Development of temperament types from infancy to adolescence: Genetic and environmental influences with an economically and racially/ethnically diverse sample

Abstract

Kagan theorized biologically-based temperament types that are present in infancy, stable across development, and essential for understanding individual differences. Despite evidence, temperament research remains focused on a few prominent dimensions of temperament, without adequately addressing covariance among dimensions and temperament types. Using longitudinal twin data, we took a person-centered statistical approach to identify temperament types and examined continuity and change across five developmental periods (N_infancy=602; N_toddlerhood=522; N_{early childhood}=390; N_{late childhood}=718; N_{early adolescence}=700). We then examined the genetic and environmental etiology of temperament types. Twins were boys and girls (51–53% female), primarily Hispanic/Latinx (23–30%) and non-Hispanic/Latinx White (56–63%), and from socioeconomically-diverse families (28–38% near-or-below the poverty line). Using latent profile analysis, we identified three temperament types at each age characterized by negative reactivity and dysregulation, positive reactivity and strong self-regulation, and moderate reactivity and regulation. Latent transition analyses revealed considerable continuity in membership type for ‘Negative Dysregulated’ beginning in infancy (log odds = 1.58(SE=.65) to 3.16 (SE=.77), p<.01, of remaining relative to transitioning to ‘Typical Expressive’, and ‘Positive Well-Regulated’ beginning in early childhood (log odds = 1.41(SE=.56) to 2.25(SE=.47), p<.05). Twin analyses revealed moderate heritability and a consistent role of the shared environment on Positive Well-Regulated, with Negative Dysregulated and Typical Expressive also moderately heritable with the shared environment being important at some ages. Findings support the presence of theorized biologically-based temperament types that develop rapidly in infancy and toddlerhood and provide a foundation for the study of individual differences and risk and resilience processes across the lifespan.

Development of temperament types from infancy to adolescence: Genetic and environmental influences with an economically and racially/ethnically diverse sample

Temperament is the foundation of socioemotional development and adjustment throughout the lifespan and has been linked with children’s cognitions about the social and physical world and reactions to stimuli including stressful experiences, new situations, and daily functioning such as sleep and coping (Rothbart & Bates, 2007; Rothbart, 2011). Researchers have long been concerned with the etiology of temperament and its key components, with different approaches converging on the importance of reactive approach and withdrawal, sociability, activity level, positive and negative emotionality, and self-regulation (Shiner et al., 2012). Temperament can be studied by investigating these components separately, and doing so has great predictive power for adaptation and adjustment (Rothbart, 2011).

However, restricting research to a few discrete dimensions may not holistically portray the complexity of temperament (Kagan, 1994), which may be best captured by typologies encompassing behavioral, emotional, and biological traits that frequently cooccur (Kagan, 1999). Such a holistic approach is important because self-regulation, positive and negative emotionality, and reactive approach and withdrawal do not exist in isolation. Rather, they act together to shape individual responses to the environment over the course of development, directly and in interaction with each other (Rothbart & Bates, 2007). For instance, a recent study found that preschoolers high in negative emotionality were most at risk for general psychopathology when they also exhibited low self-regulation (Phillips et al., 2022). Further, the combination of negative emotionality, low positive emotionality, and low effortful control may matter more for the development of depression from late childhood to adolescence than any of those dimensions alone (van Beveren et al., 2019). However, testing three-way interactions is often difficult, and testing four- or five-way interactions is infeasible in terms of statistical power and interpretation. Accounting for complex, heterogeneous patterns among dimensions of temperament requires a different, typological approach.

In this study, we used a person-centered analytic approach to temperament, and extended current understanding of its development by assessing patterns of continuity, or the identification of and transitions between types that exhibit similar characteristics (e.g., consistently high surgency, high effortful control, and low negative emotionality), and change, or the inconsistent identification of and transitions between types that exhibit differing characteristics, in each of the types across child development, from infancy to early adolescence. The continuity and change in types in our analyses accommodate heterotypic continuity, where some dimensions of temperament are expressed differently in early childhood (e.g., tantrums as normative expressions of negative emotion), or only appear later in development (e.g., activation control as normative expression of effortful control). Additionally, we examined genetic and environmental contributions to temperament types at each developmental period in a large and diverse sample of twins.

Definition and Measurement of Temperament

Multiple definitions of temperament have been proposed, each with a distinct perspective on its fundamental elements (Shiner et al., 2012). The one most relevant to this study is that of Rothbart and colleagues, stating that temperament is characterized by constitutionally-based individual differences in behavioral, emotional, and physiological reactivity and regulation (Rothbart & Bates, 2007). Based on this perspective, Rothbart and colleagues have conducted extensive research providing evidence for three central higher-order temperament dimensions (surgency, negative emotionality, and effortful control), which can be further broken down into several fundamental lower-order dimensions (e.g., impulsivity, fear, attentional focusing; Rothbart, 2011).

Although parent-report measures contain measurement error (Rothbart & Goldsmith, 1985), like most forms of measurement, temperament dimensions are commonly assessed using parent-report questionnaires in infancy through early adolescence which has some advantages too. Unlike laboratory or home-based observers, parents have opportunities to extensively observe their child in a multitude of contexts (Bates, 1994), which is important for measuring constructs that are defined by consistency across time and contexts such as temperament. Parents are also more likely to observe rare behaviors such as highly negative reactions to stressful events (Rothbart & Bates, 2007). Although parent-reports are more subjective than lab-based observer-reports, mother- and father-reports of child temperament have been found to converge with one another in addition to more objective assessments providing evidence for validity (Bates & Bayles, 1984). Further, questionnaires can minimize participant burden and increase accessibility in intensive longitudinal studies with diverse samples by avoiding the need for travel and reducing participation time (Bates, 1994). These features are especially necessary for investigations that aim to longitudinally examine several facets of a construct, such as the current investigation on temperament.

From Temperament Dimensions to Types

The development of temperament sets the stage for socioemotional development as early as infancy, with evidence that negative emotionality, surgency, and effortful control predict responsiveness to parenting styles, later personality, and risk for psychopathology (Eisenberg et al., 2009; Rothbart & Bates, 2007). Although temperament has been widely studied, research providing evidence for its importance primarily uses a dimensional approach focused on examining one or a few individual dimensions of temperament across development and in relation to child outcomes. The dimensional approach to temperament involves examining the stability, change, and/or heritability of higher- or lower-order dimensions of temperament separately and has contributed greatly to our understanding of temperament development.

For example, early researchers proposed that a key component of a temperament trait is stability over time (Goldsmith et al., 1987). Using structural equation modeling to look at autoregressive paths of temperament in early childhood, Lemery-Chalfant and colleagues (1999) showed ample change in temperament across infancy with stability, emerging around age 3. This finding was consistent with later studies on the stability of temperament within various developmental periods which suggest that patterns of stability tend to be concordant with development (see Gagne & Goldsmith, 2020 for a review). Although early researchers considered stability key, and temperament does tend to be relatively stable, change does occur even after infancy due to both genetic and environmental reasons (Shiner et al., 2012). These characteristics of temperament were unmasked using the dimensional approach which has improved our knowledge of the depth, complexity, and structure of temperament, offering parents, teachers, and researchers an empirically-based understanding of this core aspect of child development.

Despite these strengths, dimensional research largely does not account for interrelation between multiple dimensions or allow for subgroups characterized by particular constellations of traits, and thus cannot provide a holistic understanding of temperament as it takes shape in the individual (Kagan, 1994). Addressing these weaknesses requires a different, typological approach that allows researchers to identify and investigate common patterns of temperament traits that tend to co-occur.

Early work using a typological approach to examine various combinations of temperament traits used naturalistic observation of extreme temperament traits to define different types, including Easy, Difficult, and Slow to Warm Up (Thomas et al., 1963) and Behaviorally Inhibited and Uninhibited (Kagan, 1994). In these studies, a reliance on pre-determined types based on observed extremes resulted in the exclusion of roughly 35% to 67% of individuals from categorization and analyses (Kagan, 1994; Rothbart, 2011). Kagan later expanded this research from naturalistic observation into theory and examining physiological and neurological mechanisms underlying the development of temperament types (Kagan, 1994). In this work, Kagan showed that Inhibited and Uninhibited types displayed evidence of heritability and markedly different resting heart rates and amygdala excitability, supporting his theory of distinct types of temperament (Kagan, 1997).

Recently, a more comprehensive methodology for conceptualizing temperament types has emerged in the literature. The person-centered typological approach allows researchers to categorize the unique constellations of temperament dimensions into naturally occurring types that emerge from the data using cluster analyses, such as latent profile analysis (LPA; Beekman et al., 2015). Unlike the initial typological approach employed to examine differences across extreme subgroups (those particularly high or low on one or more traits for a single higher-order dimension; Kagan et al., 1989), the person-centered approach does not require a priori knowledge of specific subgroups, and instead concentrates on identifying groups of individuals who share unique patterns of temperament dimensions (Nylund-Gibson et al., 2007). This approach can also identify small unobserved groups, which are often masked when using the variable-centered approach (Nylund-Gibson et al., 2023).

While a dimensional approach is powerful for prediction and improving the reliability and validity of measurement by analyzing the covariation among variables (Laursen & Hoff, 2006), most lower-order dimensions of temperament are highly correlated with other components in the same broad dimension, but much less correlated with lower-order components across domains (Rothbart, 1981). Methods used in person-centered approaches such as LPA can include indicator variables with both low and high intercorrelations. Indeed, the person-centered approach provides the added advantage of allowing low-correlated variables to define types as opposed to excluding them (Tein et al., 2013). Person-centered methods also do not assume homogeneity among respondents, allowing unobserved heterogenous groups to emerge based on similar high, low, and moderate patterns on a set of temperament dimensions, where the dimensional approach may fall short (Laursen & Hoff, 2006). In sum, the person-centered approach can complement the dimensional approach by modeling the heterogeneity of temperament in the population. Identifying specific patterns of temperament may help identify more specific mechanisms of risk and resilience (Kagan, 1994).

Temperament Types

Only a few studies have used a person-centered approach to examine temperament types developmentally in childhood. Across this research, at least three types consistently emerge throughout development, referred to here using the naming conventions of Beekman et al. (2015): Negative Reactive, Positive Reactive, and Typical Expressive. Relative to other types, Negative Reactive is characterized by the highest anger, fear, and sometimes activity level in infancy, coinciding with the lowest attention, pleasure, and regulation (Beekman et al., 2015; Lin et al., 2021). This type also shows high anger and low effortful control in childhood (Janson & Mathiesen, 2008; Scott et al., 2016). In contrast, the Positive Reactive type in infancy is typically characterized by the highest duration of orienting (a precursor to attentional focusing), smiling and laughter, falling reactivity, and both high and low intensity pleasure, coinciding with the lowest distress to limitations/anger and fear (Beekman et al., 2015; Lin et al. 2021). After infancy, this type shows higher effortful control and lower anger (Janson & Mathiesen, 2008; Scott et al., 2016). In general, Typically-Expressive infants and children show moderate levels of dimensions related to negative reactivity, positive reactivity, and regulation. When Surgent types emerge, they may be similar to Positive Reactive in positive affect and differentiated by lower regulation and higher approach (e.g., Scott et al., 2016).

Patterns of Change in Temperament Types

Few studies using a person-centered typological approach have examined continuity of temperament types. Beekman et al. (2015) and van den Akker et al. (2010) evaluated continuity and change in temperament types in infancy and toddlerhood, respectively. In infancy, continuity varied by type, ranging from 37% to 78%, with infants in Negative Reactive and Positive Reactive types showing the greatest continuity (Beekman et al., 2015). Consistent with findings from prior dimensional research showing temperament trait continuity increasing with age (Rothbart & Bates, 2007), temperament types in toddlerhood, assessed using the same measure over time, were more stable, with stability rates between 68% to 75% from 30–42 months (van den Akker et al., 2010). Only one study has examined continuity in type across childhood. Using parent-reported temperament from 18 months to 9 years, Janson and Mathiesen (2008) clustered their participants into five types across childhood with a subset of their sample and classified the remainder into the appropriate types. Of the five, Confident, Uneasy, and Unremarkable were most conceptually similar to the Positive Reactive, Negative Reactive, and Typical Expressive types described by Beekman and colleagues (2015). The Undercontrolled type was characterized by high sociability, activity, and emotionality with lower shyness, while Inhibited was characterized by the highest shyness (Janson & Mathiesen, 2008). Cross-tabulations showed that type membership was moderately consistent across age (33% to 46%). Thus, studies to date show both continuity and change, and more research is needed to document differences by type and developmental period.

Genetic and Environmental Influences on Temperament

Twin studies using a dimensional approach support the idea of temperament as at least partially genetically-influenced. These studies help disentangle genetic and environmental factors by using differences in the resemblance of monozygotic (MZ; identical) twins relative to dizygotic (DZ; fraternal) twins to statistically parse the variance in a trait into genetic influences (which decrease the similarity of DZ but not MZ twins), shared environmental influences (which make cotwins more similar regardless of genetic relatedness; e.g., socioeconomic status and family environment), and nonshared environmental influences (which contribute to differences between twins regardless of genetic relatedness; e.g., social relationships and extracurricular activities).

The heritability of temperament (i.e., the proportion of variance in a trait within a sample broadly attributable to genetic influences, including some gene-environment interplay) tends to range from 20%-60% (Saudino, 2009), with some dimensions like effortful control showing high heritability and others, like positive emotionality in childhood, more influenced by the shared environment (Goldsmith et al., 1997). For example, a twin study evaluating lower order facets of negative emotionality (fear, anger, and sadness) in middle childhood found that all three shared genetic influences, but anger and sadness also had genetic and shared environmental influences that did not overlap with fear (Clifford et al., 2015). In another study of 310 same-sex twin pairs assessed three times from ages 3 years to 5 years, Liu and colleagues (2023) found that genetic factors contributed to both initial negative affectivity and effortful control and their growth over time, while initial levels and growth in surgency were mainly due to the nonshared environment.

Only two studies using a person-centered approach have also examined genetic and environmental influences on temperament type, using different measures at different ages. Planalp and Goldsmith (2020) used latent profile analysis to examine microanalytic coding of infants’ observed temperament at 6 and 12 months of age. The nonshared environment (which often includes daily fluctuations likely to affect observed behavior) was the strongest contributor to all types in infancy, and the sole contributor to the 6-month Positive/Active and 12-month Low Negative and Low Reactive types. Findings also supported modest genetic influences on 6 month Typical and Withdrawn/Inhibited types and moderate shared environmental influences on 12 month Withdrawn/Inhibited temperament. Microanalytic coding of temperament yields higher estimates of the nonshared environment, but genetic and environmental influences are detectable and shared across observed dimensions (Clifford et al., 2015).

Using a factor mixture modeling approach in middle childhood, Scott and colleagues (2016) examined mean composites of mother- and father-report of temperament to estimate temperament profiles while also modeling dimensional variation within profile. Heritability varied by type, but findings otherwise differed from those in infancy based on coded observations, with minimal nonshared environmental contributions and at least moderate heritability for all types. The Positive Well-Regulated and Surgent types were most highly heritable, the Typical Expressive type was least heritable and primarily influenced by the shared environment, and the Negative Dysregulated type showed moderate genetic and shared environmental influences. The dimensional approach suggests that positive emotionality is significantly influenced by the shared environment with low or nonsignificant heritability (Goldsmith et al., 1997). Yet, the typological approach suggests higher heritability in middle childhood for types that include high positive emotionality. These types, such as Positive Well-Regulated and Surgent types, may be more heritable because they also include effortful control (Scott et al., 2016). However, no studies to date have modeled genetic and environmental influences on temperament types at multiple developmental periods. Given the differences between the two extant studies, there is a need to examine the genetic and environmental contributions across developmental periods in the same sample.

The Present Study

The present study aimed to expand our understanding of temperament development in several ways. First, we used a person-centered approach with latent profile analysis to determine temperament types in infancy, toddlerhood, early childhood, late childhood, and early adolescence. Based on prior literature, we hypothesized that there would be three to five types at each developmental period. These types would minimally include a Negative Dysregulated type with average to low surgency and effortful control and high on negative emotionality, a Positive Well-Regulated type with high positive emotionality and effortful control and low on negative emotionality, and a Typical Expressive type with above average surgency and average effortful control and negative emotionality. Although we expect to identify such types at each age, we do not expect them to be directly comparable to one another given developmental changes in the expression of temperament across development and corresponding adjustments made to temperament assessments (Rothbart, 2011).

Second, we investigated patterns of continuity in type membership from infancy to adolescence with latent transition analysis. We hypothesized greater continuity in more extreme types, or types that include the highest and lowest scores on underlying temperament dimensions. For example, children classified in a type with the lowest shyness and highest anger will tend to be classified in similar types across developmental periods, whereas children in a type with average shyness and average anger will show more discontinuity. We also hypothesize increasing continuity overall as children age, with little change after early childhood. Finally, using the twin design, we estimated genetic, shared environmental, and nonshared environmental contributions to the probability of profile membership for each developmental period. We hypothesized that all types would be moderately heritable.

Methods

Participants

The Arizona Twin Project is an ongoing longitudinal study of twins born across the state of Arizona, with data collected at seven waves: baseline at twin age 12 months (Infancy; M_age=12.60, SD=1.19) and follow-ups through early adolescence at 30 months (Toddlerhood; M_age= 2.66 years, SD=0.25), 5 years (Early Childhood; M_age=5.18, SD=0.27), 7–9 years (Late Childhood; M_age=8.43 years, SD=0.68; 2016–2019), 8–10 years (Late Childhood; M_age=9.72 years, SD=0.94; 2017–2020), 9–11 years (Early Adolescence; M_age=10.88 years, SD=1.13; 2018–2021), and 10–12 years (Early Adolescence; M_age=11.81 years, SD=1.13; 2019–2022). Families were initially recruited by the Arizona State Department of Health Services using state birth records between July 2007 and July 2008, for an initial sample of 602 twins from 304 families (52.50% female; 28.33% monozygotic [MZ], 37.17% same-sex dizygotic [ssDZ], and 34.50% other sex DZ [osDZ]). Because Arizona is a closed record state, researchers do not have data on number of families contacted to yield the base infancy sample. Families were re-contacted to request continued participation in the study at twin age 30 months (522 twins, 262 families; M_age=31.89 months, SD=2.96; 2010–2011), and a limited subsample of 390 twins from 197 families were also contacted at five years (M_age=5.18 years, SD=.27; 2013) when twins were transitioning to kindergarten. More information on retention rates for the early childhood sample can be found in earlier publications (Lemery-Chalfant et al., 2013; Lemery-Chalfant et al., 2019). When the twins were 7–9 years of age, the initial sample was re-contacted, and new families from the same birth cohort were recruited from parents of twins’ groups and online postings. Specifically, the sample in late childhood to early adolescence comprised 700 twins at ages 7–9 (352 families), 798 twins at ages 8–10 (400 families), 780 twins at ages 9–11 (391 families), and 718 twins at ages 10–12 (359 families). Supplemental Table S1 shows the number of families participating in any assessment at a given wave, split according to the wave in which they were recruited.

Twenty-two twins from 14 families not included in the numbers reported above were excluded from the study sample due to developmental or cognitive disabilities that interfered with their ability to complete study procedures. There were no other exclusions.

There was no pre-designated endpoint for participant recruitment, but our goal was to include a minimum of 300 participants in each wave of data collection in order to have adequate power for multivariate twin analyses not conducted in this manuscript. A prior simulation study suggests that a sample of 300 is sufficient for latent transition analyses (Nylund-Gibson et al., 2007).

To examine temperament by developmental period rather than wave, we combined ages 7–9 and 8–10 into a Late Childhood wave including twins between 7.00 and 9.99 years (N=718, M_age=8.51 years, SD=.71), and ages 9–11 and 10–12 into an Early Adolescent wave including twins 10.00 years and older (N=700, M_age=11.10 years, SD=1.04). Data were taken from the earliest wave when the child fell within the specified age range and had temperament data, resulting in a Late Childhood group primarily drawn from ages 7–9 (N=641, 89.28%) and an Early Adolescen group primarily drawn from ages 9–11 (N=662, 94.57%). Demographics and attrition analyses are reported for these groups, rather than by the four waves. Table S1 shows the number of twins with temperament data at each wave, and the overlap between waves, with higher sample sizes in late childhood and early adolescence due to recruitment efforts.

Across all waves, twins were approximately evenly split on sex assigned at birth (50.70–53.45% female) and primarily Hispanic/Latinx (23.17% at 30 months to 30.46% in early adolescence) and non-Hispanic/Latinx White (56.30% at 12 months to 62.74% at 30 months), with the remainder being Asian American (2.73–6.77%), Black or African American (3.72–5.88%), Native American (1.04–2.88%), or multiracial or other (1.14–3.67%). Families were socioeconomically diverse, with a substantial proportion having income-to-needs ratios (United States Census Bureau, 2023) below (6.52–13.36%) or near (21.38–24.22%) the poverty line, and the rest categorized as lower middle (15.59–22.83%), middle (16.30–20.00%) and upper middle-to-upper class (20.67–33.22%).

Attrition analyses

For attrition analyses, differences in family level demographic variables (e.g., socioeconomic status) were analyzed using independent samples t-tests, while person level variables (e.g., temperament dimensions) were analyzed using mixed modeling to account for cotwin interdependence. Relative to families participating only at 12 months, independent samples t-tests revealed that a higher proportion of families participating at 30 months, t(298)=−4.20, p<.001, and 5 years, t(298)=−2.41, p=.017, were non-Hispanic White and a lower proportion were Hispanic, t(298)=3.09, p=.002. Families participating at any follow-up wave had higher income (p < .05), but no other family-level demographic differences. Mixed models showed lower 12-month negative emotionality for twins participating at 30 months (b=−.13, SE=.06, p=.040), and lower 12-month self-regulation (duration of orienting and low intensity pleasure) for twins participating in late childhood (b=−.13, SE=.07, p=.05) and early adolescence (b=−.13, SE=.07, p=.05), but no other differences in 12 month surgency (low shyness and high intensity pleasure), regulation, or negative emotionality, and no significant differences in proportion female at any wave (ps > .05).

Between adjacent waves, twins participating in Late Childhood but not Early Adolescence had lower effortful control (b=.14, SE=.06, p=.014), with no other significant differences in temperament or sex, proportion White or Hispanic, or income across adjacent waves (ps > .05). Relative to the original birth cohort, the newly recruited sample had fewer non-Hispanic/Latinx White (t(235.65)=−2.64, p=.009) and more Hispanic/Latinx families (t(246.23)=−2.11, p=.036) during Late Childhood but not Early Adolescence, with no significant differences in income, sex, or temperament (ps > .05). The most common reasons for attrition were residential mobility/loss of contact information and families being too busy to participate at a given wave.

Covid-19 Pandemic

The onset of the COVID-19 pandemic occurred during Late childhood and Early Adolescence, resulting in 162 twins (21.89%) at ages 9–11 and 342 twins (51.04%) at ages 10–12 and their families participating virtually after quarantine was declared in the state of Arizona. There were no significant differences on any demographic variable or temperament scale based on whether the family participated in these waves prior to or during the pandemic.

Procedures

Primary caregivers (>93% mothers) completed temperament questionnaires online in Infancy, Toddlerhood, Early Childhood, and the latter part of Early Adolescence, and during 2–3 hour home visits in Late Childhood and the first part of Early Adolescence (see Lemery-Chalfant et al., 2013; Lemery-Chalfant et al., 2019 for details). We requested that the same person complete the primary caregiver-report measures at each wave to the degree possible, but a small number of families did have a different primary caregiver during at least one wave. Given that data from the four middle to late childhood assessments were consolidated into two age-based waves, 510 of the 541 families included in current analyses (94.27%) had a consistent primary caregiver across all waves of data used in the study. For a further 23 families (4.25%), a different primary caregiver reported on twins’ temperament at a single wave, and for a final eight families (1.48%), the primary caregiver switched once during the study, with each primary caregiver completing multiple consecutive waves of data collection. Institutional Review Board approval was obtained, including written informed consent from primary caregivers and verbal assent from children. Families were compensated for all components of the study.

Measures

Zygosity

Zygosity was assessed using the 32-item caregiver-report Zygosity Questionnaire for Young Twins, a valid alternative to genotyping with accuracy ranging between 93% and 98% in differentiating MZ and DZ twins (Forget-Dubois et al., 2003; Goldsmith, 1991). When zygosity was difficult to determine from questionnaire, we used expert ratings based on photos and videos, hospital birth records and lab placentae reports, and genotyping.

Temperament

We used age-appropriate measures of temperament, including selected scales from the Infant Behavior Questionnaire-Revised (IBQ-R; Gartstein & Rothbart, 2003), Child Behavior Questionnaire-Short Form (CBQ-SF; Putnam & Rothbart, 2006), Temperament in Middle Childhood Questionnaire (TMCQ; Simonds, 2006), and Early Adolescent Temperament Questionnaire (EATQ; Ellis & Rothbart, 2001). These measures were consistently developed by Rothbart and her colleagues using the same theoretical framework, meticulously matching items across the questionnaires within the boundaries of each developmental period, and they show good construct and convergent validity. All temperament questionnaires were completed by primary caregivers two separate times, once for each twin. Internal consistency reliability was calculated for each of the scales within the temperament questionnaires using Cronbach’s alpha listed (see Table S3).

IBQ-R.

The IBQ-R, asked in Infancy, included items such as, “protests when being placed in a confining place (infant seat, play pen, car seat, etc.),” “seems content when left in the crib,” and “plays with one toy or object for 5–10 minutes,” where primary caregivers responded using a Likert scale from 1 (Never) to 5 (Always).

CBQ-SF.

The CBQ-SF, administered in Toddlerhood and Early Childhood, asked primary caregivers to respond to items such as “gets angry when called away from play before he/she is ready to quit,” “is among the last children to try out a new activity,” and “is very difficult to soothe when he/she has become upset,” on a Likert scale from 1 (extremely untrue) to 7 (extremely true) when considering the child’s behavior over the past six months.

TMCQ.

The TMCQ, administered in Late Childhood, involved primary caregivers responding on a Likert scale from 1 (almost always untrue) to 5 (almost always true) on items such as “likes to be physically active,” “interrupts others when they are talking,” “tends to become sad if plans don’t work out,” when considering the child’s behavior in the past six months. Activation control and sadness were included only during the 7–9 year wave. A subset of scales (activity level, impulsivity, sadness, and soothability) from the TMCQ were also used in combination with the EATQ during the early adolescent waves because the items from these scales were still developmentally appropriate and the EATQ includes no equivalent scales.

EATQ.

The EATQ, administered in Early Adolescence, asked primary caregivers to respond on a Likert scale from 1 (almost always untrue) to 5 (almost always true) when considering the child’s behavior over the past six months. Sample items include, “calls out answers before being called on by a teacher,” “pays close attention when someone tells them how to do something,” and “is nervous being home alone.”

Although item and scale content changed, Rothbart’s three higher-order dimensions were represented at each wave. A breakdown of how each scale corresponds to each higher-order dimension of temperament, along with the internal consistency reliability results for each scale, is presented in Table S3 of the Supplementary Material. Most lower-order dimensions were included at all ages (i.e., pleasure, attentional focusing, soothability, and anger), but some were not included on the questionnaires until Toddlerhood (i.e., shyness, impulsivity and inhibitory control) or Late Childhood (i.e., activation control). Additionally, activity level and sadness were included in the study starting in Late Childhood due to increased grant funding.

Analytic Plan

To address missing data, we used full information maximum likelihood (FIML; Enders, 2022), in Mplus Version 8.8 (Muthén & Muthén, 1998–2017) for all phenotypic analyses, and corrected for twin dependencies using type=complex. Age (included as a covariate in LTAs) had no missing data, and only one twin pair was missing data on sex (included as a covariate in twin analyses).

Analyses for Aim 1: Latent Profile Analysis

To examine possible temperament types across childhood, we conducted a series of latent profile analyses (LPA) using data from Infancy, Toddlerhood, Early Childhood, Late Childhood, and Early Adolescence. There is minimal guidance in conducting power analyses for LPA, but research suggests a sample size of at least 300 (Nylund-Gibson et al., 2007), though sample size may contribute less to power than number of class indicators (Tein et al., 2013). For each LPA, we have between 6 and 10 class indicators. To avoid local maxima, we used a minimum of 1000 random starts and 200 final stage optimizations, and if needed increased this number until the best log likelihood was replicated several times. We used the Akaike Information Criteria (AIC), sample-size adjusted Bayesian Information Criteria (saBIC), Adjusted Lo-Mendell-Rubin (LMR) significance test, entropy, lowest class proportion, and theoretical interpretation for model selection (Spurk et al., 2020). After selecting the best profile solution, we examined the estimated means for each dimension in each profile to label the temperament type.

Analyses for Aim 2: Latent Transition Analysis

We used Latent Transition Analysis (LTA) to capture types across development with no assumption of measurement equivalence (Nylund-Gibson et al., 2023). After identifying temperament types at each age, we conducted four LTAs across each pair of adjacent ages (i.e., Infancy to Toddlerhood, Toddlerhood to Early Childhood, Early to Late childhood, Late Childhood to Early Adolescence) to assess transitions between types, controlling for the association between age and profile membership within wave. We used a minimum of 2,000 random starts and 100 optimizations to avoid local maxima and used the best class solutions from LPAs as a guide to identify the number of profiles at each wave and provide start values for the means and variances of indicators. Although we selected the number of profiles based on LPA results rather than testing competing models, we report the log-likelihood, AIC, saBIC, and entropy. We also report the estimated log odds of latent class transitions provided by the multinomial logistic regression output for each LTA. As a post-hoc exploratory analysis, we attempted to expand the LTA to three time-points, but the model would not converge due to a large number of parameters. Thus, we created a contingency table for the three waves from Infancy to Early Childhood in SPSS, Statistical Package for Social Sciences, Version 28.0 (with listwise deletion), to examine longer-term continuity and to demonstrate the complexity that arises as more ages are considered and possible transition patterns exponentially increase (e.g., 3 waves of data yielded 27 types).

Analyses for Aim 3: Heritability

Monozygotic (MZ) twins share 100% of their segregating DNA, deoxyribonucleic acid, whereas Dizygotic (DZ) twins share on average 50%. Thus, stronger cross-twin covariance for MZ relative to DZ twins reared in the same environment suggests a role for genetic influences. The quantitative genetic ACE model is a multigroup structural equation model that uses these differences in phenotypic similarity to estimate the variance in one or more traits attributable to broad, latent additive genetic (A; fixed to correlate 1.0 for MZ pairs and .50 for DZ pairs), shared environmental (C; fixed to correlate 1.0 for both MZ and DZ pairs), and nonshared environmental (E; uncorrelated between cotwins) components. A DZ cross-twin correlation less than half the MZ correlation suggests the influence of dominant (D) or other non-additive genetic influences, which are fully shared between MZ twins but correlated only .25 on average between DZ twins. In this case, an alternate ADE model estimating both additive and non-additive genetic influences can be fit. However, because they draw on the same source of information (the difference between MZ and DZ covariance), C and D cannot be included in the same model when the sample only includes twin pairs and not parents or other relatives, for model identification reasons.

We used the structural equation modeling software OpenMx (Boker et al., 2011) to fit univariate ACE models (Neale & Cardon, 1992) estimating genetic and environmental influences on the extracted posterior probability of profile membership for each type in each developmental period. Model results were independent across different developmental periods. We used posterior probabilities rather than most likely profiles because the liability threshold model yields biased estimates of heritability at our sample size even when the assumption of an underlying normal liability distribution is met (Benchek & Morris, 2013), and modeling profile membership as a distribution rather than a fixed class better accounts for uncertainty in profile assignment (Ferguson et al., 2020). The significance of A and C or D parameters can be tested by dropping them from the model and using the likelihood ratio chi-square test, AIC, and saBIC to compare the fit of full and reduced models. However, the A parameter is never dropped in the presence of D, because it is theoretically unlikely that all genetic influences on a trait are dominant, and E is always retained because it includes measurement error. In all genetic models, we included sex (male=0, female=1) as a covariate allowed to influence the mean probability of profile membership.

Transparency and Openness

We report the sample size and how it was determined, all exclusions, attrition, and all manipulations, measures, and analyses. We follow Journal Article Reporting Standards (JARS) guidelines (Kazak, 2018). Data were analyzed using Mplus Version 8.8 (Muthén & Muthén, 1998–2017) and the R packages OpenMx Version 2.21.1 (Boker et al., 2011) and ggplot2 Version 3.4.0 (Wickham, 2016) in R Version 4.2.2. The study was not preregistered.

Sample scripts to conduct latent profile analyses, latent transition analyses, and twin ACE models are available in the online supplementary material. Deidentified data are available from study principal investigators upon reasonable request. The questionnaires used in this study are not ours to disseminate, but can be obtained free of charge by completing a request form on Dr. Rothbart’s website [https://research.bowdoin.edu/rothbart-temperament-questionnaires/request-forms/], or contacting Dr. Putnam over email [sputnam@bowdoin.edu] or postal mail [Department of Psychology, Bowdoin College, 6900 College Station, Brunswick, ME 04011].

Results

Descriptive Statistics

Descriptive statistics and bivariate correlations for all temperament dimensions at each age are summarized in Supplemental Table S2. Overall, most correlations between dimensions correspond to what we would expect given the extensive literature on the hierarchical structure of temperament, and longitudinal stability. No temperament scale at any age had skewness or kurtosis that exceeded recommended cutoffs for skewness (+/−2.00) or kurtosis (+/−7.00; Muthén & Kaplan, 1985). No transformations were applied to any of the data and no data were excluded.

Latent Profile Analysis Model Selection

Model fit indices, entropy, and profile membership percentages at each age are presented in Tables 1 and 2. All models converged without estimation problems. The AIC, saBIC and LMR p-value did not agree on the optimum number of profiles at any age. Because we aimed to differentiate theoretically meaningful profiles from undifferentiated profiles, we selected the 3-profile over the 2-profile solution at each age, after Wald Z-tests confirmed all profiles were significantly different in key indicator means (see Figure 1), because the former was supported by better class separation, adequate percent membership in each profile, and theoretically-meaningful structure.

Table 1.

Latent profile analysis model selection summary table – Infancy to Toddlerhood

Age	Profiles	No. of Parameters	Entropy	Log-likelihood	AIC	saBIC	LMR p-value	% Profile 1	% Profile 2	% Profile 3	% Profile 4	% Profile 5
Infancy	1	12		−2823.24	5670.48	5684.41		100%
	2	25	.65	−2674.04	5398.08	5427.10	.07	40.74	59.26
	3	38	.70	−2570.53	5217.06	5261.16	.09	31.51	24.03	44.45
	4	51	.76	−2518.44	5138.88	5198.07	.16	31.85	1.06	25.70	41.39
	5	64	.77	−2473.55	5075.10	5149.38	.71	38.00	1.06	12.83	23.91	24.19
Toddlerhood	1	14		−5117.42	10262.42	10277.12		100%
	2	29	.68	−4953.46	9964.91	9995.38	.01	59.79	40.21
	3	44	.72	−4852.89	9793.79	9840.01	.06	43.21	26.98	32.57
	4	59	.71	−4793.31	9704.62	9766.60	.74	26.65	31.12	15.78	26.45
	5	74	.74	−4751.07	9650.13	9727.86	.24	31.90	19.49	8.40	24.77	21.80

Table 2.

Latent profile analysis model selection summary table – Early Childhood to Early Adolescence

Age	Profiles	No. Parameters	Entropy	Log-likelihood	AIC	saBIC	LMR p−value	% Profile 1	% Profile 2	% Profile 3	% Profile 4	% Profile 5
Early Childhood	1	14		−3863.00	7754.00	7764.74		100%
	2	29	.83	−3643.30	7344.59	7366.84	.11	70.54	29.46
	3	44	.80	−3557.81	7203.63	7237.39	.01	41.73	41.93	16.34
	4	59	.78	−3492.27	7102.53	7147.81	.59	36.47	30.79	16.28	16.47
	5	74	.82	−3455.24	7058.48	7115.27	.32	29.46	10.63	7.53	16.42	35.96
Late childhood	1	20		−7206.14	14452.29	11480.37		100%
	2	41	.83	−6522.66	13127.32	13184.89	<.01	43.93	56.07
	3	62	.83	−6228.08	12580.16	12667.21	<.01	27.44	37.72	34.84
	4	83	.85	−6004.00	12174.00	12290.53	<.01	23.91	35.82	20.72	19.55
	5	104	.85	−5909.82	12027.64	12173.66	.30	23.11	16.01	32.20	20.40	8.28
Early Adolescence	1	22		−8188.40	16420.80	16451.06		100%
	2	45	.84	−7447.18	14984.35	15046.27	<.01	44.24	55.76
	3	68	.84	−7171.45	14478.90	14572.46	<.01	35.75	24.70	39.55
	4	91	.84	−7021.21	14224.41	14349.62	.12	19.32	22.81	34.53	23.33
	5	114	.85	−6891.25	14010.49	14167.34	.62	13.56	22.28	19.11	31.54	13.52

Table 1a-1b Note. Bold: “best-fitting” model for that age period based on log-likelihood, saBIC, LMR p−values, entropy, and class proportions. saBIC = sample size adjusted Bayesian Information criteria; LMR = Lo-Mendell-Rubin likelihood ratio. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Figure 1. Profile plots of temperament types.

Note. IBQ-R – Infant Behavior Questionnaire. CBQ – Childhood Behavior Questionnaire. TMCQ – Temperament in Middle Childhood Questionnaire. EATQ – Early Adolescent Temperament Questionnaire

Temperament Types in Infancy

In Infancy, while the LMR p-value was not significant (p > .05) for any model, the AIC and saBIC steadily decreased until the 3-profile model where they began to stabilize (see Table 1). The 3-profile solution did not have the best entropy, but classification probabilities for most likely latent class membership in the 3-profile model exceeded .80 for all profiles. Thus, we chose the 3-profile solution.

Figure 1 shows the profile plot of the means for each class. The Negative Dysregulated type showed the lowest soothability and the highest anger. Positive Well-Regulated had the highest duration of orienting (a precursor to attentional focusing) and high- and low-intensity pleasure. The third type, Typical Low Expressive, had the lowest distress to novelty (precursor to shyness) and was similar to Negative Dysregulated in low levels of duration of orienting and both low- and high-intensity pleasure, and similar to Positive Well-Regulated in high soothability and low anger.

Temperament Types in Toddlerhood

In Toddlerhood, the AIC and saBIC steadily decreased until the 3-profile model where they began to stabilize, supporting the 3-profile model (Table 1). The LMR p-value was only significant for the 2-profile model, but unlike the 3-profile model, entropy was insufficient. All profiles in the 3-profile model (see Figure 1) had classification probabilities for most likely latent class above .80, profile membership percentages >25%, and significant differences from one another on at least three dimensions. As in infancy, a Negative Dysregulated type emerged, uniquely characterized by the lowest pleasure and soothability, although another well-regulated type emerged in toddlerhood characterized by the highest attentional focusing, inhibitory control and soothability with the lowest anger. Additionally, this type had high shyness and low impulsivity, similar to Negative Dysregulated, thus, we termed this type Reserved Well-Regulated. The final type, the Surgent type, had the lowest shyness, the highest impulsivity, moderate soothability that differed significantly from other types, high pleasure (similar to the Reserved Well-Regulated type) and low attentional focusing and inhibitory control (similar to the Negative Dysregulated type).

Temperament Types in Early Childhood

In Early Childhood, the AIC and saBIC steadily decreased until the 3-profile solution, which was the only solution with a statistically significant LMR p-value, supporting this solution as the best model (Table 2). Entropy was highest for the 2-profile solution, but classification probabilities for most likely latent class in the 3-profile solution were all above .89. As before, we selected the 3-profile solution (Figure 1) based on the theoretical value of more clearly differentiating the two most extreme classes. Again, we found a Negative Dysregulated type with the lowest pleasure, attentional focusing, inhibitory control, and soothability and the highest anger, and a Positive Well-Regulated type with the highest pleasure, attentional focusing, inhibitory control, and soothability and lowest anger. Finally, a Typically Expressive type emerged. This type fell between the other two profiles on all indicators except shyness (similar to Positive Well-Regulated) and impulsivity (similar to Negative Dysregulated), but Wald Z-tests indicated significant differences in indicator means between Typical Expressive and the other two types (See Figure 1).

Temperament Types in Late Childhood

Unlike the previous developmental periods, the saBIC and LMR p-values yielded non-adjacent model suggestions for the best-fitting model in Late Childhood (Table 2). The AIC and saBIC steadily decreased across all of the models, but the LMR suggested that the 4-profile solution fit better than the smaller profile solutions and the 5-profile solution. Although we used the 3-profile model in late childhood for the sake of consistency with other developmental periods, we describe the 2- and 4-profile solutions in S8, both of which are supported by the data.

In the 3-profile solution, we found Negative Dysregulated and Positive Well-Regulated types similar but not identical to those in early childhood. As before, Negative Dysregulated was low and Positive Well-Regulated high on soothability and dimensions tapping effortful control, whereas impulsivity, anger, and sadness were highest for Negative Dysregulated and lowest for Positive Well-Regulated. However, both Negative Dysregulated and Positive Well-Regulated children had high activity level and pleasure. Finally, we found a third Typical Expressive profile, characterized at this age by moderate attentional focusing, inhibitory control, and anger, but also the lowest activity level and pleasure, the highest shyness, and similar activation control, soothability, and sadness to the Negative Dysregulated type.

Temperament Types in Early Adolescence

In Early Adolescence, while the AIC and saBIC steadily decreased across all models, the LMR p-value supported three profiles (Table 2). Thus, we selected the 3-profile solution. As at previous ages, we found a Negative Dysregulated type (high on impulsivity and negative emotionality, low on effortful control and soothability), a Positive Well-Regulated type (high on activity level, effortful control, and soothability, low on shyness, impulsivity, and negative emotionality), and a Typical Expressive type (generally moderate levels of indicators other than shyness and fear which were similar to Negative Dysregulated).

Latent Transition Analysis: Patterns of Change in Temperament Types

Relying on LPA results to guide model choice at each age, we tested LTAs with the 3-profile solutions at each pair of adjacent ages, with log-likelihood, AIC, saBIC, and entropy reported in Supplemental Table S4 and transition probabilities in Table 3. All models converged without estimation problems, but start values from LPAs were needed to achieve stable best log likelihood values. Entropy and class differentiation were lower for LTAs than LPAs, likely due to uncertainty introduced by missing data estimation, but profile structure at each age was largely unchanged.

Table 3.

Latent transition probabilities

	Toddlerhood				Early Childhood
Infancy	Negative Dysregulated	Reserved Well-Regulated	Surgent	Toddlerhood	Negative Dysregulated	Positive Well-Regulated	Typical Expressive
Negative Dysregulated	.73	.16	.11	Negative Dysregulated	.66	.09	.25
Positive Well-Regulated	.21	.35	.44	Reserved Well-Regulated	.21	.79	0
Typical Low-Expressive	.19	.47	.34	Surgent	.09	.21	.70
	Late Childhood				Early Adolescence
Early Childhood	Negative Dysregulated	Positive Well-Regulated	Typical Expressive	Late Childhood	Negative Dysregulated	Positive Well-Regulated	Typical Expressive
Negative Dysregulated	.72	.01	.27	Negative Dysregulated	.85	.04	.11
Positive Well-Regulated	.05	.72	.49	Positive Well-Regulated	.04	.63	.33
Typical Expressive	.23	.28	.49	Typical Expressive	.25	.12	.63

Note. Latent transition probabilities are based on the estimated model. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

In the LTA from Infancy to Toddlerhood, average probabilities for most likely latent class ranged from .78 (Positive Well-Regulated in infancy and Negative Dysregulated in toddlerhood) to .84 (Negative Dysregulated in infancy). Latent transition probabilities based on the estimated model (Table 3) and multinomial logistic regression indicated that Negative Dysregulated infants were more likely to remain in this type in Toddlerhood (log odds = 2.48, SE = 1.01, p = .014), with 27.22% of the sample estimated to fall into this transition pattern according to their most likely latent class, relative to 5.98% estimated to transition to Reserved Well-Regulated and 4.01% estimated to transition to Surgent. Negative Dysregulated infants were not more likely to transition into Reserved Well-Regulated than Surgent (log odds =0.06, SE = 1.42, p = .968). Positive Well-Regulated infants appeared most likely to transition into the Surgent type (9.60% of the sample), but had no significant difference in log odds of transitioning to Reserved Well-Regulated (7.51% of the sample) or Negative Dysregulated (4.65% of the sample) compared to the Surgent type (log odds = −0.55, SE = 0.60, p = .352; log odds = −0.19, SE = 1.01, p = .852).

In the LTA from Toddlerhood to Early Childhood, the average probabilities for most likely latent class ranged from .77 (Negative Dysregulated in Early Childhood) to .87 (Reserved Well-Regulated). Latent transition probabilities (Table 3) again suggested continuity in Negative Dysregulated, with 28.73% of the sample estimated to belong to this type at both ages according to most likely latent class. Unlike the previous latent transitions, only one person was estimated to have the most likely transition from Reserved Well-Regulated in toddlerhood to Typical Expressive in Early Childhood. As a result, we were unable to use Typical Expressive as a reference group for Early Childhood in the multinomial logistic regression. Thus, the reference groups in this analysis are Surgent in toddlerhood and Negative Dysregulated in Early Childhood.

Although a substantial group was predicted to transition from Negative Dysregulated to Typical Expressive (10.88%), Negative Dysregulated toddlers were less likely to transition to Typical Expressive (log odds = −3.16, SE = 0.77, p < .001) or Positive Well-Regulated (3.97%; log odds = −2.86, SE = 1.10, p < .001) than to remain in the Negative Dysregulated group. Reserved Well-Regulated toddlers were unlikely to transition to Typical Expressive (0%). A higher proportion of Reserved Well-Regulated toddlers was predicted to transition to Positive Well-Regulated (21.70%) than Negative Dysregulated (5.98%), but the difference was not significant (log odds =0.45, SE = 0.95, p = .636). Finally, a high proportion of Surgent children transitioned to Typical Expressive (20.37%) and relative to Positive Well-Regulated (6.05%) or Negative Dysregulated (2.48%), but the difference in likelihood of transitioning to Typical Expressive relative to Negative Dysregulated did not reach statistical significance (log odds =10.50, SE = 6.16, p = .088).

The results of the LTA from Early to Late Childhood were consistent with earlier ages. Average probabilities for most likely latent class ranged from .74 (Negative Dysregulated and Typical Expressive in Early Childhood) to .95 (Positive Well-Regulated in Late Childhood). The Negative Dysregulated type showed strong continuity (Table 3), with 30.64% of the sample predicted to remain in this type according to their most likely latent class, 3.78% predicted to transition to Typically Expressive, and only 0.25% predicted to transition to Positive Well-Regulated. The Positive Well-Regulated type also showed strong continuity, with a lower probability of transitioning to Typical Expressive (Table 3). Due to the smaller percentage of children estimated to fall into the Positive Well-Regulated type in Early Childhood, remaining in this type was the most likely transition pattern for only 56.56% of the sample, and only 1.64% were predicted to transition from Positive Well-Regulated to Typical Expressive. There was also a significantly greater likelihood of remaining in either Negative Dysregulated (log odds = 1.77, SE = .78, p = .024) or Positive Well-Regulated (log odds = 1.72, SE = .61, p = .005) relative to transitioning to Typical Expressive. Only 0.25% of Positive Well-Regulated children were predicted to transition to Negative Dysregulated. Finally, children in the Typical Expressive type were most likely to remain in this type in Late Childhood (34.30%), but a sizable group was estimated to transition to Positive Well-Regulated (17.78%) or Negative Dysregulated (4.79%).

All profiles in the final LTA from Late Childhood to Early Adolescence had moderate-to-high continuity (Table 3). Average probabilities for most likely latent class ranged from .85 (Typical Expressive at both ages) to .91 (Positive Well-Regulated in Late Childhood). Most children were predicted to have the same profile in Late Childhood and Early Adolescence (Negative Dysregulated: 32.39%; Positive Well-Regulated: 19.69%; Typical Expressive: 30.53%). Negative Dysregulated showed the least discontinuity, with 1.75% of children in this type predicted to transition to Typical Expressive, a significantly higher likelihood of remaining in Negative Dysregulated relative to Typical Expressive (log odds = 2.99, SE = .64, p < .001), and only 0.77% predicted to transition to Positive Well-Regulated. Children in the Positive Well-Regulated type were highly unlikely to transition to Negative Dysregulated (0.55%) but had a substantial probability of transitioning to Typical Expressive. However, because this type was smaller, only 5.25% of the sample were predicted to have this transition pattern, and children in the Positive Well-Regulated type were significantly more likely to remain in Positive Well-Regulated than transition to Typical Expressive (log odds = 2.25, SE = .47, p < .001). Finally, although continuity was most common, children in the Typical Expressive group also showed some likelihood of transitioning into either the Negative Dysregulated (6.13%) or Positive Well-Regulated (2.95%) types.

As a post-hoc analysis, we conducted contingency table analysis to examine continuity and change across Infancy, Toddlerhood, and Early Childhood (see Supplemental Table S5). Due to the number of transition patterns increasing from 9 to 27 when considering three time points with three profiles each, and our decision to only include twins who completed all three waves of data collection, the composition of each pattern is small. However, the contingency table supported our hypothesis that more extreme types would show greater continuity over time.

Heritability of Temperament Types

In order to test the assumptions of the twin design (equal means and variances across zygosity group, and no sex differences in means, variances, or cross-twin covariances), saturated models were tested at all waves, with results reported in Supplemental Table S6. Twin intraclass correlations are reported in Supplemental Table S7, and the model fit, parameter estimates, and 95% confidence intervals for the univariate ACE models are reported in Tables 4, 5, 6, 7, and 8. ACE model results highlight the importance of the shared environment for probability of membership in Positive Well-Regulated, which had significant, moderate-to-high shared environmental influences (C=.37- .76) in every developmental period where this type emerged (Infancy, Early Childhood, Late Childhood, and Early Adolescence). Shared environmental influences on probability of membership in Negative Dysregulated were also significant and modest-to-high in Infancy, Early Childhood, and Early Adolescence (C = .29-.62), but not Toddlerhood or Late Childhood, when there were no significant shared environmental influences. Finally, shared environmental influences were significant for Typical Low Expressive in Infancy (C=.61) and Typical Expressive in Early Adolescence (C=.29), but not any other age. Interestingly, neither Reserved Well-Regulated nor Surgent types in Toddlerhood had significant shared environmental influences.

Table 4.

Genetic and environmental influences on probability of profile membership – Infancy

Infancy (12 months)
Temperament Type	Model	−2LL	df	AIC	saBIC	Δ-2LL	Δdf	p	A	95% CI	C or D	95% CI	E	95% CI
Negative Dysregulated	ACE	402.63	556	412.63	418.21	–	–	–	.08/.49	(.05, .11)/ (.29, .73)	.05/.30	(.02, .08)/ (.12, .54)	.03/ .21	(.02, .04)/ (.15, .29)
	AE	409.82	557	417.82	422.29	7.19	1	.007
	CE	417.90	557	425.90	430.36	15.27	1	<.001
	E	548.81	558	554.81	558.16	146.18	2	<.001
Positive Well-Regulated	ACE	42.59	556	52.59	58.17	–	–	–	.07/.52	(.05, .08)/ (.42, .64)	.06/.46	(.04, .08)/ (.30, .65)	.003/.02	(.002, .003)/ (.01, .03)
	AE	69.56	557	77.56	82.02	26.97	1	<.001
	CE	171.63	557	179.63	184.1	129.04	1	<.001
	E	450.53	558	456.53	459.88	407.94	2	<.001
Typical Low Expressive	ACE	340.59	556	350.59	356.16	–	–	–	.03/.20	(.01, .06)/ (.07, .40)	.10/.61	(.07, .13)/ (.44, .80)	.03/ .19	(.02, .04)/ (.14, .26)
	AE	380.39	557	388.39	392.86	39.81	1	<.001
	CE	345.31	557	353.31	357.77	4.73	1	<.029
	E	561.98	558	567.98	571.32	221.39	2	<.001

Note. Bold: “best-fitting” model for that age period based on log-likelihood, AIC, saBIC, LMR p−values, entropy, and class proportions. AIC = Akaike information criterion; saBIC = sample size adjusted Bayesian Information criterion; LMR = Lo-Mendell-Rubin likelihood ratio. A – Additive genetic influences. C – Shared environmental influences. D – Dominant genetic influences. E – Nonshared environmental influences and measurement error. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Table 5.

Genetic and environmental influences on probability of profile membership – Toddlerhood

Toddlerhood (30 months)
Temperament Type	Model	−2LL	df	AIC	saBIC	Δ-2LL	Δdf	p	A	95% CI	C/D	95% CI	E	95% CI
Negative Dysregulated	ACE	501.05	497	511.05	516.62	–	–	–	.10/.56	(.05, .17)/ (.27, .95)	.02/.09	(.005, .10)/ (.03, .56)	.06/ .35	(.04, .09)/ (.24, .48)
	AE	501.44	498	509.44	513.9	.39	1	.533	.12/.66	(.09, .15)/ (.51, .83)	–	–	.06/ .34	(.04, .08)/ (.25, .45)
	CE	509.50	498	517.50	521.96	8.46	1	.004
	E	563.48	499	569.48	572.83	62.43	2	<.001
Reserved Well-Regulated	ACE	397.75	497	407.75	413.33	–	–	–	.06/.41	(.02, .12)/ (.14, .81)	.04/.24	(.01, .08)/ (.05, .55)	.05/ .35	(.04, .07)/ (.24, .49)
	AE	401.13	498	409.13	413.59	3.38	1	.066	.10/.69	(.08, .13)/ (.54, .85)	–	–	.05/ .31	(.03, .06)/ (.23, .41)
	CE	402.24	498	410.24	414.71	4.50	1	.034
	E	468.60	499	474.60	477.95	70.85	2	<.001
Surgent	ACE	414.96	497	424.96	430.53	–	–	–	.06/.45	(.04, .09)/ (.29, .65)	.00/.00		.08/ .55	(.06, .10)/ (.41, .71)
	ADE	413.53	497	423.53	429.11	–	–	–	.01/.10	(.05, .22)/ (.37, 1.58)	.06/.40	(.002, .19)/ (.01, 1.32)	.07/ .50	(.05, .09)/ (.35, .66)
	AE	414.96	498	422.96	427.42	1.43	1	.232	.06/.45	(.04, .09)/ (.29, .65)	–	–	.08/ .55	(.06, .10)/ (.41, .71)
	E	440.34	499	446.04	449.38	26.51	2	<.001

Note. Bold: “best-fitting” model for that age period based on log-likelihood, AIC, saBIC, LMR p−values, entropy, and class proportions. AIC = Akaike information criterion; saBIC = sample size adjusted Bayesian Information criterion; LMR = Lo-Mendell-Rubin likelihood ratio. A – Additive genetic influences. C – Shared environmental influences. D – Dominant genetic influences. E – Nonshared environmental influences and measurement error. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Table 6.

Genetic and environmental influences on probability of profile membership – Early Childhood

Early Childhood (5 years)
Temperament Type	Model	−2LL	df	AIC	saBIC	Δ-2LL	Δdf	p	A	95% CI	C	95% CI	E	95% CI
Negative Dysregulated	ACE	349.61	372	359.61	365.18	–	–	–	.04/.22	(.004, .12)/ (.02, .62)	.09/.48	(.05, .15)/ (.26, .78)	.06/ .30	(.04, .08)/ (.19, .43)
	AE	362.53	373	370.53	374.99	12.92	1	<.001
	CE	351.32	373	359.32	363.79	1.72	1	.190	–	–	.12/.62	(.09, .15)/ (.47, .80)	.07/ .38	(.06, .09)/ (.31, .46)
	E	443.15	374	449.15	452.49	93.54	2	<.001
Positive Well-Regulated	ACE	106.51	372	116.51	122.08	–	–	–	.07/.55	(.05, .09)/ (.39, .74)	.04/.37	(.02, .07)/ (.19, .62)	.01/ .08	(.01, .01)/ (.05, .11)
	AE	117.35	373	125.35	129.82	10.85	1	<.001
	CE	136.50	373	144.50	148.96	29.99	1	<.001
	E	270.07	374	276.07	279.42	163.57	2	<.001
Typical Expressive	ACE	363.62	372	373.62	379.2	–	–	–	.07/.39	(.02, .15)/ (.10, .87)	.04/.24	(.01, .11)/ (.04, .64)	.07/ .37	(.04, .09)/ (.24, .53)
	AE	366.13	373	374.13	378.6	2.51	1	.113	.12/.67	(.09, .15)/ (.51, .86)	–	–	.06/ .33	(.04, .08)/ (.23, .45)
	CE	366.87	373	374.87	379.33	3.25	1	.072	–	–	.09/.49	(.06, .12)/ (.34, .66)	.09/ .51	(.07, .11)/ (.41, .62)
	E	417.53	374	423.53	426.88	53.91	2	<.001

Note. Bold: “best-fitting” model for that age period based on log-likelihood, AIC, saBIC, LMR p−values, entropy, and class proportions. AIC = Akaike information criterion; saBIC = sample size adjusted Bayesian Information criterion; LMR = Lo-Mendell-Rubin likelihood ratio. A – Additive genetic influences. C – Shared environmental influences. D – Dominant genetic influences. E – Nonshared environmental influences and measurement error. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Table 7.

Genetic and environmental influences on probability of profile membership – Late Childhood

Late Childhood (7 to 10 years)
Temperament Type	Model	−2LL	df	AIC	saBIC	Δ-2LL	Δdf	p	A	95% CI	C/D	95% CI	E	95% CI
Negative Dysregulated	ACE	678.28	715	688.28	693.86	–	–	–	.14/.76	(.10, .19)/ (.55, 1.00)	.01/.03	(.02, .10)/ (.13, .53)	.04/.21	(.03, .05)/ (.15, .26)
	AE	678.39	716	686.39	690.85	.11	1	.742	.15/.80	(.13, .17)/ (.68, .93)	–	–	.04/.20	(.03, .05)/ (.15, .26)
	CE	714.44	716	722.44	726.9	36.16	1	<.001
	E	828.35	717	834.35	837.69	150.06	3	<.001
Positive Well-Regulated	ACE	591.60	715	601.60	607.17	–	–	–	.05/.31	(.02, .09)/ (.14, .54)	.07/.43	(.05, .11)/ (.26, .63)	.05/.26	(.03, .06)/ (.20, .34)
	AE	610.71	716	618.71	623.18	19.12	1	<.001
	CE	599.39	716	607.39	611.86	7.80	1	.005
	E	778.17	717	784.17	787.52	186.57	2	<.001
Typical Expressive	ACE	727.02	715	737.02	742.6	–	–	–	.13/.71	(.11, .16)/ (.58, .86)	.00/.00		.05/ .29	(.04, .07)/ (.22, .37)
	ADE	724.48	715	734.48	740.05	–	–	–	.07/.36	(.01, .18)/ (.05, .97)	.07/.37	(.01, .18)/ (.05, .98)	.05/ .27	(.04, .06)/ (.20, .35)
	AE	727.02	716	735.02	739.48	2.54	1	.108	.13/.71	(.11, .16)/ (.58, .86)	–	–	.05/ .29	(.04, .07)/ (.22, .37)
	E	824.34	717	830.34	833.69	97.32	2	<.001

Note. Bold: “best-fitting” model for that age period based on log-likelihood, AIC, saBIC, LMR p−values, entropy, and class proportions. AIC = Akaike information criterion; saBIC = sample size adjusted Bayesian Information criterion; LMR = Lo-Mendell-Rubin likelihood ratio. A – Additive genetic influences. C – Shared environmental influences. D – Dominant genetic influences. E – Nonshared environmental influences and measurement error. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Table 8.

Genetic and environmental influences on probability of profile membership – Early Adolescence

Early Adolescence (10 years and older)
Temperament Type	Model	−2LL	df	AIC	saBIC	Δ-2LL	Δdf	p	A	95% CI	C	95% CI	E	95% CI
Negative Dysregulated	ACE	583.42	695	593.42	598.99	–	–	–	.11/.57	(.08, .14)/ (.42, .74)	.06/.29	(.03, .09)/ (.15, .49)	.03/.14	(.02, .03)/ (.10, .18)
	AE	593.75	696	601.75	606.21	10.33	1	.001
	CE	623.06	696	631.06	635.52	39.64	1	<.001
	E	823.92	697	829.92	833.89	240.51	2	<.001
Positive Well-Regulated	ACE	250.31	695	260.31	265.89	–	–	–	.03/.16	(.02, .04)/ (.09, .24)	.13/.76	(.10, .15)/ (.62, .91)	.01/.08	(.01, .02)/ (.06, .11)
	AE	365.65	696	373.65	378.12	115.34	1	<.001
	CE	264.61	696	272.61	277.07	14.29	1	<.001
	E	732.48	697	738.48	741.83	482.17	2	<.001
Typical Expressive	ACE	623.31	695	633.31	638.89	–	–	–	.09/.51	(.06, .13)/ (.33, .72)	.05/.29	(.02, .10)/ (.13, .51)	.04/.20	(.03, .05)/ (.15, .26)
	AE	631.66	696	639.66	644.12	8.35	1	.004
	CE	645.76	696	653.76	658.22	22.45	1	<.001
	E	817.03	697	823.03	826.38	193.72	2	<.001

Note. Bold: “best-fitting” model for that age period based on log-likelihood, AIC, saBIC, LMR p−values, entropy, and class proportions. AIC = Akaike information criterion; saBIC = sample size adjusted Bayesian Information criterion; LMR = Lo-Mendell-Rubin likelihood ratio. A – Additive genetic influences. C – Shared environmental influences. D – Dominant genetic influences. E – Nonshared environmental influences and measurement error. Infancy – 12 months. Toddlerhood – 30 months. Early Childhood – 5 years. Late Childhood – 7–9 years. Early Adolescence – 10–12 years.

Genetic influences ranged from modest to high depending on type and developmental period. Specifically, there were significant genetic influences on probability of belonging to Negative Dysregulated at all ages except Early Childhood (A=.49-.80), Positive Well-Regulated at all ages where this type was identified (A=.16-.55), Reserved Well-Regulated (A=.69) and Surgent (A=.45) in Toddlerhood, and Typical Expressive in Infancy, Late Childhood, and Early Adolescence (A=.20-.71). The full ACE model for the Typical Expressive type in Early Childhood suggested moderate heritability (A=.39) and modest shared environmental influence (C=.24). Neither A nor C were significant on their own, but it was not possible to drop both without a significant loss of fit. This may suggest that familial influences are needed to explain the probability of belonging to the Typical Expressive type in Early Childhood, though we were unable to distinguish between them. Estimates of the nonshared environment were modest for Positive Well-Regulated (E=.02 to .26) and modest-to-moderate for Negative Dysregulated (E=.14 to .38) across all ages, moderate for Reserved Well-Regulated (E=.31) and Surgent (E=.55) in Toddlerhood, and modest for Typical Expressive in Infancy, Late Childhood, and Early Adolescence (E=.19 to .29), but not Early Childhood (E=.37 in the full model).

Discussion

We used a person-centered approach to examine the developmental course of different temperament types from infancy to adolescence and used the twin design to examine the genetic and environmental etiology of these types at each age. As our person-centered approach was data driven, it is common for findings to differ somewhat across samples and developmental periods, but our findings corresponded with past research in the emergence of profiles characterized by dysregulation and negative reactivity, by strong self-regulation, and by moderate regulation and reactivity (Beekman et al., 2015; Lin et al., 2021). Although Negative Dysregulated and Well-Regulated types emerged at all ages, and Typical Expressive at all ages besides toddlerhood, types were not directly comparable across age due to varying indicators corresponding with temperament development, an issue discussed further in the study limitations.

Extending person-centered temperament research to a longitudinal framework across childhood, we found considerable continuity in profile composition and membership for the moderately-heritable and environmentally-influenced Negative Dysregulated type beginning in Infancy, and the Positive Well-Regulated and Typical Expressive types beginning in Early Childhood, with greater discontinuity during the infant and toddler years. Notably, our findings included a strong consistent role for the shared environment in membership in Positive Well-Regulated across all ages, and a modest-to-moderate role of the shared environment for the other types at some but not all ages. Similar to dimensional approaches, a person-centered approach revealed that temperament rapidly develops in infancy and toddlerhood and becomes more consistent with age. Unlike dimensional approaches, the person-centered approach uncovered the importance of the shared environment for self-regulation from infancy to adolescence.

Temperament Types Within Developmental Periods

Consistent with findings from prior studies using a person-centered approach (Lin et al., 2021; van den Akker et al.,2010), three temperament types emerged at each age, with consistencies and differences depending on developmental period. As hypothesized, in each developmental period Negative Dysregulated and Well-Regulated types emerged; however, regulation corresponded with reticence in Toddlerhood but higher activity level, lower shyness and positive affect in other developmental periods. The third type, Typical Expressive, varied to some extent across development. In Infancy, we found a Typical Low Expressive profile similar to those described by others (Beekman et al., 2015; Lin et al., 2021) in infancy, characterized by low positivity and regulation but also low negativity. A Typical Expressive type was identified again in Early Childhood and continued to emerge into Early Adolescence, falling between Negative Dysregulated and Positive Well-Regulated on dimensions of regulation and negative reactivity. In contrast, a Surgent type emerged in Toddlerhood, defined by the lowest shyness and highest impulsivity, with similarly low levels of regulation to the Negative Dysregulated type. The higher percentage of Negative Dysregulated and Surgent types in Toddlerhood, in addition to the emergence of shyness and impulsivity as factors differentiating between the unregulated types, may be related to the combination of the rapidly developing self-regulation, social skills, cognition, and novel social demands during toddlerhood (Eckerman & Peterman, 2001).

Although the Reserved Well-Regulated and Surgent types only emerged in three-profile solutions in Toddlerhood, a Reserved type has been identified in a sample of 10- to 15-year olds (Hirvonen et al., 2018), another developmental stage characterized by social salience. Findings suggest self-regulation and its precursors in infancy and early childhood may coincide with greater reserve. There were some indications that our Typically Expressive type in Early Adolescence included more reticent children, with this type displaying shyness and fear similar to the Negative Dysregulated type. In addition, the four-profile solution in Late Childhood resulted in Typical Expressive separating into Reserved and Surgent types. Thus, while it is likely that some children do fall into the moderate range on all or most temperament dimensions and are well-described by the Typical Expressive type found in many person-centered studies (e.g., Beekman et al., 2015; Scott et al., 2016; van den Akker et al., 2010), it is plausible that this type may also include smaller, more rare types such as Kagan’s (1994) behaviorally inhibited type that we were under-powered to detect.

Patterns of Change in Temperament Types Across Development

We found evidence for both continuity and change in temperament type, with the greatest discontinuity from infancy to toddlerhood, though findings should be interpreted with caution given that some temperament dimensions could not be measured in all developmental periods. The Negative Dysregulated type in particular had moderate to high continuity from infancy on, with probabilities of remaining in this type across adjacent years ranging from 66% across the transition from Toddlerhood to Early Childhood, to 85% from Late Childhood to Early Adolescence. In contrast, continuity was not seen until toddlerhood for the Well-Regulated type, which may correspond to the development of self-regulation systems starting in toddlerhood (Calkins, 2007; Posner & Rothbart, 2000). For Typical Expressive, continuity was not seen until Late Childhood, when similar types were identified across Late Childhood and Early Adolescence. Findings were consistent with increasing continuity with age and stronger continuity in more extreme types. Alternatively, this finding could be a result of more consistent measurement on key indicators for the extreme types.

Prior studies have examined continuity in temperament with age, but only one has looked across more than one developmental period in childhood (Jansen & Mathieson, 2008). That investigation considered four ages between 18 months and 9 years and reported moderate continuity. Our study adds to these findings by increasing the ages considered to early adolescence and providing insight to when continuity may begin to take shape across development in different temperament types with similar characteristics. Notably, our study did find a smaller number of types than those identified by Janson and Mathieson (2008), which may be explained by the difference in statistical methods. Janson and Mathieson (2008) used I-States-As-Objects-Analysis, a type of cluster analysis that requires predetermined cut-off values on the indicators for the analysis to identify the clusters, or types, in a subsample of the data. The remainder of the sample was then classified into the types already identified in the original subsample. This approach is unlike the LPA used in the present study, which is completely data driven and identified types using the full sample at a given developmental period independently of the other developmental periods. Despite these differences, our study found similar patterns of continuity such that individuals tended to be classified in types with the same characteristics across development in both studies.

Regarding discontinuity, the probability of transitioning from Negative Dysregulated into Positive Well-Regulated was largest from Infancy to Toddlerhood, and gradually increased for Surgent or Typical from Infancy to Early Childhood, peaking from Early to Late childhood and reducing from Late Childhood to Early Adolescence. However, transitions from Surgent or Positive Well-Regulated to Negative Dysregulated were extremely unlikely at all ages after early childhood and relatively low up to early childhood. Thus, it will be important in future studies to identify factors predicting transitions out of Negative Dysregulated.

The current study extended past research in two important ways. First, we were the first to examine continuity in temperament across the entirety of childhood from infancy to early adolescence, making this study uniquely poised to observe how continuity in temperament takes shape during crucial periods of development. Infancy and toddlerhood are two important developmental periods for the development of regulatory systems (Eisenberg et al., 2010), which may explain the small percent of Positive Well-Regulated infants and the two well-regulated types in toddlerhood that lean toward reserved or surgent. The introduction of novel social situations (e.g., entering preschool) and a greater ability to independently explore in toddlerhood may also help to explain the introduction of the two well-regulated types. Negative Dysregulated also increased in size with age and showed high continuity beginning in infancy. Although the high continuity in Negative Dysregulated may reflect more consistent key indicators across development, this is not supported by the results. Specifically, the same scales were used in both toddlerhood and early childhood, yet continuity was highest in late childhood to early adolescence despite the change in questionnaires and the late inclusion of affiliation and fear.

Second, similar to developmental change in temperament dimensions (Putnam, et al., 2008), Positive Well-Regulated infants were most likely to transition to the Surgent type in Toddlerhood, and Surgent toddlers were most likely to transition to Typical Expressive in Early Childhood. Given the emergence of impulsivity and inhibitory control in Toddlerhood, reflecting the development of neurobiological self-regulatory systems (Calkins, 2007; Posner & Rothbart, 2000), the Surgent type may have been more prevalent than Typical Expressive, which may explain the high probability of Positive Well-Regulated and Typical Low Expressive infants transitioning to Surgent and Reserved Well-Regulated, respectively. With a larger sample, we may have identified both Surgent and Typical Expressive types in Toddlerhood, which might increase early continuity in Typical Expressive. Future studies should focus on continuity and change in positive emotion expressivity from infancy to early childhood to clarify.

In summary, prior research on temperament using the dimensional approach has provided evidence for the stability of individual facets of temperament over time (e.g., Lemery et al., 1999; Liu et al., 2023). Expanding on this research using a person-centered approach and a longitudinal design, the results of the present study suggest that an individual’s pattern of temperament traits also tends to be continuous over time. However, while temperament from infancy to adolescence in our sample was largely characterized by continuity, some children did transition from one type to another across development. For example, continuity did not begin to approach for the Well-Regulated and Typical Expressive types until toddlerhood whereas Negative Dysregulated showed strong continuity from infancy on. Additionally, the present study provided clear evidence supporting that temperament traits are not independent of one another. Instead, they cluster together naturally, and those patterns tend to be consistent across time (e.g. well-regulated children who were high effortful control and low on negative emotionality consistently emerged across development). The heterogeneity in both the patterns of temperament traits identified in this study and in the patterns of continuity and change highlight the need for future studies to consider person-centered approaches to better understand changes in temperament across development. This could take many forms. One example could be assessing the genetic and environmental contributions to different patterns of continuity and change in temperament types, though this would require a larger sample size than that of the current study given that some transition patterns are rare. Further exploring developmental changes in temperament identified in this study will allow future researchers both to predict maladaptive outcomes and inform the development of new prevention and intervention programs.

Heritability and Shared Environment Contributions to Temperament Types

One of the most striking findings from our twin analyses was a strong, consistent role for the shared environment on probability of membership in Positive Well-Regulated across all ages, and less consistently for other types in Infancy and Early Adolescence. Modern conceptualizations of temperament emphasize the importance and inextricability of biological and environmental influences across development (Shiner et al., 2012). Consistent with research using a dimensional approach (e.g., Lemery-Chalfant et al., 2013; Saudino, 2009), we saw evidence of genetic influences (typically moderate to high) on all three temperament types in every developmental period except early childhood. However, we also most often found that both genetic and shared environmental influences were needed to explain probability of belonging to a type. This finding stands in contrast to dimensional research on self-regulation (e.g., Lemery-Chalfant et al., 2008; Liu et al., 2023; Mullineaux et al., 2009), including past research from this sample (Rea-Sandin et al., 2023), which typically identified high heritability and little or no shared environmental influence on effortful control or its lower order dimensions. However, our Well-Regulated profiles were characterized not only by strong self-regulation, but also by low negative reactivity and higher positive reactivity, activity level, and sociability. Other twin research found a role for the shared environment in positive reactivity in infancy, early childhood (Flom et al., 2018; Goldsmith et al., 1997; Wang & Saudino, 2015), and to a lesser extent middle childhood (Allan et al., 2014). Although a typological approach allows distinct etiologies for upper and lower extremes of a dimension (e.g., effortful control in well-regulated and dysregulated types), our findings raise the question of why the different approaches can lead to different conclusions about the role of genetic and environmental influences on temperament.

One explanation is that the dimensional and typological approaches answer different questions. By averaging across subgroups, the dimensional approach considers each temperament trait along a single continuum, allowing examination of fine-grained individual differences within that dimension. The approach also provides information on genetic and environmental influences on that dimension on average in the sample as a whole. In contrast, we parsed temperament into types using person-centered analyses to identify levels of each dimension that presented simultaneously, an approach that automatically accounts for interactions between dimensions and provides information about genetic and environmental influences on a pattern of temperament dimensions for a subgroup of individuals. At the same time, within-profile variation in temperament dimensions still exists. Our examination of posterior probability scores partially addresses this variation by accounting for uncertainty in type but is not the same as examining the etiology of within-profile differences in temperament. It may be that children in environments conducive to emotional and behavioral regulation and with the expression of positive emotion (e.g., warm, consistent parenting; Grolnick et al., 2019) are more likely to belong to Positive Well-Regulated for environmental reasons, but variation in self-regulation within that type is still heritable, or even influenced by different environmental factors.

In addition, the univariate twin ACE model describes genetic and environmental variation in a sample as a whole, but not differences in etiology for individuals or subgroups. However, twin studies that consider moderation of genetic and environmental influences by measured environmental factors show that these sources of variation are not always uniform within a sample. For instance, twin studies examining moderation of the heritability of IQ, the Intelligence Quotient, and academic outcomes by socioeconomic status find variation in a trait to be largely genetic in some environments (e.g., high-SES families; countries with strong social safety nets; Tucker-Drob & Bates, 2016) but more environmentally influenced in others (e.g., impoverished families with less access to social resources; Turkheimer et al., 2003). Our person-centered analyses allowed groups to emerge based on patterns of co-occurring temperament traits without attributing them to any environmental factor but may similarly allow detection of within-sample etiological heterogeneity using a data-driven approach. In light of this possibility, it will be important in future studies to examine which environmental factors might differentiate type membership within and across time.

In taking a person-centered approach allowing for heterogeneity in constellations of traits and their etiology, we uncovered important shared environmental influences complementing our past work in another twin study where modeling gene-environment correlation and interaction revealed that dimensional temperament was less heritable and more influenced by the shared environment in safe, calm homes (Lemery-Chalfant et al., 2013). Together, these findings suggest that shared environmental influences on children’s temperament are evident when twin studies model temperament or the environment with greater nuance.

In a prior study with a different childhood twin sample, we found high heritability and limited shared environmental influences on membership in a similar Positive Well-Regulated type (Scott et al., 2016). The different findings may be due to a fourth type modeled separately in that study, Regulated-Typical, which had heritability and large shared environmental influences. Twin study findings are also influenced by both the broader sociocultural context and sample characteristics (Shanahan & Hofer, 2005), and environmental influences are less easily detected when trait relevant environments vary less within a sample (Johnson et al., 2010). Thus, the consistently strong shared environmental influences on the Positive Well-Regulated type in this study may be due, in part, to the greater socioeconomic and racial/ethnic diversity of the Arizona Twin Project. As such, findings emphasize the need for genetically-informed studies of child development in representative samples from diverse sociocultural and economic backgrounds, and examination of temperament and other key constructs using a range of methods (e.g., person- and variable-centered approaches, parent-reported and observational measurement).

The large, consistent shared environmental influences uncovered in this study highlight targets for prevention and intervention. For example, we found that the early childhood Family Check-Up (FCU) not only improved inhibitory control across childhood in a randomized control trial with a racially/ethnically diverse low-income sample (Hentges et al., 2020) but moderated the influence of genetic risk on children’s observed effortful control (Oro et al., 2023), likely protecting children from future mental health problems (Connell et al., 2019; Hentges et al. 2020).

Strengths and Limitations

This study had multiple strengths and contributions to the literature. First, while short-term longitudinal studies of temperament profiles in early development exist (e.g., Beekman et al., 2015), we were the first to examine transitions in profile membership from infancy to early adolescence, and one of the few to use a person-centered genetically-informed approach, which allowed us to uncover the importance of the shared environment for positive, well-regulated temperament. This holistic approach to temperament, allowing patterns of co-occurrence among dimensions and heterogeneity within samples, may be especially important for understanding developmental phenomena where heterogeneous groups are expected (Kagan, 1994).

Second, using a longitudinal genetically-informed design, our study was uniquely positioned to highlight the nuances of genetic and environmental influences of temperament over time. For example, the results of our study suggest that the genetic and environmental influences on temperament differ depending on type within the same developmental period and may change throughout development. Future studies can expand on these findings by examining genetic and environmental influences on growth in temperament types over time within a single developmental period to allow for measurement equivalence, as done by Liu and colleagues (2023) using the dimensional approach. Combining the dimensional and typological approach within the same study may provide further insights into the stability and change of genetic influences on temperament across development beyond our current understanding of temperament from studies only using the dimensional approach to temperament. Finally, our twin sample was racially, ethnically, and socioeconomically diverse and representative for the southwest United States, which may have enabled us to detect shared environmental influences that studies with more homogeneous samples cannot (Johnson et al., 2010).

However, there were also limitations. First, due to both attrition and new recruitment, our sample overlap across age is only partial, making comparisons between the infancy-to-early-childhood and late-childhood-to-adolescent periods difficult, and the limited subsample in early childhood increased the uncertainty of profile transitions into and out of that wave. Additionally, temperament develops over time, with some traits such as activation control only measurable at later ages (Rothbart & Bates, 2007; Rothbart, 2011). Thus, the indicators used to identify types across development are necessarily different, a limitation facing most developmental research (Nunnally, 1973). We also expanded the breadth of our measurement of temperament over time, including the addition of scales tapping sadness and activity level, as increases in funding allowed for a more comprehensive study. Thus, we limited the scope of this study to continuity and change in temperament types as opposed to stability over time. Further, these changes in measurement may have contributed to the discontinuity in type membership across development, though the transition from late childhood to early adolescence had the most continuity despite it being the transition with the most change in scales. Our measurement is also limited to parent-report, which offers a valid but partial perspective on children’s behavior in everyday contexts (Rothbart & Bates, 2007). Future research should consider multiple reporters and observational assessments of temperament and aim to better understand the implications of using person- and variable-centered methods to temperament across development.

Conclusion

We used a person-centered approach to identify temperament types and transitions from infancy to adolescence and uncovered three relatively consistent types across development: Negative Dysregulated, Positive Well-Regulated, and Typical Expressive. Similar to dimensional studies, children’s temperament rapidly developed in infancy and toddlerhood and became more consistent with age, with continuity appearing earlier for Negative Dysregulated than Positive Well-Regulated. Importantly, transitioning out of Negative Dysregulated occurred more often from toddlerhood to early childhood relative to later ages, further supporting early intervention. In contrast to dimensional research yielding moderate to high heritability estimates and a limited role for the shared environment, we found moderate genetic influences and strong and consistent shared environmental influences on Positive Well-Regulated at all ages, and less consistent but substantial estimates of the shared environment on other types. These findings support the use of family interventions to decrease temperament risk and increase resilience across childhood.

Public significance statement:

Children can be classified into three types based on their temperament: negative dysregulated, positive well-regulated, or typical expressive. These types develop rapidly across infancy and toddlerhood becoming more continuous by early childhood. Types are both heritable and influenced by the early environment, supporting a complex interplay between temperament and sociocultural context when predicting children and adolescent’s adaptation.