Sleeping Through the Ages: Developmental Trajectories of Actigraphy-Derived Sleep-Wake Patterns Across Adolescence and Emerging Adulthood

Abstract

Sleep is a fundamental domain of development and has important implications for other key aspects of adaptation. Little is known about normative changes in sleep across adolescence and emerging adulthood and whether change is linear or nonlinear. We examined growth trajectories of sleep across 9 years of development, and individual differences in trajectories according to participants’ race/ethnicity and sex assigned at birth. During each wave, we measured four key actigraphy-derived sleep parameters over one week including duration (number of minutes spent asleep), efficiency (percentage of time scored as sleep), midpoint (preferences for morningness-eveningness), and consistency/variability in duration (night-to-night fluctuations in duration over one week). Participants (N=295, 53% female, 69% European American/White, 30% African American/Black, 1% biracial) from diverse socioeconomic backgrounds completed five waves of data (M ages at each wave: 15, 16, 17, 22, 24). We observed a cubic trajectory for sleep duration whereby sleep duration declined across adolescence, increased into emerging adulthood, and declined again from 22–24. Consistency/variability in sleep duration also exhibited a cubic trajectory. Nightly fluctuations in sleep duration decreased from 15–16, increased from 16–22, and decreased again from 22–24. We found a quadratic trajectory for sleep midpoint indicating a progressively greater preference for eveningness. Sleep efficiency exhibited linear growth and improved over time. Differences in trajectories for sleep duration and sleep efficiency emerged based on participants’ race and sex. Findings indicated significant changes in developmental trajectories of four sleep parameters across adolescence and emerging adulthood and underscore the importance of nonlinear assessments.

Sleep is a fundamental domain of development and changes considerably throughout the life course (Boatswain-Jacques et al., 2023; Evans et al., 2021; Reynolds et al., 2023). Understanding normative changes in the sleep-wake cycle over development is critical for identifying individuals at greatest risk for poor cognitive, mental and physical health outcomes as well as individuals most likely to experience enhanced well-being (Bakour et al., 2020; Machado et al., 2021b; McVeigh et al., 2021; Thompson et al., 2024a). However, little is known about typical developmental trends in objective assessments (e.g., actigraphy) of sleep over time, especially across adolescence and emerging adulthood. Moreover, drawing on socioecological frameworks of sleep, developmental trajectories of sleep are likely to vary based on individual characteristics such as race/ethnicity (hereafter referred to as race) and sex (Kwon et al., 2024). Although substantial evidence documents sleep disparities across these groups (El-Sheikh et al., 2022a; Kwon et al., 2024), less is known about whether developmental trajectories of various sleep parameters differ by race or sex. Addressing these gaps, the primary goal of this study was to examine developmental growth trajectories of four key actigraphy-based sleep parameters across five waves of data (age 15 to 24). As a secondary aim, we examined race and sex assigned at birth as moderators of sleep growth trajectories to determine whether specific groups deviated from (a) the average developmental trajectory in the sample and (b) their demographic counterparts (i.e., female vs. male; African American/Black [AA] vs. European American/White [EA]).

Sleep Across Adolescence and Emerging Adulthood

Although sleep needs are known to change across the life course (Meltzer et al., 2021; Thacher, 2013), changes in typical sleep-wake patterns during the transition from adolescence to adulthood are not well understood. Focusing on changes in sleep during adolescence, the perfect storm model proposes that the interaction between biological processes and psychosocial factors creates a chronic pattern of insufficient and irregular sleep (Carskadon & Tarokh, 2013). Biological changes in sleep during this time are regulated by homeostatic (Process S) and circadian (Process C) mechanisms (Carskadon & Tarokh, 2013; Tarokh et al., 2019). Specifically, the slower accumulation of homeostatic sleep pressure during wakefulness and the lengthening of the circadian rhythm both independently and jointly contribute to delayed sleep timing and stronger evening preferences. These biological shifts are further amplified by psychosocial factors (e.g., increased media use, screen time, autonomy, extracurriculars, homework) at the same time societal constraints like early school start times limit opportunities for sufficient sleep. As adolescents transition to emerging adulthood, similar biological and psychosocial influences on sleep are thought to persist (Carskadon & Tarokh, 2013). However, this perspective risks portraying emerging adults as simply more mature adolescents, overlooking the heterogeneity of this life stage and the varied pathways individuals take into adulthood (Schwartz, 2016). Initial evidence suggests that sleep timing may serve as a biological marker of the end of adolescence, reaching an inflection point in early adulthood when individuals begin to adopt earlier sleep schedules relative to adolescents (Roenneberg et al., 2004). This shift underscores that sleep indeed changes during this transitional period and may reflect the substantial control adults have over their sleep relative to youth (Meltzer et al., 2021). Accordingly, there is a critical need to map developmental trajectories of sleep across adolescence and emerging adulthood using longitudinal data (Tarokh et al., 2019).

Sleep is a multifaceted construct (Buysse, 2014; Meltzer et al., 2021; Sadeh, 2015) and sleep parameters represent distinct facets with differing developmental and clinical implications (Boatswain-Jacques et al., 2023; Park et al., 2019; Thompson et al., 2024a). Accordingly, we draw on multidimensional models of sleep health (Peds B-SATED, RU-SATED; Buysse, 2014; Meltzer et al., 2021) and assessed trajectories of four well-recognized sleep domains derived through actigraphy: sleep duration (total number of minutes scored as sleep between sleep onset and wake time), sleep efficiency (an indicator of sleep quality calculated as the percentage of time between sleep onset and wake time scored as sleep), chronobiology (sleep midpoint, the halfway point between sleep onset and sleep offset and an indicator of an individual’s circadian rhythm or their morningness-eveningness preference), and consistency/variability in sleep duration (intraindividual variability, night-to-night fluctuations in sleep over a one-week period). Relations between actigraphy-based sleep duration, sleep efficiency, and chronobiology and other important developmental outcomes (e.g., mental health, academic functioning) have been well-documented (Bakour et al., 2020; Buysse, 2014; Cooper et al., 2023; Machado et al., 2021b; McVeigh et al., 2021; Thompson et al., 2024a; Walsh et al., 2022). Some work on nonlinear associations between youths’ sleep and adjustment suggests the benefits of longer and better-quality sleep may plateau or reverse beyond optimal thresholds (e.g., Fuligni et al., 2018; Shimizu et al., 2020). This highlights the importance of identifying and examining nonlinear effects, as associations between sleep and other developmental outcomes may depart from traditional linear, dose-response models. Sleep consistency/variability is less studied, but emerging findings show its robust associations with key developmental domains (e.g., mental health, academic functioning; Becker et al., 2017; Bei et al., 2016; Castiglione-Fontanellaz et al., 2023; Mathew et al., 2024; Thompson et al., 2024a).

Sleep Trajectories

Few studies have examined trajectories of objective sleep parameters derived through actigraphy across childhood and adolescence using more than three waves of data of a single cohort (spanning ages 8–12, Boatswain-Jacques et al., 2023; spanning ages 9–18, Thompson et al., 2024a)—comparable assessments have yet to be conducted across adolescence and emerging adulthood. Below we review the available evidence for changes in sleep duration, sleep efficiency, chronobiology, and consistency/variability in sleep duration during adolescence and emerging adulthood.

Prior work suggests that early adolescents sleep longer than late adolescents (Evans et al., 2021; Keyes et al., 2015; Machado et al., 2021a; Maslowsky & Ozer, 2014; Randler et al., 2019; Saelee et al., 2023). Differences in average sleep duration in late adolescence versus emerging adulthood are more inconsistent. Compared to emerging adults, some find that late adolescents have shorter sleep (Doane et al., 2015; Maslowsky & Ozer, 2014; Randler et al., 2019) and others find they have longer sleep (Saelee et al., 2023). Notably, some of these findings are descriptive and do not necessarily reflect a statistically significant difference, which could contribute to the inconsistencies in observed patterns. Few studies examined sleep duration trajectories across adolescence and emerging adulthood; using accelerated longitudinal designs (i.e., assessing multiple cohorts over shorter time periods to examine longitudinal patterns), the evidence from these studies is mixed. One study of actigraphy-derived sleep duration revealed linear declines in sleep duration from adolescence to emerging adulthood across three waves (ages 14–22; Park et al., 2019). Other work suggests possible nonlinear trajectories of sleep duration—across seven waves of data from age 13 to 22, self-reported sleep duration trajectories declined across adolescence (age 14 to 18) and increased from age 18 to 22 (Chen & Chen, 2021). Disparate findings may be due to variations in the number of waves (three versus seven) or sleep measure (subjective versus objective). No studies to our knowledge have examined developmental trajectories of objective sleep duration during these developmental periods across more than three waves of data.

Comparatively, fewer studies have examined change or growth in sleep efficiency, chronobiology, and consistency/variability in sleep duration. Regarding sleep efficiency, prior work does not offer conclusive evidence regarding direction of change. Some suggest that average sleep efficiency in adolescence and emerging adulthood is relatively comparable (Urner et al., 2009), and others show that adolescents have worse sleep efficiency than emerging adults (Doane et al., 2015; Evans et al., 2021). One study examined sleep efficiency trajectories utilizing an accelerated longitudinal design, revealing linear declines from adolescence to emerging adulthood across three waves of data (ages 14–22; Park et al., 2019). Notably, this is the one study to our knowledge that examined sleep efficiency trajectories during these developmental periods. Methodological study features including comparing averages versus modeling trajectories or the use of short versus longer lags between waves may have contributed to the discrepant findings.

Next, research on changes in chronobiology has primarily assessed sleep midpoint (halfway point between sleep onset and sleep offset) and has found an earlier average sleep midpoint for adolescents relative to emerging adults, indicating an increasing preference for eveningness (age 15 to 25, Fischer et al., 2017; age 16 to 22, Kuula et al., 2019; infancy to age 18, Randler et al., 2019; age 17–19 to 22–24, Urner et al., 2009). These studies further show that average sleep midpoint may begin to decrease (increasing preference for morningness) or plateau in emerging adulthood (Fischer et al., 2017; Kuula et al., 2019; Randler et al., 2019). One study examined trajectories of self-reported morningness-eveningness and found a similar pattern, showing that eveningness increased across three time points (age 12 to 17) before plateauing (age 17 to 19) (Cooper et al., 2023). The literature generally indicates a delay in sleep timing across adolescence that decreases or plateaus in late adolescence/emerging adulthood, except for Urner and colleagues (2009), whose two-wave design may have limited the ability to detect nonlinear patterns.

Lastly, one study revealed a linear decrease in consistency in sleep duration (i.e., increased variability) across three measurement occasions ranging from age 14 to age 22 using an accelerated longitudinal design (Park et al., 2019). The underlying developmental trajectory of consistency/variability in sleep duration warrants further investigation, particularly with respect to potential nonlinear patterns.

In summary, at a methodological level, prior work examining changes in sleep during adolescence and emerging adulthood has predominantly focused on comparing averages in sleep parameters across repeated measures or cohorts and does not estimate an overall pattern of change over time. The few studies examining developmental growth trajectories across these developmental periods have been characterized by either subjective assessments of sleep, short term longitudinal designs, limited measurement occasions, or accelerated longitudinal designs (e.g., Doane et al., 2015; Park et al., 2019; Saelee et al., 2023). Although these studies provide important insights, utilizing additional methods and approaches could clarify the developmental process of sleep. For example, short longitudinal designs (e.g., two years) do not capture how development unfolds across multiple developmental periods, and limited measurement occasions (e.g., two or three waves) hinder the detection of possible nonlinear growth. Because development is complex, simpler models (e.g., linear growth, change or gain) may not adequately describe some dynamic processes (Grimm et al., 2011; Whittaker & Khojasteh, 2017). Moreover, commonly used subjective sleep measures and accelerated longitudinal designs also raise questions about bias (e.g., response bias, lack of standardization; Lauderdale et al., 2008; Sadeh, 2015) and age versus cohort effects (Fischer et al., 2017; Galbraith et al., 2017), respectively.

At a conceptual level, multidimensional models of sleep suggest that sleep domains may not change uniformly across development, and it is essential to identify these domain-specific patterns over time (Meltzer et al., 2021). For example, because adults have greater control over their sleep (Meltzer et al., 2021), it is possible that sleep in one domain—especially one that is easier to modify such as timing—may improve (e.g., shift earlier) during the transition to emerging adulthood while sleep in another domain may worsen (e.g., less efficient). To our knowledge, one study examined trajectories of multiple domains of sleep (duration, efficiency, consistency/variability in duration) during this developmental period (Park et al., 2019); however, because the study included three time points, it was not possible to determine whether there was an acceleration, deceleration, or directional change in growth. More research is needed to determine typical growth trajectories for distinct sleep parameters and to assess whether these parameters follow different patterns (e.g., linear vs. nonlinear), rates (e.g., acceleration vs. deceleration), and directions (e.g., improvement vs. decrement) of change during the transition from adolescence to emerging adulthood. Notably, describing patterns of linear and nonlinear change may help pinpoint periods of heightened sensitivity, which may differ across sleep parameters. For instance, if sleep duration declines throughout adolescence and increases in emerging adulthood, late adolescence may represent a particularly vulnerable period for the negative developmental effects of short sleep—and a key window for intervention. In contrast, if sleep efficiency worsens over time, emerging adulthood may represent a period of greater susceptibility to the consequences of poor sleep quality.

Building on the literature in important ways, our primary aim was to model linear and nonlinear trajectories of several key actigraphy-derived sleep parameters across five measurement occasions spanning nine years of development (i.e., age 15 to age 24). We utilized latent growth modeling because it offers flexible modeling of nonlinear growth, which may better capture naturally occurring patterns of change in the data (i.e., letting the data speak). Additionally, latent growth modeling isolates between-person (i.e., interindividual) differences and examines rank order stability of individual differences in sleep over time (e.g., how does an individual’s sleep change over time relative to the sample average) (Selig & Preacher, 2009; Voelkle, 2007). At a translational level, examining interindividual stability can help to identify individuals experiencing either poor or more optimal sleep over time relative to sample averages in sleep trends.

Individual Differences in Trajectories of Sleep According to Individuals’ Race and Sex

Socioecological models of sleep (Kwon et al., 2024; Meltzer et al., 2021) emphasize that sleep trajectories likely vary across demographic groups, with race and sex emerging as the most frequently observed individual characteristics associated with differences in sleep (Kwon et al., 2024). Understanding normative sleep trajectories, therefore, creates a foundation for meaningful comparisons that can reveal which groups are more likely to experience disparities in sleep over time. This is particularly important given that the influence of demographic characteristics such as race and sex may differ across developmental stages (Meltzer et al., 2021), underscoring the need to examine whether these characteristics modify normative sample trajectories. Importantly, applying a multidimensional sleep framework is critical for understanding sleep through a socioecological lens. Examining multiple domains of sleep allows researchers to assess how demographic characteristics may confer unique risks or benefits across different aspects of sleep (Kwon et al., 2024).

Racial disparities in sleep are well-documented (El-Sheikh et al., 2022a; Guglielmo et al., 2018; Johnson et al., 2019) and AA individuals are at greater risk compared to other racial groups for experiencing less-optimal sleep patterns (Johnson et al., 2019). In our sample, 99% of youth identified as AA or EA. Thus, our literature review focused on AA and EA comparisons. Across adolescence and emerging adulthood, AA participants consistently experienced shorter and worse quality sleep than their counterparts (El-Sheikh et al., 2022b; Fuller-Rowell et al., 2017; Keyes et al., 2015; Maslowsky & Ozer, 2014; Saelee, et al., 2023). Notably, racial differences in trajectories of sleep efficiency, sleep midpoint, and consistency/variability in sleep duration have yet to be examined during these developmental periods. Risk for poor sleep among racially and ethnically minoritized groups is likely related to societal (e.g., discrimination, racism), institutional, and structural barriers that reduce access to protective psychosocial and material resources (Johnson et al., 2019). Toward conducting culturally specific research and identifying potential risk and protective factors within AA samples, a critical step is to characterize normative growth in sleep within AA samples (Guglielmo et al., 2018; Loyd et al., 2024; Volpe et al., 2022; Williams & Deutsch, 2016). Accordingly, we utilized multigroup moderation to examine variations in growth trajectories by race. Although this approach still maintains some degree of comparison, it permits modeling of growth trajectories separately for AA and EA participants.

Likewise, prior work reveals possible differences in sleep based on participants’ sex, though the findings are inconclusive. For adolescents and young adults, some research indicates that females had longer, higher quality sleep and greater morningness preference than males (Fischer et al., 2017; Machado et al., 2021a; Kuula et al., 2019; Park et al., 2019; Randler et al., 2019); yet other studies reported opposite findings (Chen & Chen, 2021; Fatima et al., 2017). Sex differences may be specific to developmental periods. For example, males slept longer in adolescence and females slept longer in emerging adulthood (Machado et al., 2021a; Maslowsky & Ozer, 2014). Such effects may be attributable to a range of factors including puberty and hormonal changes (e.g., related to pregnancy), socialization expectations, experiences of psychosocial stress, and sleep hygiene and attitudes toward sleep (Franco et al., 2020; Jonasdottir et al., 2021; Olds et al., 2010; Ruggiero et al., 2019). Our assessment of sex-related effects in trajectories of sleep across adolescence and emerging adulthood addresses a notable gap.

Present Study

Using data spanning five waves and two developmental periods, we examined sleep trajectories across adolescence and emerging adulthood and assessed race- and sex-related effects. Guided by theory and prior research, biological processes and psychosocial factors during the transition from adolescence to emerging adulthood may shape developmental trajectories of sleep in ways that diverge from traditional linear perspectives. For example, as proposed by the perfect storm model, biological changes in sleep regulation—including a slowing rate of accumulation of homeostatic sleep pressure and a progressively later circadian phase (Carskadon & Tarokh, 2013)—may drive a shift toward later chronotypes across adolescence, followed by a reversal toward earlier chronotypes in emerging adulthood (Roenneberg et al., 2004). Although this is only one example, it highlights the importance of examining not only overall improvements or decrements in sleep over time but also changes in the rate and direction of sleep problems. Accordingly, the current study tests the possibility of both linear and nonlinear developmental trajectories of sleep. Given the inconclusive literature on growth or change over time in sleep duration, sleep efficiency, sleep midpoint, and consistency/variability in sleep duration, we do not propose specific directional hypotheses for these variables. Likewise, we do not propose specific directional hypotheses regarding the nature of nonlinear effects for any sleep parameter. Finally, we generally expected AA participants to experience worse sleep over time; however, our assessment of race- and sex-related differences in trajectories was exploratory.

Because theoretical perspectives emphasize not only whether sleep changes over time but also how the rate and direction of change may shift across developmental transitions (Carskadon & Tarokh, 2013), analytic approaches that can capture nonlinear trajectories are necessary. Therefore, to integrate theory and methodology, we tested our hypotheses using latent growth modeling, which was selected over alternative approaches (e.g., repeated-measures ANOVA) for several reasons. Latent growth modeling directly models individual differences in developmental trajectories (Grimm et al., 2011; Whittaker & Khojasteh, 2017). By modeling change as a latent process, this approach yields developmentally interpretable latent parameters that represent the amount of change, rate of change, and direction of change over time (Grimm et al., 2011). Thus, latent growth models can flexibly accommodate both linear and nonlinear patterns of change over time, including quadratic trajectories reflecting acceleration or deceleration in rate of change and cubic trajectories capturing directional shifts in change (Grimm et al., 2011; Whittaker & Khojasteh, 2017).

To operationalize these trajectories, higher-order growth parameters (i.e., quadratic and cubic terms) were incorporated by squaring or cubing the linear factor loadings, respectively, which were specified to reflect the amount of time elapsed between the first wave of data collection and each subsequent wave. We began with the simplest model (no growth) and sequentially estimated increasingly complex models (linear, quadratic, and cubic). Notably, as model complexity increased, lower-order growth parameters were retained in estimation; thus, cubic models included intercept, linear, and quadratic terms. Finally, latent growth modeling allows for formal comparisons of competing growth trajectories to determine which pattern best explains the data. We conducted chi-square difference tests to identify the growth trajectory that best fit the data. The full analytic plan is described in detail in the Method section, and Figure 1 illustrates an example specification of a cubic growth model.

Figure 1.

Illustration of the Specification of a Cubic Growth Model

Method

Participants

Data were drawn from Waves 4–8 of a large longitudinal investigation on family stress and youth development. At Wave 1, children (N = 251) were 8 years of age and were recruited through flyers distributed at schools in the southeastern USA. Families were eligible for participation if two parents had been living together for more than two years. Exclusion criteria included children having a chronic physical illness, ADHD, or learning disability. Waves 1–3 were not included in the current study as they did not include actigraphy assessments of sleep. At the start of Wave 4, 53 new families were recruited from the same school districts as the original sample using the same inclusion/exclusion criteria. There were no differences between participants recruited at Wave 1 and Wave 4 based on demographic or primary variables. For explication, the five waves (4–8) used in analyses will be referred to by participants’ average age (15, 16, 17, 22, and 24).

Table 1 provides participants’ sociodemographic information across waves. The analytical sample at age 15 consisted of 247 adolescents. One year (M = 367 days, SD = 27 days) later, 231 participants returned for the age 16 assessment. About one year (M = 349 days, SD = 20 days) later, 220 adolescents participated at age 17. Five years (M = 5.73 years, SD = 1.92 years) after the age 17 assessment, 251 participants took part in the age 22 assessment, including 47 new individuals recruited from the same geographical region using the same inclusion/exclusion criteria. New participants did not differ from those who participated in the previous waves on study variables, except that they were slightly older (M = 23.89, SD = .63) than the existing cohort (M = 22.76 years, SD = .70), t(249) = −10.15, p < .001). Approximately two years later (M time lag = 1.87 years, SD = 26 days), 231 individuals returned for the age 24 assessment. Retention was high across waves ranging from 92% to 95%.

Table 1.

Sample Level Sociodemographic Information and Availability of Sleep Data Across Waves

	Wave 4 (N = 247)		Wave 5 (N = 231)		Wave 6 (N = 220)		Wave 7 (N = 251)		Wave 8 (N = 231)
	N (%)	M (SD)	N (%)	M (SD)	N (%)	M (SD)	N (%)	M (SD)	N (%)	M (SD)
Age, years (months)		15.79(9.70)		16.78(9.28)		17.69(8.94)		22.97(9.74)		24.88(9.60)
Sex
Female	131(53)		125(54)		121(55)		143(57)		134(58)
Male	116(47)		106(46)		99(45)		108(43)		97(42)
Ethnicity
European American	163(66)		157(68)		154(70)		173(69)		162(70)
African American	84(34)		74(32)		66(30)		73(29)		65(28)
Biracial							5(2)		4(2)
Income-to-Needs Ratio		2.39(1.30)		2.80(1.72)		3.03(1.83)		2.71(2.16)		3.32(2.05)
Sleep Data
Number of nights with data		5.58(1.67)		4.60(1.96)		5.35(1.67)		6.40(1.29)		5.14(2.18)
Number with missing data (<3 nights)	22(9)		28(12)		11(5)		5(2)		30(13)

Note. Because gender was not collected during earlier waves, we only report on sex assigned at birth.

Socioeconomic status (SES) was measured by income-to-needs ratio, which is the quotient of a family’s total income divided by the federal poverty threshold for their household size (U.S. Department of Commerce, 2025). Parents reported on family income during adolescence (ages 15–17). More than half of the sample represented low to lower-middle class (54%–66%; ratio < 3) and 34%–46% were middle-class or higher (ratio ≥ 3). In emerging adulthood (ages 22 and 24), participants reported on their own income. Over half of participants represented low to lower-middle SES (52%–66%) and 34%–48% were middle-class or higher.

Participants were included in the final analytic sample if they had participated at any point across the five waves. As a result, analyses included 295 participants (53% female, 69% EA, 30% AA, 1% biracial).

Procedures and Measures

All study procedures were approved by the university’s Institutional Review Board. At ages 15–17, parents’ written consent and adolescents’ assent were collected. At ages 22 and 24, participants provided written consent. Participants wore actigraphs at home for seven consecutive nights. At ages 15–17, actigraphy assessments occurred during the school year, excluding holidays. Sleep was examined throughout the year in young adults who were mostly not attending school. Season of watch wear (i.e., fall or spring) was not correlated with any sleep parameters. To corroborate actigraphy data, participants completed nightly sleep diaries.

Sleep

Sleep was assessed during each wave by Octagonal Basic Motionlogger actigraphs (Ambulatory Monitoring) and scored in Action W2 software. Analyses employed Octagonal Motionlogger Interface with Actme software and Action W2, 2000 Ambulatory Monitoring analysis software package. To calculate sleep variables, the analytic program employed the widely recognized Sadeh scoring algorithm in adolescence (Sadeh et al., 1994) and the Cole-Kripke scoring algorithm in emerging adulthood (Cole et al., 1992). Sleep duration represents the number of 1-min epochs scored as sleep between actigraphy-derived sleep onset and wake time. Sleep efficiency is the percentage of time scored as sleep between sleep onset and wake time (also known as % sleep in the literature; e.g., Ancoli-Israel et al., 2015). Sleep midpoint is the halfway point between sleep onset and sleep offset, and provides a measure of chronobiology, or morningness-eveningness. Lastly, consistency/variability in sleep duration was derived using the intraindividual standard deviation of sleep duration across the one-week measurement occasion. Higher scores (larger standard deviations) indicate greater variability/lower consistency in sleep duration across the week whereas lower scores (smaller standard deviations) reflect greater consistency/lower variability in sleep duration across the week.

Actigraphy data were included for those who had ≥ 3 nights of sleep data and were treated as missing for those who had fewer nights of sleep data (2% to 13% of participants across the waves), consistent with other work (Gillis & El-Sheikh, 2019; Sadeh, 2015). At each wave, each sleep variable was derived by creating an average across all available nights, thus aggregating nightly sleep data to the person level. Missing data were due to forgetting to wear the device or actigraph malfunctions (the latter was rare). Nights were also excluded when discrepancies between sleep diary reported and actigraphy-measured sleep onset and wake time exceeded 30 minutes. Average number of nights of actigraphy data and amount of sleep data considered missing are reported for each wave in Table 1.

Transparency and Openness

The study design and analysis were not pre-registered. Data and analytic syntax will be shared for verification purposes upon requests submitted by email to the corresponding author.

Plan of Analysis

Latent growth modeling analyses were utilized to examine developmental trajectories of sleep parameters across ages 15 and 24 and variations in trajectories by participants’ race and sex. Growth models were fit in a step-by-step fashion. First, unconditional univariate growth models were utilized to determine the average trajectory of change (i.e., means of the growth parameters) and the extent of between-person variability (i.e., variances of the growth parameters). Person-level aggregate scores for each sleep variable—calculated across all available nights—served as the observed repeated measures. Consistent with a structural equation modeling framework (Grimm et al., 2016), these repeated measures were treated as observed indicators of underlying latent growth factors (i.e., intercept and slopes), which capture between-person differences in initial levels and rates of change over time. Sleep intercepts were set to age 15. Specifically, factor loadings for sleep parameters were fixed at 0, .1, .2, .7, and .9 for linear slopes, 0, .01, .04, .49, and .81 for relevant quadratic slopes, and 0, .001, .008, .34, and .73 for relevant cubic slopes (see Figure 1). Chi-square difference tests were used to determine the best fitting growth trajectory. Within the best fitting model, the mean and variance of each growth parameter were assessed for significance. Significant variance in the intercepts and slopes would demonstrate between-person variability in initial levels (i.e., at age 15) and in rates of change. In contrast, nonsignificant variance in growth parameters would indicate that participants exhibited similar starting values or similar rates of change over time.

Next, best fitting univariate growth models were used in determining whether trajectories of sleep varied by race (AA vs. EA) and sex (male vs. female). We tested one sleep parameter and one moderator at a time for a total of eight models (4 sleep parameters X 2 moderators). For each sleep trajectory, we utilized the multi-group function in Mplus to compare model fit across two growth models where (a) growth model parameter estimates were constrained to be equal across groups, and (b) all growth model parameter estimates were allowed to vary freely across groups (Selig et al., 2015). A significant chi-square difference test suggests the free-to-vary model is better fitting and indicates that the growth trajectory significantly varies across groups (Werner & Schermelleh-Engel, 2010). Follow-up Wald chi-square tests were conducted to determine the source(s) of differences in growth trajectories.

Analyses were conducted in Mplus Version 8.4. Data were missing for 29% of values; we used full information maximum likelihood (FIML) estimation to handle missing data and retain the full sample for analyses (Little et al., 2014). Data were missing completely at random, based on Little’s MCAR test, χ² (271) = 292.15, p = .18 (Little, 1988; Schlomer et al., 2010). Variables were assessed for normality. Skewed variables (skewness value > 2.0) were winsorized and outlier values were recoded to the value corresponding to 3 SDs. Acceptable model fit included meeting at least two of the following three criteria: χ²/df < 3, comparative fit index (CFI) > 0.90, and root mean square error of approximation (RMSEA) < 0.08.

Results

Preliminary Analyses

Descriptive statistics and correlations among study variables are provided in Table 2. Autocorrelations for sleep duration and sleep efficiency were modest to moderate and generally significant across ages 15 to 24. Autocorrelations for sleep midpoint and consistency/variability in sleep duration were more erratic and generally not significant (40% significant for each sleep midpoint and consistency/variability in sleep duration) across ages 15 to 24. Average sleep duration exhibited an absolute decrease from age 15 to 24 whereas sleep efficiency, sleep midpoint, and variability in sleep duration increased. Sleep duration was associated with participant race and sex at all ages—shorter sleep was associated with being AA and male. Sleep efficiency and sleep midpoint were associated with race but only at age 17; specifically, worse sleep and greater eveningness were associated with being AA. Sleep efficiency was associated with sex at ages 15, 16, and 17 with males being more likely to have worse sleep. Sleep midpoint was associated with sex at ages 16 and 17 with males experiencing greater preferences for eveningness.

Table 2.

Means, Standard Deviations, and Correlations Among Study Variables

	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23
1. Race	–	–	–
2. Sex	–	–	−.09	–
3. Income	2.76	1.43	−.30*	.12*	–
4. Sleep Duration 15	403.64	56.28	−.23*	−.19*	.11	–
5. Sleep Duration 16	400.29	55.91	−.23*	−.29*	.07	.43*	–
6. Sleep Duration 17	394.34	60.68	−.27*	−.26*	.09	.48*	.59*	–
7. Sleep Duration 22	406.49	70.20	−.24*	−.22*	.14	.22*	.10	.20*	–
8. Sleep Duration 24	400.47	66.84	−.12	−.20*	.17*	.32*	.31*	.33*	.24*	–
9. Sleep Midpoint 15	165.41	45.03	.07	.11	−.03	−.11	.03	.06	−.01	.06	–
10. Sleep Midpoint 16	174.86	54.78	−.02	.26*	−.04	−.10	−.22*	−.17*	.13	−.14	.46*	–
11. Sleep Midpoint 17	191.98	65.09	−.14*	.20*	.04	−.22*	−.18*	−.03	−.02	−.11	.36*	.54*	–
12. Sleep Midpoint 22	241.88	117.29	−.01	.00	.01	−.19*	−.06	−.23*	−.16*	−.21*	.06	.18*	.14	–
13. Sleep Midpoint 24	293.06	179.66	.05	.12	−.08	−.03	.07	−.09	−.06	−.08	.10	.06	.07	.23*	–
14. Sleep Efficiency 15	90.73	7.09	−.12	−.17*	.06	.59*	.24*	.38*	.05	.25*	.09	.02	.00	−.16*	−.02	–
15. Sleep Efficiency 16	91.40	6.57	−.11	−.20*	.05	.33*	.43*	.32*	.10	.23*	.13	−.03	−.04	.01	−.01	.56*	–
16. Sleep Efficiency 17	91.04	7.41	−.18*	−.18*	.15*	.28*	.29*	.61*	.01	.33*	.09	−.10	.12	.01	.03	.39*	.56*	–
17. Sleep Efficiency 22	93.83	5.49	−.12	−.02	.16*	.14	.09	.10	.28*	.19*	.09	.04	.10	−.08	.02	.13	.23*	.26*	–
18. Sleep Efficiency 24	93.44	6.14	−.04	−.01	.15*	.24*	.23*	.16	.03	.45*	.01	−.00	.06	−.19*	−.06	.31*	.22*	.24*	.42*	–
19. SD Sleep Duration 15	60.78	27.37	.06	−.05	−.19*	−.10	.05	.07	.02	−.17*	.23*	.11	.11	.06	−.00	−.01	−.00	.10	.10	−.13	–
20. SD Sleep Duration 16	56.57	27.58	−.05	.01	−.03	−.01	.07	.03	.12	.07	.14	.21*	.27*	−.02	.08	.09	.12	.17*	.08	.08	.29*	–
21. SD Sleep Duration 17	60.49	29.55	−.04	−.03	−.00	−.12	−.18*	−.03	−.08	.07	.00	.11	.23*	.15	.08	.01	.07	.17*	.09	.15	.17*	.21*	–
22. SD Sleep Duration 22	78.77	31.98	.08	.03	.02	.00	−.01	−.17*	−.06	.05	.10	.10	.00	.27*	.16*	.03	.11	−.09	−.04	.02	.13	.04	.16*	–
23. SD Sleep Duration 24	69.83	30.68	−.06	−.10	.02	−.06	.04	−.13	−.00	.01	.01	−.08	−.05	.11	.09	−.03	.01	−.02	−.16	.05	−.01	−.10	.10	.15	–

Note. Race was coded as 0 = European American/White, 1 = African American/Black; sex was coded as 0 = female, 1 = male. SD = intraindividual standard deviation across nights; higher values indicate greater night-to-night variability (i.e., lower consistency).

^*p < .05.

Developmental Trajectories of Sleep Parameters

We fit a series of unconditional univariate growth models for sleep duration, sleep efficiency, sleep midpoint, and consistency/variability in sleep duration to determine the trajectory of growth for each variable (model building details are provided in Supplemental Materials, see Table S1). We accepted: (a) a cubic growth model for sleep duration (Figure 2A; Model 1d in Table S1), (b) a linear growth model for sleep efficiency (Figure 2B; Model 2g in Table S1) that specified a covariance between the residuals of the observed sleep efficiency scores at ages 15 and 16, (c) a quadratic growth model for sleep midpoint (Figure 2C; Model 3d in Table S1), and (d) a cubic growth model for consistency/variability in sleep duration (Figure 2D; Model 4d in Table S1). All sleep models constrained observed residual variances to be equal over time, except for sleep midpoint, where observed residual variances were allowed to vary freely. This suggests at least some variation in the meaning of the observed measurements of sleep midpoint. A summary of final model selection is provided in Table 3.

Figure 2. Developmental Trajectories for Sleep Duration, Sleep Efficiency, Sleep Midpoint, and Consistency/Variability in Sleep Duration.

Note. Sleep duration exhibited a cubic growth trajectory (Panel A). Sleep efficiency exhibited a linear growth trajectory (Panel B). Sleep midpoint exhibited a quadratic growth trajectory (higher values correspond with a later sleep midpoint or greater eveningness preference; Panel C). Consistency/variability in sleep duration exhibited a cubic growth trajectory (higher values [larger standard deviations] correspond to greater average night-to-night fluctuations in sleep duration; Panel D). The bolded black line represents the average growth trajectory for each sleep parameter. The dotted grey lines represent individual data for each sleep parameter.

Table 3.

Best Fitting Growth Trajectories for the Full Sample and Demographic Groups

Sleep Variables	Full Sample	European American/ White	African American/ Black	Female	Male
Duration	Cubic	Cubic	Linear	Cubic	Linear
Efficiency	Linear	Linear	Linear	Linear	Linear
Midpoint	Quadratic	N/A	N/A	N/A	N/A
SD Duration	Cubic	N/A	N/A	N/A	N/A

Note. SD = intraindividual standard deviation across nights; higher values indicate greater night-to-night variability (i.e., lower consistency).

After establishing the best fitting unconditional models, we examined whether the growth parameters of each sleep variable were significantly different from zero and demonstrated significant between-person variability (see Table 4). Because complex slopes (e.g., quadratic) are built upon precursory, simpler slopes (e.g., linear), we examined the means and variances for all growth parameters estimated within the final model. Below, we describe the means and variances of the growth parameters for each sleep variable, organized by growth parameter.

Table 4.

Model Fit and Unstandardized Parameter Estimates for Final Unconditional Univariate Growth Models of Sleep Parameters

	Model Fit					Unstandardized Parameter Estimated
Sleep Parameter	χ2	df	χ2/df	RMSEA	CFI	Intercept Mean (μ)	Intercept Variance (σ2)	Linear Slope Mean (μ)	Linear Slope Variance (σ2)	Quadratic Slope Mean (μ)	Quadratic Slope Variance (σ2)	Cubic Slope Mean (μ)	Cubic Slope Variance (σ2)
Duration	3.40	5	.68	.00	1.00	403.62***	26.16***	−97.92*	1588.98**	285.66*	21842.58***	−201.18*	14874.84***
Efficiency	17.88	13	1.38	.04	.96	9.05***	.34***	.37***	.43***	–	–	–	–
Midpoint	5.33	6	.89	.00	1.00	165.95***	1260.98**	10.67***	688.68	.24	10.75	–	–
SD Duration	6.86	5	1.37	.04	.93	60.28***	18.34	−43.85**	716.51**	246.08***	9149.60	−206.29***	6709.66

Note. Intercepts centered at age 15. SD = intraindividual standard deviation across nights; higher values indicate greater night-to-night variability (i.e., lower consistency).

^{* =}p ≤ .05.

^{** =}p < .01.

^{*** =}p < .001.

Intercept

The intercept of each sleep parameter was significantly different from zero and showed significant between-person variability, except for consistency/variability in sleep duration (Table 4). This suggests that participants had similar starting levels of consistency/variability in sleep duration, but differed significantly in their initial sleep duration, sleep efficiency, and sleep midpoint at age 15 (e.g., some slept longer or shorter at 15 years of age than the sample average).

Linear Slope

The linear slope mean for each sleep variable was significantly different from zero, and apart from sleep midpoint, displayed significant between-person variability (Table 4). Sleep duration decreased from age 15 to 24, and rate of change over time varied across participants (e.g., some exhibited more steep decreases whereas others exhibited more stable growth over time). Sleep efficiency, sleep midpoint, and variability in sleep duration increased from age 15 to 24. Rate of change for sleep efficiency and variability in sleep duration varied across participants. In contrast, rate of change for sleep midpoint was comparable across participants.

Quadratic Slope

We estimated a quadratic slope for sleep duration, sleep midpoint, and consistency/variability in sleep duration. The quadratic slope for sleep duration and consistency/variability in sleep duration, but not sleep midpoint, was significantly different from zero, indicating an acceleration/deceleration in rate of change (Table 4). Only sleep duration exhibited significant variance in the quadratic slope, indicating between-person variability in the rate of acceleration (or deceleration) over time.

Cubic Slope

We estimated a cubic slope for sleep duration and consistency/variability in sleep duration, which was significantly different from zero for each, indicating a directional shift in change over time (Table 4). Sleep duration, but not consistency/variability in sleep duration, exhibited significant variance in the cubic slope, indicating between-person variability in when and how rapidly directional shifts occurred.

Moderation Analyses Based on Participants’ Race and Sex

Based on the best fitting univariate growth models described above, we investigated possible individual differences in trajectories of each sleep parameter based on participants’ race (AA or EA) and their sex (female or male) (see Table 3 for the final models). Specifically, we utilized the multigroup function in Mplus to determine whether sleep trajectories differed across groups. Model building details are provided in Supplemental Materials (Table S2 for multigroup based on race; see Table S3 for multigroup based on sex). If the groups significantly differed, we conducted Wald chi-square tests to identify the specific sources of variation in growth trajectories. A significant Wald test would indicate a significant difference in the specific parameter tested across groups.

Variations in Sleep Trajectories by Race

Sleep Duration.

Results from chi-square difference testing demonstrated that the growth trajectory for sleep duration varied across AA and EA participants (Model 5d in Table S2). First, we observed a difference in the overall best fitting trajectory—whereas a cubic growth model was the best fitting for EAs, a linear growth model was the best fitting for AAs (Figure 3A). Next, Wald tests revealed several additional differences (Table 5). The intercept means significantly differed for AA and EA participants (χ²(1) = 24.13, p < .001). AAs experienced shorter sleep (intercept μ = 377.76 minutes [6:18 hrs:min]) than EAs (intercept μ = 412.56 minutes [6:53 hrs:min]) at age 15. Additionally, the linear slope mean for sleep duration significantly differed for AA and EA participants (χ²(1) = 4.56, p = .03). EAs showed a decrease in sleep duration over time (slope μ = −88.08), whereas AAs showed a slight increase (slope μ = 9.42). Last, the linear slope variance significantly differed by race (χ²(1) = 15.25, p < .001), showing that EA participants (slope σ² = 2283.00) had more variability around change in sleep duration than AA participants (slope σ² = 52.20).

Figure 3. Variations in Developmental Trajectories for Sleep Duration and Sleep Efficiency by Participants’ Race.

Note. Individual differences in trajectories of sleep duration by participants’ race (Panel A). Individual differences in trajectories of sleep efficiency by participants’ race (Panel B). The dotted blue lines represent individual data for European American/White (EA) participants. The dashed red lines represent individual data for African American/Black (AA) participants. Respective bolded lines represent the average trajectory for EAs and AAs.

Table 5.

Conditional Univariate Growth Models of Sleep Parameters Moderated by Race

		Intercept Mean (μ)	Intercept Variance (σ2)	Linear Slope Mean (μ)	Linear Slope Variance (σ2)	Quadratic Slope Mean (μ)	Quadratic Slope Variance (σ2)	Cubic Slope Mean (μ)	Cubic Slope Variance (σ2)
Sleep Durationabc	European American/White Youth	412.56***	24.54***	−88.08*	2283.00***	292.32*	26,004.36***	−226.56*	17,019.9***
	African American/Black Youth	377.76***	22.44**	9.42	52.20†	–	–	–	–
Sleep Efficiencya	European American/White Youth	91.22***	3.01***	3.27***	4.56***	–	–	–	–
	African American/Black Youth	89.30***	3.21**	4.14***	4.48†	–	–	–	–

Note. Intercepts centered at age 15.

^{† =}p < .10.

^{* =}p ≤ .05.

^{** =}p < .01.

^{*** =}p < .001.

^aIntercept means significantly vary by race based on the Wald test.

^bSlope means significantly vary by race based on the Wald test.

^cSlope variances significantly vary by race based the Wald test.

Sleep Efficiency.

The growth trajectory for sleep efficiency significantly differed across AA and EA participants (Model 6c in Table S2). A linear growth model was the best fitting for both AAs and EAs (Figure 3B). A follow-up Wald test revealed that the intercept means significantly differed for AA and EA participants (χ²(1) = 4.42, p = .036) (Table 5). Specifically, AAs had less efficient sleep on average (intercept μ = 89.30) than EAs (intercept μ = 91.22) at age 15.

Neither the growth trajectory for sleep midpoint nor consistency/variability in sleep duration varied by race.

Variations in Sleep Trajectories by Sex

Sleep Duration.

The growth trajectory for sleep duration significantly varied by sex (Model 9d in Table S3). First, a cubic growth model was the best fitting for females whereas a linear growth model was the best fitting for males (Figure 4A). Follow-up Wald chi-square tests revealed several additional differences (Table 6). The intercept means significantly differed for female and male participants (χ²(1) = 17.75, p < .001). Females experienced longer sleep (intercept μ = 413.10 minutes [6:53 hrs:min]) than males (intercept μ = 385.08 minutes [6:25 hrs:min]) at age 15. The intercept variance also significantly differed based on sex (χ²(1) = 6.84, p = .009). Females had more variability in their sleep duration at age 15 (intercept σ² = 36.84) than males (intercept σ² = 15.12). Additionally, linear slope variance was significantly different for female and male participants (χ²(1) = 15.80, p < .001). Females exhibited more variability in change in sleep duration over time (slope σ² = 2358.72) than males (slope σ² = 35.34), suggesting females exhibited greater between-person variability in rate of change over time compared to males.

Figure 4. Variations in Developmental Trajectories for Sleep Duration and Sleep Efficiency by Participants’ Sex.

Note. Individual differences in trajectories of sleep duration by participants’ sex (Panel A). Individual differences in trajectories of sleep efficiency by participants’ sex (Panel B). The dotted blue lines represent individual data for female participants. The dashed red lines represent individual data for male participants. Respective bolded lines represent the average trajectory for females and males.

Table 6.

Conditional Univariate Growth Models of Sleep Parameters Moderated by Sex

		Intercept Mean (μ)	Intercept Variance (σ2)	Linear Slope Mean (μ)	Linear Slope Variance (σ2)	Quadratic Slope Mean (μ)	Quadratic Slope Variance (σ2)	Cubic Slope Mean (μ)	Cubic Slope Variance (σ2)
Sleep Durationabc	Female Youth	413.10***	36.84***	−64.86	2358.72***	214.5	28,262.4***	−158.52	19,249.98***
	Male Youth	385.08***	15.12**	−0.60	35.34†	–	–	–	–
Sleep Efficiencyad	Female Youth	91.72***	2.71***	2.38**	4.17**	–	–	–	–
	Male Youth	89.12***	3.73**	5.45***	3.80*	–	–	–	–

Note. Intercepts centered at age 15.

^{† =}p < .10.

^{* =}p ≤ .05.

^{** =}p < .01.

^{*** =}p < .001.

^aIntercept means significantly vary by sex based on the Wald test.

^bIntercept variances significantly vary by sex based on the Wald test.

^cSlope variances significantly vary by sex based on the Wald test.

^dSlope means significantly vary by sex based on the Wald test.

Sleep Efficiency.

The trajectory for sleep efficiency also varied by sex (Model 10b in Table S3). A linear growth model was the best fitting model for both females and males (Figure 4B). Wald testing indicated significant differences in growth parameters (Table 6). First, the intercept means significantly differed for female and male participants (χ²(1) = 9.54, p = .002). Females exhibited higher sleep efficiency (intercept μ = 91.72) than males (intercept μ = 89.12) at age 15. Second, the slope means significantly differed for female and male participants (χ²(1) = 6.38, p = .012). Males showed a steeper increase (slope μ = 5.45) in sleep efficiency compared to females (slope μ = 2.38).

Neither the growth trajectory for sleep midpoint nor consistency/variability in sleep duration varied by sex.

Sensitivity Analyses

We conducted a set of sensitivity analyses that controlled for racial differences in the sex-based multigroup moderation analyses and vice versa. The previously described multigroup moderation findings replicated with one exception—in the sleep duration model, a Wald chi-square test indicated that the linear slope mean was no longer significantly different across racial groups when controlling for sex, (χ²(1) = 1.11, p = .29).

Discussion

Sleep is a multi-dimensional construct and relations between sleep duration (Bakour et al., 2020; Machado et al., 2021b), sleep efficiency (McVeigh et al., 2021; Thompson et al., 2024a), chronotype (Cooper et al., 2023; Walsh et al., 2022), and consistency/variability in sleep (Castiglione-Fontanellaz et al., 2023; Mathew et al., 2024; Thompson et al., 2024a) and other developmental outcomes have been well-documented. During adolescence and emerging adulthood, biological and psychosocial forces converge to adjust the sleep-wake cycle while also increasing barriers to achieving adequate sleep (Carskadon & Tarokh, 2013; Nicholson et al., 2023). Few studies have assessed developmental trajectories of sleep across adolescence and young adulthood. At a methodological level, prior work examining sleep trajectories during these developmental periods has largely utilized subjective assessments of sleep, short longitudinal designs, limited measurement occasions, accelerated longitudinal designs, or measured linear growth. At a conceptual level, there is a significant gap in the literature examining sleep trajectories from a multidimensional perspective—particularly with respect to possible nonlinear change over time—with one study, to our knowledge, assessing trajectories across distinct sleep parameters (actigraphy-derived duration, efficiency, consistency/variability in duration) during this developmental period (Park et al., 2019).

Toward identifying variations around normative trends and linking sleep health with broader population health agendas (Buysse, 2014), examining interindividual differences in change in sleep over time and across developmental periods are critical to better understand development in sleep and identify individuals who are at greater risk for poor health outcomes or more likely to experience enhanced well-being. Accordingly, we utilized latent growth modeling to examine trajectories of four well-acknowledged domains of sleep (duration, efficiency, midpoint, and consistency/variability; Buysse, 2014; Meltzer et al., 2021; Sadeh, 2015) across nine years of development (age 15 to age 24). In order of simplest to most complex patterns of change: (a) sleep efficiency showed a linear pattern indicating improvements over time; (b) sleep midpoint followed a quadratic pattern suggesting a progressively greater preference for eveningness over time; and (c) sleep duration and consistency/variability in sleep duration demonstrated cubic patterns suggesting a directional shift in change over time. Guided by socioecological models of sleep, a secondary aim of the study was to examine variations in trajectories based on participants’ race and sex. Growth trajectories for both sleep duration and sleep efficiency varied significantly across these demographic factors.

Sleep Trajectories

Sleep Duration

The cubic growth model was the best fitting model for capturing sleep duration trajectories spanning adolescence to emerging adulthood. Participants experienced a decrease in sleep duration from age 15 to 17, followed by an increase from age 17 to 22, and another decrease from age 22 to 24. Sleep was shortest at age 17 and longest at age 22. Extending prior work, the cubic trajectory suggests a developmental patterning of change in sleep duration. As suggested by the perfect storm model, declines in sleep duration across adolescence may be the result of biological (e.g., circadian phase delay) and social (e.g., extracurriculars, school schedules) factors that delay sleep onset while simultaneously reducing opportunities for sufficient sleep (Carskadon & Tarokh, 2013; Tarokh et al., 2019). Although it has been hypothesized that these influences persist into emerging adulthood (Carskadon & Tarokh, 2013), our findings suggest meaningful developmental phase shifts. Specifically, we observed increases in sleep duration during emerging adulthood that may be attributed to increased freedom over sleep schedules—for example, college students may choose class times that align with their personal preferences (Maslowsky & Ozer, 2014; Onyper et al., 2012) or working adults may opt for jobs with greater schedule flexibility (Dugan et al., 2022). As emerging adulthood continues, declines in sleep duration may be due to increasing responsibilities related to work and family (e.g., marriage, parenthood) (Maslowsky & Ozer, 2014).

Within the best fitting cubic growth model, we also found significant linear and quadratic slopes. These findings are consistent with prior work that has noted linear declines in sleep duration (Keyes et al., 2015; Park et al., 2019; Saelee et al., 2023). The linear slope indicates that, despite directional shifts over time, sleep duration was ultimately shorter at age 24 than at age 15 in this sample. Notably, this finding does not reflect a decrease in sleep need over time; rather, it underscores a persistent issue of insufficient sleep.

Sleep Efficiency

The linear growth model provided the best fit, indicating that sleep efficiency significantly increased steadily from age 15 to 24. Research on changes in sleep efficiency is inconclusive, but our findings extend prior work showing higher average sleep efficiency in emerging adulthood relative to adolescence (Doane et al., 2015; Evans et al., 2021). Despite improvements in sleep efficiency, we observed co-occurring decrements in other sleep parameters (e.g., duration, midpoint) over this timeframe. This may also reflect a compensatory process whereby individuals adapt to poor sleep in one facet by improving sleep in other domains (Doane et al., 2015).

Chronobiology

Results showed that a quadratic model fit best for sleep midpoint, suggesting a change in the rate of growth over time. Specifically, there was a progressively greater preference for eveningness as adolescents transitioned to emerging adulthood. Notably, the quadratic slope was not significant, despite observing a slight acceleration in the rate of change between each measurement occasion. Still, the best fitting quadratic model for sleep midpoint corroborates and extends prior research comparing average sleep midpoint across repeated measures and cohorts, contributing to a growing body of literature suggesting a shift to greater eveningness across adolescence and emerging adulthood (Fischer et al., 2017; Kuula et al., 2019; Randler et al., 2019; Urner et al., 2009). These findings are consistent with the perfect storm model (Carskadon & Tarokh, 2013; Tarokh et al., 2019) and help address some of the limited understanding of how sleep-related biological and psychosocial processes unfold into emerging adulthood. As noted above, previous work has hypothesized that these influences persist during this transitional period (Carskadon & Tarokh, 2013), though it has remained unclear whether this persistence reflects a plateau in delayed sleep timing following the initial shift in mid-adolescence, or a continued progression toward increasingly later timing. Our findings support the idea that these processes not only persist but may also intensify, contributing to a progressively later sleep midpoint across adolescence and into emerging adulthood.

Relatedly, prior work hypothesizes that chronotype may reflect a biological marker for the end of adolescence with a peak lateness in late adolescence and emerging adulthood (Cooper et al., 2023; Fischer et al., 2017; Kuula et al., 2019; Randler et al., 2019; Roenneberg et al., 2004). Notably, we did not observe a decline or plateau. This raises an important point: the significant lower-order linear slope in the quadratic model suggests a steady delay in sleep timing over time, with no evidence of an inflection point. Compared to prior work, the absence of an inflection point may be attributable to sample or cultural differences (e.g., USA-based vs. Australian-based sample; Cooper et al., 2023), or methodological differences such as modeling estimated trajectories versus comparing average differences across cohorts (e.g., Fischer et al., 2017; Kuula et al., 2019; Randler et al., 2019). It remains possible that further extending assessments would reveal a different trajectory.

Consistency/Variability in Sleep Duration

A cubic growth model provided the best fit to the data for consistency/variability in sleep duration. Variability in sleep duration decreased from age 15 to 16, which was followed by an increase from age 16 to 22 and another decrease from age 22 to 24. Sleep duration was most consistent at age 16 and most variable at age 22. This is the first study to our knowledge to test possible nonlinear trajectories in consistency/variability in sleep duration. Expanding upon previous work reporting an overall increase in variability in sleep duration (Park et al., 2019), the current findings support changes in consistency/variability in sleep duration over time following a developmental phase patterning. Parent-set rules around sleep and monitoring are proposed to support a more consistent sleep routine in adolescence (Nicholson et al., 2023). However, adolescents may experience a spike in independence at age 16 (e.g., related to driving; Keating & Halpern-Felsher, 2008). Greater independence over discretionary activities may be associated with increases in night-to-night fluctuations in sleep (Nicholson et al., 2023). In addition to increasing independence, the brain, including the prefrontal cortex, has yet to reach full maturation during late adolescence and emerging adulthood resulting in underdeveloped executive functions (Daddis, 2011; Taber-Thomas & Pérez-Edgar, 2015; Taylor et al., 2015). Combined with novel environments, this developmental period may see an increase in irregular lifestyle behaviors introducing greater variability in night-to-night sleep (Nicholson et al., 2023). As the brain continues to mature throughout emerging adulthood, fine-tuning of the prefrontal cortex may increase sensitivity to negative consequence and future-oriented decision making (Taber-Thomas & Pérez-Edgar, 2015). These changes coincide with increasing adult responsibilities and may support the adoption of long-term advantageous behaviors such as greater consistency in sleep routines. Nevertheless, these interpretations are tentative and warrant further research.

Within the cubic growth model, we also found a significant linear slope, revealing an absolute decrement in consistency (i.e., increase in variability) in sleep duration from age 15 to 24. This overall increase in variability conforms to the one study to our knowledge that examined trajectories of consistency/variability in sleep duration during this developmental period (Park et al., 2019). We also observed a significant quadratic slope in this sleep parameter, indicating a slight deceleration in the rate of change from age 15 to 17, which was followed by an acceleration in the rate of change into emerging adulthood.

Variance in Sleep Growth Parameters

We found significant variance in multiple sleep growth parameters, reflecting substantial between-person differences in initial levels and rates of change over time. Although the sources of this variability are not fully understood, finding significant variability underscores the need for future research to identify predictors of individual differences in sleep trajectories. Moreover, this variability highlights the utility of sleep as a potential biological marker for identifying individuals who may be at greater risk for—or protected from—maladaptive developmental outcomes. For example, individuals experiencing steeper declines in sleep duration relative to the sample average may be at heightened risk and stand to benefit most from targeted intervention efforts.

Multidimensional Perspective

The divergent trajectories observed across four key sleep parameters have important implications for multidimensional models of sleep health. The current findings underscore prior work suggesting that sleep parameters do not change uniformly over time (Meltzer et al., 2021). Notably, sleep efficiency improved over time, whereas other parameters—sleep duration, sleep midpoint, consistency/variability in sleep duration—showed absolute declines based on their linear slopes. Interestingly, despite a progressive delay in sleep midpoint, we also observed improvements in sleep duration (ages 17–22) and consistency/variability in sleep duration (ages 15–16 and 22–24), suggesting that as sleep timing shifts later, youth may be adapting in ways that help them achieve adequate sleep (e.g., by waking later, improving regularity) during some developmental phases. These discrepancies emphasize the importance of identifying domain-specific patterns of change in sleep over time to advance a more comprehensive understanding of sleep health.

Individual Differences in Trajectories Based on Participants’ Race and Sex

Guided by socioecological models of sleep (Kwon et al., 2024; Meltzer et al., 2021), we examined variations in sleep growth trajectories by participants’ race and sex. By using normative sleep trajectories in the full sample as a meaningful foundation, multigroup moderation analyses provide insight into understanding which groups are more likely to experience sleep disparities. The salience of these demographic characteristics may vary across development, underscoring the importance of identifying when and how race- and sex-based disparities emerge or widen over time.

We found significant differences in sleep duration and sleep efficiency trajectories by participants’ race and sex, whereas no differences emerged for chronobiology or consistency/variability in sleep duration. Related to sleep duration and variations by race, we observed differences in the best fitting growth model and significant differences in the intercept and linear slope. Specifically, we observed a cubic pattern of change for EA participants and a linear pattern for AA participants. Although AAs had shorter sleep at age 15 relative to EAs, AAs experienced a nonsignificant increase in sleep duration over time whereas EAs experienced a significant decline in sleep duration, reducing the racial disparity in sleep duration at age 24. Regarding sleep efficiency and variations by race, we observed significant differences in the intercept. Specifically, AAs experienced significantly less efficient sleep at age 15 relative to EAs. Although rate of change was significantly different across racial groups, AA participants’ sleep efficiency increased at a slightly faster rate than EA participants’ sleep efficiency, minimizing the discrepancy in sleep efficiency by age 24. Across both sleep parameters, there was considerable individual variation in how sleep duration and sleep efficiency changed over time for EA, but not AA, participants.

Extending the well-documented literature on racial disparities in sleep (El-Sheikh et al., 2022a; Guglielmo et al., 2018; Johnson et al., 2019), our findings suggest that AA participants experience worse initial sleep (short, inefficient) in adolescence, and despite experiencing improvements over time, continue to experience less optimal sleep into emerging adulthood. Moreover, we found limited variability in changes in sleep for AA participants. These findings underscore the prevalence of institutional and structural barriers experienced by AA samples and potential impacts on sleep disparities. For instance, AA samples historically experience greater socioeconomic risk and have access to fewer psychosocial and material resources (Shrider & Creamer, 2023; Syed & Mitchell, 2016; Williams & Deutsch, 2016), which have been associated with less optimal sleep (Fuller-Rowell et al., 2021; Etindele Sosso et al., 2021). Relatedly, prior work suggests that even if AA samples acquire more and higher quality education, they do not reap the same health benefits as their EA counterparts (Assari, 2018; Gaydosh et al., 2018; Whiting & Bartle-Haring, 2022). Studies have also shown that neighborhood safety in childhood (Fuller-Rowell et al., 2021) and experiences of discrimination and racism are associated with sleep disparities in adolescence and young adulthood (Davenport et al., 2021; Slopen & Williams, 2014; Slopen et al., 2016; Yip et al., 2020). There are also indications that AA samples may not experience a prototypical emerging adulthood (e.g., less likely to attend college, more likely to be working class and contributing to the family household; Syed & Mitchell, 2016), which may give rise to a different developmental pattern of growth. Toward reducing sleep disparities, it will be necessary to identify the wide array of institutional and structural barriers likely associated with sleep disparities (e.g., SES, labor market, regional culture, neighborhood; Johnson, 2019; Fuller-Rowell et al., 2017; Mayne et al., 2021; Slopen et al., 2016). Moreover, prior work has shown considerable within-group variability in sociocultural experiences (e.g., SES, discrimination; Abdallah et al., 2024; Loyd et al., 2024; Volpe et al., 2022) that would suggest individual differences in risk. Identification of risk and protective factors associated with improvements and decrements in sleep within AA samples is warranted for offsetting risk and rectifying institutional and structural barriers (Guglielmo et al., 2018; Loyd et al., 2024; Thompson et al., 2024b; Volpe et al., 2022; Williams & Deutsch, 2016).

Turning to differences in trajectories based on participants’ sex, we observed differences in the best fitting growth model and significant differences in the intercept and linear slope for sleep duration. Specifically, we observed a cubic pattern of change for female participants and a linear pattern for male participants. Males also had substantially shorter sleep at age 15 and a less steep slope relative to females. Because neither group exhibited significant linear change over time, sex-based disparities in sleep duration remained relatively stable over time. There was also significant variation in sleep duration over time for female, but not male, participants. Turning to sleep efficiency, males experienced considerably worse sleep at age 15 than females; yet males experienced a faster rate of improvement in sleep efficiency over time compared to females, narrowing sex differences by age 24.

Research on differences in sleep based on participants’ sex is inconclusive. Adding to this literature, our findings are consistent with prior work that shows females experience longer, more efficient sleep than male emerging adults (Machado et al., 2021a; Kuula et al., 2019; Park et al., 2019; Randler et al., 2019). Sex differences in sleep trajectories may be related to various biological and environmental factors. Males may stay up later despite having to wake up at similar times as a result of hormonal shifts (Olds et al., 2010). Additionally, greater screen time (Olds et al., 2010) and more negative attitudes toward sleep (Ruggiero et al., 2019) among males may increase risk for less optimal sleep patterns. Despite females generally experiencing better sleep, we found a more rapid improvement in sleep efficiency among males. At a biological level, ovarian steroid secretion during puberty, the menstrual cycle, and pregnancy have been associated with sleep problems and may explain the faster improvement in sleep efficiency among male participants (Franco et al., 2020). Notably, explanations are speculative, and more research is needed to fully understand sex differences.

Limitations and Future Directions

Several limitations should be noted. First, although the sample was sociodemographically and racially diverse, findings may not be generalizable to other samples (e.g., clinical samples, urban samples, affluent families, individuals of other racial groups). Relatedly, the small sample sizes of AA female (n = 53) and AA male (n = 36) participants limited our ability to examine sleep trajectories across intersecting identities. Although we conducted sensitivity analyses that controlled for race in gender moderation analyses and vice versa, these findings do not capture the full complexity of intersectionality. We encourage future research to investigate sleep disparities at the intersection of race, sex, and other salient identities that may underlie differential sleep trajectories. Second, there was a 5-year gap in data collection between Time 3 (age 17) and Time 4 (age 22), and we cannot account for changes in sleep during this time. Likewise, our measurement occasions were bound by ages 15 and 24, so changes in sleep before or after these ages remain unexamined. Although this study extends prior work examining developmental trajectories of sleep from childhood to adolescence (Boatswain-Jacques et al., 2023; Thompson et al., 2024a), it will be necessary for future work to further extend growth trajectories, especially into adulthood. For instance, prior work comparing average sleep midpoint found earlier midpoints in adulthood compared to early adulthood (Fischer et al., 2017; Kuula et al., 2019). Third, some changes in sleep parameters over time appeared small. However, these differences may still correspond to disruptions in sleep architecture or stages (e.g., reduction in slow wave sleep [SWS] and rapid eye movement [REM] sleep; Burman & Muzumdar, 2020; Colten & Altevogt, 2006), which are typically examined via polysomnography. Individuals who awaken frequently, even if briefly, during early sleep stages may have difficulty progressing into deeper stages of SWS and REM sleep (Carskadon & Dement, 2005; Colten & Altevogt, 2006). Moreover, time spent in deeper sleep stages declines with age and arousal thresholds are lower in early sleep stages, which may increase the frequency of brief awakenings with age (Colten & Altevogt, 2006). These awakenings, even if brief, disrupt sleep architecture and sleep efficiency, preventing individuals from obtaining adequate sleep needed to support emotional and cognitive functioning (Fuligni et al., 2021; Goldstein & Walker, 2014; Simon et al., 2020).

Finally, we measured trajectories of change in sleep utilizing latent growth modeling. Because latent growth models isolate between-person differences, these models test rank-order stability in interindividual differences in sleep over time. Although these models are important for identifying individuals most consistently experiencing less optimal sleep, and as a result facing increased risk for poor health outcomes, important questions remain. For example, how does an individual’s sleep change over time relative to their own average or score at a previous time point (Bainter & Howard, 2016)? Underscoring the importance of within-person differences, it is possible that some individuals exhibit declines in sleep efficiency over time while still achieving better sleep efficiency relative to the sample mean. However, these within-person differences are not captured in latent growth modeling, and examining such differences may reveal a different trajectory.

Conclusions

Findings add to a growing literature documenting changes in sleep during the transition from adolescence to emerging adulthood. Extending prior work, we provided a novel assessment of developmental trajectories of objective sleep spanning five waves of data across ages 15 to 24. In summary, we found linear and nonlinear trajectories of growth in sleep parameters. During mid-adolescence, sleep duration declined, consistent with a period of increasing biological pressure toward later sleep timing coupled with externally imposed constraints on sleep opportunity. As individuals transitioned into emerging adulthood, sleep duration partially rebounded; however, this rebound was not sustained and sleep duration again declined across the early twenties. Across this same period, sleep efficiency steadily improved, indicating that although sleep duration fluctuated, individuals became increasingly efficient at sleeping. Sleep timing also shifted later over time, reflecting a growing preference for later bedtimes and wake times as adolescents moved into emerging adulthood. Finally, sleep duration was most consistent during mid-adolescence, became increasingly variable across the transition to emerging adulthood, and then became more consistent again in the early twenties. Notable variations in trajectories also emerged based on participants’ race and sex, highlighting that the salience of demographic characteristics in relation to sleep may vary across developmental stages. Findings (a) underscore developmental effects on sleep-wake patterns during the transition from adolescence to emerging adulthood and (b) show that some individuals consistently experience less optimal sleep over time and may especially need intervention and prevention efforts to reduce risk for negative outcomes.

Public Significance Statement.

Sleep parameters exhibited significant linear and nonlinear growth across adolescence and emerging adulthood. Findings underscore developmental effects on sleep-wake patterns and highlight important between-person variability in growth trajectories, which may help to identify individuals most at risk for subsequent health problems as well as individuals who may be protected from negative outcomes. Moreover, findings show the pervasiveness of racial/ethnic- and sex-based sleep disparities and emphasize the importance of developing therapeutic tools and targets to rectify institutional and structural barriers.

Data Availability and Transparency:

Data will be made available after completion of each longitudinal study in accordance with National Institutes of Health data sharing guidelines. The study design and analysis were not pre-registered. Data and analytic syntax will be shared upon requests submitted by email to the corresponding author.