Sleep and Movement Behaviours in Preschool Children: A Cross‐Sectional Study With Compositional Data Analysis

ABSTRACT

Reciprocal relations between physical activity and sleep in early childhood are inconsistent relative to reports in adolescents and adults. Compositional data analysis research highlights the need to examine 24‐h behaviours holistically. Yet, studies often focus on daytime metrics, neglecting sleep components. This study aimed to determine if the compositions of overnight sleep, 24‐h sleep, and nap sleep stages are associated with physical activity in preschool children and if behaviours of a complete 24‐h cycle (sleep and wake) vary by sex or nap habituality. Actigraphy data pooled from two studies (n = 272 children; 4.2 ± 0.8 years) provided estimates of sleep and physical activity. Night and 24‐h sleep composition included sleep onset latency, duration, and wake after sleep onset. Nap sleep was measured in 31 children with polysomnography (wake and non‐REM sleep stages 1, 2 and 3). Nap sleep stage compositions were not associated with movement behaviours. Six multivariate regression models explored differences in compositional 24‐h time use between sex and nap habitually groups. Time‐use compositions that included both wake and sleep components varied by age, sex, and nap habituality for all components except total sleep time. This study demonstrates the value of CoDA for understanding 24‐h behaviour patterns, revealing that nap habits, age, and sex are linked to specific sleep and activity components in preschoolers. Future research should explore these compositional associations in more diverse populations, consider additional physical activity indicators, and incorporate overnight polysomnography assessments.

1. Introduction

Understanding the impact of sleep and other daily behaviours on health outcomes requires analytic methods that reflect the co‐dependent nature of time‐use (Willumsen and Bull 2020; Pedišić et al. 2017). Because time spent in one behaviour constrains the time available for others, compositional data analysis (CoDA) has emerged as a statistical approach to capture the relative distribution of behaviours within a 24‐h day (Pedišić et al. 2017; Dumuid et al. 2020). Unlike traditional models that treat behaviours as independent variables, CoDA accounts for the constant‐sum constraint of time‐use data, reducing multicollinearity and allowing for more accurate interpretations. The CoDA approach reduces the risk of multicollinearity by transforming the relative behaviours into ratios (e.g., isometric log ratios) before entering them into traditional statistical models (Chastin et al. 2015; Dumuid et al. 2018). Additionally, isotemporal substitution modelling enables researchers to examine how theoretical time reallocations between behaviours affect health outcomes, providing actionable insights for intervention and public health messaging (Dumuid et al. 2019).

While CoDA has gained traction in modelling time‐use behaviours, its application in sleep science remains limited and underdeveloped. A key gap is the tendency to treat sleep as a single, uniform behaviour, typically represented by total sleep time or time in bed, despite the multidimensional nature of sleep (Brown et al. 2024; Buysse 2014). Core sleep metrics such as sleep onset latency, wake after sleep onset (WASO), sleep efficiency, and time spent in different sleep stages (e.g., NREM, REM) are inherently co‐dependent and thus could be more appropriately modelled as a composition (Pawlowsky‐Glahn et al. 2015; Dumuid, Stanford, et al. 2017). Although sleep efficiency considers multiple sleep metrics, it is an aggregated measure, and CoDA allows researchers to consider a sleep profile. Additionally, most studies overlook daytime sleep (e.g., napping), which plays a critical role in cognitive and physical development in younger populations and may hold unique associations with health outcomes independent of overnight sleep (Thorpe et al. 2015; Kurdziel et al. 2013). Objective sleep measures, such as those derived from actigraphy or polysomnography, are underutilised in CoDA‐based studies, despite offering the granularity needed to explore complex intra‐sleep dynamics (Brown et al. 2024; Zahran et al. 2023; Janssen et al. 2020; Kuzik et al. 2025). Furthermore, few studies leverage isotemporal substitution within CoDA to examine how reallocating time across different sleep components may influence health (Brown et al. 2024). These gaps underscore a significant opportunity for sleep scientists to use CoDA to more fully capture the compositional and interdependent nature of sleep behaviour, leading to more nuanced and actionable insights into sleep‐health relationships.

The overarching aim of this study was to demonstrate the utility of CoDA to analyse co‐dependent sleep metrics and their potential for advancing sleep health research through a holistic lens. Sleep and movement behaviours (physical activity and sedentary time) are ‘time‐use’ behaviours often explored in isolation, despite being inherently co‐dependent components of a daily cycle (Dumuid et al. 2020; Rosenberger et al. 2019). Research in adults and adolescents has shown that these behaviours interact in meaningful ways: physical activity can influence circadian rhythms, thermoregulation, and mood, while sleep quality and quantity can affect energy levels, alertness, and motivation to be active (Chennaoui et al. 2015; Kline et al. 2021). Additionally, sleep physiology (e.g., non‐rapid eye movement stage 3, or N3 sleep) could influence physical activity levels (e.g., greater sleep depth and quality may contribute to less daytime sleepiness and better mood and energy to engage in physical activity) (Chennaoui et al. 2015; Dworak et al. 2008).

However, the relations between sleep and physical activity in preschool‐aged children are less well understood (Antczak et al. 2020; St Laurent et al. 2021). Existing studies have produced mixed findings, with most research focusing on how physical activity influences sleep, rather than examining sleep's potential role in shaping physical activity patterns. Also notable, most previous 24‐h CoDA studies in preschool children do not include or separate daytime sleep in their analyses. During the preschool years (ages ~3–5), children undergo a key transition from biphasic sleep (naps plus overnight sleep) to monophasic sleep (overnight only), though this shift varies widely among individuals and may reflect a broader developmental milestone (Spencer and Riggins 2022). While most children stop napping by age five, the reasons for continued or discontinued napping differ, ranging from physiological need to behavioural or contextual factors (Newton and Reid 2023; Newton et al. 2020).

Napping patterns are also shaped by environmental influences, including family routines, childcare attendance, and potentially physical activity behaviours. Further, relations of sleep macrostructure (proportion of sleep time in sleep stages) and movement behaviours have been scarcely explored in preschoolers (St Laurent et al. 2022). Notably, studies have shown that physical activity levels and sedentary behaviours differ by sex in early childhood, suggesting that sex‐specific patterns may also shape the development of sleep and activity patterns (Nilsen et al. 2019; Schmutz et al. 2018; Kwon et al. 2022; Hinkley et al. 2010). Given that both sleep and physical activity undergo rapid developmental changes during this period, age and sex are important factors to consider when examining 24‐h behaviour. These factors may influence not only the timing and duration of sleep and activity but also the nature of their interaction during early childhood.

Therefore, in the present study, we explore relations of movement behaviours, sleep subcomponents, and nap behaviour in preschool children to illustrate some applications with CoDA. First, we demonstrate an approach to assess differences in compositions by exploring whether behaviours of a complete 24‐h cycle (i.e., including both sleep and wake components) varied by nap habituality in preschool children. To exemplify associational modelling with CoDA, we explored relations between nap sleep stages and daytime movement behaviours.

2. Participants and Methods

2.1. Study Overview

The current cross‐sectional study is a secondary data analysis of data stemming from two parent studies that followed similar protocols to examine how daytime naps affect memory consolidation (ClinicalTrials.gov ID: NCT03285880). The parent study protocols were approved by the University of Massachusetts Amherst Institutional Review Board and the University of Maryland Institutional Review Board. Parents or legal guardians (caregivers) completed written consent and parental permission, and verbal assent was obtained from the child participants. Both parent studies used a within‐subjects design over 2 weeks (Spencer et al. 2016). Briefly, both included two conditions (1 day with nap promotion and 1 day with wake promotion). Children were asked to wear an activity monitor watch on their non‐dominant wrist for the full study period, and caregivers completed a questionnaire. In the Maryland study, sleep physiology was also measured during the nap on the nap promotion day (Allard et al. 2021).

2.2. Participants

Children of US preschool age (i.e., 2 years, 9 months to 5 years, 11 months) were recruited from childcare centres and database lists in Massachusetts and Maryland (Figure 1). Inclusion criteria for the parent studies were (1) normal or corrected‐to‐normal vision and hearing, (2) no current or past diagnosis of a developmental disability or sleep disorder, (3) no use of psychotropic or sleep‐affecting medication, and (4) no history of neurological injury or presence of metal in the body (Maryland study only). For the present study, only participants with sufficient actigraphy data were included.

FIGURE 1.

Participant flow diagram. (PSG = polysomnography).

2.3. Measurements

2.3.1. Actigraphy

Daytime movement behaviours and sleep were derived from actigraphy with Actiwatch Spectrum watches (Philips Respironics, Bend, OR). Sleep and wake were estimated from these water‐resistant watches through a combination of triaxial accelerometry (to estimate movement), off‐wrist detection (to determine non‐wear time, which was excluded from analyses), light detection, and an event marker button. These monitors are a common instrument for sleep estimation in children (Meltzer et al. 2012), and a validity study conducted in preschool children by Sitnick et al. (2008) reported overall agreement, sensitivity, and specificity metrics of 94%, 97%, 24% with video‐polysomnography, respectively. The validity of this monitor to estimate daytime energy expenditure relative to indirect calorimetry was initially examined in preadolescent children (Ekblom et al. 2012). In a sample of preschool children, Alhassan et al. (2017) did a cross‐validation study with direct observation using threshold activity counts proposed by Ekblom et al. (2012) study to estimate intensity categories.

2.3.2. Actigraphy Protocol

Actiwatch data was configured in the Actiware software (Philips Respironics, Bend, OR) with a sampling rate of 32 Hz, sensitivity < 0.01 g, and 15‐s epochs. Children and caregivers were asked to press the event marker button on the watch at ‘lights out’ and ‘lights on’ to mark the time that children tried to start falling asleep and when they woke up. Caregivers were also asked to record times in and out of bed, bedtimes, and wake times on a daily sleep diary. Data from the monitors were downloaded and then processed in Actiware with the default algorithm (Oakley 1997), which has been validated with preschool children (Meltzer et al. 2015; Hyde et al. 2007) to classify each epoch as sleep, wake, or off‐wrist. A daily cycle was defined as the morning wake time to the next morning wake time, and therefore, the average daily cycle time for each day may not be exactly 24 h. Each daily cycle was then split into overnight rest, daytime rest (if applicable), and daytime wake periods. Participants with at least three daily cycles of data were included (Hinkley et al. 2012).

2.3.3. Sleep Measures

Overnight and daytime sleep periods were scored for sleep metrics using a combination of event markers and sleep diaries in Actiware to set the time in bed periods. If neither of these was present or the times did not correspond, the first three consecutive minutes and the last five consecutive minutes, defined as sleep by the Actiware algorithm, were used to set the sleep onset and sleep offset, respectively. However, in the current study, we only included actigraphy for participants who had either event markers or sleep diary data, as sleep onset is necessary to better estimate sleep onset latency. These intervals were then manually set as rest intervals and rescored in Actiware to calculate sleep onset latency (the time it took to fall asleep in minutes), total sleep time (actual time asleep after ‘lights out’ in minutes), and wake after sleep onset (WASO: sum of time awake during the interval in minutes). These variables were calculated for sleep over a daily cycle (referred to here as 24‐h sleep, although daily cycles of morning wake to wake may span more or less than 24 h). Sleep intervals that had excluded periods (e.g., subsequent off‐wrist epochs) were not included in the analysis.

2.3.4. Movement Behaviours

Although at least 600 min/day is a commonly recommended minimum accelerometry wear time in young children for daytime studies (Migueles et al. 2017), most of our sample had 60 to 120 min of daytime sleep. Therefore, days with daytime wake actigraphy for at least 480 min were included in this study. The wake intervals not classified as rest or off‐wrist were further processed to derive waking movement behaviours. Total activity was calculated as the daily average actigraphy‐measured activity counts per minute (Bassett et al. 2015; Renninger et al. 2020). Each 15‐s wake epoch was also classified into an intensity category using the Ekblom cut points (less than 70 counts = physical inactivity, 80–261 counts = light intensity physical activity, and 262 or more = MVPA) (Ekblom et al. 2012). These epochs were then summed for each day and expressed as the total minutes inactive, in light physical activity, and in MVPA.

2.4. Nap Sleep Stages

In the Maryland parent study, nap sleep stages were measured on the afternoon of the nap‐promotion condition in the participants' home. The Embletta MPR (Natus) ambulatory system, with a 14‐electrode polysomnography (PSG) montage applied with the 10–20 system, was used. The montage included two EOG leads, two chin EMG leads, and 10 cortical EEG leads (F3, F4, C3, C4, CZ, O1, O2, M1, M2, Ground), with electrodes referenced to Cz. Following the American Academy of Sleep Medicine guidelines for sleep scoring, each 30‐s epoch was scored as N1, N2, or N3, rapid eye movement (R), or wake during nap time in bed (Berry et al. 2012).

2.5. Covariates

To describe the characteristics of the sample and include the models as potential confounders, sociodemographic variables were obtained from the caregiver questionnaire. Variables included age, sex, race and ethnicity. Nap frequency was calculated from the actigraphy data (number of days with nap sleep/number of days with usable actigraphy data × 7) or obtained from the caregiver questionnaire if there was missing actigraphy data, and then categorised into the following groups: rarely or never (non‐nappers: 0–1 naps/week), sometimes (2–4 naps), or regular nappers (5+ naps/week). Although nap frequency categories are not assigned consistently across studies, we selected this common classification to align with previous work (Kurdziel et al. 2013; Newton and Reid 2023; St Laurent et al. 2022).

2.6. Statistical Analyses

Analyses were conducted in Stata (Version 18.5, StataCorp LLC, College Station, TX) and in R version 4.1.1 (Team RC 2017), using the ‘compositions’ (van den Boogaart and Tolosana‐Delgado 2008), ‘car’ (Fox and Weisberg 2011), ‘codaredistlm’ (Nakazawa 2024), and ‘fsmb’ (Nakazawa 2024) packages with an alpha level of 0.05.

2.6.1. 24‐h and Nap Compositions

We first created the ‘time‐use’ compositions of interest to explore co‐dependent variables in a CoDA approach (Dumuid et al. 2018). Compositions were created for the 24‐h cycle (6 parts: inactive time, light physical activity, MVPA, sleep onset latency, total sleep time, and WASO) and nap sleep stages (4 parts: N1, N2, N3, wake). Compositional components with zeros present cannot be directly transformed into isometric‐log ratio (ilr) coordinates. Although methods to address components with zero values have been proposed (Rasmussen et al. 2020), given that only 9% (n = 3) of the current sample with PSG measures had time in R sleep (i.e., 1.5, 2 and 7.5 min) and that R sleep during naps in preschool children is not common (Kurth et al. 2016), we opted to exclude R sleep from the nap sleep compositions in the present study. Additionally, wearing the PSG equipment, along with the presence of the research staff, may have artificially increased sleep onset latency, so that metric was also excluded. Therefore, sleep onset latency and R sleep (if present) were subtracted from the total nap time in bed.

Proportions of each component were calculated for each day using total wear time as the denominator for the 24‐h cycle, and time in bed was used for the nap composition. These proportions were averaged across each participant's wear period. Geometric means were calculated for each behaviour within a composition and then used to normalise the component to sum to 1440 min for the 24‐h composition and to 100% of sleep time for the nap sleep composition (Pawlowsky‐Glahn et al. 2015). Components of each composition were then expressed as a set of ilr coordinates (Egozcue et al. 2003). This transformation results in one less ilr coordinate than the number of components. Thus, the 24‐h cycle had a set of five ilr coordinates, and the nap sleep composition had a set of three ilr coordinates.

2.6.2. Participant Characteristics

Sociodemographic variables were summarised with descriptive statistics (i.e., means, standard deviations, and proportions) for the overall sample and stratified by nap habituality and sex. Summary statistics of the compositional components included arithmetic means and standard deviations for the non‐transformed data and geometric means and variance matrices for the compositional data. To visually explore variations in time use by the above‐mentioned group factors, radar plots were created.

2.6.3. Differences in 24‐h Compositions by Nap Habitually and Sex

To identify differences in compositional time‐use data between sex and nap habitually groups, a set of six multivariate regression models was used (one for each 24‐h behaviour of interest). The dependent variables were the ilr‐transformed time‐use components (i.e., the 24‐h cycle composition). Six sets of ilr coordinates were created using sequential binary partitioning, with each behaviour being entered as the numerator of the first ilr coordinate (and thus, serving as the behaviour of interest in that model while accounting for the remaining behaviours). Age, sex, and nap habitually group were the independent variables.

2.6.4. Nap Sleep Composition and Movement Behaviours

To explore associations between nap sleep compositions and movement behaviours, three multiple linear regression models followed the approach described in the methods paper by Dumuid, Pedišić, et al. (2017). Movement behaviour outcomes included total physical activity (expressed as mean activity counts/min), MVPA (% wake time), and inactive time (% wake time). The nap sleep composition ilr coordinates were entered into each model as independent variables. Although it may depend on the research question of interest, a common approach for CoDA is to first determine if a composition overall is associated with an outcome, and if so, is temporal analysis is conducted as a post hoc analysis for interpretation (Dumuid et al. 2019). Therefore, in the present paper, if the composition was significantly associated with any of the movement behaviour outcomes, we planned to explore theoretical time reallocation predictions. Specifically, we planned to explore estimated changes in movement behaviour outcomes for 5‐min time reallocations from one sleep stage to one of the other stages.

3. Results

3.1. Participant Characteristics

Summary descriptives of the participant characteristics and 24‐h behaviours are presented in Table 1. In the overall sample, the average actigraphy data collection period was 10.2 ± 3.3 days for wake behaviours (7.4 ± 2.3 weekdays and 2.7 ± 1.3 weekend days) and 10.9 ± 3.4 nights for sleep metrics (8.1 ± 2.6 weekday nights and 2.9 ± 1.2 weekend nights). Participants wore the actigraphy watches for an average of 732.9 ± 63.1 min during the day. The average nap total sleep time was 89.9 ± 23.3 min, and the average nap frequency was 3.8 ± 2.1 days/week, ranging from no naps to every day of the week. Among those with nap PSG (n = 31), 15 were males, and 16 were females. In the full sample, 58% met the World Health Organization (WHO) recommendation of at least 10 h of daily sleep duration, and 94.2% met the WHO recommendation of at least 180 min of total physical activity (with at least 60 min of MVPA).

TABLE 1.

Summary statistics of characteristics and sleep metrics for the overall sample and stratified by sex and nap habits.

Variable	Full sample (n = 274)	Males (n = 157)	Females (n = 117)	Rarely naps (n = 52)	Sometimes naps (n = 112)	Regularly naps (n = 110)
Socio‐demographic
Age (years)	4.2 (0.8)	4.2 (0.7)	42 (0.8)	4.6 (0.7)	4.3 (0.7)	4.0 (0.7)
Sex
Male	157 (57.3)	N/A	N/A	28 (53.8)	64 (57.1)	65 (59.1)
Female	117 (42.7)	N/A	N/A	24 (46.2)	48 (42.9)	45 (40.9)
Study
Massachusetts	241 (88.0)	140 (58.1)	101 (41.9)	44 (84.6)	99 (88.4)	98 (89.1)
Maryland	33 (12.0)	17 (51.2)	16 (48.5)	8 (15.4)	13 (11.6)	12 (10.9)
Race
White	172 (62.8)	98 (62.4)	74 (63.2)	38 (73.0)	60 (53.6)	74 (67.3)
Black/African American	23 (8.4)	14 (8.9)	9 (7.7)	1 (1.9)	15 (13.4)	7 (6.4)
HA/PI	1 (0.4)	0	1 (0.8)	0	0	1 (0.9)
Asian	6 (2.2)	3 (1.9)	3 (2.6)	2 (3.9)	2 (1.8)	2 (1.8)
2+ races	31 (11.3)	18 (11.5)	13 (11.1)	3 (5.8)	14 (12.5)	14 (12.7)
Other	16 (5.8)	11 (7.0)	5 (4.3)	2 (3.9)	11 (9.8)	3 (2.7)
Missing	25 (9.1)	13 (8.3)	12 (10.3)	6 (11.5)	10 (8.9)	9 (8.2)
Hispanic
Yes	57 (20.8)	34 (21.7)	23 (19.7)	4 (7.7)	36 (32.1)	17 (15.5)
No	194 (70.8)	112 (71.3)	82 (70.1)	42 (80.8)	68 (60.7)	84 (76.4)
Missing	23 (8.4)	11 (7.0)	12 (10.3)	6 (11.5)	8 (32.2)	9 (8.2)
Nap habits
Weekly nap frequency	3.8 (2.1)	3.8 (2.1)	3.7 (2.1)	0.5 (0.5)	3.3 (0.8)	5.7 (0.7)
Nap frequency category
Regular	110 (40.2)	65 (41.4)	45 (38.5)	0	0	110 (100)
Sometimes	112 (40.8)	64 (40.8)	48 (41.0)	0	112 (100)	0
Non‐nappers	52 (19.0)	28 (17.8)	24 (20.5)	52 (100)	0	0
24‐h sleep
Sleep onset latency (min)	19.4 (13.2)	18.9 (13.2)	19.9 (13.2)	15.7 (12.3)	18.9 (12.1)	21.5 (14.4)
Total sleep time (min)	608.3 (37.8)	605.3 (36.3)	612.4 (39.5)	604.2 (48.1)	609.4 (36.7)	609.2 (33.4)
WASO (min)	55.9 (15.0)	54.9 (14.0)	57.3 (16.3)	54.6 (15.6)	55.7 (15.3)	56.8 (14.6)
Nap sleep stages (n = 31)
N1 (min)	9.8 (7.6)	8.3 (8.0)	11.5 (7.0)	7.7 (6.0)	12.3 (9.2)	8.6 (6.7)
N2 (min)	30.3 (12.9)	32.1 (14.1)	28.5 (11.7)	26.5 (10.7)	30.7 (14.9)	32.7 (12.5)
N3 (min)	47.6 (16.0)	51.5 (16.1)	43.5 (15.3)	42.2 (12.1)	48.5 (16.2)	50.6 (18.5)
WASO (min)	22.7 (12.7)	21.2 (13.6)	24.5 (12.0)	18.8 (11.4)	23.0 (15.5)	25.5 (10.6)
Daytime movement behaviours
Physical activity (counts/min)	581.0 (110.6)	599.7 (118.5)	555.8 (93.6)	601.2 (120.2)	571.2 (111.1)	581.4 (104.8)
Physical inactivity (min)	303.7 (71.1)	304.8 (76.8)	302.2 (62.9)	292.0 (73.7)	323.5 (71.8)	289.1 (64.7)
Light physical activity (min)	311.9 (40.0)	306.3 (38.3)	319.4 (41.2)	324.6 (44.9)	312.7 (37.0)	305.1 (39.3)
MVPA (min)	118.0 (40.5)	126.8 (42.3)	106.3 (34.8)	124.8 (40.4)	119.4 (41.9)	113.4 (38.8)

Note: Means (standard deviations) presented for numeric variables and sample sizes (proportions) presented for categorical variables.

Abbreviations: HA/PI = Native Hawaiian or Pacific Islander; MVPA = moderate to vigorous intensity physical activity; WASO = wake after sleep onset.

3.2. 24‐h and Nap Sleep Compositions

The means of the 24‐h and nap sleep compositions for the overall sample and the sex and nap habitually groups are presented in Table 2. Variation matrices were created to present the spread of the compositional parts (Table 3). Values closer to zero are more dependent on each other, and zero values are exactly proportional (similar to correlations with non‐compositional data). Thus, in the 24‐h behaviour composition, total sleep time and light physical activity were the most co‐dependent. In the nap sleep composition, the highest co‐dependency was between N2 and N3 sleep. Radar plots were used to illustrate the slight variations in the geometric means by nap habituality groups and sex (Figures 2 and 3).

TABLE 2.

Compositional means of the time‐use compositions.

	Full sample	Males	Females	Rarely naps	Sometimes naps	Regularly naps
24‐h behaviours (n = 274)
Sleep onset latency (min)	15.1	14.8	15.6	11.5	14.9	17.4
Total sleep time (min)	628.4	625.8	631.4	625.2	619.2	638.8
WASO (min)	55.9	55.1	56.9	54.7	54.6	57.8
Inactive time (min)	305.9	305.8	305.7	293.8	321.1	296.4
Light physical activity (min)	320.1	314.7	327.3	333.7	316.1	317.7
MVPA (min)	114.6	123.8	103.1	121	114	112.1
Nap sleep stages (n = 31)
N1 (%)	7.1	5.6	8.9	6.7	8.9	5.6
N2 (%)	28.1	28.7	27.1	28.6	27.4	28.3
N3 (%)	45.7	49.1	41.9	47.4	45.9	44
WASO (%)	19.2	16.6	22	17.3	17.8	22.2

Abbreviations: MVPA = moderate to vigorous intensity physical activity; WASO = wake after sleep onset.

TABLE 3.

Variation matrices of the time‐use compositions.

	SOL	Duration	WASO	P‐IN	LPA	MVPA
24‐h behaviours (n = 274)
SOL	0	0.679	0.744	0.827	0.707	0.832
TST	0.679	0	0.075	0.062	0.024	0.152
WASO	0.744	0.075	0	0.149	0.093	0.192
ST	0.827	0.062	0.149	0	0.092	0.292
LPA	0.707	0.024	0.093	0.092	0	0.127
MVPA	0.832	0.152	0.192	0.292	0.127	0

	N1	N2	N3	WASO
Nap sleep stages (n = 31)
N1	0	1.001	1.16	0.812
N2	1.001	0	0.366	0.822
N3	1.16	0.366	0	0.551
WASO	0.812	0.822	0.551	0

Abbreviations: LPA = light intensity physical activity; MVPA = moderate to vigorous intensity physical activity; P‐IN = physically inactive time; SOL = sleep onset latency; total PA = activity counts/min; TST = total sleep time; WASO = wake after sleep onset.

FIGURE 2.

Radar plots for mean nap habituality groups and sex.

FIGURE 3.

Radar plots for mean 24‐h compositions of sex (a) and nap habituality (b) groups.

3.3. Sex, Nap Habituality, and 24‐h Behaviour Composition

Results from the six multivariate regression models run to identify differences in compositional time‐use data between sex and nap habitually groups, while adjusting for age, are presented in Table 4. For each model, the overall MANOVA revealed a significant multivariate effect of age, sex, and nap group on the time‐use composition.

TABLE 4.

Multivariate regression model results for the 24‐h behaviours as compositional outcomes (n = 247).

	MANOVA model				ilr 1 univariate effects			ilr 1 model effects
	Wilk's Λ	df	F	p	df	F	p		Estimate	95% CI
Sleep onset latency
Age	0.84	5, 265	10.1	< 0.001	1, 269	2.8	0.09	Age	−0.05	−0.18 to 0.07
Sex	0.86	5, 265	8.3	< 0.001	1, 269	0.5	0.49	Female	0.07	−0.11 to 0.26
Nap group	0.90	10, 530	2.8	0.002	2, 269	3.6	0.03	Sometimes nappers	−0.13	−0.33 to 0.08
								Non‐nappers	−0.36	−0.62 to −0.09
Total sleep time
Age	0.84	5, 265	10.1	< 0.001	1, 269	1.4	0.23	Age	−0.02	−0.05 to 0.005
Sex	0.86	5, 265	8.3	< 0.001	1, 269	0.02	0.41	Female	0.01	−0.03 to 0.06
Nap group	0.90	10, 530	2.8	< 0.001	2, 269	0.06	0.12	Sometimes nappers	−0.001	−0.05 to 0.05
								Non‐nappers	0.05	−0.003 to 0.12
Wake after sleep onset
Age	0.84	5, 265	10.1	< 0.001	1, 269	6.8	0.009	Age	−0.06	−0.11 to 0.02
Sex	0.86	5, 265	8.3	< 0.001	1, 269	1.4	0.24	Female	0.04	−0.03 to 0.11
Nap group	0.90	10, 530	2.8	0.002	2, 269	0.8	0.45	Sometimes nappers	−0.02	−0.09 to 0.06
								Non‐nappers	0.04	−0.06 to 0.14
Physical inactivity
Age	0.84	5, 265	10.1	< 0.001	1, 269	0.4	0.54	Age	−0.03	−0.09 to 0.03
Sex	0.86	5, 265	8.3	< 0.001	1, 269	0.03	0.86	Female	0.004	−0.08 to 0.00
Nap group	0.90	10, 530	2.8	0.002	2, 269	3.5	0.03	Sometimes nappers	0.12	0.03–0.21
								Non‐nappers	0.07	−0.04 to 0.19
Light physical activity
Age	0.84	5, 265	10.1	< 0.001	1, 269	5.3	0.02	Age	0.02	−0.007 to 0.06
Female	0.86	5, 265	8.3	< 0.001	1, 269	5.1	0.03	Female	0.05	0.002–0.09
Nap group	0.90	10, 530	2.8	0.002	2, 269	5.0	0.007	Sometimes nappers	0.01	−0.04 to 0.07
								Non‐nappers	0.1	0.04–0.17
Moderate to vigorous physical activity
Age	0.84	5, 265	10.1	< 0.001	1, 269	29.7	< 0.001	Age	0.14	0.08–0.2
Female	0.86	5, 265	8.3	< 0.001	1, 269	17.1	< 0.001	Female	−0.19	−0.2 to −0.09
Nap group	0.90	10, 530	2.8	0.002	2, 269	0.8	0.43	Sometimes nappers	0.009	−0.08 to 0.1
								Non‐nappers	0.08	−0.05 to 0.2

Note: Models were run with the six sets of five ilr coordinates (rotating the 24‐h behaviour of focus) with the same independent variables. In the ilr 1 model effects section, male and regular nappers served as the reference groups for sex and nap group, respectively. Model effects for ilr 2 through ilr 5 are not shown. Bold text indicates a p‐value < 0.05.

3.3.1. Sleep Onset Latency

In the model with sleep onset latency as the focus, follow‐up univariate tests indicated that nap group was significantly associated with ilr 1 (F(2, 269) = 3.6, p = 0.03), which considered sleep onset latency while accounting for the other time‐use components. Neither age (F(1, 269) = 2.77, p = 0.097) nor sex (F(1, 269) = 0.47, p = 0.495) was significantly associated with sleep onset latency. Post hoc pairwise comparisons using estimated marginal means (with Tukey adjustment) showed that regular nappers had significantly higher ilr 1 values than non‐nappers (mean difference = 0.23, 95% CI [0.04, 0.67], p = 0.02) while accounting for the other compositional components. A back transformation to the original composition scale to aid with the interpretation showed that the geometric mean of sleep onset latency for regular nappers was 14.4 min, whereas for non‐nappers it was 12.9 min.

3.3.2. Total Sleep Time

In the model with total sleep time as the focus, follow‐up univariate tests indicated that none of these independent variables were associated with total sleep time, indicating this sleep component did not significantly differ by levels of age, sex, or nap category. Therefore, no post hoc comparisons were calculated for total sleep time.

3.3.3. Wake After Sleep Onset

In the model for WASO as the focus, follow‐up univariate tests indicated that age was significantly associated with ilr 1 (F(1, 269) = 6.8, p = 0.009). Post hoc analyses indicated that age was significantly associated with the ilr 1 coordinate (B = −0.06, 95% CI [−0.110, −0.016], p = 0.009), implying that for each additional year of age, the WASO ilr decreased by approximately 0.063 units, holding sex and nap group constant. This translates to a predicted compositional mean of 57.3 and 53.6 min of WASO for 4‐ and 5‐year‐old children, respectively.

3.3.4. Inactive Time

In the model with inactive time as the focus, follow‐up univariate tests indicated that nap group was significantly associated with ilr 1 (F(2, 269) = 3.5, p = 0.03). Neither age (F(1, 269) = 0.04, p = 0.54) nor sex (F(1, 269) = 0.03, p = 0.86) was significantly associated with inactive time. Post hoc pairwise comparisons using estimated marginal means showed that regular nappers had higher ilr 1 values than sometimes nappers (mean difference = −0.12, 95% CI [−0.23, −0.01], p = 0.02) while accounting for the other compositional components. A back transformation to the original composition scale to aid showed that the predicted geometric mean of inactive time for regular nappers was 337.28 min, whereas for sometimes nappers it was 283.68 min.

3.3.5. Light Physical Activity

In the model with light physical activity as the focus, follow‐up univariate tests indicated that age (F(1, 269) = 5.3, p = 0.02), sex (F(1, 269) = 5.1, p = 0.03), and nap group (F(2, 269) = 5.0, p = 0.007) were significantly associated with ilr 1. However, post hoc analyses revealed that only sex remained significant with light physical activity. Pairwise comparisons using estimated marginal means showed that males had higher ilr 1 values than females (mean difference = −0.05, 95% CI [−0.09, −0.003], p = 0.03). Back transformations showed that the predicted geometric mean of light physical activity for males was 361.44 min, versus 349.92 min for females.

3.3.6. Moderate to Vigorous Physical Activity

In the model with MVPA as the focus, follow‐up univariate tests indicated that age (F(1, 269) = 29.8, p < 0.001) and sex (F(1, 269) = 17.1, p < 0.001) were significantly associated with ilr 1. Post hoc analyses indicated that age was significantly associated with the ilr 1 coordinate (B = 0.14, 95% CI [0.08, 0.20], p < 0.001), indicating that for each additional year of age, the MVPA ilr increased by approximately 0.14 units, holding sex and nap group constant. This translates to a predicted compositional mean of MVPA for 4‐ and 5‐year‐old children of 110.9 and 128.1 min, respectively. Pairwise comparisons using estimated marginal means showed that males had lower ilr 1 values than females (mean difference = 0.18, 95% CI [0.09, 0.27], p < 0.001). Back transformations showed that the predicted geometric mean of MVPA for males was 80.6 min, versus 139.7 min for females.

3.4. Nap Sleep Compositions and Movement Behaviours

Although ilr‐transformed nap composition varied among participants, it was not significantly associated with total physical activity (R ² = 0.018, F(3, 270) = 1.67, p = 0.17), MVPA (R ² = 0.0177, F(3, 270) = 1.62, p = 0.18), or inactive time (R ² = 0.0089, F(3, 270) = 0.81, p = 0.49) in any model (Table 5). Therefore, isotemporal substitutions were not calculated.

TABLE 5.

Type II ANOVA results of multiple linear regression models exploring movement behaviours as the outcome (n = 31).

Total activity		MVPA		Physical inactivity
df, F	p	df, F	p	df, F	p
3, 2.5	0.059	3, 1.6	0.18	3, 0.81	0.49

Abbreviations: df = degrees of freedom; MVPA = moderate to vigorous intensity physical activity.

4. Discussion

4.1. Overall Findings

In this cross‐sectional study of preschool children, as an illustrative example of CoDA in preschool children, we explored variations of 24‐h behaviour components by sex and nap habits, as well as nap sleep stage compositions and movement behaviours, through a CoDA approach. In this sample of young children, sleep onset latency and inactive time differed significantly by nap group, with regular nappers showing longer sleep onset and more inactive time than non‐nappers or sometimes nappers. Wake after sleep onset decreased with age, while light physical activity was higher in males, though only sex remained significant in post hoc tests. Moderate to vigorous physical activity increased with age and was also higher in females than males. However, nap composition was not significantly associated with total physical activity, MVPA, or inactive time, and therefore, no isotemporal substitution models were conducted.

4.2. Differences in 24‐h Behaviour Compositions

In the current study, children demonstrated variations in 24‐h compositions based on their typical nap habits when considering components of both waking movement behaviours and time in bed. Preschool children who regularly napped were more likely to take longer to fall asleep than children who did not nap, and more likely to engage in more inactive time than children who sometimes napped. It is possible that regular nappers engaged in quiet rest or play during nap opportunities, rather than always sleeping, particularly in childcare settings. Given that nap habituality was derived from a mix of actigraphy and parent report, this cannot accurately be inferred in the current study. Although we are not aware of other preschool studies to make a direct comparison to, de Souza et al. (2022) recently reported on differences in movement behaviour compositions among 270 Brazilian preschoolers. They found that children categorised as short sleepers, relative to adequate sleepers, spent more time in light physical activity (p = 0.0005, Cohen's d = 0.44, Bayes Factor: 46.4).

Both sex and age were associated with several 24‐h behaviour components in our study.

However, some associations varied from the findings of other preschool studies. Although sex differences may not appear until later in childhood (Kwon et al. 2022), some reports have noted higher MVPA and total physical activity among males (Nilsen et al. 2019; Schmutz et al. 2018). However, in our sample, males engaged in less MVPA but more light physical activity than females. Comparable to other studies, age was positively associated with MVPA levels (Schmutz et al. 2018; Hinkley et al. 2017). Although sociodemographic factors such as age and sex have been established as strong correlates of inactive behaviour, physical activity, and sleep in research that has explored these behaviours independently, limited 24‐h behaviour research in young children has included information on these key factors. The presence of variations at this early age may be important to consider when using 24‐h behaviour approaches. Indeed, through another 24‐h behaviour approach (i.e., compliance with 24‐h movement behaviour guidelines), a recent study by Kracht et al. (2019) supported that sociodemographic factors such as poverty level and race were linked to engagement of all three behaviours.

Interestingly, none of the factors included in our multivariate compositional models resulted in differences in total sleep time. Although ranges are not always reported in other similar preschool studies, total sleep time in these other samples appeared more variable, which may be more indicative of general population norms (Dolinsky et al. 2011; Schmutz et al. 2017; Yu et al. 2011; Kohyama 2007). Due to the exclusion criteria, children in our sample were generally healthy sleepers (e.g., no sleep disorders or sleep‐related medications). While our sample's total sleep time range of 8.5–12.2 h over 24 h is comparable to other studies with actigraphy‐measured sleep (Cairns and Harsh 2014), approximately 50% of children in the general population may experience sleep problems (Carter et al. 2014). Additionally, children in the current study were also very active (i.e., mean of 1.9 h of daily MVPA and 5.5 h of total physical activity), which is higher than average levels commonly reported in this age group (Hnatiuk et al. 2014). Therefore, the proportion of children obtaining the physical activity recommendations in our sample could have contributed to a potential ceiling effect.

4.3. Nap Sleep Compositions

In older populations, physical activity has been shown to elicit changes in sleep architecture, such as increases in overnight N3 sleep (Chennaoui et al. 2015; Dworak et al. 2008). Relatedly, given the importance of N3 sleep on cognitive performance, alertness, mood, and recovery, we anticipated that the proportion of time spent in this sleep stage would be related to activity levels (Chennaoui et al. 2015). However, in the current study, PSG data were only available from a single daytime nap in each participant in a small subsample. Although N3 sleep has been associated with various performance markers in other preschool studies (Spencer 2013), for some outcomes, such as physical activity, one nap may not be representative of daily sleep physiology and we may not have been sufficiently powered to explore these relations. The inclusion of nighttime sleep staging, as well as daily variations or influences, may be important to further explore the potential contributions of each sleep stage to waking behaviours in this age group.

4.4. Limitations

Though we aimed to illustrate the use of compositional data analysis in sleep health research, it is important to note some limitations to our study and this statistical analysis approach overall. As with most measures of physical activity and sleep, there is a risk of potential misclassification, particularly between inactive time and time in bed. Relatedly, the devices used in this study did not measure posture, which is recommended to estimate sedentary behaviour. Thus, we referred to our measure as physical inactivity and used this as a proxy for sedentary behaviour. Specific to our protocol, the wrist location for actigraphy is generally preferred for estimation of sleep metrics, but the hip or waist is generally more accurate for physical activity measures (Migueles et al. 2017). Further, movement during sleep in children is more common than in adults, and thus, WASO can be overestimated via actigraphy in young children, and therefore, there could have been some misclassification in our study that would influence the compositional makeup (Bélanger et al. 2013). As previously noted, the current sample consisted of generally healthy children. Combined with the other highlighted limitations, these considerations restrict the generalizability of our findings. An additional consideration to this is the lack of diversity in race, ethnicity, and socioeconomic status of the small sub‐sample of children with PSG data.

Although CoDA has become a common statistical analysis approach adopted by time‐use epidemiologists, as with other methods, there are some challenges and constraints to consider. Although this approach allows multiple, co‐dependent parts of a composition to be explored in the same model, researchers should be aware that general time‐use itself may only present one aspect of an association. For example, the context of the behaviours (e.g., types of physical activities, social or cognitive engagement, physiological sleep markers such as spindles) may also play a role. Additionally, even though approaches for dealing with zeros in compositional parts have been suggested (Rasmussen et al. 2020), CoDA appears to work better when there is some actual time spent in each of the components. Researchers should be prudent about reporting details on both their data processing steps and CoDA model development, as many methodological steps are often not disclosed and therefore cannot be replicated (Brown et al. 2024). Finally, given the present cross‐sectional design, causality should not be inferred without longitudinal or experimental data.

4.5. Summary and Conclusions

This study highlights the utility of CoDA in examining 24‐h movement behaviour patterns among preschool‐aged children. Our findings suggest that habitual napping is associated with longer sleep onset latency and greater inactive time, while age and sex influence specific behaviours such as WASO, light physical activity, and MVPA. However, total sleep time remained consistent across groups, likely reflecting the overall healthy and active nature of our sample. Although no associations were found between nap sleep stage composition and physical activity levels, this may be due to limited PSG data and the constraints of a single‐nap measurement.

These results underscore the importance of integrating behavioural and physiological dimensions in early childhood sleep health research. CoDA offers a complementary approach to traditional methods, such as evaluating compliance with 24‐h movement guidelines, by enabling researchers to assess how changes in one behaviour relate proportionally to others. Future longitudinal studies incorporating diverse samples, comprehensive sleep staging, and multiple analytic frameworks are warranted to better understand the interplay between sleep and activity patterns in early development.

Author Contributions

Conceptualization: Christine W. St. Laurent, Tracy Riggins and Rebecca M.C. Spencer. Methodology: Christine W. St. Laurent, Tracy Riggins and Rebecca M.C. Spencer. Formal analysis: Christine W. St. Laurent, Fatemeh Yousefi and Pardis Parvizi. Investigation: Jennifer F. Holmes, Sanna Lokhandwala, Tracy Riggins and Rebecca M.C. Spencer. Resources: Tracy Riggins, Rebecca M.C. Spencer. Data curation: Christine W. St. Laurent, Jennifer F. Holmes, Sanna Lokhandwala, Tracy Riggins and Rebecca M.C. Spencer. Writing – original draft: Christine W. St. Laurent. Writing – review and editing: Fatemeh Yousefi, Pardis Parvizi, Jennifer F. Holmes, Sanna Lokhandwala, Tracy Riggins and Rebecca M.C. Spencer; Visualisation: Christine W. St. Laurent, Fatemeh Yousefi and Pardis Parvizi. Supervision and project administration: Tracy Riggins and Rebecca M.C. Spencer. Funding acquisition: Christine W. St. Laurent, Tracy Riggins and Rebecca M.C. Spencer.

Funding

This work was supported by the National Institutes of Health, R01 HD079518, R21 HD094758, F32 HD105384.

National Science Foundation, BCS 1749280.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.