Early Childhood Behavioral Inhibition Predicts Altered Social and Emotional Functioning in Early Adulthood: An Ecological Momentary Assessment Study

Abstract

Early childhood behavioral inhibition (BI), the tendency to display inhibited behavior in the face of novelty, has been associated with a number of negative outcomes in later childhood and early adolescence. However, few studies have examined young adult outcomes of BI, and none have used intensive assessment methods like ecological momentary assessment (EMA) to capture both inter- and intraindividual patterns of social and emotional functioning. In the current study, BI at age 3 (2004–2007) was assessed through both behavioral observation in the laboratory and parent-report questionnaire. At age 18 (2019–2023), 330 participants completed an EMA study that involved surveying participants’ current emotions and recent (past hour) social experiences 5x daily for 14 days. Age 3 BI was correlated with lower average positive affect and the tendency to appraise social interactions as less positive at age 18. Additionally, while BI was not correlated with average negative affect, nor with frequency of social interaction, it was associated with a tendency to report heightened anxiety following less positive, more negative, and more nervous/uncomfortable social interactions. Results suggest that young adults who were behaviorally inhibited as children tend to experience lower levels of positive emotion in general and more anxiety specifically in the context of uncomfortable and less pleasant social interactions. Further, while they appear to engage in social interactions with equal frequency, they tend to find social interactions less rewarding. Our sample is relatively homogeneous with respect to race and ethnicity, limiting generalization to more diverse samples.

Introduction

In one of the foundational developmental texts, Kagan first observed that young children who appeared inhibited in the face of new people, places, and situations often retained this trait into adolescence and adulthood (Kagan & Moss, 1962). This observation spurred decades of research into the temperamental trait Kagan termed Behavioral Inhibition (BI). There is now a large literature showing that a substantial proportion of inhibited children experience a range of social and emotional problems in later childhood and early adolescence (Kagan, 2018; Pérez-Edgar & Fox, 2018). For example, behaviorally inhibited children are more likely to demonstrate social withdrawal (Pérez-Edgar et al., 2010; Pérez-Edgar et al., 2011) and lower social competence (Bohlin, Hagekull, & Andersson, 2005; Penela, Walker, Degnan, Fox, & Henderson, 2015; Walker, Henderson, Degnan, Penela, & Fox, 2014). They are also more likely to experience peer victimization and ostracism (Affrunti, Geronimi, & Woodruff-Borden, 2014; Deater-Deckard, 2001; Hanish & Guerra, 2004), and are at increased risk of developing an anxiety disorder (Hudson & Dodd, 2012; Mumper & Klein, 2021; Sandstrom, Uher, & Pavlova, 2020), particularly social anxiety (Clauss & Blackford, 2012).

Despite Kagan’s initial suggestion that many inhibited children continue to differ from their peers in adulthood, relatively few prospective studies have assessed outcomes in later developmental periods, which present markedly different social and emotional demands than childhood and early adolescence. Inhibited 3-year-olds in the Dunedin cohort were more likely to be depressed at 21 and have low trait positive affect and high trait constraint at 23 (Caspi et al., 2003; Caspi, Moffitt, Newman, & Silva, 1996). Another study found that highly inhibited 4–6 year-olds were more delayed in important social transitions (e.g., first stable partnership and full-time job) at age 26 (Asendorpf, Denissen, & Van Aken, 2008). Finally, a recent study found that BI in infancy was prospectively associated with more reserved personality, lower social functioning, fewer romantic relationships, and higher levels of internalizing psychopathology in early adulthood (Tang et al., 2020).

Findings from these studies suggest that early BI continues to have ramifications for social and emotional functioning into later adolescence and early adulthood. However, these studies are few in number and there are important gaps and limitations in our understanding of inhibited children’s socioemotional functioning later in life. For example, the Dunedin cohort study’s measure of BI diverges substantially from Kagan’s definition and traditional measures of BI. Close inspection of the factor loadings in Tang et al.’s (2020) psychopathology bifactor model reveals that the “internalizing” factor is composed exclusively of somatic symptoms of anxiety, while depression and other measures of anxiety loaded on the general psychopathology factor. Moreover, the outcomes examined in these studies were largely based on retrospective self-report of traits, symptoms, and life experiences. Retrospective reports often represent impressionistic summaries of experiences over extended periods of time, and as such are vulnerable to recall biases (Scollon, Kim-Prieto, & Diener, 2003).

In contrast, “real-time” intensive longitudinal methods, like experience sampling or ecological momentary assessment (EMA; Shiffman, Stone, & Hufford, 2008) involve repeated assessment of recent (e.g., within the last few minutes/hour) events, behaviors, and subjective experiences while individuals navigate their natural environments, capturing both inter- and intraindividual processes in an ecologically valid way. EMA studies provide a rich picture of real-time and daily experiences that often differs substantially from self-report questionnaires (Shiffman et al., 2008), and capture both how individuals differ from each other as well as how individuals differ from what is typical for them in everyday life. EMA research has proliferated in recent years, particularly with the advent and widespread use of smartphones, which drastically improves the feasibility of these studies. Much of this research has been interested in the dynamic processes of emotion and socialization. For example, one meta-analysis identified 79 articles on the link between emotion dynamics and markers of wellbeing (Houben, Van Den Noortgate, & Kuppens, 2015). Another meta-analysis identified 23 studies examining the within-person associations between emotions and various aspects of social interaction (e.g., uniqueness of partner, perceptions of the interaction; Liu, Xie, & Lou, 2019). Despite the sizable literature examining social and emotional functioning using EMA, to the best of our knowledge, no prior study has explored links to BI in childhood.

The purpose of the current study is to test prospective associations between early childhood (age 3) BI and social and emotional functioning in daily life in 18-year-olds, an age that marks the transition to young adulthood. Temperament has long been presumed to be at least moderately stable over time (Rothbart & Bates, 2006; Shiner & Caspi, 2003), hence we expected to observe continuity between early childhood BI and momentary emotional and social functioning at age 18. More specifically, BI is defined in terms of fearfulness, a core component of trait negative affectivity in virtually all temperament frameworks (De Pauw & Mervielde, 2010; Rothbart & Bates, 2006) and social withdrawal, which is a significant component of the low end of the temperament trait of extraversion, or surgency (Watson & Clark, 1997). Descriptions and measures of BI often also include low positive affect (Kagan, Snidman, & Arcus, 1998; Laptook, Klein, Olino, Dyson, & Carlson, 2010; Pfeifer, Goldsmith, Davidson, & Rickman, 2002), which is generally considered to be the emotional core of extraversion (Watson & Clark, 1997). Thus, in this study we focus on momentary experiences of negative affect, and specifically anxious affect, as well as positive affect and social interactions, as these are the areas in which we would expect BI to be most likely manifested in the daily lives of young adults. Importantly, the affective and interpersonal domains are likely related. For example, from the perspective of social learning theories of anxiety (e.g., Zinbarg, Williams, & Mineka, 2022), children who experience greater negative affect and less positive affect in social situation should be less likely to participate in subsequent social interactions, and this avoidance will be positively reinforced by the absence of negative affect, leading to further withdrawal.

Based on theory and the available literature, we had three sets of hypotheses at the between-person level: young adults with a history of early childhood BI would experience: (1) greater average negative and anxious affect and lower average positive affect; (2) fewer social interactions; and (3) greater proportions of interactions that are unpleasant and nervous or uncomfortable, and a lower proportion of interactions that are pleasant. In addition, as BI is manifested most clearly in challenging social contexts, we also had a set of hypotheses regarding cross-level interactions. In the context of EMA, cross-level interactions represent interactions between variables at the person and observation levels, capturing between-person differences in within-person processes. In our study, the between-person difference of interest was childhood BI, while the within-person process of interest was the connection between emotions and recent social interactions. We hypothesized that: (4) young adults with a history of childhood BI would experience elevations in negative affect, and specifically anxious affect, following social interactions that are more unpleasant and nervous/uncomfortable, and less pleasant than usual.

In the current study, parent-report and laboratory behavioral observation were combined to create a multimethod index of BI at age 3. At age 18, participants completed an EMA study that involved repeated assessments of emotions and social experiences, with surveys being completed 5 times daily for 14 consecutive days. We used two different approaches to operationalizing social and emotional functioning using the EMA data. First, we aggregated all ratings across the 2 weeks of EMA to create an index of each individuals’ typical emotion levels and social patterns to test our between-person hypotheses. We also conducted multilevel regression analyses to capture interindividual (between-person) differences in intraindividual (within-person) links between emotions and social experiences. This latter approach allowed us to capture emotional patterns that are contextualized by recent social experiences, whereas the former captured emotional patterns that were general across contexts.

Methods

Participants

Participants were part of an ongoing study of temperament and the development of psychopathology that has followed children and their families since participants were 3 years old (Klein & Finsaas, 2017). At the onset of the study, families with a 3-year-old child living within a 20-mile radius of Stony Brook, New York were recruited via commercial mailing lists. At least one biological parent who spoke English was required to participate alongside their child. Children with a history of significant medical disorders or developmental disability were excluded. Parents provided informed consent on behalf of themselves and their child at age 3 and participants provided consent on behalf of themselves at age 18. Of the 559 participants from the initial age 3 wave of the study, 362 returned for the age 18 wave and began the EMA study, and 330 provided sufficient EMA data to be included in the current analyses. Returning participants did not differ in race, ethnicity, or family income reported at age 3, however, they were more likely to be female. Additionally, BI scores did not differ for returning participants compared to the original sample, t(df) = 0.23(465.83), p = .811. The final sample was predominantly white and non-Hispanic (9% Hispanic, 2% Black, 2% Asian, and <1% other race) and approximately evenly split between males and females (52%). Greater detail on sample characteristics and attrition are reported in Supplement Table 1.

Procedures

At age 3 (data collected between 2004–2007), families were invited into the laboratory to complete a series of standardized behavioral tasks, as well as a battery of parent-report questionnaires about the index child. The child was invited to complete a follow-up EMA study at age 18 (data collected between 2019–2023). EMA surveys were administered using the RealLife Exp app by LifeData Corporation. Participants were sent 5 surveys randomly throughout the day for 14 consecutive days. To maximize compliance, participants were allowed to pick among different survey timing windows. Staff monitored compliance on a daily basis, and payment doubled after the first 40 surveys from $1.00 to $2.00 per survey. The Stony Brook University Institutional Review Board approved all study procedures.

Measures

Behavioral Inhibition (age 3)

BI was assessed through both parent-report questionnaire and behavioral observation in the laboratory. For the laboratory observation measure, 3 episodes were selected from the Laboratory Temperament Assessment Battery (Lab-TAB; Gagne, Van Hulle, Aksan, Essex, & Goldsmith, 2011): Risk Room, Stranger Approach, and Exploring New Objects. BI was coded using an approach that was similar to most previous studies (e.g., Durbin, Klein, Hayden, Buckley, & Moerk, 2005; Kagan, 1989; Pfeifer et al., 2002). The three episodes were divided into 20 or 30 second epochs, and a series of affective and behavioral codes were rated for each epoch. Within each epoch, a maximum intensity rating of facial, bodily, and vocal fear was coded on a scale of 0 (absent) to 3 (highly present and salient). Following previous studies, BI was computed as the average standardized ratings of latency to fear (reversed); and facial, vocal, and bodily fear (Risk Room, Stranger Approach, and Exploring New Objects); latency to touch objects; total number of objects touched (reversed); tentative play; referencing parent; proximity to parent; referencing experimenter; and time spent playing (reversed) (Risk Room and Exploring New Objects); startle (Exploring New Objects); sad facial affect (Exploring New Objects and Stranger Approach); and latency to vocalize; approach towards the stranger (reversed); avoidance of the stranger; gaze aversion; and verbal/nonverbal interaction with the stranger (Stranger Approach). Thirty-five participants were rated by a second coder; interrater reliability for the total BI score, computed using the intraclass correlation, was 0.88.

One parent (predominantly mothers) completed the Behavioral Inhibition Questionnaire (BIQ; Bishop, Spence, & McDonald, 2003) about their 3-year-old child. The BIQ contains 30 items assessing inhibited behavior across domains (e.g., approach of unfamiliar peers or novel physical activities) rated on a 7-point Likert scale from 1 (hardly ever) to 7 (almost always). In the current study, the BIQ had excellent internal consistency (α = .96) and good convergent and discriminant validity (Kim et al., 2011).

Parent-report and behavioral observation BI scores were standardized and then averaged to create a single, multimethod BI score. As expected, the two measures were modestly correlated with each other (r = .29, p < .001) and strongly correlated with the combined score (r = .79-.82, p < .001).

Social and Emotional Functioning (age 18)

Emotion Variables.

The EMA surveys included questions about current emotions and recent social interactions designed for this study. Emotion items were administered before social interaction items to prevent the latter from influencing ratings of the former. Emotion items were preceded by a prompt asking participants to think about how they were feeling right before starting the survey to reduce reactivity effects. The emotion items instructed participants to rate “To what extent did you feel (emotion)” on a Likert-scale from not at all (1) to extremely (5). To balance survey brevity with maximized coverage of valence domains, some items combined several similar emotions. Four questions assessed positive valence emotions: 1) happy, 2) excited, 3) cheerful, and 4) content or peaceful. Four questions assessed negative valence emotions: 1) sad, down, or depressed, 2) anxious, worried, or nervous, 3) irritated, annoyed, or angry, and 4) upset. Items of the same valence were averaged to create positive (PA) and negative affect (NA) composites. A prior methodological study exploring the measurement properties of these affect items in a subset of current participants found that the emotion items can be combined into internally consistent PA and NA composites along the proposed item groupings(Hawes, Olino, & Klein, 2022). Specifically, reliability of NA/PA factors was adequate at the within-person level (omega = .72/.79) and excellent at the between-person level (omega = .91/.93). The item assessing anxious, worried, or nervous affect was also used as a stand-alone variable in the present analyses due to its theoretical importance to the construct of BI.

Social Interaction Variables.

Two questions asked participants to indicate how many people they interacted with in the last hour by 1) texting/messaging app (not counting email) and 2) in-person or over the phone/video chat. Both items were rated on an ordinal scale from 0 (I had no interactions) to 4 (4 or more people). The items were combined into a single dichotomous variable representing whether participants had at least one social interaction of either type (0 = No, 1 = Yes). Following this, a prompt screen asked participants to think about the most significant social interaction they had in the last hour, followed by 3 questions eliciting ratings of how 1) pleasant or positive and 2) unpleasant or negative the interaction was, along with how 3) nervous or uncomfortable they felt during the interaction. The three items were rated on a Likert scale from 1 (not at all) to 5 (extremely).

Currently, there are no widely used conventions for demarcating valid from potentially invalid responses in EMA, and thresholds are recognized to be largely arbitrary and idiosyncratic. However, a few considerations are recommended (Viechtbauer, 2021), and we set the following thresholds with these in mind. Individual survey responses were excluded if survey completion time was equivalent to an average response of less than one second per item or a total completion time greater than 45 minutes. Participants were excluded if they completed fewer than 14 surveys (< 2.5 days of surveys). Five surveys were excluded for too short completion times, 135 surveys were excluded for too long completion times, and 32 participants were excluded for too few surveys.

Data Analysis

First, all EMA variables were averaged across surveys to create aggregates representing each individual’s typical positive, negative, and anxious emotion intensities, sociability (i.e., proportion of surveys reporting at least one recent social interaction), and social appraisals (i.e., how pleasant/positive, unpleasant/negative, and nervous/uncomfortable they tended to find these social experiences). Correlations were computed between these person-level aggregates and the multimethod BI score to examine non-contextualized relationships between BI and patterns in affectivity and socialization (i.e., between-person differences).

Next, a series of multilevel regression analyses were conducted to test individual differences in the emotions experienced following recent social interactions. Multilevel regression models relationships at multiple levels, preserving the hierarchical structure of data, such as the repeated surveys nested within persons in EMA. In the current study, each multilevel model regressed an emotion variable (PA, NA, anxious affect) on one of the social interaction variables (any interaction and appraisals of how positive, negative, and nervous or uncomfortable participants felt during a recent interaction) from the same survey (level 1), their BI score from age 3 (level 2), and the aggregate of that social interaction variable (i.e., that person’s average score across all surveys; level 2). Including the person-level aggregate alongside the individual survey ratings ensured that between-person effects were separated from within-person effects (Curran & Bauer, 2011). Cross-level interaction effects were specified between the survey-level social interaction variables and person-level BI scores. These cross-level interactions were the primary effect of interest in these models, as they represent differences in how individuals of varying BI levels rate their current emotions following a recent social interaction, or lack thereof (i.e., between-person differences in within-person processes). Participant sex (level 2) was also included as a covariate in all models.

In these models, survey-level social interaction appraisal ratings were centered around their person-level means so that effects could be interpreted in terms of deviations from the person average (e.g., reporting a more pleasant interaction than typical for that individual). The variable representing report of a recent interaction was not recoded because it is dichotomous and more easily interpreted in its original form. Person-level aggregates were grand-mean centered so these effects could be interpreted in terms of deviations from the sample average. Finally, as BI is a standardized score, effects could be interpreted in terms of standard deviation units.

We only tested models in which emotion variables were outcomes and social interaction variables were predictors because the timing window of the instructions (i.e., rating current emotions vs. social interactions within the last hour) made the opposite direction of effect theoretically impossible. We conducted sensitivity analyses exploring the impact of replacing the combined BI score with scores from the individual measures (i.e., laboratory observation and parent report) to ensure that our results are not method-specific, which are displayed in Supplement Tables 2 and 3. All analyses were conducted in R version 4.2.2 (R Core Team, 2023). The multilevel models were estimated using the lme4 package for R (Bates, Maechler, Bolker, & Walker, 2015) and significance tests were conducted with the lmerTest package (Kuznetsova, Brockhoff, & Christensen, 2017).

Transparency and Openness

Data were not made publicly available, as consent to do so was not obtained from study participants. This study’s design and its analysis were not pre-registered. Annotated code and output can be found at the following link: https://osf.io/fasqw/?view_only=c91a65d32597461ab1b3aa61a7bd164e

Results

Between-Person Differences (Hypotheses #1–3)

Table 1 displays correlations between BI and the EMA aggerates of the emotion and social interaction variables. Age 3 BI was negatively correlated with age 18 PA and positive appraisals of recent social interactions. BI was not significantly associated with average intensity of NA or anxious affect, nor with sociability (the proportion of surveys a participant reported having any social interaction), negative appraisals of recent social interactions, or reports of nervousness/discomfort during a recent social interaction.

Table 1.

Correlations between Child Behavioral Inhibition and Young Adult Emotion and Social Interaction Patterns

	M	SD	1	2	3	4	5	6	7
1. Behavioral Inhibition	0	1
2. Positive Affect	2.83	0.75	−.12*
3. Negative Affect	1.43	0.45	.10	−.31**
4. Anxious Affect	1.55	0.63	.07	−.30**	.88**
5. Any Socialization	0.76	0.20	.04	.10	−.01	.04
6. Positive Socialization	3.61	0.62	−.12*	.73**	−.25**	−.21**	.13*
7. Negative Socialization	1.23	0.28	−.01	−.20**	.65**	.51**	−.18**	−.27**
8. Nervous Socialization	1.20	0.27	.02	−.17**	.72**	.62**	−.16**	−.24**	.70**

^*p<.05

^**p<.01

^***p<.001

Note: M = mean, SD = standard deviation; cells contain zero-order bivariate correlations among all variables; behavioral inhibition scores were standardized; all emotion and social interaction variables are person-level aggregates based on the average of individual survey ratings; Any socialization = proportion of surveys reported socializing in the past hour; Positive Socialization = average rating of how positive/pleasant a recent interaction was; Negative Socialization = average rating of how negative/unpleasant a recent interaction was; Nervous Socialization = average rating of how nervous/uncomfortable one felt during a recent interaction.

Between-Person Differences in Within-Person Processes (Hypothesis #4)

Table 2 displays results for the multilevel models. Results for associations between the emotion variables and social interaction variables are not described in detail, as these results are not particularly germane or novel. Briefly, these associations were predominantly in the expected directions (e.g., interactions perceived as more negative and less positive were associated with greater NA and lower PA). Table 3 displays all simple slopes results.

Table 2.

Associations Between Momentary Affect and Past-Hour Social Interaction Moderated by Child Behavioral Inhibition

	Positive Affect			Negative Affect			Anxious Affect
Model 1	B(S.E.)	t	p	B(S.E.)	t	p	B(S.E.)	t	p
Any Social Interaction (L1)	0.28(0.02)	16.16	< .001	−0.01(0.01)	−0.92	.357	−0.01(0.02)	−0.32	.747
Behavioral Inhibition (L2)	−0.07(0.04)	−1.63	.105	0.04(0.03)	1.55	.121	0.04(0.04)	1.13	.260
Any Social Interaction Average (L2)	0.15(0.20)	0.74	.462	−0.12(0.12)	−1.01	.315	−0.02(0.17)	−0.11	.914
BI Moderation Effect	−0.03(0.02)	−1.76	.080	0.00(0.01)	−0.11	.916	0.00(0.02)	−0.02	.982
Model 2
Positive Social Interaction (L1)	0.40(0.01)	35.13	< .001	−0.18(0.01)	−16.76	< .001	−0.12(0.01)	−9.62	< .001
Behavioral Inhibition (L2)	−0.04(0.03)	−1.29	.198	0.03(0.02)	1.28	.202	0.03(0.03)	0.88	.381
Positive Interaction Average (L2)	0.90(0.04)	20.31	< .001	−0.21(0.03)	−6.29	< .001	−0.26(0.05)	−4.98	< .001
BI Moderation Effect	0.00(0.01)	0.09	.933	−0.02(0.01)	−2.23	.027	−0.03(0.01)	−2.74	.006
Model 3
Negative Social Interaction (L1)	−0.35(0.02)	−20.31	< .001	0.35(0.02)	22.25	< .001	0.21(0.02)	11.36	< .001
Behavioral Inhibition (L2)	−0.10(0.04)	−2.37	.019	0.04(0.02)	2.38	.018	0.04(0.03)	1.51	.131
Negative Interaction Average (L2)	−0.58(0.14)	−4.00	< .001	1.15(006)	17.98	< .001	1.27(0.11)	11.92	< .001
BI Moderation Effect	0.00(0.02)	0.02	.985	0.02(0.02)	1.37	.173	0.06(0.02)	2.99	.003
Model 4
Nervous Social Interaction (L1)	−0.18(0.02)	−8.71	< .001	0.24(0.02)	13.08	< .001	0.29(0.02)	12.89	< .001
Behavioral Inhibition (L2)	−0.10(0.04)	−2.33	.021	0.04(0.02)	2.57	.011	0.04(0.03)	1.65	.101
Nervous Interaction Average (L2)	−0.45(0.15)	−3.02	.003	1.20(0.06)	19.95	< .001	1.44(0.10)	14.67	< .001
BI Moderation Effect	0.00(0.02)	0.01	.994	0.05(0.02)	2.70	.008	0.08(0.02)	3.35	.001

Note: L1 = level one predictor. L2 = level two predictor. B(S.E.) = coefficient(standard error). Sex is coded as 1 = male. Any Social Interaction is dichotomously coded as 1 = had a social interaction in the past hour. Positive, Negative, and Nervous Social Interaction are 5-point Likert rating of how positive, negative, or nervous/uncomfortable a recent interaction was appraised to be, respectively. The Interaction Average variables are person averages of the respective level 1 social interaction variable. BI Moderation Effect is the cross-level interaction between the level 1 social interaction variable and level 2 behavioral inhibition (BI). All models controlled for participant sex.

Table 3.

Simple Slopes Results for Significant Interactions

Outcome = Negative Affect
	Est.(S.E.)	t	p
BI x Positiveness of Social Interaction
Low BI (−1 SD)	−0.15(0.02)	−10.18	< .001
Average BI	−0.18(0.01)	−16.76	< .001
High BI (+1 SD)	−0.20(0.02)	−13.32	< .001
BI x Nervousness of Social Interaction
Low BI (−1 SD)	0.19(0.03)	7.22	< .001
Average BI	0.24(0.02)	13.08	< .001
High BI (+1 SD)	0.30(0.03)	10.82	< .001
Outcome = Anxious Affect
BI x Positiveness of Social Interaction
Low BI (−1 SD)	−0.08(0.02)	−4.81	< .001
Average BI	−0.12(0.01)	−9.62	< .001
High BI (+1 SD)	−0.15(0.02)	−8.67	< .001
BI x Negativeness of Social Interaction
Low BI (−1 SD)	0.15(0.03)	5.83	< .001
Average BI	0.21(0.02)	11.36	< .001
High BI (+1 SD)	0.27(0.03)	10.27	< .001
BI x Nervousness of Social Interaction
Low BI (−1 SD)	0.21(0.03)	6.57	< .001
Average BI	0.29(0.02)	12.89	< .001
High BI (+1 SD)	0.37(0.03)	11.17	< .001

Note: BI = Behavioral Inhibition; Est.(S.E.) = estimate(standard error).; SD = standard deviation.

Positive Affect Following Recent Social Interactions

No significant interaction effects were observed for models where PA was the outcome, however, significant (negative) main effects were observed for BI in the models including negative and nervous/uncomfortable appraisals of recent interactions as predictors.

Negative Affect Following Recent Social Interactions

For the models where NA was the outcome, significant interactions were observed between BI and positive and nervous/uncomfortable appraisals. Figure 1 displays plots for these two significant interactions, with simple slopes estimated at average, low (−1 standard deviation), and high (+1 standard deviation) levels of BI. As Figure 1 shows, all groups reported heightened negative emotions following interactions they appraised as less positive and more nervous/uncomfortable than typical. However, for individuals higher in BI at age 3, the slopes of these relationships were steeper, reflecting reports of even more intense negative emotional reactions to less positive and more nervous/uncomfortable social interactions. Significant interactions were not observed for the models with any socialization and negative appraisals of social interactions as predictors, however, there was a significant (positive) main effect of BI in the model with negative appraisal as a predictor. Of note, these main effect relationships may differ from the zero-order bivariate correlations due to the presence of the social interaction variable averages, which are moderately-strongly associated with the affect variables.

Figure 1. Simple Slopes of Significant Interactions between Behavioral Inhibition and Appraisals of Past-Hour Social Interactions on Negative Affect.

Note: The solid line represents low behavioral inhibition (BI) (−1 standard deviation), the dashed line represents average BI, and the dotted line represents high BI (+1 standard deviation). All lines are shaded with 95% confidence bands. The graph on the left displays BI-moderated associations between ratings of how positive/pleasant a recent social interaction was and current negative emotion intensity. The graph on the right displays BI-moderated associations between ratings of how nervous/uncomfortable one felt during a recent social interaction and current negative emotion intensity.

Anxious Affect Following Recent Social Interactions

Lastly, in the models where anxious affect was the outcome, significant interactions were observed between BI and positive, negative, and nervous/uncomfortable appraisals. No main or interaction effects were significant in the model with the any social interaction variable. Figure 2 displays plots for the three significant interactions, with simple slopes estimated at average, low (−1 standard deviation), and high (+1 standard deviation) levels of BI. Similar to the results for NA, slopes for the relationships between social interaction appraisals and affect were steeper for individuals higher in BI at age 3, reflecting reports of even more intense anxiety following less positive, more negative, and more nervous/uncomfortable social interactions.

Figure 2. Simple Slopes of Significant Interactions between Behavioral Inhibition and Appraisals of Past-Hour Social Interactions on Anxious Affect.

Note: The solid line represents low behavioral inhibition (BI) (−1 standard deviation), the dashed line represents average BI, and the dotted line represents high BI (+1 standard deviation). All lines are shaded with 95% confidence bands. The graph on the top left displays BI-moderated associations between ratings of how positive/pleasant a recent social interaction was and current anxious emotion intensity. The graph on the top right displays BI-moderated associations between ratings of how negative/unpleasant a recent social interaction was and current anxious emotion intensity. The graph on the bottom left displays BI-moderated associations between ratings of how nervous/uncomfortable one felt during a recent social interaction and current anxious emotion intensity.

Sensitivity Analyses

Sensitivity analyses replacing the combined BI score with laboratory observation only (Supplement Table 2) or parent-report only (Supplement Table 3) are displayed in the supplement. Results of the sensitivity analyses do not differ substantively, with the exception that some effects dropped into the marginal significance range (p < .10), supporting the use of the combined score to dampen method-specific noise and/or maximize signal strength.

Discussion

Despite Kagan’s long-ago observation that behaviorally inhibited children continued to show signs of inhibition from the subtle and not necessarily impairing or obvious to observers, to the more overt (e.g., shyness) and impairing (e.g., anxiety disorders) into adulthood, very few studies have examined outcomes of BI past late childhood/early adolescence. The current investigation leveraged a long-running study following a cohort of individuals from age 3 through age 18 to test prospective associations between early childhood BI and social and emotional patterns in young adulthood. To the best of our knowledge, this is the first study on young adult outcomes of BI to use real-time assessment methods (EMA). Overall, our findings suggest that young adults who were behaviorally inhibited as children continue to display differences in social and emotional functioning relative to their less inhibited peers, however, these differences are highly nuanced.

Our first three hypotheses were focused on between-person effects. We hypothesized that individuals who were behaviorally inhibited as children would, on average, experience greater negative and anxious affect, lower positive affect, and fewer social interactions in young adulthood. We found that childhood BI was associated with lower average positive affect, but not with average levels of negative and anxious affect. BI was also not correlated with our measure of frequency of socialization, suggesting that young adults with a history of early childhood BI socialize to a similar degree as their peers. We also hypothesized that young adults with a history of early childhood BI would report greater proportions of interactions that are experienced as unpleasant and nervous or uncomfortable, and a lower proportion of interactions that are experienced as pleasant. What we found was that BI was indeed negatively associated with average positive appraisals of social interactions, however, it was not associated with negative or nervous/uncomfortable appraisals, suggesting that these formerly inhibited young adults tend to find their social interactions less pleasant but not necessarily unpleasant and uncomfortable, on average.

BI is distinguished from many other common personality trait models (e.g., the Big Five) in that these traits describe patterns that generalize across contexts, while BI is specific to the context of novel people, places, and situations. For example, while negative emotionality/neuroticism describes a tendency towards experiencing negative emotions, BI is thought to reflect a tendency towards experiencing negative emotions, particularly fearfulness, specifically in the face of novelty (Klein & Mumper, 2018). Our findings with regards to between-person differences in within-person processes, corresponding to our fourth hypothesis, are consistent with this notion. We found that young adults who were more behaviorally inhibited as young children reported more intense negative emotions (relative to their typical level) when they had social interactions they appraised as less positive/pleasant and more nervous/uncomfortable than typical for them. When anxious affect was examined individually, the same pattern was observed, and an additional effect emerged suggesting that these individuals also experienced more intense anxiety following social interactions appraised as more negative/unpleasant than typical, consistent with our hypotheses regarding the cross-level interactions. The fact that there were more effects for anxious affect individually than for negative affect overall aligns with the assumption that fearfulness is the primary component of negative affectivity associated with temperamental BI (Kagan, Reznick, Snidman, Gibbons, & Johnson, 1988). While our EMA surveys did not directly assess whether a social interaction involved an unfamiliar person or social setting, the way we specified our multilevel models allowed us to separate atypical social appraisals from typical appraisal patterns. Thus, while young adults who were behaviorally inhibited as children may not experience more anxiety and negative emotions in general, they appear to react more strongly to social interactions that do not go well. Importantly, this effect may be a manifestation of BI in adulthood rather than an outcome, which is consistent with both Kagan’s observations and subsequent research supporting the stability of BI over shorter developmental spans (Kagan, 2018).

The findings for positive emotions differed in important respects. While significant correlations were found between childhood BI and low average PA in early adulthood, individuals who were more behaviorally inhibited as children did not differ in their PA following recent social interactions they appraised as more pleasant or unpleasant than usual. The one prior study testing associations between childhood BI and adult personality observed a negative association between BI and trait positive emotionality and no association between BI and trait negative emotionality (Caspi et al., 2003), suggesting that this is not a spurious pattern. Concurrent studies in childhood likewise find associations between BI and positive emotionality/extraversion, and though they often observe associations with negative emotionality/neuroticism too, these relationships tend to be weaker (e.g., Muris et al., 2009).

The connection between BI and trait positive emotionality has been attributed more to the sociability/approach domains of the overlapping trait, extraversion, than the tendency towards positive affect (Klein & Mumper, 2018). However, we observed essentially the opposite in the current study, with BI negatively correlated with average PA intensity and unrelated to our EMA index of sociability (proportion of surveys reporting a recent interaction). It is possible that our measure of sociability was not sensitive enough to capture the more subtle differences in social approach behaviors that persist into adulthood. For example, behaviorally inhibited individuals may be less expressive, make fewer contributions to the conversation, and take little action to prolong a novel social interaction. As a result, the interaction may be more superficial and experienced as less enjoyable, consistent with the negative correlation we observed between BI and average positive social appraisal. Thus, the tendency to experience less momentary positive emotion in general may be a consequence of a chronically reduced enjoyment of novel social experiences. Similar theories have been proposed to explain the robust association between low positive affect and social anxiety (Kashdan, 2007). However, BI extends beyond social contexts and there are many areas of life that may be limited in enjoyment due to avoidance of novelty (e.g., restricting one’s work and leisure activities, modes of travel and places visited, etc.). Unfortunately, our EMA study did not include measures of non-social behaviors to test this conjecture. In addition, these findings raise the possibility that the cumulative experience of inhibition over childhood and adolescence leads to a general dampening of positive affect and enthusiasm by late young adulthood.

Strengths & Limitations

This study possessed a number of strengths, including a long follow-up period extending from early childhood into young adulthood, the availability of multiple methods of assessing BI in early childhood, and the use of real-time assessment methods for capturing social and emotional functioning in early adulthood. Moreover, our approach to analyzing the EMA data allowed us to separate general emotional patterns from emotions contextualized by social experiences.

However, our findings should also be evaluated in the light of several limitations. First, without a third measure of BI available, we were unable to model BI using latent variable techniques that can account for measurement error (e.g., factor analysis). The assessment of daily emotions and behavior relies on the individual’s appraisals and reports of their own emotional states, which can be influenced by both current affect and propensities to experience negative or positive moods. Additionally, appraisals of social interactions could have referenced virtual or in-person interactions, however, the question wording did not allow for distinction between the type of interaction. Such differences may be particularly relevant to the social functioning of behaviorally inhibited individuals and warrant further study. Although BI has both social and non-social components, we did not have a measure of specifically non-social behavior available at age 18. Finally, due to data constraints, this study does not explore mechanisms underlying the relationships between early childhood BI and adulthood social and emotional problems. As suggested previously, mechanisms could include negative feedback loops outlined in social learning theories of anxiety (Zinbarg et al., 2022) or, if our EMA is capturing adulthood manifestations of BI, other processes that contribute to the stability of traits. As we demonstrate associations between early BI and later functioning, questions about mechanisms supporting these associations are important for future research.

Constraints on Generality

Our sample is relatively homogeneous with respect to race and ethnicity, limiting generalization to more diverse samples.

Conclusion

Research over the last several decades has supported the significance of BI in early childhood as a marker of negative outcomes in later childhood and adolescence, however, we are just beginning to understand what those outcomes are in adulthood. Results of the current study suggest that behaviorally inhibited children may not grow up to be less social as young adults but may nevertheless have more negative emotional reactions, particularly anxiety, to some social experiences. Further, they tend towards lower positive emotion in general. This may reflect chronically reduced enjoyment of social experiences, but it may also extend to non-social behaviors and experiences (e.g., performance at work, hobbies). Future work should examine the associations between BI and emotions in both social and non-social contexts. Identifying these sources of blunted enjoyment and heightened negative affect may suggest targets for interventions and compensatory strategies to improve wellbeing for individuals with a behaviorally inhibited temperament style in childhood, such as cognitive behavioral interventions for modifying perceptions of threat and negative self-evaluation during social interactions.