Developmental stability and change in emotion regulation strategies and strategy repertoires across adolescence

Abstract

The ability to manage emotions using emotion regulation (ER) strategies is a core competency developed across childhood and into adolescence. Youth are expected to develop more diverse ER repertoires—the range of strategies that adolescents use—as they approach adulthood. However, to date, an examination of longitudinal change and stability in normative ER strategy use or repertoires across early to late adolescence has yet to be conducted. The present study reports on two longitudinal samples with measures of six ER strategies. Reports in the Younger Sample (N = 201, aged 11–12 at Wave 1) were repeated once a year for 5 years, and reports from the Older Sample (N = 187, aged 13–15 at Wave 1) were repeated twice a year for 3 years. Growth curve analyses revealed that Distraction, Rumination, and Suppression increased in the Younger Sample, whereas Reappraisal, Relaxation, and Engagement increased in the Older Sample. Latent Markov models showed four ER repertoires in both samples (low/average, suppression propensity, engagement propensity, and high diversity) that showed moderate to high stability across waves. Across adolescence, there were increasing transitions into the high diversity profile, with some nuances by sample. Results are discussed in terms of normative emotional development and implications for understanding adolescent polyregulation and ER flexibility.

INTRODUCTION

A core competency that develops over childhood and adolescence is the ability to self‐regulate emotions (Allen & Sheeber, 2008; Dahl, 2004; Hollenstein & Lougheed, 2013). In adolescence, youth face new social and emotional challenges which provide opportunities to further their acquisition and mastery of emotion regulation (ER; Fombouchet et al., 2023; Silvers, 2022). Furthermore, normative maturation of the physiological and psychological processes underlying ER undergo rapid and significant change during adolescence (Steinberg, 2008). Thus, ER capacities are expected to change and improve during this period (Fombouchet et al., 2023; Silvers, 2022) which may be reflected in developmental changes in ER repertoires (i.e., the range of strategies youth use; De France & Hollenstein, 2017; Zimmermann & Iwanski, 2014). However, robust assessment of longitudinal, normative development of ER strategy use in adolescence is strikingly scarce, making generalizations about normative strategy use difficult. Further complicating this, most research in this area assesses wide‐ranging lists of strategies categorized as “adaptive” or “maladaptive” for well‐being (e.g., Aldao et al., 2010; Cracco et al., 2017). Although it is important to investigate how ER strategies relate to psychopathology, these approaches are often favored at the neglect of important normative functions and development of ER strategies. Consequently, we lack a comprehensive description of what ER strategies adolescents typically use. Thus, the purpose of the present study is to address these research gaps by revealing the normative trajectories of individual ER strategies and broad ER repertoires across early to late adolescence. By better establishing normative patterns of stability and change in the use of various strategies, we can begin to build evidence‐based theories about the development of ER strategy use across the critical formative period of adolescence.

ER strategies in adolescence

ER is an ongoing process (Cole et al., 2004; Kappas, 2011) often operationalized as the strategies used to modify emotions (Gross, 2015; Ochsner & Gross, 2005). Because emotion is generally conceptualized as loosely coupled changes in emotion components (i.e., cognition, expression, arousal), ER strategies can be conceptualized as the means through which emotion components are modified (Gross, 2008). Some research has suggested that individual ER strategies differ in effectiveness (i.e., the strategy's ability to produce the intended hedonic or instrumental effect) and adaptiveness (i.e., the consequences of frequent strategy use on mental health or functioning; for reviews, see Aldao et al., 2010; Webb et al., 2012). However, strategies' effectiveness can be in conflict with their adaptiveness. For example, distraction is relatively effective for lowering negative emotional intensity in the moment (Webb et al., 2012), but habitual use is associated with poor well‐being (e.g., Aldao et al., 2010). Furthermore, not only can adaptiveness and effectiveness be in conflict, but they can also be in conflict with themselves. For instance, an adolescent may choose to chronically suppress negative emotions at school to focus on schoolwork and have positive peer interactions (i.e., good functioning), but this may have unintended negative consequences for their mental health.

Given the immense variability in conceptualizations of strategies and differing strategies proposed to be relevant to ER, we focus on six self‐regulation strategies used to downregulate negative emotions¹ which have direct impact on the central components of emotion: cognition (i.e., distraction, rumination, reappraisal), expression (i.e., expressive suppression, expressive engagement) and arousal (i.e., relaxation). We did not examine indirect strategies (e.g., social support) as these may influence strategy selection itself.

Adolescence has unique potential for normative developmental change in ER, as it is a period when emotional maturation and new social challenges (e.g., first jobs, dating) coincide. Despite this, there is little research on how the use of ER strategies develops longitudinally across adolescence in non‐clinical samples. Only four studies to date have examined individual strategies longitudinally across two time points in a non‐clinical sample: over 6 months (Wang & Hawk, 2020), over 1 year (Chervonsky & Hunt, 2019) and over 2 years (Gullone et al., 2010; Voon et al., 2014). As such, we describe the six strategies of interest below and cautiously summarize normative developmental trends for each using the few existing longitudinal studies and, indirectly, supplemental cross‐sectional age comparisons when available.

Distraction

Distraction is a cognitive strategy which directs attention away from emotion components (e.g., browsing social media). Distraction's adaptiveness is contentious; although some research characterizes it as a uniformly maladaptive strategy due to its negative associations with well‐being (Aldao et al., 2010), others suggest that distraction is an effective strategy for down‐regulation (Webb et al., 2012). There is also evidence that distraction has benefits over more “adaptive” strategies under certain contexts (Sheppes, 2020). Research on how use of distraction develops is limited. Normative development of distraction has been widely studied in infancy and childhood, with distraction present in ER repertoires by early childhood (Ratcliff et al., 2021). However, there are no longitudinal studies examining use of distraction in adolescence. Cross‐sectionally, Cracco et al. (2017) found distraction to be lower in early adolescence (ages 12–15) compared to late childhood (ages 8–11) before rising later in adolescence (ages 16–18). Additionally, adolescents use distraction more than do adults (De France & Hollenstein, 2019; Smith et al., 2023), suggesting that this rise in later adolescence may be temporary.

Rumination

Rumination is a cognitive strategy which involves persistently re‐evaluating the causes and consequences of emotion components (e.g., reflecting; Nolen‐Hoeksema, 1991). Although rumination is often considered a maladaptive regulation strategy, rumination can be done with the adaptive intent to work through a negative emotion or trying to find a solution to the emotion‐eliciting event (Watkins, 2008). Rumination as a regulation strategy has not been examined longitudinally. Cross‐sectional work suggests that rumination increases nonlinearly around ages 12–14 for boys but linearly from ages 8–18 for girls (Cracco et al., 2017). The opposite has also been found, with linear increases for boys aged 10–17 and cubic (i.e., S‐shaped) increases for girls aged 10–17 (Jose & Brown, 2008). Adolescents report using rumination less than adults (De France & Hollenstein, 2019), possibly indicating that it increases into adulthood. Additionally, girls report more rumination than boys across age (Cracco et al., 2017; Johnson & Whisman, 2013; Sanchis‐Sanchis et al., 2020).

Reappraisal

Reappraisal is a cognitive strategy which modifies the evaluative meaning of the emotion components (e.g., looking for a “silver lining”). Reappraisal is considered a cognitively demanding strategy to deploy (Sheppes, 2020) and thus may require greater cognitive maturity. Reappraisal is often construed as an adaptive ER strategy based on positive associations with emotional functioning and well‐being (Aldao et al., 2010; Webb et al., 2012). However, reappraisal also has drawbacks depending on the circumstances in which it is used (e.g., Ford & Troy, 2019). Reappraisal appears to emerge sometime after age nine (DeCicco et al., 2012, 2014), potentially due to burgeoning cognitive capacities that begin to mature in adolescence (Buhle et al., 2014; McRae et al., 2012). Longitudinally, reappraisal use appears stable across the first half of adolescence (Wang & Hawk, 2020 [ages 10–11 to 13–14]; Chervonsky & Hunt, 2019 [ages 11–12]; Voon et al., 2014 [ages 14–16]; Gullone et al., 2010 [ages 9–15]) but may increase in the second half of adolescence (ages 15–18; Wang et al., 2015). However, other cross‐sectional work has found no age differences in later adolescence (ages 14–18; Verzeletti et al., 2016). Furthermore, some forms of reappraisal increase across late childhood to early adolescence (ages 8–13) before flattening out (ages 16–18; Cracco et al., 2017). Cross‐sectional comparisons between adolescents and adults have been mixed: some find that adolescents rely on reappraisal as much as adults (De France & Hollenstein, 2019), whereas others find increases across the lifespan (ages 13–80; Le Vigouroux et al., 2017). Mean‐level gender differences have been found, with girls reporting higher use of reappraisal than boys (Wang et al., 2015).

Expressive suppression

Expressive suppression is a behavioral strategy which reduces outward displays of the emotion (e.g., pretending not to be upset). This strategy is frequently painted as uniformly negative despite evidence that it has some benefits (Webb et al., 2012) and may not always be detrimental to mental health (Geisler & Schröder‐Abé, 2015; Wylie et al., 2025). Expressive suppression emerges early in childhood and may plateau around late childhood (Gross & Cassidy, 2019). However, longitudinal studies on suppression in adolescence have been mixed. Gullone et al. (2010) found that suppression decreased over 2 years across ages 9–15, whereas Voon et al. (2014) found that suppression increased over 2 years across ages 14–16. Other longitudinal studies found no changes across early adolescence (Chervonsky & Hunt, 2019; Wang & Hawk, 2020). Cross‐sectional comparisons have also been mixed, some finding no age differences in expressive suppression in adolescence (Verzeletti et al., 2016; Wang & Hawk, 2020) and some finding that it may not reach its peak until young adulthood (De France & Hollenstein, 2019; Zimmermann & Iwanski, 2014). Mean‐level gender differences have been found in adolescents' reports of suppression, with boys reporting higher use of suppression than girls (Wang et al., 2015). However, gender differences in longitudinal trajectories during adolescence have been small to non‐existent (Gullone et al., 2010). All in all, these findings may point to expressive suppression having a nuanced developmental pattern: rising in childhood, temporarily plateauing in late childhood and adolescence before rising again in young adulthood.

Expressive engagement

Expressive engagement is a behavioral strategy—distinct from expressive suppression—which involves enhancing or displaying outward displays of the emotion (e.g., crying; Cameron & Overall, 2018). Expressive engagement has been associated with many benefits, such as facilitating insights into emotions and eliciting support (Kennedy‐Moore & Watson, 2001). However, expressive engagement is not uniformly adaptive. When used without awareness and self‐reflection, it may amplify distress (Kennedy‐Moore & Watson, 2001) and, in certain social contexts, may elicit negative reactions from others. To date, expressive engagement has received little research attention. Wang and Hawk (2020) found no changes over 6 months or differences in expressive engagement between 10–11‐year‐olds and 13–14‐year‐olds. Similarly, Sanchis‐Sanchis et al. (2020) did not find age differences between 9–12‐year‐olds and 13–15‐year‐olds. Additionally, gender differences have been observed in adolescence, with girls reporting higher expressive engagement than boys (Pascual et al., 2016; Sanchis‐Sanchis et al., 2020).

Relaxation

Relaxation is a physiological strategy which directly modifies emotional arousal (e.g., deep breathing). Although relaxation strategies are commonly used in interventions which indirectly target ER (e.g., mindfulness; Deplus et al., 2016) and research suggests that arousal control is effective for downregulating emotions (Hamdani et al., 2022), there have been no studies explicitly examining relaxation strategies within the ER literature itself. Given the dearth of literature, it is difficult to draw conclusions about the development of relaxation strategies during adolescence. One cross‐sectional study found that relaxation was used more frequently in early/middle adolescence (ages 12–15) than in both young (ages 20–25) and middle adulthood (ages 40–60; De France & Hollenstein, 2019), very tentatively suggesting normative declines between early/middle adolescence and early adulthood.

In sum, very little is known about the developmental trajectories of ER strategy use into and across adolescence. However, some evidence points to different strategies being characterized by distinct trajectories (i.e., different starting points and change), with some potential variation by gender. These limited findings also cover a wide age range which makes it difficult to interpret fine‐grained developmental timing of change across different periods of adolescence (Zimmermann & Iwanski, 2014, 2018).

ER repertoires

Following several decades of examining ER strategies in relative isolation from one another, the broader (i.e., adult) ER field has increasingly shifted to examining the range of ER strategies individuals use (i.e., repertoire; Dixon‐Gordon et al., 2015), combinations of strategies employed simultaneously or in close succession (i.e., polyregulation; Ford et al., 2019), and the situationally contingent use of strategies (i.e., ER flexibility; Bonanno & Burton, 2013). There are two general principles underlying these approaches. First, no individual strategy is inherently “bad” or “good”, but rather functional if it is appropriate to the situation and achieves regulatory goals (Bonanno & Burton, 2013; Haines et al., 2016). Second, experience sampling studies show that use of only one ER strategy for a given emotional event is rare (Heiy & Cheavens, 2014; Ladis et al., 2022). Indeed, in experience sampling, 50%–90% of adolescent ER's variance is at the within‐person level (Benson et al., 2019; De France & Hollenstein, 2022; McKone et al., 2024). This implies that youth select from a range of ER strategies to implement based on situational context. Thus, examining individual strategies in isolation may be less informative than examining youth's ER repertoire (De France & Hollenstein, 2017; Grommisch et al., 2020).

Relying on just one or a few strategies may reduce opportunities for context‐sensitive use, polyregulation, and regulatory success. That is, a limited ER repertoire constrains regulatory possibilities and, by extension, a more diverse repertoire provides a more sufficient “toolbox” of strategies (De France & Hollenstein, 2019; Dixon‐Gordon et al., 2015; Lougheed & Hollenstein, 2012). Rather than having high levels of “adaptive” strategies and low levels of “maladaptive” strategies, it may be better to have a diverse repertoire of strategies which can be flexibly selected and deployed based on situational demands. Thus, understanding the structure and content of individuals' ER repertoires is a necessary precursor to understanding polyregulation and context‐contingent strategy use. Repertoires represent the range of options from which one may select and deploy ER strategies. As such, identifying adolescents' normative ER repertoires is a preliminary step necessary to investigate contingent and flexible ER deployment dynamics.

To date, a handful of studies have examined ER repertoires as individual differences in social–emotional functioning in adults (e.g., Dixon‐Gordon et al., 2015; Grommisch et al., 2020; Southward et al., 2018) and youth (Cummings et al., 2023; De France & Hollenstein, 2017, 2019; Lougheed & Hollenstein, 2012; Thomsen & Lessing, 2020). Having a more diverse repertoire has been associated with better functioning (Cummings et al., 2023; De France & Hollenstein, 2017; Lougheed & Hollenstein, 2012), with some nuances depending on the composition of the diverse repertoires (Grommisch et al., 2020). Among studies focusing on adolescence, most have used Latent Profile Analysis (LPA) to extract repertoires of dispositional ER strategy use at one point in time. In LPA, each adolescent is identified with membership in one ER repertoire (i.e., latent profile) that reflects their relative use of each strategy, based on the untested assumption that repertoire membership is stable over time. To date, only one study has examined normative change in ER repertoires, although in early childhood (Thomsen & Lessing, 2020). As adolescents' ER capacities mature and they experience opportunities to practice and refine ER strategies, it is likely that their ER repertoires will change (De France & Hollenstein, 2017; Zimmermann & Iwanski, 2014). However, no study thus far has tested this assumption.

The present study

This study's goal was to examine stability and change of within‐person ER trajectories across two samples of adolescents assessed longitudinally, collectively spanning ages 11 to 17. This was done using two complementary approaches: first, by examining trajectories of individual strategies and, second, by examining trajectories of repertoires. For the first approach, we estimated individual slopes of six strategies: distraction, rumination, reappraisal, suppression, engagement, and relaxation. Given limited longitudinal research, mixed findings, and limited work in normative samples, we did not have a‐priori hypotheses for this approach. The second approach, and primary goal of this study, was to examine the stability and change of adolescents' repertoires over time. At each time point, each participant's repertoire was used to estimate the probability of remaining in that repertoire (i.e., stability) or transitioning to a different repertoire at the next time point (i.e., change). Given that youth may develop greater context sensitivity across adolescence (De France & Hollenstein, 2022; McKone et al., 2024), we predicted that ER repertoires would shift from a reliance on fewer strategies towards more diverse repertoires. However, given the paucity of longitudinal research on repertoire change, our hypotheses as to whether and how repertoires may change longitudinally were relatively open‐ended. Finally, given probable mean‐level gender differences in the use of some strategies, we supplementally examined gender differences in the trajectories of both individual strategies and repertoires. Because prior findings have been mixed for individual strategies and non‐existent for repertoires (Cummings et al., 2023; Lougheed & Hollenstein, 2012), these analyses were exploratory.

METHODS

Participants

Younger sample

Adolescents (N = 201, 46.3% girls; 11–12 years old; M _age = 11.5 years) were recruited through a database of families who signed up to hear about studies at the university in a small Canadian city. 81.1% of adolescents reported their ethnic/racial background as White, 3.0% as Asian, 3.0% as Indigenous, 1.0% as Black, 0.5% as Non‐White Latin American, 8.0% as multi‐ethnic, and 3.5% unknown or preferred not to say. 7.5% of mothers reported their annual family income as <50,000 CAD; 11.5% as 50,000–75,000; 20.5% as 75,000–100,000; 34.5% as 100,000–150,000; and 26.0% as >150,000. Across all Waves, retention was 78.1–91.5%.

Older sample

Adolescents (N = 187, 50.3% girls, 13–15 years old, M _age = 13.9 years) were recruited through the same database. 79.1% of adolescents reported their ethnic/racial background as White, 5.3% as Asian, 0.5% as Black, 0.5% as Latin American, 0.5% as Indigenous, 7.0% as multi‐ethnic, and 7.0% as unknown or preferred not to say. 94.0% of adolescents lived above the poverty line. Across all Waves, retention was 69.5%–92.5%.

Procedures

Younger sample

The study was approved by the university's Health Sciences Research Ethics Board (blinded for review). Wave 1 recruitment spanned September 2018 to March 2019. Adolescents visited the lab with their mother and, following consent, completed questionnaires separately on computers, followed by lab tasks. Participants returned 1 year later to complete the same protocol. After Wave 2, lab visits ceased due to COVID‐19. The next three annual waves were completed online using Limesurvey (LimeSurvey GmbH, n.d.) with data secured on a lab server. Adolescents' compensation was $15 at Wave 1 and increased by $10 each following wave.

Older sample

The study was approved by the same university's Internal Review Board (blinded for review). Wave 1 recruitment spanned October 2017 to April 2018. At Wave 1, adolescents visited the lab with a caregiver and, following consent and assent, adolescents completed questionnaires on a computer with their caregiver present but out of view. For the subsequent five bi‐annual waves, the questionnaires were completed online using Limesurvey as described above. The Older Sample completed four biannual waves prior to the COVID‐19 pandemic. Adolescents' compensation at each wave was, chronologically, $10, $15, $20, $30, $20, and $25.

Measures

Emotion regulation strategies

Participants completed the Regulation of Emotion Systems Survey (RESS; De France & Hollenstein, 2017), which assesses distraction (e.g., “Engaging in activities to distract myself”), rumination (e.g., “Thinking repeatedly about what was bothering me”), reappraisal (e.g., “Thinking of other ways to interpret the situation”), suppression (e.g., “Hiding my feelings”), engagement (e.g., “Vocalizing my feelings”), and relaxation (e.g., “Taking deep breaths”). The Older Sample completed the original 38‐item RESS and the Younger Sample completed the 24‐item short form RESS. Participants indicated how frequently they use each item in response to negative emotions on a scale from 1 (never) to 5 (always). Across waves, αs = .76–.93 (Younger Sample) and αs = .86–.96 (Older Sample) for each strategy, indicating high reliability.

Analysis plan

Standard longitudinal latent growth curve modeling was used to obtain estimates of the latent trajectory of each ER strategy across waves using MPlus 8.10 (Muthén & Muthén, 2017) with the maximum likelihood estimator. For each strategy, a series of growth models were fitted using structural equation modeling, comparing intercept only (i.e., no growth), linear growth, and quadratic growth models. To rule out COVID‐19 pandemic effects, growth was explored before and after the onset of the pandemic. Specifically, significant quadratic slopes were followed up by spline models with two linear slopes with the knot point coinciding with the first pandemic wave: Wave 3 for the Younger Sample, Wave 4 for the Older Sample. Chi‐square difference tests indicated whether adding each growth factor improved model fit. For each strategy, the final model was selected based on both parsimony and satisfactory model fit (Browne & Cudeck, 1993; Hu & Bentler, 1999). To explore potential gender differences, we added gender (0 = boy, 1 = girl) as a covariate predicting the growth factors.

Latent Markov models (LMMs) were used to examine changes in adolescents' ER repertoires over time. LMMs represent a longitudinal extension of LPA, modeling changes in latent profile membership across adjacent measurement occasions (i.e., waves). The latent profiles at each wave are referred to as latent states and represent the categories of a categorical latent variable, each characterized by distinct response patterns across a set of observed variables (Bartolucci et al., 2014). LMMs consist of two components: the measurement model and the structural model. The measurement model links the observed variables to the underlying, time‐varying latent state. In the present study, all observed variables were continuous, thus the measurement model is equivalent to that used in LPA. The structural model includes two sets of probabilities: initial probabilities (i.e., the probability of being in a particular latent state at Wave 1) and transition probabilities (i.e., the probability of transitioning to a given latent state at a subsequent wave, conditional on latent state membership at the previous wave).

Based on previous cross‐sectional ER repertoire profiles (De France & Hollenstein, 2017, 2019; Lougheed & Hollenstein, 2012), we computed a series of LMMs with one to six latent states. The data matrix consisted of five rows per person for the Younger Sample and six rows per person for the Older Sample (i.e., one row per wave). Missed waves were coded as missing values. The six ER strategies were entered as continuous observed variables. As recommended (Bartolucci et al., 2012), the optimal number of latent states was determined using basic LMMs with time‐homogeneous measurement models. We evaluated model fit using the Bayesian information criterion (BIC), Akaike information criterion (AIC), and sample‐size adjusted BIC (SABIC). Lower values indicate better fit, with the BIC typically prioritized in LMM applications (Bartolucci et al., 2012). In addition, we inspected the results of the Vuong‐Lo–Mendell–Rubin test (VLMR; Lo et al., 2001), which compares the fit of a k‐state to a k‐1‐state model. Of note, as the number of latent states increases, information criteria and the VLMR test can continue to favor models with increasing numbers of states, even if additional states may not be meaningfully distinct (Masyn, 2013). In such cases, we examined whether fit indices flattened out at a certain point (Nylund et al., 2007). To further inform model selection, we computed the approximate Bayes factor (BF) as an index of relative model fit between neighboring models. The BF reflects the ratio of the probability that the k‐state model is the correct model relative to the k + 1‐state model, with values >10 indicating strong evidence for the k‐state model (Masyn, 2013). Finally, if an additional state was only a minor variation of an existing state in a k − 1 state model, we favored the simpler solution. After selecting the number of latent states k, we compared LMMs with time‐invariant versus time‐varying measurement model parameters and tested whether transition probabilities could be constrained to be equal across waves (Bartolucci et al., 2012). All models were estimated using LatentGOLD version 6 (Vermunt & Magidson, 2013, 2021) with 1000 sets of random starting values. To explore potential gender differences, we included gender (0 = boy, 1 = girl) as a covariate predicting initial state membership and transition probabilities (Vermunt, 2010). To evaluate gender coefficients, we used Wald tests.

Sample size considerations

During the study planning phase, the sample sizes were determined based on power analyses for other research questions related to each broader study. For LMMs, estimation precision and the power to identify the correct number of latent states depend not only on design factors (e.g., number of measurement occasions, number of observed indicators), but also on population characteristics such as latent state separation, latent state proportions, and the size of transition probabilities (Gudicha et al., 2016). Given the lack of previous research examining the longitudinal stability (vs. change) of ER profiles in adolescents, we did not have estimates of these population characteristics and thus were unable to conduct a priori power analyses. Instead, we drew on findings from simulation studies that examined the quality of estimation of LMM parameters (Crayen et al., 2017), as well as the power to identify the correct number of latent states/profiles in LMMs (Bacci et al., 2014) and longitudinal (multilevel) latent class models (Lukočienė et al., 2010). Based on these findings, we considered our two samples—201 individuals with six observed indicators measured across five occasions (Younger Sample) and 191 individuals with six observed indicators measured across six occasions (Older Sample)—large enough to achieve high (>.80) power to correctly identify the number of latent states and obtain adequately precise parameter estimates in our LMMs.

RESULTS

Distributions of all ER strategies were visually inspected using histograms to inspect normality. Distributions were normal and no extreme (i.e., ±3.5 SD) outliers were identified, so raw mean values were used for each analysis below. Initial strategy means at Wave 1 can be found in the Supplemental Materials (Table SM1).

Growth trajectories of each ER strategy

For each sample and ER strategy, a series of latent growth models were fitted, comparing intercept only (no growth), linear growth, and quadratic growth models. Full test statistics can be found in the Supplemental Materials (Tables SM2–SM4). The linear growth factor improved model fit for all strategies, except for Reappraisal and Relaxation in the Younger Sample and Distraction in the Older Sample, for whom the intercept‐only model was retained, Δχ ²(3) = 4.94–7.36, p = .06–.18. Adding a quadratic factor either did not improve model fit, had nonsignificant quadratic mean and variance, and/or did not converge. Therefore, no quadratic growth models were retained and thus spline models were not considered.

Younger sample

The top of Table 1 reports the final latent growth curve estimates for the Younger Sample. For visual depiction of these analyses, see Figure 1. Each strategy's intercept mean and variance were significant, indicating that adolescents generally used all strategies at Wave 1 with between‐person variation in how much each strategy was used. Adolescents reported significant increases in Distraction, Rumination, and Suppression. Though including a linear slope improved model fit for Engagement, the slope was not significant, suggesting the Younger Sample's Engagement use was relatively stable. Adolescents' use of Reappraisal and Relaxation was also stable. Gender differences were then explored by regressing the final models' growth factors on gender. Girls reported lower use of Reappraisal at Wave 1 than boys (b = −.16, SE = .08, p < .05), as well as greater increases in Rumination use over time (b = .20, SE = .04, p < .001). No other gender differences were found in intercepts or slopes.

TABLE 1.

Final latent growth curve models.

	Distraction		Rumination		Reappraisal		Suppression		Engagement		Relaxation
	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p	b (SE)	p
Younger sample
I mean	2.60 (0.06)	<.001	2.36 (0.05)	<.001	2.18 (0.04)	<.001	2.42 (0.07)	<.001	2.42 (0.05)	<.001	2.24 (0.05)	<.001
S mean	0.14 (0.02)	<.001	0.19 (0.02)	<.001	—	—	0.18 (0.02)	<.001	−0.02 (0.02)	.28	—	—
I variance	0.47 (0.09)	<.001	0.32 (0.06)	<.001	0.23 (0.03)	<.001	0.57 (0.10)	<.001	0.25 (0.06)	<.001	0.40 (0.05)	<.001
S variance	0.03 (0.01)	.02	0.03 (0.01)	.001	—	—	0.06 (0.13)	<.001	0.02 (0.01)	.02	—	—
I‐S covariance	−0.05 (0.03)	.09	−0.02 (0.02)	.30	—	—	−0.09 (0.03)	<.01	−0.03 (0.02)	.14	—	—
Model fit
χ 2 (df), p	2.10 (10), 1.00		18.18 (10), .05		14.43 (13), .34		12.92 (10), .23		12.54 (10), .25		41.07 (13), <.001
CFI/TLI	1.00/1.00		.96/.96		.99/.99		.99/.99		.98/.98		.87/.90
RMSEA [CI]	.00 [.00, .00]		.06 [.00, .11]		.02 [.00, .08]		.04 [.00, .09]		.04 [.00, .09]		.10 [.07, .14]
SRMR	.02		.09		.07		.06		.07		.10
Older sample
I mean	2.96 (0.05)	<.001	3.23 (0.07)	<.001	2.21 (0.05)	<.001	2.85 (0.07)	<.001	2.45 (0.05)	<.001	2.11 (0.06)	<.001
S mean	—	—	0.02 (0.02)	.23	0.08 (0.02)	<.001	0.01 (0.02)	.56	0.03 (0.01)	.02	0.04 (0.02)	.004
I variance	0.40 (0.05)	<.001	0.69 (0.10)	<.001	0.32 (0.06)	<.001	0.60 (0.09)	<.001	0.39 (0.06)	<.001	0.46 (0.08)	<.001
S variance	—	—	0.03 (0.01)	<.001	0.02 (0.01)	<.001	0.03 (0.01)	<.001	0.01 (0.00)	.001	0.01 (0.01)	<.01
I‐S covariance	—	—	−0.06 (0.02)	.001	−0.01 (0.01)	.305	−0.06 (0.02)	.001	−0.02 (0.01)	.10	−0.02 (0.02)	.19
Model fit
χ 2 (df), p	18.25 (19), .51		37.43 (16), <.01		28.46 (16), .03		26.00 (16), .05		34.24 (16), <.01		23.75 (16), .10
CFI/TLI	1.00/1.00		.95/.96		.96/.96		.97/.98		.95/.96		.98/.98
RMSEA [CI]	.00 [.00, .06]		.08 [.05, .12]		.06 [.02, .10]		.06 [.00, .10]		.08 [.04, .11]		.05 [.00, .09]
SRMR	.07		.06		.08		.09		.09		.07

Note: Bold indicates significant values.

Abbreviations: I, intercept; S, linear slope.

FIGURE 1.

Growth trajectories for each ER strategy for Younger (a) and Older Samples (b).

Older sample

The bottom of Table 1 reports the final latent growth curve estimates for the Older Sample with corresponding plots in Figure 1. Again, intercept means and variances were significant for all strategies, indicating that adolescents generally used all strategies at Wave 1 with between‐person variation in how much each strategy was used. Adolescents reported significant increases in Engagement, Reappraisal, and Relaxation. Though including the slope improved model fit for Distraction, the slope was not significant, suggesting the Older Sample's use of Distraction was stable. Adolescents' use of Rumination and Suppression also showed no significant change over time. Adding gender as a covariate revealed that girls reported higher initial levels of Rumination (b = .44, SE = .14, p = .001) and Engagement (b = .27, SE = .11, p = .01) than boys. No other gender differences were found.

Latent Markov models

Latent state enumeration process

Younger sample

The top of Table 2 summarizes the fit coefficients and classification diagnostics for LMMs with up to six latent states. The VLMR test was significant for all models except the three‐state model. The AIC and SABIC continued to decrease with increasing model complexity, while the BIC was lowest for the four‐state model. The approximate BF also favored the four‐state model (LMM4). Therefore, we selected the four‐state solution. A version of the four‐state model with time‐varying measurement model parameters (i.e., no measurement invariance across waves; LMM4a) showed worse fit and was therefore not retained. Next, we tested whether transition probabilities could be set equal across occasions. A version of the four‐state model with time‐homogeneous transition probabilities (i.e., constant wave‐to‐wave change; LMM4b) had a lower BIC than the same model with time‐heterogeneous transition probabilities (i.e., varying rates of change), suggesting that model LMM4b was more parsimonious and thus preferred. Classification quality for the final model was high. The smallest average posterior probability for the four‐state model with time‐homogeneous transition probabilities was .82, indicating that individuals were assigned to their most likely latent state at each occasion with high certainty. Figure 2a displays the measurement model results for model LMM4b. To aid interpretation, the conditional means of each observed indicator (i.e., ER strategy) are depicted as SDs from the sample mean. The four latent states can be characterized as representing different configurations of ER strategy use. Based on strategies beyond ±0.5 SDs, we labeled the states: “Suppression Propensity” (State 1; n = 74), “Engagement Propensity” (State 2; n = 52), “Low/Average” (State 3; n = 49), and “High Diversity” (State 4; n = 26).

TABLE 2.

Model fit statistics and classification diagnostics for basic LMMS with different numbers of latent states.

Model	# of states	LL	BIC	AIC	SABIC	# of par	VLMR	BF	Classif error	Entropy R 2	Smallest AvePP	Smallest state size
Younger sample
LMM1	1	−7103.04	14269.71	14230.07	14231.69	12	—	—	—	—	—	—
LMM2	2	−6795.02	13733.23	13644.04	13647.69	27	616.03***	—	.10	.68	.89	.42
LMM3	3	−6638.95	13543.06	13377.90	13384.65	50	312.15***	—	.15	.68	.84	.24
LMM4	4	−6529.00	13487.57	13220.01	13230.95	81	219.89**	>10.00	.16	.70	.83	.14
LMM5	5	−6437.42	13511.24	13114.84	13131.06	120	183.17***	>10.00	.17	.73	.78	.12
LMM6	6	−6366.71	13619.07	13067.42	13089.99	167	141.42**	—	.17	.74	.80	.09
LMM4a	4	−6477.49	13766.38	13260.97	13281.65	153	—	—	.14	.74	.85	.11
LMM4b	4	−6540.25	13319.14	13170.50	13176.58	45	—	—	.17	.68	.82	.13
Older sample
LMM1	1	−7719.90	15502.84	15463.81	15464.82	12	—	—	—	—	—	—
LMM2	2	−7339.80	14831.92	14737.61	14740.06	29	760.20***	—	.08	.72	.89	.39
LMM3	3	−7055.15	14404.43	14222.30	14227.04	56	569.31***	—	.11	.75	.88	.23
LMM4	4	−6853.17	14194.81	13892.35	13900.22	93	403.95**	>10.00	.12	.78	.88	.21
LMM5	5	−6753.74	14242.80	13787.48	13799.33	140	198.87**	>10.00	.12	.80	.87	.06
LMM6	6	−6647.03	14328.76	13688.07	13704.74	197	213.41***	—	.12	.81	.84	.07
LMM4a	4	−6804.32	14569.81	13974.65	13990.13	183	—	—	.11	.80	.88	.10
LMM4b	4	−6879.35	13995.06	13848.71	13852.52	45	—	—	.12	.77	.87	.21

Note: Bold indicates the selected model. All models except LMM4a include time‐homogeneous measurement model parameters. LMM4a is the four‐state model with time‐varying measurement model parameters (i.e., no measurement invariance over time). LMM4b is the four‐state model with time‐homogenous transition probabilities.

Abbreviations: # of par, number of parameters; AvePP, average posterior probability for latent states; BF, Bayes factor; Classif Error, classification error; LL, log‐likelihood; LMM, latent Markov model; TP, transition probabilities.

^** p < .01.

^{*** p}  < .001.

FIGURE 2.

Overall latent states of ER strategies for Younger (a) and Older Samples (b).

Older sample

The bottom of Table 2 summarizes the fit coefficients and classification diagnostics for LMMs with up to six latent states. The VLMR test was significant for all models, and the AIC and SABIC continued to decrease with increasing model complexity. The BIC was lowest for the four‐state model. The approximate BF also favored the four‐state model (LMM4). Therefore, we selected the four‐state solution. A version of the four‐state model with time‐varying measurement model parameters (LMM4a) showed worse fit. The four‐state model with time‐homogeneous transition probabilities (LMM4b) had a lower BIC than the same model with time‐heterogeneous transition probabilities. Therefore, model LMM4b was more parsimonious and selected as our final model. The smallest average posterior probability for this model was .87, indicating that the accuracy with which individuals were assigned to their most likely latent state at each occasion was very high. Figure 2b displays the measurement model results for model LMM4b. Based on strategies beyond ±/−0.5 SDs, we labeled the states “Suppression Propensity” (State 1; n = 53), “Engagement Propensity” (State 2; n = 43), “Low/Average” (State 3; n = 52), and “High Diversity” (State 4; n = 39).

Stability versus change in emotion regulation profiles over time

Younger sample

Table 3 shows the initial and transition probabilities for the 4‐state LMM. At the first occasion, adolescents were most likely to be in “Low/Average” and least likely to be in “High Diversity”. ER profiles were moderately stable over time, with probabilities of remaining in the same state from one wave to the next ranging between .64 to .87. The probabilities of switching to “Suppression propensity” from another state at the previous occasion were moderate (i.e., .16–.21) and higher than all other cross‐state transitions. Figure 3a depicts the state probabilities calculated for each wave based on the estimated initial probabilities and the estimated transition probabilities. Over time, the probability of being in “Suppression Propensity” and the probability of being in “High Diversity” increased, and the probability of being in “Low/Average” and the probability of being in “Engagement Propensity” decreased. Gender did not significantly predict the initial state, χ ²(3) = 0.32, p = .96, or transition probabilities, χ ²(12) = 15.01, p = .24.

TABLE 3.

Initial and transition probabilities in the four‐state LMM with time‐homogeneous transition probabilities (model LMM4b).

Initial state probability	State—1	6‐month interval				1‐year interval
		State				State
		1	2	3	4	1	2	3	4
Younger sample
.18	1 (Suppression propensity)	—	—	—	—	.87	.03	.02	.08
.34	2 (Engagement propensity)	—	—	—	—	.16	.69	.07	.08
.41	3 (Low/Average)	—	—	—	—	.21	.11	.64	.04
.07	4 (High Diversity)	—	—	—	—	.17	.12	.02	.69
Older sample
.33	1 (Suppression Propensity)	.84	.05	.05	.06	.71	.08	.09	.12
.28	2 (Engagement Propensity)	.08	.82	.04	.06	.13	.69	.07	.12
.32	3 (Low/Average)	.05	.04	.84	.08	.09	.07	.71	.14
.09	4 (High Diversity)	.00	.01	.06	.92	.01	.02	.11	.86

Note: Transition probabilities indicate switching from the previous state (State −‐1, rows) to the current state (State, columns). Probability of staying in the same state is depicted in bold. Probabilities may not add up to 1 due to rounding. Transition probabilities for a 1‐year interval were calculated using the transition probabilities as estimated in model LMM4b for a 6‐month interval.

FIGURE 3.

Latent state probabilities for each wave in the Younger (a) and Older Samples (b).

Older sample

Table 3 shows the initial and transition probabilities for the 4‐state LMM with time‐homogeneous transition probabilities. At Wave 1, adolescents were most likely to be either in “Suppression Propensity” or “Low/Average” and least likely to be in “High Diversity”. The estimated transition probability matrix shows that ER profiles were very stable over time, with probabilities of remaining in the same ER profile ranging between .82 to .92. Given that intervals between waves were only 6 months for the Older Sample (compared to 1 year for the Younger Sample), we also calculated the Older Sample's transition probabilities for a 1‐year interval to ensure samples were comparable. Figure 3b shows the state probabilities calculated for each wave based on the estimated initial probabilities and transition probabilities. Over time, the probability of being in “High Diversity” increased, and the probability of being in other repertoires slightly decreased. Gender significantly predicted the initial state, χ ²(3) = 11.93, p = <.01, but not the transition probabilities, χ ²(12) = 11.09, p = .52. Boys were more likely than girls to be in “Suppression Propensity” at Wave 1 (i.e., .45 vs. .20). Girls were more likely than boys to be in “Engagement Propensity” at Wave 1 (i.e., .41 vs. .19). Boys and girls were similarly likely to be in “Low/Average” (i.e., .25 vs. .32) and “High Diversity” at Wave 1 (i.e., .12 vs. .07).

DISCUSSION

Given a considerable lack of longitudinal research on the normative development of adolescent ER strategy use, the goals of this study were twofold: to examine stability and change of within‐person (1) strategy use and (2) repertoire trajectories across adolescence. This study was the first to track longitudinal changes in the use of ER strategies across more than 2 years of adolescence. Most strategies examined increased linearly, though the strategies that increased differed in the Younger (i.e., Distraction, Rumination, Suppression) and Older (i.e., Reappraisal, Relaxation, Engagement) samples. Repertoires were moderately to highly stable across waves, indicating more between‐ than within‐subject variance. Nonetheless, as expected, adolescents transitioned from limited repertoire into the High Diversity repertoire as they got older, although the younger sample were more likely to transition into the Suppression Propensity repertoire.

Developmental changes in ER strategy use

The Younger Sample displayed both stability and change in ER strategy use. Distraction, Rumination, and Suppression increased from early to mid adolescence, with girls experiencing a greater increase in rumination than boys. In contrast, Reappraisal, Relaxation, and Engagement remained relatively consistent, with girls reporting more reappraisal at Wave 1 than boys. The increase in Distraction contrasts with previous cross‐sectional work that found decreases during that period (Cracco et al., 2017). Growth in Rumination, however, was consistent with previous findings (Cracco et al., 2017), as was greater growth in girls' rumination (Cracco et al., 2017; Jose & Brown, 2008). Past suppression findings have been mixed, yet our findings were consistent with two‐year longitudinal studies which found increases during this period (Gullone et al., 2010; Voon et al., 2014). Additionally, Reappraisal's stability is also consistent with past cross‐sectional and longitudinal research in early to middle adolescence (Gullone et al., 2010; Voon et al., 2014; Wang & Hawk, 2020), as was girls' reporting higher initial levels of reappraisal (Wang et al., 2015). Relaxation and Engagement are the least studied of the six ER strategies, with no prior evidence either cross‐sectionally or longitudinally across this age period. Thus, these stable trajectories from early to middle adolescence were novel.

The Older Sample displayed both stability and change in ER strategy use as well, although not in the same strategies. Reappraisal, Relaxation, and Engagement increased over 3 years from mid to late adolescence, with girls reporting higher initial levels of Engagement than boys. Distraction, Rumination, and Suppression remained consistent, with girls reporting higher initial levels of Rumination than boys. The increases, particularly in Reappraisal, a more cognitively demanding ER strategy, are consistent with some previous research (Wang et al., 2015) and may be supported by underlying cognitive capacities which increase during this period as well (Fombouchet et al., 2023; Silvers, 2022). As mentioned above, there is little to no research on Relaxation and Engagement's trajectories; therefore, these increases from middle to late adolescence were relatively novel—although the gender difference in Engagement is consistent with past work (Pascual et al., 2016; Sanchis‐Sanchis et al., 2020). Distraction's stability contradicts some cross‐sectional work which found increases after age 15 (Cracco et al., 2017). Research on Rumination in this age period is mixed; our finding of stability is consistent with one cross‐sectional study (Jose & Brown, 2008) but contradicts the other (Cracco et al., 2017). Research on suppression is also very mixed; however, this stability is consistent with some past work, which found plateaus during this period (Verzeletti et al., 2016).

Synthesizing results across both samples reveals an interesting pattern. The two samples overlap in age trajectories, such that older sample began their study at about the same age as Waves 3/4 in the younger sample's study. Thus, it is possible to cautiously infer a broader developmental pattern from ages 11 through 17 by integrating both samples' ER. It may be that Distraction, Rumination, and Suppression rises across early to mid‐adolescence and then plateaus by mid‐ to late adolescence. Correspondingly, Reappraisal, Relaxation, and Engagement appear to be stably underused in early to mid‐adolescence followed by a surge in mid‐ to late adolescence. Taken together, our findings point towards a potential rise in less cognitively and emotionally demanding (typically considered “maladaptive”) strategies from early to mid‐adolescence, followed by a shift towards more demanding strategies (typically considered “adaptive”) in mid to late adolescence. Despite some differences in individual strategy trajectories, an early “maladaptive shift” is in line with the conclusions drawn from previous cross‐sectional work which spanned a similarly broad age range (i.e., Cracco et al., 2017). There are many potential explanations for this shift, such as the sudden increase in new social and academic pressures during this age period (Fombouchet et al., 2023; Silvers, 2022), an overlap with the average age of anxiety disorder onset (Kessler et al., 2005), and puberty‐driven cognitive and emotional changes (Somerville et al., 2010). In contrast, the “adaptive shift” observed in the older sample in this study likely reflects the gradual cognitive maturation and experience navigating emotional events that occurs across adolescence, as many of the more “adaptive” strategies are considerably more challenging to deploy than the “maladaptive” strategies (e.g., Sheppes, 2020).

Developmental change in ER repertoires

Next, we examined whether ER strategy repertoires changed across adolescence, with more diverse repertoires expected to develop as youth age. Many youth experienced stability, remaining in their initial repertoire across the course of the study. In the Older Sample, girls were initially more likely to be in the Engagement Propensity repertoire and less likely to be in the Suppression Propensity repertoire than boys. We also found evidence for a developmental shift from repertoires constituting reliance on a single strategy to a repertoire of multiple strategies, as expected. This is particularly clear in the Older Sample; the High Diversity repertoire was least common initially and ended up being the most common repertoire by Wave 6. This suggests that youth who experience change in their repertoires are not simply replacing one strategy with another but rather expanding their repertoires to include different methods of regulation.

There are two caveats to this interpretation. First, Younger Sample repertoires were characterized by reliance on a single strategy, and there was a large shift toward the Suppression Propensity repertoire. Interestingly, this developmental pattern mirrors the findings for the individual strategy trajectories, with a “maladaptive shift” in the younger sample (i.e., a shift towards a repertoire characterized by reliance on a single strategy) contrasting an “adaptive shift” in the older sample (i.e., a shift towards diverse repertoires). The High Diversity repertoire increased moderately across waves in the younger sample but remained less common than repertoires reflecting reliance on one strategy. Early adolescence might be a divergence point when individual differences become amplified (Caspi & Moffitt, 1991), as some adolescents' repertoires narrowed while others expanded. Alternatively, this could reflect different maturation rates (Foulkes & Blakemore, 2018), with some adolescents maturing into more diverse repertoires earlier than others. Second, repertoires were very stable across waves, especially in the older sample. Developmentally, this stability implies that youth are forming coherent and consistent regulatory patterns by late adolescence, like other domains of competence (e.g., Masten et al., 1995). Another possibility is that repertoire formation begins before adolescence, when there may be more change year‐to‐year than in the Younger Sample.

Limitations and future directions

This study had some limitations. First, both samples were conducted in a Canadian city with limited demographic diversity (e.g., White, middle/high SES). Thus, replications are necessary to contextualize the finding's generalizability. Second, youth's participation traversed pre‐ to post‐pandemic. Although quadratic modeling indicated that this may not have affected their ER trajectories, it is important to contain inferences made from these results. Third, this study relied entirely on survey self‐reports. Therefore, we cannot disentangle self‐impressions of strategy use (i.e., frequency estimates) from true incidence. Therefore, multi‐method approaches (e.g., burst design: intensive period of sampling two or more times separated by months or years; Benson et al., 2019) are needed to assess true incidence given that correspondence between surveys and experience sampling is modest at best (Medland et al., 2020; Wylie et al., 2023). Additionally, participants reported their frequency of using each strategy, however, the strategies' regulatory success may be more important (Gross, 2015; Sheppes, 2020; Wylie et al., 2023). Disentangling attempts from success will be necessary to achieve greater understanding of adolescent ER development, especially for work that aims to study the impacts on well‐being. Lastly, the COVID‐19 pandemic occurred during the span of both studies. Although the present study did not find evidence that this had an impact on the stability or change of ER strategies, researchers should still attempt to replicate these findings outside the context of a pandemic.

CONCLUSION

Relative to childhood and adulthood, adolescence is considered a normatively emotional age period (Rosenblum & Lewis, 2006), driven by the convergence of various social, cognitive, and biological transformations (Dahl, 2004). The path to greater emotional competence is paved with myriad emotional experiences for which youth can fail and succeed when using various regulatory strategies. Over time, this practice, supported by various socialization inputs (Morris et al., 2017), may make ER repertoires more diverse, potentially providing opportunities for more effective ER. The present study aimed to document whether and how this may be the case as an early step toward developing a more explicit developmental theory of ER.

Additionally, the present study aimed to fill a neglected gap in adolescent development. The foundation of developmental psychopathology concerns deviations from normative development (Cicchetti & Toth, 2009; Masten, 2006; Rutter & Sroufe, 2000), yet almost no research delineates normative trajectories without reference to pathology. Despite typical cursory acknowledgements that ER is an ongoing and ubiquitous process, ER is usually conceptualized in relation to symptoms and dysfunction (e.g., Compas et al., 2017; Klein et al., 2022). If the adolescent experience is fundamentally emotional (e.g., “storm and stress”; Buchanan et al., 2023; Hall, 1904), then the challenge of regulation is ubiquitous as well (Hollenstein & Lougheed, 2013). By documenting patterns of ER strategy use across adolescence, this study put forth the initial steps towards a clearer understanding of how youth navigate their inevitably emotional worlds.

AUTHOR CONTRIBUTIONS

Emma Galarneau: Writing – original draft; formal analysis; visualization. Tom Hollenstein: Conceptualization; investigation; funding acquisition; writing – original draft; formal analysis; resources. Xiaomei Li: Formal analysis; writing – review and editing. Tanja Lischetzke: Formal analysis; writing – review and editing. Jessica P. Lougheed: Investigation; writing – review and editing. Kalee De France: Investigation; writing – review and editing.

FUNDING INFORMATION

This research was supported by a Social Sciences and Humanities Research Council grant (435‐2018‐0099) and a Natural Sciences and Engineering Research Council (2016‐03734).

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to report.

ETHICS STATEMENT

This research was approved by Queen's University's Health Sciences Research Ethics Board (PSYC‐181‐16) and Internal Review Board (29/09/2017; GPSYC‐817‐17).

CONSENT STATEMENT

Informed consent and assent were obtained from all participating caregivers and their adolescents, respectively.

Supporting information

Data S1.

DATA AVAILABILITY STATEMENT

Data are not available in a third‐party repository as we did not gain approval from the REBs to distribute the data publicly when this research began; however, data and analysis syntax can be provided upon request by contacting the first or last authors.