al., 2015)	Assessment of child’s initial understanding, developing interpretation, personal reflection, and critical stance	1 if standard score above 16th percentile
Cognitive	Vocabulary	PPVT-III (Dunn & Dunn, 1997)	Assessment of child’s receptive vocabulary	1 if standard score above 16th percentile
	Fluid reasoning	WISC-IV Matrix Reasoning subscale (Wechsler, 2003)	Assessment of child’s fluid reasoning	1 if standard score above 16th percentile
Social-emotional	Internalizing problems	Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001)	Parent report of internalizing problems	1 if t-score below the borderline range were coded as 1; .5 if t-score in the borderline range
	Externalizing problems	CBCL (Achenbach & Rescorla, 2001)	Parent report of externalizing problems	1 if t-score below the borderline range were coded as 1; .5 if t-score in the borderline range
	Attention problems	CBCL (Achenbach & Rescorla, 2001)	Parent report of attention problems	1 if t-score below the borderline range were coded as 1; .5 if t-score in the borderline range
	Infrequent bullying	Panel Study of Income Dynamics – Child Development Supplement, Wave 2 (PSID-CDS2) Bullying Scale	Child’s self-report of the frequency of bullying at school	1 if bullied 0–2 times in the last month
	Infrequent delinquent behaviors	Developed by EHSRE researchers; drew from the work of Loeber, Stouthamer-Loeber, Van Kammen, and Farrington (1991) and the National Institute of Child Health and Human Development Study of Early Child Care and Youth Development	Child’s self-report of delinquent behaviors, e.g., stealing, property damage, smoking	1 if fewer than 3 delinquent behaviors .5 if 3 delinquent behaviors

Data Analytic Plan

We used Latent Class Growth Analysis (LCGA) to identify distinct trajectory groups for emotion regulation, using MPlus version 7 (Muthén & Muthén, 2009). LCGA is a semiparametric group-based approach that can estimate the mean parameter level at a given point in time (i.e., intercept) and the rate of increase or decrease over time (i.e., linear slope). This approach assumes that the population consists of a number of groups with different trajectories and seeks to identify them (Nagin, 1999). As it is unlikely that the population falls into truly distinct groups, the patterns should be viewed as the best approximation of generally distinct experiences (Kamp Dush et al., 2008).

We used a single conditional LCGA that tested uncentered predictors (i.e., significant covariates, child negative emotionality, positive parenting behaviors, negative parenting behaviors, environmental risk score) of trajectory class membership using multinomial logistic regression to estimate odds ratios (Jung & Wickrama, 2008). The model also included a distal outcome (i.e., resilient functioning in 5^th grade), predicted by class membership. Membership in each trajectory group was compared with a reference trajectory group (i.e., the group with most members). By simultaneously including continuous latent variables (intercept and growth slopes) as well as categorical latent variables (trajectory class) in the same conditional LCGA model, we reduced Type 1 error by minimizing multiple testing (Muthen, 2004). Furthermore, multinomial logistic regression within a conditional LCGA enabled predictions of the categorical latent variable “class” by using posterior probabilities to assign each individual fractionally to all classes, rather than forcing a 0/1 classification (Muthen, 2004). This is important because it is unlikely that any participant has a 100% probability of membership in any particular class, which is an infrequently acknowledged assumption when conducting separate follow-up multinomial logistic regressions outside of the LCGA model. That is, separate multinomial logistic regressions that predict “trajectory class” are actually predicting the most likely trajectory class, without accounting for each individual’s probability of not being a member of this class. By employing multinomial logistic regression within a conditional LCGA framework, the present models directly account for this error (Muthen, 2004).

We compared LCGA models to identify the optimal number of groups, the shape of the trajectory of each group, and the proportion of the sample belonging to each group. We determined the number of groups that best fit the data by evaluating the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Lo, Mendell, and Rubin (2001) likelihood ratio test (LMR-LRT), and the bootstrap likelihood ratio test (BLRT) for a series of models that varied in number of trajectory groups. We started with a model with 1 trajectory group and increased the group number by 1 in a stepwise fashion. Improved model fit is indicated by smaller AIC and BIC values and significant LMR-LRT and BLRT test statistics, while also maintaining no less than 1% of the sample in each group (Jung & Wickrama, 2008). In terms of missing data, as mentioned above, all included participants had data present for at least one key measure. All participants had complete data for negative emotionality, positive parenting, and negative parenting. Otherwise, about a quarter of participants were missing one of the emotion regulation time points, and less than half of the participants were missing the 5^th grade resilience score. Other missing data included race/ethnicity (1%) and environmental risk (4%). Little’s test (Little, 1988) determined data missingness to be Missing Completely at Random, x²(206, N = 1901) = 231.61, p = ns. Thus, we utilized full information maximum likelihood (FIML) to account for missing data. Finally, Wald chi-square tests were used to detect differences in the distal 5^th grade outcome between the trajectory classes.

Results

Descriptive statistics

Table 2 shows means, standard deviations, and Pearson correlations among key variables. Negative emotionality was negatively associated with positive parenting, emotion regulation at 14 and 24 months, and long-term resilience, and positively associated with negative parenting and environmental risk. Positive and negative parenting were inversely correlated. Environmental risk was negatively related to emotion regulation at all time points. Of note, these significant correlations were relatively weak (.07 – .37; Taylor, 1990). Emotion regulation demonstrated some stability over time, with correlations ranging from .27 to .37.

Table 2.

Means, standard deviations, and correlations among key variables

Variable	M	SD	1	2	3	4	5	6	7
1. Negative emotionality – T1	2.96	.95	1
2. Positive parenting – T1	3.94	1.06	−.13*	1
3. Negative parenting – T1	1.46	.79	.08*	−.38*	1
4. Environmental risk index – T1	2.64	1.19	.10*	−.27*	.15*	1
5. Emotion regulation – T1	3.69	.69	−.16*	.16*	−.10*	−.07*	1
6. Emotion regulation – T2	3.64	.80	−.09*	.22*	−.13*	−.10*	.27*	1
7. Emotion regulation – T3	3.93	.76	−.05	.16*	−.11*	−.12*	.12*	.37*	1
8. Resilience index – 5th grade	12.60	2.47	−.13*	.19*	−.13*	−.23*	.12*	.23*	.30*

Notes. T1 at child age 14 months. T2 at child age 24 months. T3 at child age 36 months.

^*p < .05.

LCGA results

Growth class characterization.

Goodness of fit comparisons for the full LCGA model are shown in Table 3. Goodness of fit tests indicated that a three-class model best fit the data. Our largest group (n = 1225; 64.2%) was characterized by inclining emotion regulation (z = 0.38; intercept B = 3.95, SE = .03, 95% CI [3.90, 4.00]; linear slope B = 0.07, SE = .02, 95% CI [0.03, 0.10]) across the three time points. Another group of children (n = 354; 18.6%) appeared to start relatively high (z = −0.17; intercept B = 3.57, SE = .06, 95% CI [3.46, 3.69]) but declined over time (linear slope B = −0.22, SE = .06, 95% CI [−0.34, −0.09]). The third group (n = 326; 17.1%) started relatively low (z = −1.35; intercept B = 2.76, SE = .09, 95% CI [2.52, 2.94]) but improved over time (linear slope B = 0.56, SE = .05, 95% CI [0.46, 0.67]). See Figure 1 for depiction of trajectory group classes, henceforth referred to as the Steady Incline, Decline, and Catch-Up groups, respectively.

Table 3.

Fit indices for LCGA models with 1 to 4 classes

Number of classes	AIC	BIC	LMR Test p	BLRT Test p
1	56156.54	56312.51	--	--
2	15015.96	15138.11	<.001	<.001
3	14859.05	15047.83	<.001	<.001
4	14794.54	15049.94	.53	<.001

Notes. Bold indicates the best fitting model. AIC = Akaike information criterion; BIC = Bayesian information criterion; LMR = Lo-Mendell-Rubin likelihood ratio test.

Figure 1.

Developmental patterns of emotion regulation in the first three years of life as identified by latent class growth analysis.

Predicting trajectory group class membership.

We used latent class multinomial logistic regression to model the probability of membership in each trajectory class from negative emotionality, positive parenting, negative parenting, and environmental risk index, controlling for child sex and ethnicity. Initial evaluation of program status (Early Head Start vs. services as usual) as a covariate revealed that it was not predictive of class membership; thus, program status was excluded from subsequent models for parsimony and power. Table 4 provides the statistics from the multinomial logistic regressions used to predict class membership. Negative emotionality increased likelihood of membership in the Catch-Up group, when compared to the Stable-Incline group (B = −0.67, SE = .10, 95% CI [0.47, 0.87] and the Decline group (B = −0.30, SE = .13, 95% CI [0.04, 0.56]). Positive parenting increased likelihood of membership in the Steady Incline group when compared to the Catch-Up group (B = −0.44, SE = .12, 95% CI [0.21, 0.67]) and the Decline group (B = −0.48, SE = .10, 95% CI [0.27, 0.69]); but positive parenting did not differentiate between the Catch-Up group and Decline group. Negative parenting increased likelihood of membership in the Decline group when compared to the Steady Incline group (B = 0.26, SE = .12, 95% CI [0.02, 0.50]) and the Catch-Up group (B = 0.25, SE = .12, 95% CI [0.01, 0.49]), but did not distinguish the Steady Incline group from the Catch-Up group. Lastly, the environmental risk index increased likelihood of membership in the Decline group as compared to the Steady Incline (B = 0.38, SE = .10, 95% CI [0.19, 0.57]) and Catch-Up (B = 0.26, SE = .10, 95% CI [0.06, 0.54]) groups, but did not differentiate the Steady Incline group from the Catch-Up group.

Table 4.

LCGA multinomial logistic regression predicting class membership

	Catch-Up vs. Steady Incline			Catch-Up vs. Decline			Decline vs. Steady Incline
	B	SE	OR	B	SE	OR	B	SE	OR
Sex (covariate)a	−0.84***	0.20	0.43	−0.40	0.28	1.49	−1.23**	0.25	3.44
Ethnicity (covariate)
Black	−0.08	0.24	0.92	0.29	0.30	0.75	0.20	0.25	0.82
Latino	0.73**	0.27	2.08	−0.12	0.33	1.12	0.62*	0.27	0.54
Other	0.39	0.47	1.47	1.85	2.10	0.16	2.23	2.05	0.11
Negative Emotionality	−0.67***	0.10	0.51	−0.30*	0.13	0.74	−0.37**	0.12	1.44
Positive Parenting	0.44***	0.12	1.56	0.04	0.14	0.97	0.48***	0.11	0.62
Negative Parenting	−0.01	0.13	0.97	−0.25*	0.12	1.28	−0.26*	0.12	1.30
Environmental Risk Index	−0.12	0.09	0.89	−0.26*	0.10	1.29	−0.38***	0.10	1.46

Notes. The first class mentioned in each main column is the reference class. ^a0=female, 1=male. OR = odds ratio.

^*p < .05

^**p < .01

^***p < .001

Predicting long-term resilience from class membership.

As depicted in Figure 2, Wald chi-square tests were used to compare the trajectory group classes on a long-term resilience variable. Wald tests revealed that the mean resilience score in the Steady Incline group (M = 13.59, SD = 1.82) was significantly higher than the Catch-Up group (M = 12.81, SD = 1.82) (x² = 10.74, p < .001) and Decline group (M = 9.09, SD = 1.82) (x² = 182.80, p < .001). Further, the Catch-Up group was determined to have a higher mean resilience score than the Decline group (x² = 105.61, p < .001).

Figure 2.

Global resilience index in the 5^th grade as predicted by developmental patterns of emotion regulation in toddlerhood. Significant differences between all groups.

Discussion

In a prospective study of toddlers from the Early Head Start Research and Evaluation (EHSRE) project, we explored whether distinct trajectory groups of emotion regulation change could be identified across the first three years of life. We examined predictors of membership in those classes and explored class membership as a predictor of long-term resilient functioning. Our analyses revealed several main findings. First, latent class growth analysis (LCGA) supported a three-class model: (1) the Steady Incline group started highest (z = 0.38) at 14 months and continued to steadily improve over time; (2) the Catch-Up group started out lowest (z = −1.35) at 14 months but demonstrated the fastest rate of growth over time; (3) the Decline group started close to the overall group mean (z = −0.17) at 14 months, but was the only group to show negative growth in emotion regulation over toddlerhood. Second, when examining predictors of group membership, child negative emotionality, positive parenting, negative parenting, and environmental risk significantly predicted group membership. Children in the Steady Incline group were the most likely to have high ratings of positive parenting. Children in the Catch-Up group were most likely to have high ratings of negative emotionality. Lastly, children in the Decline group were more likely to have high ratings of negative parenting and higher environmental risk index scores. Third, trajectory group classes across the first three years of life were predictive of resilient functioning in the 5^th grade, on our measure spanning physical, academic, cognitive, and social-emotional functioning. Specifically, children in the Steady Incline group exhibited the most resilient functioning, followed by the Catch-Up group, and children in the Decline group had the lowest scores of resilient functioning.

The three trajectory groups revealed by our exploratory LCGA provide interesting insights. First, over half of the toddlers were in the Steady Incline group, suggesting that most children exhibit high levels of emotion regulation that are maintained across at least the first three years of life. The remaining toddlers were just about evenly split between the other two groups, which were quite disparate in their trajectories. While the Catch-Up group started out the lowest of the three, they improved steadily over the next two years and appeared to almost catch up to the Steady Incline toddlers. As previous investigations have found increases in emotion regulation over this period (Blandon et al., 2008; McRae et al., 2012), it is possible that these toddlers are experiencing developmentally expected improvement in emotion regulation over time. On the other hand, the Decline group looked quite similar to the Steady Incline group at 14 months, but exhibited a precipitous decline in emotion regulation over the next two years. This group is particularly perplexing, given that they contrast developmental expectations. Examining predictions of membership in this group could have important implications for risk factors and intervention targets.

When examining predictors of group membership, an even more intriguing story unfolds. Toddlers in the Steady Incline group were characterized by higher ratings of positive parenting and lower ratings of child negative emotionality. Thus, this group had the lowest internal risk (temperament) and low risk in the proximal environment (parenting), both of which portend adaptive emotion regulation development in previous studies (Calkins, 1994). It makes sense, then, that these children who are dispositionally more positive and have more positivity in their parenting environment exhibit good emotion regulation initially and continue in the same vein.

In contrast, children in the Catch-Up group had ample room for improvement in emotion regulation. Likelihood of membership in this group was predicted by high ratings of child negative emotionality. Interestingly, this group did not differ from the Steady Incline group in terms of negative parenting or environmental risk. These findings suggest that while these children had internal risk that had apparent early effects, the environment afforded them opportunities to develop strong emotion regulation over time. With low rates of negative parenting and low environmental risk, it is likely that toddlers in the Catch-Up group were exposed to a family environment typified by the factors long-established to promote emotion regulation development, such as a higher proportion of sensitive and responsive parenting behaviors and modeling of emotion regulation (Feldman, 2007; Leerkes et al., 2009; Amanda Sheffield Morris et al., 2007). Further, children with higher dispositional negative emotionality have been found to be more susceptible to parenting influences (Chang et al., 2012; Kim & Kochanska, 2012); thus, children in the Catch-Up group may be poised to benefit the most from their family environments.

Likelihood of membership in the Decline group was associated with the poorest constellation of environmental factors (i.e., low positive parenting, high negative parenting, and high environmental risk). These results suggest that the effects of such environmental factors are perhaps not immediately evident in early toddlerhood, but instead are revealed over time. Thus, our data suggest that consequences of environmental risks are incurred after young children interact with their environment over time and are faced with contingencies that harm emotion regulation development. For example, when infants in the Decline group express distress, our findings suggest that their caregivers are less likely to respond sensitively and more likely to respond with negativity, perhaps ignoring or expressing anger at their distress. Furthermore, early exposure to higher levels of environmental risks suggests that children in this group may demonstrate emotion regulation deficits compared to children in the other two groups because of differences in prefrontal cortex-limbic system connectivity and HPA-axis reactivity (Blair, 2010). While not a primary objective of the study, results also suggested that male toddlers were most likely to be in the Decline group. This group membership may presage boys’ higher rates of behavior problems and internalizing and externalizing symptoms that emerge in childhood (Demmer et al., 2017; Munkvold et al., 2011; Sterba et al., 2007).

Another interpretation of these Decline group findings comes from previous propositions that, under certain risk conditions, emotion regulation poses as a “double-edged sword” (Thompson & Calkins, 1996). In their thought-provoking piece, Thompson and Calkins discuss that some children, such as those who witness domestic violence, utilize strategies that regulate emotions in the short-term but have long-term negative consequences. In relation to the current findings, toddlers in the Decline group may be among the group of children who engage in regulatory strategies that serve immediate goals but impair long-term functioning. As an example, one dimension on the Bayley Behavior Rating Emotion Regulation Scale measures resistance to change. Toddlers with higher environmental risk may evidence more distress and resistance when transitioning between activities because of instability and unpredictability in their rearing environment. Transitioning from an activity that feels safe and calm elicits distress and resistance because they do not know if the next situation will be similarly safe and calm. This strategy may effectively increase exposure to safe environments in the short run, but repeated difficulties with change may confer long-term psychological distress, particularly anxiety and traumatic stress (Arvidson et al., 2011).

These interpretations are consistent with our long-term resilience findings. We found that the trajectory classes of emotion regulation in toddlerhood were predictive of a global resilience factor in the 5^th grade. Unsurprisingly, the Steady Incline group had the highest average resilience ratings. Importantly, improvements and declines made in the Catch-Up and Decline groups, respectively, extended to functioning seven years later. These results are consistent with previous findings that emotion regulation may serve as a protective factor that fosters resilience in at-risk populations of children (Buckner et al., 2003; Cicchetti et al., 1993; Herbers, 2011).

Several theoretical implications should be noted. First, our findings highlight the importance of developmental investigations of emotion regulation, as emotion regulation was determined to be dynamic over the course of early childhood, and trajectory groups were meaningful predictors of long-term resilience. Second, positive and negative parenting, though moderately correlated, had separable effects on emotion regulation trajectories. This underscores the importance of examining these constructs independently, as opposed to the spectrum approach which suggests that parenting lies on a continuum of negative to positive, and is consistent with previous studies who have parsed the two apart (Dallaire et al., 2006; Eamon, 2002). Third, our environmental risk index had independent effects on emotion regulation, over and above the effects of parenting. While it is certainly the case that environmental risk affects children through parenting (Pinderhughes et al., 2001), our findings suggest that it has direct effects on child development as well, though a mediation model is needed to explicitly test this effect. Future research should continue to examine environmental factors more distal than parenting behaviors, as it may provide important nuance to our understanding of child development.

Strengths, Limitations, and Future Directions

The present findings should be interpreted in the context of several limitations that generate implications for future research. First, though there was variability in the sample in terms of environmental risk, the children and families enrolled in the Early Head Start Research and Evaluation project were of higher environmental risk, as that is the target population of the program. Thus, it is possible that our trajectories of emotion regulation are specific to children with high environmental risk and that children in low-risk families may follow different developmental patterns. It is critical for future studies to extend these research questions to a more representative sample of children. Second, overlap between our measurement of temperament and emotion regulation should be acknowledged. Though theoretically distinct in that temperament is considered a relatively stable trait across the lifespan whereas emotion regulation is dynamic over time, temperament (i.e., emotional reactivity) and regulation are difficult to differentiate from a methodological perspective in young children. It should be noted that the correlation between temperament and emotion regulation in the current sample was below .2 at the three time points examined, suggesting that these variables are measuring different constructs. Third, and relatedly, our measure of emotion regulation (i.e., Bayley BRS) is based on global toddler behavior during a developmental assessment, which may not capture nuances in the child’s regulatory behaviors and may be more reflective of self-regulation in the context of compliance. A methodological improvement for future studies to address these two limitations is use of a second-to-second microcoding system for observed emotional expression to ensure disaggregation of reactivity from regulation.

Another limitation is that we measured emotion regulation at three ages in toddlerhood, so our trajectory analysis was restricted to linear growth. Researchers with more repeated measurement of emotion regulation in toddlerhood and early childhood are encouraged to examine higher-order (e.g., quadratic) trajectories. In addition, the current study did not account for negative life events that may have occurred during toddlerhood or between toddlerhood and 5^th grade. It is possible that children in the Decline group experienced traumatic or stressful life events, though we were not able to examine this possibility. Therefore, a needed next step is to study the relation between negative life events and change in emotion regulation over time. Lastly, the current study involved examination of main effects of internal and external influences on emotion regulation; however, early emotion regulation development may be influenced by moderation or mediation effects among these different factors. Future research should involve more complex predictor modeling to examine these additive, multiplicative, or pathway effects.

There are also notable strengths and novel aspects of the current study that must be acknowledged. First, we leveraged a low-income, racially and ethnically diverse sample of families and a prospective, longitudinal design with multi-informant, multi-method measurement of child and parent behavior to identify person-centered developmental trajectories of emotion regulation in toddlerhood. Our empirical classes suggest that toddler development can be captured by three trajectory groups that are particularly dynamic in the first two years of life. Further, membership in these groups is influenced by different levels of systems, ranging from within the child (temperament), to the child’s proximal environment (parenting), and finally the context in which the family is embedded (environmental risk). In addition, developmental trajectories of emotion regulation in toddlerhood appear to have important consequences for long-term resilient functioning among children living in poverty. The study and dissemination of prevention and early intervention services to support emotion regulation development in families living in poverty is needed, and our findings suggest that focus on the first 2 years of life may be most fruitful. While considerable evidence supports emotion regulation as a protective factor across many domains, much more remains to be uncovered about how to provide children with optimal support for emotion regulation development.