Private Speech and the Development of Self-Regulation: The Importance of Temperamental Anger

Abstract

In this study (N=160), we observed children’s private (i.e., self-directed) speech (PS) during a challenging puzzle task at age 3 and assessed whether the amount and maturity of their PS predicted their inhibitory control (IC) at age 4 and indirectly emotion regulation at age 9. Additionally, we examined whether the direct and indirect effects of PS were moderated by children’s temperament. As expected, the maturity of children’s PS was positively associated with IC and this association was stronger when children were reported as higher in anger reactivity by mothers (the interaction accounting for 11% of the explained variance). Children low in temperamental anger tend to have good IC and may not need to use PS. When children were at or above the mean on anger reactivity, PS maturity was indirectly associated with better emotion regulation at age 9 through an influence on IC at age 4 (index of moderated mediation =1.03 [.10, 3.60]). Findings suggest that PS is an important self-regulatory tool for 3-year-olds who typically experience and express anger.

Adaptive functioning in a social world requires self-regulation, a capacity to modulate one’s own internal states and behaviors appropriately for a context. Inhibitory control (IC) is a higher-order cognitive skill associated with the prefrontal cortex (PFC) that involves suppressing or temporarily withholding inappropriate responses (e.g., rule-incompatible), and may therefore support children’s ability to manage behavior and emotions (e.g., frustration) appropriately in social settings (Bell & Wolfe, 2004; Fox & Calkins, 2003). Children with better IC tend to have fewer externalizing behavior problems (e.g., misconduct, aggression; Utendale & Hastings, 2011; Brophy et al., 2002; Hughes et al., 2000), and IC in preschool distinguishes between children who show normative declines in these problems across childhood from those who exhibit stable high or increasing linear trajectories (Perry et al., 2018). Thus, understanding how IC develops, and identifying factors that may contribute to individual differences in the preschool period, are important.

Rudimentary forms of IC are observable late in the first year when the PFC begins to functionally mature (Diamond, 1990). It is not until around 4 years of age, however, that children can suppress or delay behavior in accordance with rules and verbal instructions (Carlson, 2005; Diamond & Taylor, 1996; Jones et al., 2003). Words may provide children with new ways of internally representing information (Zelazo, 2004), which may support the development of new ‘verbally-mediated’ IC skills. Thus, although it may build on existing capacities that are more basic (Cuevas & Bell, 2014), children’s language skills may play an important role in the continued development of IC from infancy to preschool. Many studies have reported positive associations between vocabulary size and IC in preschoolers (Blair & Razza, 2007; Wolfe & Bell, 2007), but the contribution of expressive language skills in this regard has received far less attention. Considering that regions of the PFC are uniquely specialized for speech (Broca’s area), empirical investigation into the role of children’s speech in the continued development of IC is warranted.

Private (i.e., self-directed) speech (PS) has long been implicated in self-regulation (Kopp, 1982; Vygotsky, 1978) and may support children’s emerging ability to use language internally to manage their behaviors and emotions appropriately in social settings. Indeed, Vygotsky (1978) thought that children’s PS during problem-solving activity becomes internalized over time as ‘inner speech,’ a mental tool associated with action planning. However, despite a large body of empirical research on PS in children (Winsler, 2009), very few studies have been longitudinal or focused on preschool. Additionally, no previous study on PS has considered the role of children’s temperament, which serves an organizing role in the development of self-regulation (Calkins, 1994). In particular, a young child’s propensity to react with anger, an approach-related negative emotion (Carver & Harmon-Jones, 2009), may serve as the foundation from which higher-order cognitive skills, such as IC, develop. As such, the associations between PS and IC may depend on children’s tendency to express anger. Thus, among children higher in temperamental anger, PS may contribute to the development of emotion regulation in middle childhood through an earlier influence on IC. In the current study we assessed 1) whether characteristics of children’s PS during problem-solving at age 3 were associated with their IC at age 4, 2) whether these associations were moderated by temperamental anger, and 3) whether children’s PS was linked to subsequent emotion regulation abilities in middle childhood (through IC at age 4) at varying levels of anger reactivity.

Private Speech and the Development of Inhibitory Control

PS is defined as speech that is not directed to another person or that is poorly adapted to a listener (Diaz & Berk, 1992). Although it is often produced in social contexts, PS is distinguishable from social speech by a variety of indicators (e.g., pronouns, eye contact; see Winsler et al., 2005). By the early preschool period, many children begin to use PS as a tool for resolving conflicts and solving problems. For example, to perform a difficult task (e.g., opening a birthday present), a young child may verbalize the steps to herself out loud (e.g., First I untie it!) as a way to organize her activity towards the goal. Similarly, a child may verbalize her frustration (e.g., I can’t find it!) as a way to remain focused on a problem. Emotion regulation involves using skills or strategies to manage emotional arousal so that goals are accomplished and successful interpersonal functioning is possible (Calkins, 1994). By turning negative emotion towards the self, PS may support young children’s ability to manage frustration appropriately in order to persist with a difficult task on their own.

However, this behavior is also thought to have developmental significance. Specifically, by repeatedly pairing PS that is self-relevant and self-guiding with their experiences of conflict and challenge, young children may develop a tendency to pause and self-reflect when conflicts arise or when there are problems to be solved (Vygotsky & Luria, 1994). This theory is not inconsistent with neurobiological models of self-regulation (Lewis, 2005; Posner & Rothbart, 2000, Rothbart et al., 2011), which suggest that by age 4, IC relies on neural circuitry in the PFC and anterior cingulate cortex (ACC), a brain region involved in conflict monitoring. When conflicts in information processing (e.g., goal-blockage) are detected, the ACC ‘alerts’ the PFC to the need for top-down control (Bush et al., 2000). Considering that speech involves PFC activation, young children’s PS may involve coactivation of these brain regions, which may support the formation of neural circuitry (Lewis, 2005). Thus, in addition to serving a self-regulatory function ‘in the moment,’ by guiding children’s attention and behavior in accordance with goals, PS may support children’s emerging ability to use language internally to inhibit behaviors and control emotions appropriately in novel or challenging contexts.

Importantly, for children’s PS to become internalized as inner speech that supports self-reflection and self-control, it may need to be meaningfully related to their activity, goals, or feelings; they may need to be talking about themselves as objects or what they are trying to do. In the literature, children’s PS is often characterized in terms of audibility and maturity. In general, mature PS is meaningfully related to children’s activities, goals, or feelings about a task (e.g., asking questions, defining a problem, coordinating speech with manual actions) whereas immature PS is self-stimulating (e.g., making sounds, repeated words or phrases), exclamatory, or task-irrelevant (Berk, 1986; Winsler et al., 2005). A child who routinely asks themselves goal-relevant questions when confused, for instance, may develop a tendency to pause and self-reflect in challenging situations. In contrast, PS that is self-stimulating (i.e., repeating words/phrases, making sounds), exclamatory (e.g., Bah!), or task-irrelevant could represent impulsive motor behavior that reinforces impulsive action in similar contexts over time. For these reasons, a greater amount of mature PS and a lesser amount of immature PS during problem-solving are often expected to be associated with better self-regulation in young children.

Although many studies have used Berk’s (1986) coding scheme to distinguish between mature and immature forms of PS in preschoolers, the findings do not always support this hypothesis. For example, the number of mature and immature PS utterances young children produced during a challenging Lego construction task were negatively correlated with teacher-reported self-regulation abilities in the classroom (Bono & Bizri, 2014). Similarly, 5-year-olds who produced a greater amount of overt, task-relevant PS during problem-solving performed more poorly on executive function tasks and had more teacher-reported behavior problems than children who produced less PS (Winsler et al., 2000). Of note, half of the original sample (N = 82; M = 46.4 months; 80% male) was considered at risk for ADHD based on teacher-reported behavior problems, and those children produced a greater amount of overt task-relevant PS throughout the preschool period. Across groups, however, most children showed a decline in their reliance on PS during this time.

Very few studies have assessed PS in children younger than 3.5 years, which is a notable gap in the empirical literature considering that most children are showing declines in PS use across the fourth year. In a study with preschoolers (unselected for behavior problems), 3.5-year-olds produced more overt PS during problem-solving tasks than they did 6 months later, and 4.5-year-olds produced less PS than 4-year-olds (Winsler et al., 2003). Further, in one of the first empirical studies on PS, ‘intellectually bright’ children produced more overt task-relevant PS during problem-solving at age 4 than other children, but less than other children at older ages (Kohlberg et al., 1968). Thus, by 5 years, a greater amount of PS (even mature PS) during problem-solving may be negatively associated with IC because it is a marker of increased risk for behavior problems; but these associations may not be observed at younger ages (e.g., 3 years) when PS is more developmentally normative.

In young children, the relative maturity of PS may be important to consider in addition to the overall amount. Controlling for age, the amount of PS 3.5- to 5-year-olds produced was not associated with self-regulation, but the proportion of PS utterances classified as immature was negatively associated with teacher-reported social skills and positively associated with externalizing behavior problems (Winsler et al., 2003). Thus, although individual children produced very different amounts of PS, when it was more meaningfully related to what they were doing, feeling, or thinking, they had better self-regulation in the preschool classroom. Therefore, in examining longitudinal associations between PS and IC in young children, it is necessary to consider both the amount and maturity of PS as predictors. Further, considering that rapid improvements in IC are observed between 3.5 and 4 (Jones et al., 2003), and that children may begin to decrease reliance on PS across the fourth year (Winsler et al., 2000, 2003), observing PS in 3-year-olds is important.

The Moderating Role of Temperamental Anger

Intrinsic factors, such as temperament, shape the ways children regularly engage with their environments and may provide context or meaning to patterns of behavior children display in laboratory settings. Thus, it is important to consider how children’s temperament could moderate the associations between PS and IC. Temperament is often defined as biologically based differences in reactivity to the environment within the realms of affect, activity, and attention (Derryberry & Rothbart, 1997, 1988). Anger is a particularly salient feature of children’s temperament that is observable early in life and modestly stable over time (Gagne & Goldsmith, 2011). Importantly, infants high in anger reactivity are at increased risk for externalizing behavior problems (e.g., ADHD) in childhood (Auerbach et al., 2008; Nigg et al., 2004; Sullivan et al., 2015) and anger reactivity is often negatively correlated with IC in young children (Gagne & Goldsmith, 2011; Hughes et al., 2000; Kochanska & Knack, 2003; Rothbart et al., 2001). Thus, an underlying biological disposition to experience anger intensely may interfere with the development of IC, a higher-order cognitive skill that could be used to manage anger appropriately (Calkins, 1994; Fox & Calkins, 2003).

Importantly, however, anger is relatively normative in young children and may even serve an adaptive function of increasing persistence and motivation to overcome obstacles (He, Xu, & Degnan, 2012; Lewis et al., 2015; Tan & Smith, 2018). Further, in contrast to toddlers, who tend to become aggressive and disruptive when frustrated (e.g., Calkins & Dedmon, 2000), preschoolers often engage in context-appropriate goal-directed actions to resolve the situation (Dennis et al., 2009). Problem-solving was the context in which young children’s PS was thought to become internalized to support self-regulation over time (Vygotsky, 1978). Thus, because anger may serve as an organizer of problem-solving activity in novel or challenging situations, anger reactivity may provide young children with opportunities to benefit from using PS in the real world. As such, although children higher in temperamental anger may have lower IC than children lower in temperamental anger (on average), the associations between PS maturity and IC may be stronger when children are more prone to expressing anger.

Although no previous study on PS has incorporated measures of temperament, there is some support for this hypothesis. Preschoolers rated by mothers as higher in anger reactivity (but not other components of temperament) exhibited greater PFC activation during a conflict inhibition task (Fishburn et al., 2019), suggesting anger-prone children may have stronger neural responses to cognitive conflict, an internal signal that IC may be needed. Interestingly, and consistent with that view, children’s PFC activation was also positively correlated with behavioral slowing on trials requiring inhibition, suggesting that temperamental anger is a foundation from which IC-related neurocircuitry may develop; it could facilitate the internalization of PS during problem-solving at a neural level. Thus, although biological factors associated with temperament may help explain the associations, research suggests that the effects of PS on IC in the preschool period may depend on children’s tendency to express anger.

Implications for Emotion Regulation

Through an influence on IC, characteristics of young children’s PS during problem-solving may indirectly support their ability to manage anger and frustration appropriately in later childhood. In infants and toddlers, emotion regulation is largely accomplished with avoidant (e.g., attention shifting) or self-stimulating behaviors that serve to reduce negative arousal (Buss & Goldsmith, 1998; Stifter & Braungart, 1995). As children get older, however, emotion regulation additionally involves managing behavior in ways that are socially appropriate (Saarni, 1999), which may require a more sophisticated set of underlying skills. Upon receiving a disappointing gift, for example, children are expected to respond in a socially acceptable manner (e.g., Thank you!) which may require inhibiting or suppressing inappropriate expressive displays (Carlson & Wang, 2007; Cole et al., 1994; Hudson & Jacques, 2014). Similarly, if a dominant response to frustration is approach (e.g., aggression), then children may need to use IC to manage frustration appropriately in social settings. For these reasons, the acquisition of cognitive skills (IC in particular) is thought to contribute to further developments in emotion regulation beyond infancy (Fox & Calkins, 2003).

Describing the individual differences that exist among young children in emotion regulation is not simply a matter of identifying the different strategies that they use in particular emotionally arousing situations; the developmental process by which these strategies are acquired is itself subject to individual differences (Calkins, 1994, p. 54).

Thus, because PS may support the development of IC in early childhood, it may be indirectly associated with emotion regulation in middle childhood. That is, if children can use PS to facilitate greater development of IC, this may also be important for the emergence of sophisticated emotion regulation skills needed to handle intense emotional experiences and expressions in socially appropriate ways. This developmental process may be particularly evident among children higher in anger reactivity given the evidence that they may develop stronger neural links underlying PS and IC, and that greater regulatory skills are needed to manage intense negative emotions.

Though no empirical study has tested this hypothesis, there is some evidence to support the associations. For example, in a longitudinal study, IC in preschool was important for distinguishing between children who exhibited normative declines in externalizing problems across childhood from those who displayed stable high or increasing linear trajectories (Perry et al., 2018). Additionally, however, proneness to anger at age 2 was associated with a decreased likelihood of being in the ‘stable low’ group. Furthermore, Raaijmakers and colleagues (2008) found that IC was the only dimension of executive function significantly related to aggression in 4-year-olds. However, in a study with 855 preschoolers, the highest aggression levels were reported in children with low IC and high negative affect (Suurland et al., 2016). Thus, in considering the indirect effects of PS on emotion regulation in middle childhood (through IC in preschool), it is important to consider children’s temperament as a moderator.

Current Study

The first aim of the current study was to investigate whether characteristics of children’s PS during problem-solving at age 3 were associated with their IC at age 4. Although previous studies have relied on structured paradigms to observe PS in children, we observed children’s PS from a relatively unstructured puzzle task designed to be challenging for 3-year-olds, which is effective in eliciting PS in young children (Behrend et al., 1989). Based on previous research, we expected that the amount of PS (utterances per minute) children produced during the task would be negatively associated with IC, but that the maturity of children’s PS (i.e., mature/total PS) would be positively associated with IC. To better understand how PS might contribute to the development of IC across the fourth year, we controlled for children’s IC at age 3.

The second aim of the study was to assess whether the associations between PS and IC were moderated by children’s temperament. In this regard, we focused on anger reactivity because it is a risk factor for externalizing problems associated with IC deficits (e.g., ADHD; Nigg et al., 2004). In longitudinal studies, parent report of children’s anger reactivity is stable from infancy to early childhood whereas observed measures are not (Gagne & Goldsmith, 2011). Thus, we focused on parent-report of child anger reactivity in this study. Based on theory and previous research, we expected that temperamental anger reactivity at age 3 would be negatively associated with IC at age 4 and that the positive association between PS maturity and IC would be stronger when children were higher in anger.

A final aim of the study was to assess whether characteristics of children’s PS contribute to the development of emotion regulation in childhood through an influence on IC and whether this indirect effect was moderated by children’s temperament. In general, because children lower in anger reactivity may be less likely to experience difficulties with managing anger and frustration appropriately in middle childhood, the hypothesized positive association between IC and emotion regulation may not be significant among children lower in anger reactivity. Additionally, because children lower in anger reactivity tend to have better IC, they may not benefit as much from PS. As such, we hypothesized that an indirect (positive) association between PS maturity at age 3 and emotion regulation at age 9 (through IC at age 4) would be stronger at higher levels of temperamental anger, but non-significant at lower levels of anger.

Method

Participants

Children were part of a longitudinal study investigating the integration of cognition and emotion in development. The full sample was recruited at 5 months (4 were added at 10 months; N = 410) from two locations (Blacksburg, Virginia; Greensboro, North Carolina) via commercial flyers and mailing lists generated from county birth records. Because children’s PS was only coded at the Greensboro site, only children from the Greensboro cohort were considered for inclusion in the current study (N = 199). These children (53% male) were diverse in terms of race and ethnicity: 26% were Black, 8% were of mixed race, 59% were racially White and non-Hispanic. Approximately 60% had 1 or more parents with a 4-year college degree. All demographic information was self-reported on by mothers. One-hundred and seventy-five children participated at age 3 (M = 3.07, SD = .08), 166 participated at age 4 (M = 4.07, SD = .08), and 149 participated at age 9 (M = 9.16, SD = .13).

Procedures

Upon arrival to the laboratory at each visit, families were greeted by a research assistant who explained the study procedures and obtained signed consent. After an initial warm-up period, children participated in a battery of tasks designed to assess emotion and cognition processes while mothers completed questionnaires. All lab-based assessments were completed on the same day. Families were compensated $50 and children were given a small gift at the end of each visit. All data collection protocols were approved by the university institutional review board.

Measures

Private Speech.

Children’s speech was observed at age 3 from a ‘Difficult Puzzle’ task. After completing a non-challenging puzzle with their mothers, children were given an alphabet puzzle to complete on their own; mothers were given a magazine and instructed not to interact with their child. The experimenter took all the pieces (letters) out of the puzzle before leaving the room for 2 minutes (M = 110 s; SD = 26 s). No child successfully completed the puzzle. For control purposes, task engagement (i.e., working on the puzzle) was coded on a 1 to 4 scale for each 30 s epoch and averaged to yield a composite (ICC = .89).

Children’s speech from the puzzle task was later transcribed and coded according to established guidelines (Winsler et al., 2005). Utterances were operationally defined as a sentence, sentence fragment, clause, or any string of speech that was separated from another by at least 2 seconds. An utterance could not contain 3 seconds of silence or a major shift in semantic content. Accuracy of speech transcription was accomplished on 20 percent of the sample. Coding of speech involved classifying utterances as social or self-directed and then further classifying utterances based on content. Utterances were considered social if they were accompanied by eye contact or gestures, or if they contained a vocative or pronoun reference. Utterances that occurred within 2–3 seconds of any previous social utterance were considered social if they were not accompanied by a significant change in tone or content.

Self-directed utterances were coded according to Berk’s (1986) scheme, which categorized PS utterances into categories subsumed under three ‘levels’ based on overtness (i.e., audibility) and task-relevance. Level 1 (i.e., immature) PS was defined as self-stimulating or task-irrelevant and included four categories: wordplay, repetition, comments to imaginary others, and utterances that were semantically unrelated to the task. Level 2 (i.e., mature) PS was defined as task-relevant externalized PS, such as statements or questions about the task, describing one’s own activity, and task-relevant affect expression (i.e., verbalization of feelings). Level 3 PS (i.e., partially internalized) included inaudible muttering, whispers, and silent lip movements.

Using Berk’s (1986) scheme and the Winsler et al. (2005) coding manual, self-directed utterances were classified into one of 13 categories (see Table 1 for a complete description of coding scheme with examples). Categories 8 through 13 were conceptualized as mature PS (i.e., Level 2/3) and categories 1 through 7 were conceptualized as immature PS (i.e., Level 0/1). Summing across children, there were 1,315 PS utterances (M = 8.48, SD = 8.87, Max = 42); approximately 70% were classified into one of the mature PS categories. Among children who produced PS (N=128), 11% produced only immature PS and 27% produced only mature PS. However, most children produced a combination of mature and immature PS. Inter-rater reliability was evaluated for the mature PS composite and was found to be within acceptable range (ICC = .99). Thus, following Winsler et al. (2003), a PS maturity score was calculated by dividing the number of mature PS utterances by the total number of PS utterances.

Table 1.

Private Speech Coding Scheme

Category		Level	Definition (example)
1.	Unintelligible	0a	Unable to transcribe due to poor articulation but spoken at full volume
2.	Nonword sounds	1b	Nonlinguistic wordforms or sounds (e.g., clucking; bing-bing-bong)
3.	Exclamations/transitions	1b	Served to bridge silence or occurred after an event without a meaningful semantic connection to activity (e.g., Darn!, Um...)
4.	Wordplay/repetition	1	Repeating words or phrases for the sake of saying them (e.g., B-B-B-B-B)
5.	Fantasy/imaginary	1	Talking to the puzzle or speaking for the puzzle pieces as part of pretend play
6.	Off-task/irrelevant	1	Utterances irrelevant to the puzzle or task (e.g., Do we eat grapes? I got football)
7.	Describing materials	1	Labeling the puzzle pieces or clues (e.g., [points] A moon!)
8.	Feedback/self-correction	2d	Evaluative statements following task-relevant actions (e.g., [tries piece]—No, that’s not it)
9.	Action coordination	2d	Words or labels occurring in conjunction with task-relevant actions (e.g., This goes [puts piece on] here)
10.	Affect expression	2	Defining the problem or describing internal states (e.g., I don’t know where this goes; I’m mad at this one [points] ‘cause I can’t put it on!)
11.	Asking questions	2	(e.g., Where’s the U? How does put this in?)
12.	Describing activity/planning	2	(e.g., I wanna put them back in their places; I’m looking for a ‘S)
13.	Partially-internalized muttering	3	Unable to transcribe due to low-volume or inaudibility, but occurred in the context of focused attention and/or task-relevant activity (e.g., putting pieces on)

Notes

^aWinsler et al. (2005) considered this as ‘Level 0;’

^bthese were not part of Berk’s (1986) ‘Level 1’ class but were recognized in the Winsler et al. (2005) manual

^dthese were not part of Berk’s (1986) ‘Level 2’ class but were recognized by Winsler et al. (2005).

Anger Reactivity.

At age 3, mothers reported on children’s anger with the Children’s Behavior Questionnaire (CBQ), a survey designed to assess general patterns of behavior associated with temperament in children of 3–7 years of age (Rothbart et al., 2001). The shortened form of the CBQ consists of 94 items that load onto 15 scales (Putnam & Rothbart, 2006). For each item, mothers indicated how accurately it described their child’s behavior (1 = extremely untrue, 7 = extremely true) during the 2 weeks prior to the laboratory visit. The Anger/Frustration scale consists of 6 items (e.g., Gets quite frustrated when prevented from doing something s/he wants to do) and had an internal consistency of .78 in the current sample.

Inhibitory Control.

Children’s IC was observed at age 4 from three experimental tasks and experimenter report.

In the ‘Three Pegs’ task (Balamore & Wozniak, 1984), children are given a wooden mallet and instructed to tap a set of colored pegs in non-canonical order; to succeed, they need to inhibit the dominant tendency to tap the pegs in the order they are presented (from left to right). To ‘pass’ the task, children needed to tap the correct color sequence twice in a row. If the child tapped correctly, a second trial was conducted. If children tapped incorrectly, another set of instructions was given. Children received at least two and as many as six opportunities to tap the pegs depending on their pattern of responding. A pretest was given before the initial instruction to ensure that children could distinguish between the colors (blue, green, yellow). A trained coder later scored the child’s performance by watching the DVD recording and assigning an ordinal score. Specifically, a 0 was given if they failed the task, 1 if they required two repeated instructions, 2 if they required one additional instruction, and 3 if they successfully performed the task after the first instruction.

In the ‘Hand Game’ (Hughes, 1998), children are asked to place a flattened hand on the table whenever the experimenter presents her fist and to make a fist whenever the experimenter places her flattened hand on the table. Children were given at least two practice trials (during which they were praised or corrected), and then 16 test trials were administered. The experimenter’s hand was in a fist for half of the trials, and trials were arranged in a pseudorandom order. A trained coder later scored the child’s performance by watching the DVD recording. A point was awarded if the child responded to the experimenter’s hand gesture with the opposite gesture; only the initial response was scored. The percentage of correct trials was calculated and used in the analyses. Reliability coding for this task was accomplished on 14% of the sample and was within acceptable range (ICC = .94).

In the ‘Gift Delay’ task (Kochanska, Murray, & Harlan, 2000), an experimenter brings a gift and wrapping supplies into a room where the child is waiting. During the wrapping phase, the child is told to sit with his or her back to the experimenter for 1 minute without peeking while the gift is wrapped. Then, the experimenter leaves the room (for 1 minute) to retrieve a bow; during this time, the child is told not to touch the gift box. A trained coder later scored the child’s behavior using the DVD recording. Latency to touch the gift box (in seconds) was calculated and used in the analyses. Reliability coding for this task was accomplished on 15% of the sample and was found to be within acceptable range (ICC = .96). This task was also administered at age 3 and served as a control for earlier IC in the analyses (ICC = .85).

Experimenters reported on children’s IC with the Preschool Self-Regulation Assessment (PSRA; Smith-Donald et al., 2007), a survey designed to assess self-regulation from structured laboratory tasks. The PSRA contains 25 items that load onto two broad factors (Attention/Impulse Control, Positive Emotion). For each item, raters circle the number (0 – 3) that best describes the child’s behavior during the entire laboratory session. The Attention/Impulse Control factor consists of 11 items (e.g., Child has difficulty waiting; Lets examiner finish before starting task; Does not interrupt) and had an internal consistency of .91 in the current sample. Completion of the PSRA was accomplished by the experimenter who interacted with the child within 30 minutes of the family leaving.

Emotion Regulation.

Children’s emotion regulation was assessed at age 9 from a combination of questionnaire and observed measures. Lower scores reflect better emotion regulation.

The Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001) is a 112-item screening instrument for emotional and behavioral problems in children aged 6 to 18. Each item describes a problem that children may have (e.g., acts too young for his/her age) and loads onto one of eight scales and two broad factors (Internalizing, Externalizing). For each item, mothers indicated how well the statement described their child within the past 6 months using a 3-point scale (0 = Not True, 1 = Somewhat/Sometimes True, and 2 = Very/Often True). The Aggressive Behavior scale (17 items; e.g., Argues a lot, Gets in many fights, Temper tantrums) was used in the analyses (Cronbach’s alpha = .87).

The Strengths and Difficulties Questionnaire (SDQ) is a screening instrument for mental health problems in children (Goodman, 1997; Goodman & Goodman, 2009). The self-report version (Goodman, Meltzer, & Bailey, 1998) contains 25 items that ask about the child’s behavior and feelings in the past 6 months, covering a broad range of difficulties commonly observed in children and adolescents. Responses are provided on a 3-point scale (0 = Not True, 1 = Somewhat True, and 2 = Certainly True). Each item loads onto one of five scales (5 items each): Emotional Problems, Conduct Problems, Hyperactivity/Inattention, Peer Problems, and Prosocial Behavior. Items reflecting problems are summed to generate a total difficulties score (20 items; Ruchkin, Jones, Vermeiren, & Schwab-Stone, 2008), which has been shown to discriminate well between children with and without psychopathology (Goodman, 2001). In the current study, items from the Emotional Problems (e.g., I am often unhappy; I worry a lot) and Conduct Problems scale (e.g., I get very angry and often lose my temper; I fight a lot) were used in the analyses (Cronbach’s alpha = .64).

In the ‘Rigged Posner’ task, children were told they could play a computer game designed by another lab with an opportunity to win ten dollars (if they scored 1,000 points). To maximize emotional investment in the game, the experimenter placed a ten-dollar bill on the table (in front of the child) and asked the child what they would spend the money on if they won. To play the game, children needed to respond (by pressing a computer key) to the location of a circle displayed on the computer screen as fast and as accurately as possible. If children made a mistake, a loud sound (“Wrong!”) resulted. Negative feedback in the first block was accurate and children rarely made a mistake. However, in subsequent blocks, the computer program was rigged to provide negative feedback regardless of children’s performance (20% of trials in block 2 and 22.5% of trials in block 3). At the end of the third block, a message appeared on the screen to let children know they had lost the game; they were then given 1 minute to ‘stew’ over their loss before being debriefed. Children’s emotion was later coded by a trained research assistant using the DVD recording. The task was broken up into 10 s epochs and negative affect was coded for each epoch with a Likert scale (1 = Low, 2 = Moderate, 3 = High). The mean negative affect code was calculated and used in the analyses. Reliability coding was accomplished on 18% of the sample and was within acceptable range (ICC = .90).

Sample Reduction and Missing Data

Children were included in the current study if they came into the lab at 3 or 4 years. The reduced sample (N = 160) had a similar demographic profile as the full sample (54% male, 63% White and non-Hispanic, 66% had 1 or more parents with a 4-year college degree). Children excluded from the study were more likely to be non-white or Hispanic, t = 2.12(57), p = .04), but did not significantly differ in terms of sex, t = .56(56), p = .58, parent education, t = 1.73(57), p = .09, or infant negative temperament, t = .18(26), p = .86).

Most (74%) children provided some data at all three timepoints. Five children were missing data from the puzzle task at age 3 (3 did not come into the lab, 1 could not be scored due to a recording error, 1 was voided due to experimenter error); an additional 28 were missing data for PS maturity (10 did not talk during the task and 17 produced only social speech). Among children who came into the lab at age 4 (N=145), most (82%) had usable data for all four IC measures. Twenty-five children did not play the Hand Game (due to child refusal or inability to understand the rules); five were missing one of the other IC measures. Among children who participated at age 9 (N=133), most (77%) had usable data for all three measures. Seventeen children did not complete the SDQ because it was not yet in the study protocol; an additional 11 were missing the SDQ and observed measure because they did not come into the lab (mothers completed questionnaires only). Nearly two-thirds (64%) of children included in the study had complete data at age 9.

Analysis Plan

Analyses were conducted in R Studio using the ‘lavaan’ package (Rosseel, 2012) and full information maximum likelihood (FIML) was used to handle missing data (Kline, 2011). To address the first study aim, a latent factor was created from the four IC measures (using the ‘cfa’ function) and then regressed onto PS variables (using the ‘sem’ function). To evaluate the overall goodness of model fit, we report the chi-square goodness-of-fit statistic, degrees of freedom (df), corresponding p-value, comparative fit index (CFI), and the root mean square error of approximation (RMSEA; McDonald & Ho, 2002). Hu and Bentler (1999) suggested that for continuous variables, CFI > .95 and RMSEA < .06 are indicative of a good model fit.

To address the second study aim, Anger Reactivity (mean-centered) was added to the model as a predictor of IC along with two ‘Anger x PS’ interaction terms. Path coefficients of the interaction terms were examined and significant interactions (p < .05) were further probed via simple slopes analysis. Specifically, conditional effects at 1 SD above and 1 SD below the moderator (Anger) were estimated. To provide an index of effect size, we report the changes in R² (explained variance) accounted for by the main effects and interaction terms (Warner, 2013).

Finally, to address the third study aim, a latent factor was created from the three emotion regulation measures (using the ‘cfa’ function) and added to the model as a dependent variable (using the ‘sem’ function). Conditional indirect effects of PS on emotion regulation (through IC) were then tested at 1 SD above and below the mean on Anger Reactivity using bias-corrected boot-strapping (1,000 draws). If the 95% confidence interval surrounding the estimate does not contain zero at the selected level of confidence the result is interpreted as statistically significant. Regarding effect size, we report the index of moderated mediation (Hayes, 2015), which provides an estimate of the statistical weight of a moderator on an indirect association. For clarity, unstandardized path estimates are reported in-text and standardized estimates are depicted in figures.

Results

Preliminary Analyses

Bivariate correlations between study variables are displayed in Table 2; all variables are normally distributed. IC task performance at age 3 is positively correlated with two of four IC variables at age 4, which are positively correlated with one another. PS Amount is not significantly correlated with PS Maturity or any other study variables. However, PS Maturity is positively corelated with two of four IC measures. Emotion regulation variables are positively correlated, and some are significantly correlated with IC. Anger Reactivity is marginally negatively correlated with IC at age 3, significantly correlated with two of four IC variables at age 4, and with Aggressive Behavior (CBCL) at age 9. However, Anger Reactivity is not correlated with PS variables. Puzzle task engagement is marginally positively correlated with both PS variables, indicating that children tended to produce more mature PS when they were actually working on the puzzle (vs. noncompliant, looking around). Thus, to ensure that any associations between PS maturity and IC are not attributable to children’s behavioral engagement with the task, task engagement (along with Age 3 IC) served as controls in the analyses.

Table 2.

Descriptive Properties and Bivariate Correlations among Study Variables

			1	2	3	4	5	6	7	8	9	10	11	12
1.	PS (utterances per minute)			.05	†.13	−.04	−.09	.05	−.10	−.08	−.09	.08	−.07	−.09
2.	PS (% maturity)				†.16	−.06	−.08	*.24	.09	*.20	.08	−.14	.00	.02
3.	Task Engagement					−.11	*.18	.07	†.17	.11	**.23	−.03	*−.25	−.10
4.	Anger Reactivity						−.14	−.08	−.12	**−.31	**−.31	.10	.03	**.23
5.	IC (‘Gift Delay’ latency)							*.21	.07	.10	**.34	*−.11	**−.29	−.15
6.	IC (‘Gift Delay’ latency)								.15	**.22	**.32	−.20	†−.20	−.13
7.	IC (‘Hand Game’)									**.34	**.29	−.06	−.13	.03
8.	IC (‘Three Pegs’)										**.36	−.08	†−.11	−.14
9.	IC (PSRA)											**−.25	**−.29	−.11
10.	ER (‘Rigged Posner’)												**.30	.13
11.	ER (SDQ)													*.25
12.	ER (CBCL)
		Mean	4.58	.69	3.38	4.98	69.20	41.60	.82	.51	2.11	1.38	15.19	4.29
		SD	4.60	.31	.68	1.06	51.54	23.71	.20	.41	.64	.83	3.20	4.33
		Range	19.53	1	2.75	5.83	120	60	.94	1.00	2.91	3.00	19.00	23.00
		Skew	1.05	−.81	−1.48	−.71	−.22	−.69	−1.61	−.01	−.77	.14	1.22	1.68
		N	155	128	155	158	140	141	120	143	144	121	105	132

Note

^**p < .01

^*p < .05

^†p < .10.

Variables 1 – 5 were collected at age 3; variables 6 – 9 were collected at age 4; variables 10 – 12 were collected at age 9; PSRA = Preschool Self-Regulation Assessment; SDQ = Strengths & Difficulties Questionnaire; CBCL = Child Behavior Checklist

Analyses Addressing Research Questions

To address the first study aim, a structural equation model was conducted to assess the relations between PS and IC. First, a latent IC factor was created from the four measures. This model fit well, χ2(2) = 1.87, p = .39; RMSEA = .00 [.00 - .16]; CFI = 1.0, and all factor loadings were significant (see Table 3). Subsequently, IC was regressed onto both PS variables (amount, maturity) and both controls. This model also fit well: χ²(14) = 17.33, p = .24; RMSEA = .04 [.00 - .09]; CFI = .96. The amount of PS was not significantly associated with IC (B = −.02, p = .18), but PS maturity was positively associated with IC (B =.62, p = .01). Thus, regardless of how many PS utterances were produced, when 3-year-olds’ PS was proportionally more mature (i.e., relevant to their activity, goals, or feelings about the task), their IC one year later tended to be greater. Together, PS variables accounted for approximately 9% of the variance (R²) in IC (31% of the total variance explained in this model).

Table 3.

Unstandardized and Standardized Factor Loadings for Latent Factors

Observed Variable	Latent Factor	β	B	SE	p
PSRA scale	IC	.69	1	-	-
‘Three Pegs’ score	IC	.54	.78	.20	<.01
‘Hand Game’ score	IC	.53	.78	.22	<.01
‘Gift Delay’ latency	IC	.44	.64	.17	<.01
SDQ ‘Emotion/Conduct Problems’	ER	.55	1		-
CBCL ‘Aggressive Behavior’	ER	.33	.82	.36	.02
‘Rigged Posner’ affect	ER	.44	.25	.11	.03

Note: PSRA = Preschool Self-Regulation Assessment; IC = Inhibitory Control; CBCL = Child Behavior Checklist; SDQ = Strengths and Difficulties Questionnaire; ER = Emotion Regulation

To address the second aim, Anger Reactivity and two ‘Anger x PS’ interaction terms were added to the model as predictors of IC. Both interaction terms were created by multiplying the mean-centered Anger Reactivity variable by the mean-centered PS variables (amount, maturity). This model fit well: χ²(33) = 39.22, p = .21; RMSEA = .03 [.00 - .07]; CFI = .94. Anger Reactivity was not correlated with PS variables or IC at age 3 but was negatively associated with IC at age 4 (B = −.22, p < .01). The interaction between Anger and PS amount was not associated with IC. However, the interaction between PS maturity and Anger Reactivity was positively associated with IC (B = .50, p = .02). The addition of Anger Reactivity and both interaction terms resulted in a change in R² of .16, suggesting that temperament in combination with PS accounted for approximately 25% of the variance in IC (57% of the total variance explained in this model). The PS Maturity x Anger interaction term alone accounted for 5% of the variance (11% of the total variance explained).

To probe the significant interaction of PS maturity and Anger Reactivity on IC, a simple slopes analysis was conducted. Specifically, the conditional effects of PS maturity on IC were estimated at 1 SD below and 1 SD above the mean on Anger. Results indicated that the slope was significant at the mean (B = .48, p < .01) and 1 SD above the mean (B = 1.01, p < .01), but not 1 SD below the mean (B = −.05, p = .90; see Figure 1). As depicted in Figure 2, when Anger Reactivity (mean-centered) was inside the interval [−.56, 6.45], the slope of PS maturity to IC was significant (p < .05), and most children had Anger Reactivity scores that fell within the significance interval (see Figure 3). Thus, the beneficial effects of PS on IC were observed for most children but were not observed among children low in Anger Reactivity and were stronger among children higher in Anger Reactivity.

Figure 1.

Interaction of Private Speech Maturity and Anger Reactivity on Inhibitory Control

Note: The effect is significant at the mean and 1 SD above the mean on Anger Reactivity; plot was created in R-Studio using the ‘sim_slopes’ function (IC was converted to a composite).

Figure 2.

Johnson-Neyman Plot of Interaction between Private Speech Maturity and Anger Reactivity in the Prediction of Inhibitory Control

Note: Anger Reactivity is mean-centered; the Y-axis represents the conditional slope of the predictor (PS maturity) on IC. The dotted line indicates at what level of the moderator (Anger Reactivity) the conditional slope differs significantly from zero. As depicted, the positive association between PS maturity and IC was only significant when children were above the mean (0 = mean of a centered variable) on Anger Reactivity.

Figure 3.

Distribution of Anger Reactivity Scale Scores

Note: Anger Reactivity is mean-centered (M = 4.98); 112 children (71%) were within 1 SD of the mean, 21 (13%) were > 1 SD above the mean, and 25 (16%) were > 1 SD below the mean.

To address the final study aim, emotion regulation was estimated as a latent factor and added to the structural model as a dependent variable. Subsequently, indirect (unconditional and conditional on Anger Reactivity) effects from PS to emotion regulation through IC were examined with bootstrapped confidence intervals (1,000 draws). The measurement model for the latent emotion regulation factor fit well, χ²(2) = .91, p = .64; RMSEA = .00 [.00 - .14]; CFI = 1.0, and all factor loadings were significant (see Table 3). The full structural model also fit well: χ²(73) = 76.02, p = .38; RMSEA = .02 [.00 - .05]; CFI = .98. IC was positively associated with emotion regulation (B = 1.57, p = .03), but other study variables were not directly associated with emotion regulation. Thus, to conserve parameter estimates, the ‘Anger x PS amount’ interaction term was taken out of the model and direct paths between age 3 controls and emotion regulation were removed (see Figure 4 for a path diagram).

Figure 4.

Path Diagram depicting Associations between Private Speech, Inhibitory Control, and Emotion Regulation

Note: ** p < .01; * p < .05. All modeled paths are depicted, but only significant covariances are depicted for simplicity. Dashed lines = nonsignificant paths. All estimates are standardized.

To assess whether an indirect effect of PS maturity on emotion regulation through IC was conditional on children’s tendency to express anger, parameters were created via syntax and bootstrapped confidence intervals (1,000 draws) around those parameters were estimated. Specifically, direct, indirect, and total effects were assessed at 1 SD above and 1 SD below the mean on Anger. The unconditional indirect effect from PS maturity to emotion regulation through IC was significant (B = 1.03 [.05, 3.78]). Additionally, the confidence interval surrounding the indirect effect did not contain zero at 1 SD above the mean on Anger Reactivity (B = 2.09 [.31, 6.15]). The index of moderated mediation was also significant (B = 1.03 [.10, 3.60]). Thus, when children were higher in Anger Reactivity, the maturity of their PS during problem-solving at age 3 was positively associated with their emotion regulation at age 9 through an influence on IC at age 4, but this effect was not significant at lower levels of Anger Reactivity.

Discussion

The purpose of this study was to investigate the role of children’s PS in the development of self-regulation. Although it has long been thought that PS is important in this regard (Barkley, 1997; Vygotsky, 1978), very few studies on PS in children have been longitudinal or focused on preschool, a time when PS is maximally observable (Berk & Winsler, 1995) and when self-regulation skills are rapidly coming online (Carlson, 2005). Rapid improvements in IC are observed towards the end of the fourth year of life (Jones et al., 2003), which may serve as a critical foundation for more sophisticated forms of emotion regulation in middle childhood (e.g., expressive suppression). Although some studies have explored the role of children’s PS in ER (Day & Smith, 2013), none have done so longitudinally and no previous study on PS has considered the role of children’s temperament. The lack of longitudinal work on children’s PS is especially surprising because scholars have argued that the beneficial effects of PS are likely to occur over time rather than ‘in the moment’ (Berk, 1985; Fraunglass & Diaz, 1985). With our longitudinal design and temperament perspective, this study fills a critical gap in the empirical literature on children’s PS and results provide information regarding which aspects of self-regulation PS may influence and for whom it is most beneficial.

The first aim of the study was to assess whether characteristics of children’s PS at age 3 were associated with their IC at age 4. Our hypothesis that the amount of PS would be negatively associated with IC was not supported. We expected this because in previous studies, young children who produce lots of PS during problem-solving are perceived by teachers as having disruptive behavior problems and poor self-regulation (Bono & Bizri, 2014; Winser et al., 2000). Importantly, however, very few children in those studies were younger than 3.5 and children may begin decreasing their reliance on overt PS across the fourth year (Winsler et al., 2003). In this study, the amount of PS was marginally positively correlated with task engagement and rarely (if ever) occurred during off-task behavior. Thus, at 3 years, the overall amount of PS a child produces during problem-solving may not be negatively associated with later IC because this is a time when PS is maximally observable and developmentally normative. However, much additional work is needed to understand the nature of developmental change in PS across the preschool period.

As hypothesized, and consistent with previous work (Winsler et al., 2003), the maturity of children’s PS was positively associated with IC. That is, regardless of how many utterances were produced, when 3-year-olds’ PS was proportionally more mature, they were better able to inhibit inappropriate behaviors in accordance with rules and social standards one year later. Mature PS is meaningfully related to children’s activity, goals, or feelings about the task whereas immature PS is self-stimulating, exclamatory, or task irrelevant. Thus, mature forms of PS may have been more intentionally produced. A potential reason for why young children’s PS maturity was positively associated with IC then is because pairing immature PS with problem-solving activity serves to promote impulsive responding (vs. self-reflection) in challenging contexts over time. The precise mechanism underlying the effect cannot be determined from this study. However, because we controlled for children’s engagement with the puzzle and IC at age 3, this finding suggests that characteristics of young children’s PS at 3 years may play a role in the development of IC across the fourth year.

Importantly, however, the positive association between PS maturity and IC was moderated by children’s temperament. Specifically, the effect was significant when children were at the mean on anger, and was stronger when children were higher in anger, but was not significant among children lower in anger. Children low in anger may have an inherent advantage in the development of IC and possibly do not need to use PS in this regard. However, anger was not correlated with the amount of PS. Rather, anger reactivity may have provided children with the necessary opportunities to benefit from using PS in the real world. Indeed, problem-solving is the context in which young children’s PS becomes internalized over time (Vygotsky, 1978) and anger motivates young children to resolve conflicts and solve problems in their daily lives (Dennis et al., 2009). Thus, even if children high in anger did not produce a lot of PS in this study, they may have been more actively engaged in problem-solving across the fourth year than children low in anger; for this reason, characteristics of their PS may have been more strongly associated with their IC over time. Much additional longitudinal work is needed to elucidate the underlying mechanisms of this developmental process. However, findings are consistent with Vygotsky’s (1978) theory of PS internalization and further suggest that the effects of PS may depend on children’s temperament.

Assuming temperament has a biological basis, a biological explanation for this finding is also possible. Indeed, studies have suggested there is overlap in neural circuitry associated with anger and IC in young children (Fishburn et al., 2019), and the neural network that supports IC in childhood is thought to come online across the fourth year (Rothbart et al., 2011). Theoretically (e.g., Lewis, 2005), 3-year-olds’ PS during problem-solving activity could contribute to the formation of this neural circuitry. However, biological factors associated with temperament may help explain the associations. In a study with 4-year-olds, for instance, the amount of mature PS during problem-solving was positively associated with IC when children also had higher levels of respiratory sinus arrythmia (Hassan et al., 2018), a biological marker of anger reactivity (Stifter & Fox, 1990; Stifter & Jain, 1996). Thus, incorporating biological measures into longitudinal research on children’s PS is important.

Finally, among children higher in anger, the maturity of PS during problem-solving at age 3 was indirectly positively associated with emotion regulation at age 9 through an influence on IC at age 4. Specifically, when 3-year-olds were rated as higher in anger by mothers, and when their PS was relatively more mature, they were better able to manage anger and frustration appropriately at age 9, because they had better IC at age 4. Importantly, although children lower in anger tended to have better IC than children higher in anger (on average), IC did not help explain an association between PS and emotion regulation among children lower in anger. Thus, even PS that is negatively valanced in tone may promote self-reflection and controlled responding in emotionally charged situations over time. Much additional longitudinal work is needed to elucidate the underlying mechanisms of this developmental process. However, findings are consistent with Vygotsky’s (1978) theory of PS internalization and further suggest that the beneficial effects of PS may depend on children’s temperament.

Strengths, Limitations, and Future Directions

This study adds meaningfully to the existing literature on children’s PS. First, few if any studies have observed children’s PS at 3 years, a time when it may be critical to do so considering that normative declines in PS are observed across the fourth year (Winsler et al., 2003) alongside rapid advances in IC (Jones et al., 2003). Although some studies have observed PS in children as young as 3 (Bono & Bizri, 2014; Winsler et al., 2003), most children in the samples (or subgroups) were approaching 4 or 5. In this study, PS was observed from a large sample of children who were almost exactly 3 years old (90% within two months of their birthday). Further, only two previous studies, to our knowledge, have examined children’s PS longitudinally in the preschool period (Winsler et al., 2000; 2003). By observing characteristics of 3-year-olds’ PS as predictors of their IC at age 4, and by considering temperamental anger as a moderator, our study builds meaningfully on that work and has theoretical implications for the role of PS in the development of emotion regulation in childhood (Cole et al., 2010).

A limitation of the study, however, was that speech from the puzzle task was not further analyzed for aspects of language complexity (e.g., mean length utterance) or volubility (e.g., total words produced). An utterance is a unit of word meaning; some utterances resemble full sentences (e.g., I can put this right here) whereas others consist of only a single word (e.g., Yes!). Thus, although a child who produced many PS utterances arguably used PS more than a child who produced fewer utterances, they may or may not have actually produced more PS in terms of vocal activity. Classifying PS into utterances is standard practice in the field and necessary to code PS based on functional meaning. However, other aspects of PS may be important to consider. Thus, to better quantify the amount of PS children produce during problem-solving, future studies should incorporate additional metrics.

The measurement context of children’s PS is a strength of the study. In most previous studies, PS was observed from highly structured problem-solving tasks administered by an experimenter, the goal being to relate children’s use of PS to their success/failure across a series of trials (e.g., Fernyhough & Fradley, 2005). Although this approach is well-suited for microgenetic analyses, to understand the role of children’s PS in development, it is essential to observe the behavior spontaneously in situations where it occurs naturally (Michel, 2007). Following others (Day & Smith, 2013), we observed children’s PS from a relatively unstructured task designed to elicit frustration, which is more consistent with Vygotsky’s (1962) original studies and the real-world contexts in which young children are solving problems. That children’s mothers were nearby but unavailable to help with the puzzle, in particular, may have increased children’s production of PS during the task in a way that was ecologically valid (McGonigle-Chambers et al., 2014). Future studies on PS should similarly incorporate more naturalistic approaches to measurement.

A related limitation of study, however, was that children were able to address their mothers during the task, but their social speech was not analyzed. The focus of this study was children’s PS. However, approximately half of utterances transcribed from the puzzle task were classified as social, most (42%) of which were further classified as ‘help-seeking.’ Children’s PS is theorized to develop from their verbal exchanges with adults (Vygotsky, 1987), and children sometimes alternate back and forth between appealing to their mothers for help and talking themselves through the problem. Longitudinal studies have suggested that social support-seeking in certain contexts may also promote the development of emotion regulation (Roben et al., 2013). Further, a recent study found that maternal behaviors in toddlerhood are predictive of young children’s usage of PS during challenging tasks (Day & Smith, 2019). Thus, incorporating children’s social speech and caregivers’ speech into research on PS is another important future direction.

In conclusion, results from this study provide support for the role of PS in the development of emotion regulation (Cole et al., 2010) and for the integration of cognition and emotion processes in early childhood (Bell & Wolfe, 2004). Findings specifically suggest that among 3-year-olds prone to experiencing anger and frustration, the maturity of PS during problem-solving activity may contribute to the development of IC across the fourth year, a cognitive skill that may contribute to emotion regulation in middle childhood (Calkins & Marcovitch, 2010; Fox & Calkins, 2003). Much additional longitudinal work is needed to understand the complex biological and social processes that may help explain the associations. However, results fill a critical gap in the empirical literature on children’s PS. In addition to incorporating measures of neural functioning, future studies should analyze children’s social speech, as well as caregivers’ speech, during problem-solving, as they likely play important roles in the larger developmental process (Calkins, 1994; Kopp, 1982, 1989; Vygotsky, 1978).

Highlights.

• Most 3-year-olds produced some private speech during a challenging puzzle task.

• When children’s private speech was more mature (i.e., meaningfully related to what they were doing, thinking, or feeling) they had better inhibitory control at age 4.

• The effectiveness of private speech may depend on children’s temperament: a stronger effect was observed at higher levels of anger reactivity.

• Children low in temperamental anger tend to have good inhibitory control and may not need to use private speech.

• Private speech may play a role in the development of emotion regulation among 3-year-olds who typically experience and express anger.