Physical activity, sedentary time, and sleep from childhood to young adulthood: a seven-wave cohort study of within-person relations

Abstract

Study Objectives

To determine whether within-person changes in total physical activity (PA), moderate-to-vigorous physical activity (MVPA), and sedentary time from ages 6 to 18 predict changes in sleep duration and insomnia symptoms, and vice versa.

Methods

Seven waves of biennially collected data from a birth cohort study were used, capturing ages 6–18 years (n = 880). Every second year, objective data on PA, sedentary time, and sleep duration were collected using accelerometers, while insomnia symptoms were assessed through clinical interviews. Random Intercept Cross-Lagged Panel Models were estimated to test the within-person relations between PA/MVPA/sedentary time and sleep. Potential sex and age differences were also examined.

Results

We found no evidence for within-person relations between the study variables, nor for any sex or age differences.

Conclusions

Children and adolescents who become more physically active or spend less time in sedentary activities are probably not more likely to sleep longer or better than they typically would.

Statement of Significance

Increased physical activity (PA) or reduced sedentary time is assumed to improve sleep, and improved sleep is expected to promote PA. While some short-term studies on daily variations in these behaviors partially support this assumption, longer-term studies are few and have substantial methodological limitations. This study is the first to test within-person relations between PA/sedentary time and sleep in children and adolescents, analyzing seven waves of data assessed by objective measures and clinical interviews. No evidence for long-term relations between PA and sleep at the individual level was revealed, thus the findings suggest that individual-level interventions targeting one behavior may not improve the other over the long term.

Graphical Abstract

Graphical Abstract.

Introduction

During the transition from childhood to adolescence and further into young adulthood, physical activity (PA) levels and sleep duration decline [1–4], whereas sedentary time and sleep problems increase [5–7]. The decline in sleep duration is linked—but not equivalent—to the corresponding increase in sleep problems [8, 9]. Although efforts to promote these health behaviors are partly based on the assumption that an increase in PA or a decrease in sedentary time leads to longer sleep, and vice versa [10], research substantiating this assumption is still limited due to several methodological shortcomings. First, studies are predominantly cross-sectional, and existing prospective studies are short-termed (follow-ups over 3–8 consecutive days). Thus, it is yet unknown how PA and sedentary time relate to sleep throughout childhood and adolescence [3, 10]. Second, studies have mainly applied self-reports of PA, sedentary time, and sleep duration, which only moderately correspond to objective measures [11], thus leaving us with less valid results [12], possibly inflating the relation between PA and sleep. Regarding sleep problems, clinical interviews of diagnostically defined symptoms of sleep disorders are the gold standard but have rarely been used in longitudinal research examining the relation between PA, sedentary time and sleep. Third, although preventive efforts are based on the premise that changes in one behavior affect the other at the individual level, existing studies have not separated within-person from between-person effects and are thus not positioned to properly test such assumption [12–16]. Between-person effects can only speak to whether an individual’s level of PA or sedentary time—compared to others, predicts their future level of sleep—relative to others, not whether children who become more physically active or less sedentary sleep longer and better than they typically would (i.e. within-person effects, using the individual as his/her own control). To the best of our knowledge, the present study is the first to address the latter research question, and we do so by using seven waves of data from a birth-cohort assessed every second year from age 6 to 18, employing objective measures of PA, sedentary time and sleep duration, and separating within- from between-person effects. Clinical interviews are used to assess symptoms of insomnia, which is the most prevalent sleep disorder in youth [7] and characterized by persistent difficulties in falling asleep, staying asleep, and waking up earlier than desired time (Diagnostic and Statistical Manual of Mental Disorders—Fifth Edition [DSM-5]) [7].

Reciprocal Relationships between Physical Activity, Sedentary Time, and Sleep

Theoretical assumptions

Although a well-established theoretical framework for a bidirectional relationship between PA/sedentary time and sleep in children and adolescents is missing [14, 17], a range of neurobiological, psychosocial, and behavioral mechanisms have been suggested [10, 18–21]. Only the latter two focus on how PA/sedentary time and sleep may affect each other over time and are thus of relevance here. Psychosocial mechanisms include intrapersonal and interpersonal factors [18], such as social connectedness, including parental control and autonomy. For example, increased bedtime autonomy (i.e. less parental control) is assumed to cause more bedtime-related screen time use, which may forecast reduced sleep duration [22]. Over time, reduced sleep duration and sleep problems (e.g. insomnia symptoms) may cause daytime sleepiness or fatigue and such lack of energy may hinder proper levels of PA [17]. On the other hand, engaging in moderate-to-vigorous physical activity (MVPA) can promote psychological well-being and stress reduction, thereby promoting improved sleep [23]. These examples suggest that the interplay between PA/sedentary time and sleep may take time to transform into long-term changes, daily routines, or habits [24] and may therefore evolve throughout development. However, this assumption has not yet been tested, which is the primary aim of the current research.

As illustrated above, the psychosocial mechanisms address factors (e.g. autonomy) that may predict and thus precede changes in PA levels and sleep patterns. The behavioral hypothesis, on the other hand, states that these health behaviors are co-dependent in the first place, because time spent in one behavior (e.g. sleep) necessarily reduces or displaces the time available for other behaviors (e.g. PA or sedentary time), drawing on the time-use research paradigm, which acknowledges that a day has 24 h [19, 21, 22, 25]. This also aligns with the displacement hypothesis [26]. For instance, screen time (i.e. sedentary time) may compromise sleep time, potentially leading to sleep problems [27–29], or increasing the risk of insomnia [30]. Also, increased sleep duration might compromise the time for PA and leave less time to spend on sedentary activities such as screen use [30].

Thus, in addition to the changes that occur within each of these health behaviors from childhood to young adulthood, the dynamic interplay between them is assumed to be subject to change [10, 24]. These behaviors are likely to reciprocally fuel each other (e.g. low levels of PA promote poor sleep, and vice versa) and such dynamic interaction constitutes a mechanism explaining why both PA and sleep duration decrease, and sedentary time and sleep problems both increase during adolescence [10]. The present study is the first to test this assumption.

Empirical evidence

Systematic reviews and meta-analyses, primarily of cross-sectional studies, show that more PA/less sedentary time is associated with longer sleep duration and a reduced risk of sleep problems or insomnia symptoms [10,11,15,27,31,32]. Furthermore, a recent systematic review and meta-analysis of 11 short-term studies (i.e. follow-ups over 3–8 consecutive days) that employed objectively measured PA in participants aged 5–15 years [15] found a weak association between increased daily PA/reduced sedentary time and improved sleep, but only for those below age 13. Similarly, improved sleep on the previous night was weakly associated with more PA/less sedentary time the following day.

A handful of longer-term studies on the within-person relationship exist, but with two exceptions [12, 13], they have relied on self-reported measures for both PA and sleep patterns. For example, one study spanning 1 year with quarterly assessments found that increased self-reported PA in adolescent girls forecasted a reduction in insomnia symptoms at the individual level, whereas more insomnia symptoms predicted decreased levels of PA [17]. Another study using self-report at four time points from ages 13–22 [14] found no significant within-person relation between vigorous PA and sleep problems in early and middle adolescence.

The first within-person study that used objective measures found a small but negligible relation between sleep duration and PA in 7- to 12-year-olds assessed at three time points over a 2-year period [12], but the direction of the relation is equivocal because a parallel process latent growth curve modeling analysis was applied, which does not estimate directionality [33].

The second within-person study, conducting yearly assessments over a 3-year period in 12- to 14-year-olds revealed no evidence for a long-term relationship between MVPA and sleep patterns [13]. Importantly, as noted by the authors themselves, the sample size (n = 214) might not have been large enough to afford the needed statistical power to detect long-term within-person relations [13]. Furthermore, these studies are limited to a 2- and 3-year follow-up, respectively. Therefore, knowledge is still needed on how the interplay between PA, sedentary time, and sleep evolves throughout childhood and adolescence. Diagnostically defined sleep problems were also not captured in these studies.

The Present Study

We extend existing research by examining seven waves of data spanning age 6–18 years and using clinical interviews to assess insomnia symptoms, which is a more valid measure of sleep problems compared to questionnaires [34]. Because PA at the moderate-to-vigorous level is associated with beneficial health outcomes in youth [35, 36], we examine both overall daily PA and MVPA. Finally, because sedentary time is associated with health irrespective of PA [37]—although they are indeed related [12]—we also examine the within-person relation between sedentary time and the two sleep outcomes. More specifically, we examine whether individuals who increase their levels of PA/MVPA or reduce sedentary time also experience increased sleep duration and fewer symptoms of insomnia during the years from age 6 to 18. The opposite direction of influence is also considered.

Furthermore, boys are more physically active and spend less time being sedentary than girls [2, 3]. However, increases in sedentary time are more prevalent in boys than girls [5]. In addition, some studies suggest that girls at certain ages (14–17 years) sleep less and face a higher risk of insomnia [9, 38]. Thus, the relations between these health behaviors may also differ by sex and age, which we therefore examine.

Methods

Participants, recruitment, and procedure

This research utilized longitudinal data from the Trondheim Early Secure Study (TESS), which began in 2007 [39]. All parents of children born in 2003 and 2004, residing in Trondheim, Norway (n = 3016) were invited to participate. Most consenting parents were of Norwegian origin (93% mothers, 91% fathers). Figure 1, reprinted from Steinsbekk et al. [40], illustrates the recruitment procedure and participant flow.

Figure 1.

The cohort study design and flow of participants.

Because the initial purpose was to study mental health, children who displayed more emotional and behavioral difficulties were oversampled to increase statistical power. Emotional and behavioral difficulties were measured by the Strengths and Difficulties Questionnaire (SDQ) [41]. Children were divided into four strata based on their SDQ scores, with cutoff points of 0–4, 5–8, 9–11, and 12–40, respectively. The probability of selection increased with higher SDQ scores, with respective selection probabilities of 0.37, 0.48, 0.70, and 0.89 for the four strata. This oversampling was corrected for in the analyses.

Data collection has occurred biennially from age 4 (T1) and is still ongoing. The present study uses data from age 6 (T2)—which was the first time point for the assessment of PA and sedentary time—to age 18 (T8), resulting in seven waves of data assessments: T2 (n = 795; M_age = 6.72, SD = 0.17; 49.8% girls), T3 (n = 699; M_age = 8.8, SD = 0.24; 51.1% girls), T4 (n = 702; M_age = 10.51, SD = 0.17; 52.3% girls), T5 (n = 668; M_age = 12.49, SD = 0.15; 51.9% girls), T6 (n = 628; M_age = 14.33, SD = 0.59; 53.0% girls), and T7(n = 666; M_age = 16.98, SD = 0.31; 55.0% girls). Table 1 presents sample descriptives at age 6 (baseline of the current inquiry).

Table 1.

Sample descriptives at age 6

Variable	%
Child gender
Boy	48.0
Girl	52.0
Informant parent gender
Male	15.2
Female	84.8
Ethnic origin of biological mother
Norwegian	93.0
Western countries	6.8
Other countries	0.3
Ethnic origin of biological father
Norwegian	93.0
Western countries	6.5
Other countries	0.5
Cohabitating status of biological parents
Cohabitating	84.6
Not cohabitating	15.4
Highest education level of informant parent
Not completed junior high school	0.0
Junior high school (10th grade)	0.3
Some education after junior high school	3.9
Senior high school (13th grade)	9.4
Some education after senior high school	2.1
Some college or university education	3.8
Bachelor’s degree	4.1
College degree (3–4 years of study)	21.4
Master’s degree or similar	13.7
PhD completed or ongoing	3.2
Occupational status of informant parent
Leader	7.8
Professional, higher level	26.3
Professional, lower level	40.5
Formally skilled worker	22.2
Farmer/fisherman	0.1
Unskilled worker	3.0

Note: The displayed sample characteristics are based on data for age 6 years, which is the baseline assessment of the present study.

The analytical sample of the present inquiry included 880 participating children with valid measures on at least one time point. An attrition analysis was conducted prior to the main analysis to test whether dropout was systematically related to the study variables. We found that higher total PA (OR = 1.002 [95% CI, 1.001, 1.003], p = .015) and more sedentary time at age 10 (OR = .9958 [95% CI, 0.9925, 0.9991], p = .012) showed a very weak association with attrition at age 12. However, the practical significance of this association is likely negligible. From age 16 to 18 years, attrition was higher in males (OR = 1.7809 [95%CI, 1.3570, 2.3372], p < .001) and lower in those displaying higher sedentary time (OR = .9967 [95% CI, 0.9948, 0.9986], p = .001). It is important to acknowledge that the explained variance, according to Cox and Snell [42], was low, ranging from R² = .008 to R² = .020, indicating a minimal combined effect of these variables on attrition.

Measurements

Physical activity, sedentary time, and sleep duration

At all timepoints, participants wore hip-mounted ActiGraph GT3X accelerometers (Manufacturing Technology Incorporated, Fort Walton Beach, FL, USA) for 7 consecutive days (24 h a day) except when bathing or showering. The accelerometer is a small monitor that has demonstrated high validity and reliability in measuring acceleration in children, as compared to direct observation and energy expenditure measurements of PA [43, 44]. PA levels and sleep duration were calculated using Actilife software (Version 6.13.3) [45]. Data analysis included only those participants who had at least 3 days of recordings with a minimum of 480 min of hip movement per day. Non-wear time was defined as periods of consecutive zero counts lasting 20 min or more.

Daily PA was estimated based on overall daytime activity (from 06:00 to 23:59, excluding periods of sleep at night), whereas the time spent in MVPA was based on a cutoff point of ≥ 2296 counts per minute [46]. This threshold is considered superior for children and adolescents [47]. Sedentary time was defined as minutes of activity ≤ 100 counts per day, a commonly used cutoff point that has shown excellent classification accuracy [47]. Sleep duration was calculated by using Tudor–Locke’s auto sleep period detection algorithm [48]. This algorithm automatically detects PA levels and determines whether a person is asleep or awake during each epoch by considering the PA levels in the 5 min before and after. It scores the minutes as asleep between bedtime and rise time, excluding any minutes awake.

Insomnia symptoms

At age 6, the sum of insomnia symptoms was assessed by the Preschool Age Psychiatric Assessment (PAPA) [49], a semi-structured parent-based clinical interview. From age 8, the corresponding Child and Adolescent Psychiatric Assessment (CAPA) [50] was employed, where both the child and the parent were interviewed separately. Participants first visited the clinic to complete the insomnia assessment, and upon finishing the assessment, they were provided with an accelerometer and instructed to wear it for the following week. This procedure ensured that the ActiGraph data collection closely followed the insomnia assessment, minimizing any potential temporal discrepancies between the two measures.

The PAPA/CAPA captures the DSM-IV criteria [51]. Therefore, at ages 6 and 8, the following diagnostic criteria were used: (1) difficulties initiating asleep/sleep onset latency (age 6 ≥ 30 min and age 8 ≥ 1 h, because information about minutes was missing); (2) difficulties staying asleep/nightly awakenings (≥ 20 min wake after sleep onset) at least three times a week; and (3) non-restorative sleep (insufficiently rested after sleep). To align the assessment of insomnia with the more recent DSM-5 criteria [52], additional interview questions were included from age 10 onward. The present criteria were therefore used at the remaining assessment points: (1) difficulties initiating sleep/sleep onset latency (≥ 30 min) or always parental intervention, at least three times a week; (2) difficulties staying asleep/nightly awakenings (≥ 20 min wake after sleep onset) or always parental intervention, at least three times a week; and (3) early morning awakenings without being able to return to sleep and being in need of more sleep, at least three times a week. For all criteria, a 3-month primary period was used, and a symptom was considered present if reported by either the participants (from age 8 onward) or the parent. All the interviewers (n = 7) had relevant qualifications (e.g. a bachelor’s degree) and experience working with children and families. Blinded raters recoded a subset of the audio-recorded interviews (9%) with a sufficient inter-rater reliability (k = 0.75) for insomnia.

Sex was coded based on the biological sex assigned at birth, which can be identified using the participant’s national identification number.

Statistical analysis

For descriptive purposes, we estimated Pearson’s correlations between all study variables at each time-point, before conducting the main analyses. The false discovery rate was not corrected, as these analyses were preliminary. To test whether within-person changes in PA/sedentary time predicted within-person changes in sleep duration/insomnia symptoms and vice versa, we applied a Random Intercept Cross-Lagged Panel Model (RI-CLPM) [53] using Mplus version 8.9 [54]. We ran six separate models, one for each potential combination of PA/sedentary time and sleep (e.g. total PA and sleep duration; MVPA and insomnia symptoms). The RI-CLPM consists of the following parts: a random intercept for each of the two factors (e.g. PA and sleep duration) loading on the observed scores at each time point, with the factor loading set to 1. A study variable’s random intercept (e.g. PA) represents the person’s average level over the study period, and the correlations between these random intercepts reflect the between-person associations between them (e.g. between PA and sleep duration). Each observed measure is assigned a latent variable with a loading of 1 and an error variance fixed at 0, thereby transferring the observed variable’s variance to the latent counterpart. These latent variables, one for each study variable at each time point, capture deviations from the subject’s mean value over the study period, using each participant as their own control. Concurrent correlations between the residuals of the latent variables are allowed. Finally, the latent changes are regressed on values from the preceding time point (i.e. 2 years earlier).

To determine if the within-person cross-lagged coefficients differed by sex, we conducted multiple-group analyses [53]. For each of the potential relations (e.g. PA and sleep duration, sedentary time, and insomnia), we thereby estimated a RI-CLPM with within-person paths constrained to be identical for boys and girls and compared it to a model with paths freely estimated between the sexes, using the Satorra–Bentler scaled χ² test [55]. If this test shows that the fit deteriorates when the paths are constrained so, it indicates that sex differences do exist (i.e. the paths cannot be constrained to be equal for boys and girls). The same approach was applied when examining age differences. More specifically, a freely estimated model was compared to a model with cross-lagged paths constrained to be of equal strength across ages. If these constraints do not impair model fit, it indicates that the strengths of the paths are not significantly different, and thus no age differences are evident [53].

Model fit was estimated based on Hu and Bentler’s criteria [56], using the following thresholds: comparative fit index (CFI > 0.90) and the Tucker–Lewis index (TLI > 0.90), root mean square error of approximation (RMSEA < 0.06), and standardized root mean square residuals (SRMR < 0.08). We applied recent guidelines for RI-CLPM in interpreting effect sizes, where standardized estimates of 0.03, 0.07, and 0.12 indicate small, medium, and large effects, respectively [57].

We estimated parameters using the robust maximum likelihood estimator with robust standard errors [58] and handled missing data using a full information maximum likelihood procedure [59]. As noted in the “Methods” section, children with emotional and behavioral problems were oversampled based on their SDQ scores. In the analyses, we therefore applied probability weights to account for this oversampling. These were calculated by dividing the number of children in the population within each stratum by the number of children selected to participate from that stratum. This adjustment was implemented using a sandwich estimator, which accounts for the stratified sampling design and provides robust standard errors. By applying these population weights, we ensured that our estimates were generalizable to the broader population. To adjust for the false discovery rate (87 p-values in total), we applied the Benjamini–Hochberg approach [60].

Finally, although the COVID-19 restrictions in Norway were less comprehensive compared to several other countries, data collected during the pandemic (i.e. T7 in 2020–2022) could have been influenced by these restrictions. Therefore, we conducted sensitivity analyses by re-running all the main models, omitting T7 to test if the findings would differ.

Results

Bivariate correlations between the main study variables are shown in Supplementary Table S1. Sleep duration was cross-sectionally and negatively associated with total PA and MVPA at ages 6 and 8, as well as with sedentary time at ages 6, 10, 14, 16, and 18. Likewise, insomnia symptoms were cross-sectionally and negatively associated with total PA and MVPA at age 18.

Table 2 displays the descriptive statistics of the study variables. Figures 2–7 present the results of the freely estimated RI-CLPMs, and all the estimated model fitting indices are shown in Table 3, indicating a good fit for all six models. As can be seen, within-person changes in total PA, MVPA, and sedentary time did not predict changes in sleep duration or vice versa, except that decreased sedentary time at age 14 (i.e. compared to the individual’s mean level) predicted increased sleep duration at age 16. However, when we adjusted for multiple tests using the Benjamini–Hochberg approach (i.e. accepting a false discovery rate of 5%), this path was no linger significant. Furthermore, no significant between-person associations were found between PA levels (i.e. total PA, MVPA, and sedentary time) and sleep duration.

Table 2.

Means (M) and standard deviations (SDs) for all study variables

	Total (N = 880)		Girls (n = 457)		Boys (n = 423)
Variables	M	SD	M	SD	M	SD
Sleep duration (minutes per day) age 6	577.30	44.03	581.14	39.78	572.90	47.58
Sleep duration (minutes per day) age 8	546.21	33.31	551.41	34.23	542.02	32.32
Sleep duration (minutes per day) age 10	551.39	39.99	555.42	38.44	546.94	41.16
Sleep duration (minutes per day) age 12	521.34	39.30	529.42	35.68	515.16	41.22
Sleep duration (minutes per day) age 14	510.83	48.85	515.12	49.48	502.84	48.63
Sleep duration (minutes per day) age 16	527.60	79.15	529.56	72.68	528.67	87.93
Sleep duration (minutes per day) age 18	508.15	77.17	508.12	71.99	509.73	85.14
Total physical activity (counts per day) age 6	616.69	180.80	588.81	166.84	643.95	190.47
Total physical activity (counts per day) age 8	588.11	199.21	547.23	186.72	633.86	205.26
Total physical activity (counts per day) age 10	516.64	178.69	483.49	178.75	554.54	173.35
Total physical activity (counts per day) age 12	453.74	160.37	431.87	135.64	494.72	179.13
Total physical activity (counts per day) age 14	392.90	156.73	375.36	146.63	430.46	164.34
Total physical activity (counts per day) age 16	328.73	148.34	311.57	131.90	349.56	163.81
Total physical activity (counts per day) age 18	333.98	129.96	314.13	119.78	347.95	142.05
Moderate-to-vigorous physical activity (minutes per day) age 6	71.51	23.93	64.78	19.81	78.31	25.97
Moderate-to-vigorous physical activity (minutes per day) age 8	70.51	26.11	61.55	20.91	80.76	36.86
Moderate-to-vigorous physical activity (minutes per day) age 10	65.46	24.05	58.54	21.09	73.79	45.23
Moderate-to-vigorous physical activity (minutes per day) age 12	59.01	23.11	56.39	20.56	64.73	63.76
Moderate-to-vigorous physical activity (minutes per day) age 14	53.54	22.26	51.17	20.97	57.93	43.41
Moderate-to-vigorous physical activity (minutes per day) age 16	36.15	21.64	33.71	19.71	38.31	33.74
Moderate-to-vigorous physical activity (minutes per day) age 18	42.53	20.02	39.56	13.14	44.53	27.55
Sedentary time (minutes per day) age 6	514.32	54.86	519.31	52.47	509.48	57.26
Sedentary time (minutes per day) age 8	552.10	64.19	559.44	65.32	546.16	62.50
Sedentary time (minutes per day) age 10	594.82	66.30	599.30	62.04	588.91	70.71
Sedentary time (minutes per day) age 12	619.35	72.09	625.52	64.47	611.62	79.25
Sedentary time (minutes per day) age 14	668.48	81.46	671.69	77.07	656.81	86.96
Sedentary time (minutes per day) age 16	609.03	104.78	611.29	100.74	601.63	108.45
Sedentary time (minutes per day) age 18	658.70	99.13	665.03	96.85	653.15	103.07
Insomnia symptoms age 6	0.12	0.34	0.12	0.34	0.12	0.34
Insomnia symptoms age 8	0.37	0.67	0.31	0.61	0.43	0.71
Insomnia symptoms age 10	0.16	0.38	0.16	0.39	0.16	0.37
Insomnia symptoms age 12	0.12	0.35	0.14	0.37	0.09	0.33
Insomnia symptoms age 14	0.13	0.35	0.14	0.37	0.11	0.32
Insomnia symptoms age 16	0.21	0.42	0.24	0.44	0.17	0.38
Insomnia symptoms age 18	0.22	0.49	0.26	0.54	0.17	0.42

Figure 2.

Within-person relations between total physical activity (PA) and sleep duration (n = 880).

Figure 7.

Within-person relations between total sedentary time and insomnia symptoms (n = 880).

Table 3.

Results of fitting indices for random intercept cross-lagged panel models

Models	x 2	df	p	RMSEA (90% CI)	SRMR	CFI	TLI
1. Total PA ↔ Sleep duration	110.45	57	.001	0.03 (0.02, 0.04)	0.061	0.935	0.896
2. MVPA ↔ Sleep duration	109.68	57	.001	0.03 (0.02, 0.04)	0.055	0.946	0.914
3. Sedentary time ↔ Sleep duration	105.75	57	.001	0.03 (0.02, 0.04)	0.070	0.952	0.923
4. Total PA ↔ Insomnia symptoms	83.48	57	.013	0.02 (0.01, 0.03)	0.043	0.965	0.944
5. MVPA ↔ Insomnia symptoms	87.42	57	.006	0.03 (0.01, 0.04)	0.040	0.964	0.943
6. Sedentary time ↔ Insomnia symptoms	77.93	57	.034	0.02 (0.01, 0.03)	0.052	0.972	0.955

Note: CFI = comparative fit index, RMSEA = root mean square error of approximation, SRMR = standardized root mean square residual, TLI = Tucker–Lewis index.

Figure 3.

Within-person relations between moderate-to-vigorous physical activity (MVPA) and sleep duration (n = 880).

Figure 4.

Within-person relations between sedentary time and sleep duration (n = 880).

Figure 5.

Within-person relations between total physical activity (PA) and insomnia symptoms (n = 880).

Figure 6.

Within-person relations between moderate-to-vigorous physical activity (MVPA) and insomnia symptoms (n = 880).

The Satorra–Bentler chi-square test showed that fit was not impaired when the within-person paths were constrained to be equal across sex, thus indicating that the within-person relations did not differ between the sexes (total PA and sleep duration [Δχ² = 7.58, df = 12, p = .817]; MVPA and sleep duration [Δχ² = 12.44, df = 12, p = .411]; and sedentary time and sleep duration [Δχ² = 9.73, df = 12, p = .640]). We also found no evidence for age differences in the relationships between total PA and sleep duration (Δχ² = 7.55, df = 10, p = .673), between MVPA and sleep duration (Δχ² = 8.27, df = 10, p = .603), or between sedentary time and sleep duration (Δχ² = 22.89, df = 10, p = .011). Thus, in sum, we found no evidence for age differences in the strengths of the relations between the study variables.

The same pattern of null-findings was observed for insomnia symptoms. Within-person changes in PA, MVPA, and sedentary time did not predict changes in insomnia symptoms, nor did insomnia symptoms predict changes in PA, MVPA, and sedentary time. Although increased total PA at age 16 predicted less insomnia symptoms at age 18, and increased insomnia symptoms at ages 8 predicted less sedentary time at ages 10, these relationships were no longer significant after adjusting for the false discovery rate using the Benjamini–Hochberg method. Furthermore, weak negative between-person associations were observed between total PA/MVPA and insomnia symptoms, whereas no such associations were found between sedentary time and insomnia symptoms.

No sex differences were found (total PA and insomnia symptoms [Δχ² = 18.78, df = 12, p = .094]; MVPA and insomnia symptoms [Δχ² = 13.45, df = 12, p = .337]; and sedentary time and insomnia symptoms [Δχ² = 14.30, df = 12, p = .282]). Similarly, no significant age differences (i.e. the strengths of the path did not significantly differ) were indicated by the Sattorra–Bentler chi-square tests (total PA and insomnia symptoms [Δχ² = 12.57, df = 10, p = .248]; MVPA and insomnia symptoms [Δχ² = 11.41, df = 10, p = .327]; or sedentary time and insomnia symptoms [Δχ² = 4.96, df = 10, p = .894]).

The estimated model fitting indices and results of the sensitivity analysis excluding T7 (age 16) for possible COVID-19 effects are provided in Supplementary Tables S2 and S3, respectively. The indices indicated an acceptable to good fit for all the models examined. Within-person changes in total PA, MVPA, and sedentary time did not predict changes in either sleep duration or insomnia symptoms, nor did sleep duration and insomnia symptoms predict changes in total PA, MVPA, and sedentary time, thus confirming the main results.

Discussion

This study aimed to test if changes in objectively measured total PA, MVPA, and sedentary time across ages 6–18 forecasted changes in sleep duration and insomnia symptoms at the individual level, and vice versa. Age and sex differences were also tested. Overall, we found that children and adolescents who increased their PA levels or displayed declined sedentary time were not more likely to sleep longer or have fewer insomnia symptoms 2 years later. Likewise, neither changes in sleep duration nor insomnia symptoms predicted future changes in total PA, MVPA, or sedentary time at the within-person level. There were no significant age or sex differences. At the between-person levels, no significant relationships were found between total PA/MVPA/sedentary time and sleep duration, respectively, indicating that these factors were also not related at the group level. There were weak negative between-person associations between total PA/MVPA and number of insomnia symptoms, whereas no such associations were found between sedentary time and insomnia symptoms. Using seven waves of data, objective measures, clinical interviews, and separating within-person from between-person effects, the findings extend previous research that has been limited to cross-sectional designs, short-term follow-ups, self-reported data, and a focus on unidirectional effects and between-person changes [2, 3, 10, 13].

No within-person relations

The study did not find evidence supporting that children and adolescents who become more physically active or less sedentary show corresponding increases in sleep duration or have fewer insomnia symptoms from childhood throughout adolescence. Within-person changes in sleep duration and insomnia symptoms did also not predict changes in total PA, MVPA, or sedentary time. Our findings from this seven-wave inquiry are consistent with two 3-wave studies spanning ages 7–12 years [12, 13], which, like the present study, applied objective measures and found no evidence for the within-person relations between PA/MVPA and sleep. The findings also align with a recent longitudinal study relying on self-reports [14], which found no significant within-person relation between vigorous PA and sleep problems in early adolescence (ages 13–14). Systematic reviews and meta-analyses of short-term studies [15, 17] and cross-sectional research [10,11,27,31,32] on the other hand, have found significant but weak associations between more PA/less sedentary time and a reduced risk of sleep problems. In alignment with this research, we also found that weak to moderate cross-sectional associations were evident at some time points (e.g. T2 and T8). These studies, along with the current findings, suggest that while there may be short-term or day-to-day associations [13], there is no support for the assumption that in the long-term increased PA/MVPA or decreased sedentary time leads to longer sleep duration or less sleep problems. This challenges the assumption of a direct or dynamic long-term interplay between PA/sedentary time and sleep in childhood and adolescence [10, 24]. Our findings also not support the behavioral model known as the time-use research paradigm [19] or the displacement hypothesis [26], which posits that time spent in one behavior (e.g. sleep) necessarily reduces the time available for other behaviors (e.g. PA). One potential explanation is that PA/MVPA typically occurs at earlier times of the day than when children and adolescents are not about to sleep, thus time-displacement might not be very likely. Although sedentary behavior, such as screen use, lying idly on the bed, and sitting while engaging in activities such as reading, doing homework, socializing, or having dinner, could more typically take place in the evening and thus affect sleep, we also found no relation between sedentary time and sleep. Given that we examined the total amount of daily sedentary time, not differentiating between sedentary time related to school hours versus evening hours, for example, we were not positioned to properly test the time–displacement hypothesis. Building on the present results, future research could aim to do such differentiations. Future studies are also needed to replicate the present null findings, preferably by comparing shorter to longer time lags. Because RI-CLPM is power-demanding [61], studies with more power are particularly needed.

Developments at the population level

Children in our sample averaged 72 min of MVPA per day, meeting the World Health Organization’s recommendation of at least 60 min [62]. However, by age 18, a notable developmental trend was observed with MVPA falling 43 min short of the recommended daily average [62]. This shortage is consistent with the typical decline observed with age from childhood to adolescence [10]. Sedentary time also increased with age, exceeding the recommended level of less than 2 h, particularly during adolescence [63]. This aligns with global developmental trends showing a rise in sedentary time among teenagers [28,64]. During childhood, participants had adequate sleep duration, falling within the recommended range of 9–12 h per night [65,66]. However, during adolescence, their sleep duration decreased to 8 h and 47 min, slightly above the minimum recommended range of 8–10 h [65,66]. This decrease is consistent with population trends observed during adolescence [1,7,67]. Finally, the number of insomnia symptoms remained low across all ages in our sample, consistent with normative data [68].

Limitations

This study boasts numerous strengths, including its longitudinal design, representative sample, seven-wave data collection and analysis (ages 6–18), objective measures of PA/sedentary time and sleep, the use of clinical interviews in assessing insomnia symptoms, and the separation of between- from within-person effects. However, some limitations should be acknowledged. The use of a hip-worn accelerometer might introduce measurement error in sleep estimates, compared to wrist devices [69–71]. For example, hip-worn devices may overestimate sleep duration, due to their reduced sensitivity to subtle movements during sleep and their reliance on movement-based sleep inference [69]. Nonetheless, hip-worn devices provide valid estimates of sleep duration, particularly when methodological consistency across multiple behaviors (e.g. PA, sedentary time, and sleep) is a priority [70]. We did not apply gold-standard measures of PA (e.g. doubly labeled water) [72] and sleep durations (e.g. polysomnography) [73], as these methods are extremely time-and resource-demanding and thus not very applicable in longitudinal birth-cohort studies such as the present. Moreover, sleep duration was measured exclusively during nocturnal periods. Future research may benefit from employing a 24-h monitoring approach to more comprehensively capture the full distribution of sleep.

Furthermore, because RI-CLPM is power-demanding [60], the study might have been underpowered to reveal sex and age differences, increasing the risk of type II errors. Future studies should therefore replicate our sex- and age-specific findings before drawing firm conclusions. Since we applied 2-year lags, future research should also examine whether our findings hold when shorter timeframes are used. It is also possible that effects are nonlinear; however, such effects are typically modest in size once main effects are accounted for, and detecting them would require substantially larger samples than the one available in the present study. Also, while the RI-CLPM adjusts for all time-invariant confounders, future studies should consider incorporating time-varying factors (e.g. stressful life events). However, we found no evidence of within-person relationships between PA and sleep; it is unlikely that adding such covariates would have affected the results.

Finally, while it remains uncertain whether these findings are applicable to low- and middle-income countries where longitudinal research is lacking [27], there is no immediate reason to assume that the relations between PA/sedentary time and sleep would differ significantly between locales, even though the levels and characteristics of these variables may vary across different countries and cultures.

Conclusions

It is commonly believed that increasing PA or reducing sedentary time will lead to improvements in sleep, and vice versa. Some short-term studies on daily variations in PA and sleep provide initial support for this assumption, but longer-term studies are few and with methodological limitations that hinder firm conclusions. Employing objective measures of sleep and physical activity and clinically assessed insomnia symptoms in a birth cohort followed biennially from age 6 to 18, we found no evidence for such long-term relations, and no sex or age differences were evident. In sum, youth who became more physically active or spent less time in sedentary activities are not more likely to sleep longer or better in the future.

Disclosure statement

Financial disclosure: This project was funded by the Research Council of Norway (301446) and the Liaison Committee between Central Norway RHA and NTNU (2024-36 863). The last two authors received these funds.

Non-financial disclosure: None declared.

Data availability

The data used are confidential and unavailable due to participant consent restrictions.