How much does sleep vary from night-to-night? A quantitative summary of intraindividual variability in sleep by age, gender, and racial/ethnic identity across eight-pooled datasets

Summary

Habitual sleep duration and efficiency vary widely by age, gender, and racial/ethnic identity. Despite growing research on the importance of night-to-night, intraindividual variability (IIV) in sleep, few studies have examined demographic differences in sleep IIV. The present study describes typical sleep IIV overall and by demographics among healthy sleepers. Eight datasets of healthy sleepers (N = 2,404; 26,121 total days of sleep data) were synthesised to examine age, gender, and racial/ethnic identity differences in sleep IIV measured via diaries, actigraphy, and electroencephalography (EEG). Sleep IIV estimates included the intraindividual standard deviation (iSD), root mean square of successive differences (RMSSD), coefficient of variation (CV), and a validated Bayesian Variability Model (BVM). There was substantial IIV in sleep across measurement types (diary, actigraphy, EEG) for both sleep duration (iSD: 85.80 [diary], 77.41 [actigraphy], 67.04 [EEG] minutes; RMSSD: 118.91, 108.89, 91.93 minutes; CV: 19.19%, 19.11%, 18.57%; BVM: 60.60, 58.20, 48.60 minutes) and sleep efficiency (iSD: 5.18% [diary], 5.22% [actigraphy], 6.46% [EEG]; RMSSD: 7.01%, 7.08%, 8.44%; CV: 5.80%, 6.27%, 8.14%; BVM: 3.40%, 3.58%, 4.16%). Younger adults had more diary and actigraphy sleep duration IIV. Gender differences were inconsistent. White and non-Hispanic/Latinx adults had less IIV in sleep duration and efficiency compared to racial/ethnic minority groups. Even among healthy sleepers, sleep varies widely from night-to-night. Like mean sleep, there also may be disparities in IIV in sleep by demographic characteristics. Study results help characterise normative values of sleep IIV in healthy sleepers.

INTRODUCTION

Obtaining sufficient sleep duration (7–9 ho/24 h; American Academy of Sleep Medicine and Sleep Research Society Joint, Consensus Conference Panel et al., 2015) and sleep efficiency (85%–100%; Buysse, 2014; Desjardins et al., 2019) is critical for health and well-being across the lifespan (Buysse, 2014). Research has revealed demographic differences in typical sleep duration and sleep efficiency patterns (Desjardins et al., 2019; Grandner et al., 2016; Ohayon et al., 2004). However, even within the same individual, sleep can fluctuate substantially from night-to-night due to a combination of environmental, psychosocial, and physiological factors (Slavish et al., 2019). Moreover, two individuals could have identical typical, or mean, sleep duration and efficiency, but be quite different in terms of regularity or variability of their sleep patterns. There is still a surprising lack of consensus for what constitutes a normative or typical amount of intraindividual variability (IIV) in sleep, as well as how levels of IIV in sleep may differ by demographic characteristics such as age, gender, and racial/ethnic identity (American Academy of Sleep Medicine and Sleep Research Society Joint, Consensus Conference Panel et al., 2015). This represents an important gap in the literature to address, as research shows variability in sleep is a determinant of health above the influence of mean sleep patterns (Bei et al., 2016).

Despite a growing appreciation for the importance of IIV in sleep, compared to mean sleep, relatively little work has examined demographic differences in IIV in sleep. A small but growing literature shows younger age is associated with greater IIV in sleep duration and efficiency. Specifically, younger age has been consistently associated with greater IIV in actigraphy sleep duration and efficiency (Bei et al., 2019). However, findings on age differences in IIV in self-reported sleep duration are less consistent (Dillon et al., 2015; Merklinger-Gruchala et al., 2008; Minors et al., 1998). Younger adults may have more variable social, work, and/or school schedules that interfere with the ability to maintain consistent sleep schedules compared to older adults (Logan & McClung, 2019). Older adults typically exhibit increased morningness (Logan & McClung, 2019) and may have fewer work-related obligations than both young and middle-aged adults, both of which may decrease IIV in sleep as individuals age (Adan et al., 2012).

Gender differences in IIV in sleep are less consistent than findings with age. Men and women exhibit differences in circadian rhythms across the lifespan, which may contribute to gender differences in IIV in sleep (Bailey & Silver, 2014). Women are also more likely to be primary caregivers, have a harder time adjusting to shift work, and have a greater risk of chronic disorders like depression, all of which may impair their ability to obtain consistent sleep duration and efficiency (Chung et al., 2009; Mallampalli & Carter, 2014). Studies examining gender differences in self-reported (Kaufmann et al., 2016) or actigraphy IIV in sleep duration (Diem et al., 2016; Jonasdottir et al., 2020; Mezick et al., 2009; Ogilvie et al., 2016; Xu et al., 2018) have revealed inconsistent results. While most studies have found no gender differences in IIV in sleep duration (Diem et al., 2016; Jonasdottir et al., 2020; Kaufmann et al., 2016; Ogilvie et al., 2016; Xu et al., 2018), one study found women had greater IIV in actigraphy-assessed sleep duration than men after controlling for the mean (Mezick et al., 2009). To our knowledge, gender differences in IIV in sleep efficiency have not been thoroughly explored.

Of the few studies that have examined racial/ethnic differences in IIV in sleep duration and sleep efficiency, two studies found that minority racial/ethnic groups had greater actigraphy-assessed IIV in sleep efficiency compared to White individuals (Knutson et al., 2007; Mezick et al., 2009). Findings on differences in IIV in sleep duration between Black and White individuals have been mixed and may depend on whether mean sleep is controlled for (Knutson et al., 2007; Mezick et al., 2009). Racial/ethnic disparities in IIV in sleep may be attributed to cultural factors (e.g., multi-generational homes, caregiving expectations, daily traditions) and/or socio-political factors (e.g., institutional racism, daily discrimination, neighbourhood disadvantage, work schedule, occupational stress; Grandner et al., 2016). People from minority racial/ethnic groups tend to experience more daily stressors, less predictable work hours, and more caregiving responsibilities, all of which may make it more difficult to maintain consistent sleep schedules (Grandner et al., 2016).

Despite growing attention on the importance of IIV in sleep, findings on demographic differences are still inconclusive, and several limitations exist. A growing list of metrics to define IIV in sleep may result in different statistical conclusions and confusion. For example, in their recent systematic review of IIV in sleep, Bei et al. (2019) found heterogeneity in how studies calculated IIV in sleep. While an overwhelming majority of studies computed intraindividual standard deviation (iSD) to estimate IIV in sleep, other commonly assessed IIV metrics included root mean squared successive differences (RMSSD) and coefficient of variation (CV). While each of these approaches estimate variability in sleep, they each have unique strengths and weaknesses (refer to Bei et al., 2019 for a review). Second, results differ based on whether sleep IIV is assessed by self-report, actigraphy, or electroencephalography (EEG). Comparisons of IIV across sleep measurement types are needed. Finally, studies use different operational definitions for demographic variables, which weakens the ability to generalise results across studies. Given these limitations, a quantitative summary of IIV in multiple facets of sleep is needed. Such results will provide essential epidemiological information about IIV in sleep, allow for better comparisons across studies, and may help identify more precise intervention targets.

The present study

We quantitatively synthesised a compilation of eight datasets consisting of healthy sleepers to examine age, gender, and racial/ethnic identity differences in multiple metrics of IIV in sleep duration and efficiency. Specifically, the goals of the study were to: (i) calculate multiple metrics (e.g., iSD, CV, RMSSD, Bayesian Variability Model [BVM]) to examine IIV in self-reported, actigraphy, and EEG sleep duration and sleep efficiency; and (ii) examine mean differences in IIV in self-reported, actigraphy, and EEG sleep duration and sleep efficiency by age, gender, and racial/ethnic identity.

METHODS

Participants

This study was a secondary analysis of eight datasets that assessed repeated measures of sleep across multiple days. For a summary of all studies and the demographics of each sample, see Table S1 (Dietch & Taylor, 2021; Le et al., 2021; Molzof et al., 2018; Slavish et al., 2020; Taylor et al., 2013; Taylor et al., 2017; Yap et al., 2020). To focus on healthy sleepers, participants were excluded from analyses if: (i) they had <3 days of usable sleep diary data; or (ii) they reported a mean diary-determined sleep duration <300 or >600 min (indicating very short or very long sleep duration) or mean diary-determined sleep efficiency <80% (indicating possible insomnia). This approach was taken to increase the reliability of IIV estimates (Slavish et al., 2019), as well as to eliminate any individuals who may be experiencing severe or probable clinical levels of sleep disturbances (Lichstein et al., 2003). We chose to exclude participants solely based on sleep diary data over actigraphy or EEG data for several reasons. First, sleep diaries were assessed in all eight datasets, rendering it a common variable we could use to systematically apply our exclusion criteria. Second, sleep diaries were used to assess the usability of both actigraphy and EEG data during scoring. Therefore, poor sleep diary compliance could hypothetically influence the quality of these other types of data. Finally, unlike sleep diaries, there is corrently no established quantitative clinical cut-offs for either actigraphy- or EEG-assessed sleep duration or efficiency to identify individuals who may be experiencing severe or probable clinical levels of sleep disturbances.

Measures

Self-report demographics (eight studies)

All studies collected self-report participant demographic information on age, gender, and racial and ethnic identities; however, each study varied in how these demographics were captured, categorised, and coded. These demographic variables were recoded to allow for comparisons across study. In addition to these primary demographic variables of interest, commonly collected demographic variables across studies included: education status, employment status, relationship status, number of children, and height and weight.

Sleep diary (eight studies)

Of the included studies, all studies used a variation of the Consensus Sleep Diary (Carney et al., 2012). For each study, participants were instructed to provide an estimate of their sleep the previous night (e.g., bedtime, sleep onset latency, wake after sleep onset, terminal wakefulness, rise time). From these variables, sleep duration was calculated by subtracting total wake time (sleep onset latency + time awake after sleep onset + terminal wakefulness, i.e., the period of time spent in bed awake after initial awakening) from time in bed (i.e., the interval between bedtime and rise time). Sleep efficiency was calculated by taking sleep duration divided by time spent in bed (with the intention of sleeping) multiplied by 100.

Actigraphy (six studies)

Actigraphs are wrist-worn, watch-like devices that prospectively capture motion as a proxy for activity, and some models also capture light exposure. Actigraphy has high sensitivity (0.97) and accuracy (0.86) compared to in-laboratory polysomnography (Marino et al., 2013). Three studies used the Philips Respironics Actiwatch Spectrum devices and Respironics Actiware version 6.0 to score raw sleep data. All three of these studies had data scored by two trained scorers using a previously-validated manual scoring hierarchy that relies on a combination of event markers, sleep diaries, light levels, and activity levels and Philips Actiware software algorithms (Rijsketic et al., 2020; Williams et al., 2018). The other three studies used the ActiGraph wGT3X-BT worn on the non-dominant wrist and ActiLife software version 6.13.3 to score raw sleep data. All three of these studies followed an established protocol based on activity, light, and daily sleep diary as well as integrated approximations from the Cole-Kripke scoring algorithm (Cole et al., 1992). In these studies, trained scorers decided the rest interval based on the sleep diary data window, and within that window, the Cole-Kripke algorithm determined whether an epoch was sleep or wake. In all studies, actigraphy data was determined usable if there was a continuous record of light and motion activity across a participant’s main rest interval and that the data within the participant’s main rest interval was scorable (i.e., devoid of data glitch artefacts or periods of no recording).

Electroencephalography (two studies)

Both studies used the Zmachine Insight+ (General Sleep, Inc.; Cleveland, OH, USA) that processes a single channel of EEG data using information from two mastoid-placed electrodes at A₁ and A₂ and one neck-placed ground electrode. Previous validation studies have shown that the sensitivity and specificity for detecting sleep using the Zmachine algorithm (compared to a polysomnography technologist) are 95.5% and 92.5%, respectively, indicating the Zmachine can accurately discriminate between time asleep and time awake (Wang et al., 2015). In both studies, EEG data were determined usable if there was a continuous record of electrophysiological recording of brain activity across a participant’s main rest interval (based on sleep diary and/or actigraphy) and that the data within the participant’s main rest interval was scorable (i.e., devoid of data glitch artifacts or periods of no recording).

Intraindividual variability calculations

The R packages dplyr (Wickham et al., 2015) and varian (Wiley, 2016) were used to compute the intraindividual mean and the four IIV measures for each person, across repeated measures of daily sleep: (i) the standard deviation of sleep observations across days by ID (iSD; Equation 1), (ii) coefficient of variation (CV; Equation 2), (iii) the root mean square of successive differences (RMSSD; Equation 3), and (4) Bayesian IIV (BVM; Equation 4):

$$\text{iSD}=\sqrt{\frac{\sum _{i=1}^N{\left(x_i-iM\right)}^2}{N-1}}$$ (1)

$$\text{CV=}\frac{\text{iSD}}{\text{iMean}}$$ (2)

$$\text{RMSSD}=\sqrt{\frac{{\sum }_{i=1}^{N-1}{\left(x_i-x_{i+1}\right)}^2}{N-1}}$$ (3)

where N is the sample size, x is the sleep variable, i is the number of days of observations, and iM is the intraindividual mean derived from the sleep variable.

The BVM uses a Bayesian multilevel location scale model. The observed sleep data are assumed to come from a normal distribution, with mean (location) equal to the predicted value based on any covariates and a random intercept and the residual variance (scale) allowed to vary randomly by individual and assumed to follow a gamma distribution, a strictly positive, continuous probability distribution appropriate for variance parameters. For an individual and scale with a location model for the mean:

$$\text{BVM}=Y_{ij}\sim N\left({\mu }_{j,}{\sigma }_j\right)\text{\hspace{0.17em}where} {\sigma }_j\sim \text{Γ}\left(\mathrm{\text{α}},\mathrm{\text{β}}\right)$$ (4)

In the BVM, sleep IIV is captured by the random residual standard deviation for each participant, conditioned on any predictors plus their own mean via the random intercept. Using multiple draws from the posterior allows the BVM to address uncertainty in estimating individual standard deviations and to shrink these estimates towards the population mean, particularly for people with few data points, as in other multilevel models with random effects.

Analysis plan

All analyses were pre-registered on the Open Science Framework: https://osf.io/hs97a/. For Aim 1, we calculated mean and IIV (i.e., iSD, CV, RMSSD, BVM) for sleep duration and sleep efficiency for the entire combined sample for each measurement type (i.e., sleep diary, actigraphy, EEG). Bivariate correlations were run to compare each of the four IIV metrics within the two sleep parameters (sleep duration and efficiency) across the three measurement types (sleep diary, actigraphy, EEG). For Aim 2, one-way analyses of covariance (ANCOVAs) were used to examine mean differences in IIV in sleep duration and efficiency by age, gender, race, and ethnicity, controlling for study. Study was included to account for any heterogeneity between studies. All models were fit to a gamma residual distribution with a log link to accommodate the non-normality of residuals in the IIV metrics. All models accounted for Type III sums of squares. Given the large number of analyses, in alignment with recommendations from Benjamini and Hochberg (1995), the false discovery rate (FDR) was accounted for by ranking each of the 96 p values obtained (i.e., four IIV metrics × two sleep parameters × three measurement types × four demographic variables) from smallest to largest, dividing each ranking by 96 (i.e., the number of statistical analyses assessed), and then multiplying the value by an expected FDR of 5%. The p values that remained <0.01 after adjusting for FDR were considered to be indicative of statistically robust effects. Tukey’s honestly significant difference (HSD) post hoc comparisons were conducted for all significant models. Back-transformed (i.e., transforming means to the original scale) estimated marginal means (EMMs) and standard errors were computed adjusting for study heterogeneity. Post hoc comparisons were not interpreted for groups with <10 individuals to decrease the chance of making a Type 1 error. Additional post hoc, one-way ANCOVAs were conducted examining mean differences in IIV in sleep duration and efficiency by age, gender, race, and study simultaneously. This approach was taken as a sensitivity test to assess if similar patterns of results emerged when all demographic variables were accounted for in the same model. The ethnicity factor was excluded from these sensitivity analyses as it was only measured in four out of the eight studies.

RESULTS

The merged dataset resulted in a total of 3,053 participants, with a total 32,231 days of data collected across all studies (per study: mean [SD] 4,028.88 [3,758.57] days; range 7–15 days). A total of 915 participants were then excluded from this dataset based on exclusion criteria (Figure 1), resulting in a total final sample size of 2,404 participants. Across studies, participants provided a total of 26,121 days of sleep diaries (per study: mean [SD] 3,265.12 [2,749.21] days), 10,828 usable days of actigraphy data (per study: mean [SD] 1,353.50 [1,586.69] days), and 2,481 usable days of EEG data (per study: mean [SD] 1,240.50 [1,167.43] days). The average rate of days of usable data across studies was 96.59% for sleep diaries (SD = 3.36%), 82.61% for actigraphy (SD = 40.59%), and 69.98% for EEG (SD = 33.06%). The overall sample was composed of primarily younger adults, women, and individuals who identified as White and non-Hispanic/Latinx (Table 1). There were significant demographic differences across studies: age, χ²(35) = 1809.70, p < 0.001; gender, χ²(7) = 230.44, p < 0.001; race, χ²(21) = 1360.60, p < 0.001; and ethnicity, χ²(4) = 132.07, p < 0.001. Although it was beyond the scope of this study, future studies using multiple samples may benefit from using meta-analytical procedures to better identify sources of study heterogeneity.

FIGURE 1.

Participant inclusion flow chart by study

TABLE 1.

Descriptive demographics of overall sample and stratified by study

Study	Overall	1	2	3	4	5	6	7	8
N (%)	2,404 (100)	360 (15.0)	123 (5.1)	552 (23.0)	904 (37.6)	70 (2.9)	182 (7.6)	120 (5.0)	93 (3.9)
Age brackets, n (%)
18–24 years	1,101 (52.7)	28 (7.8)	81 (95.3)	50 (9.1)	587 (93.2)	21 (30.0)	155 (85.6)	88 (73.3)	91 (98.9)
25–34 years	322 (15.4)	134 (37.3)	4 (4.7)	85 (15.4)	32 (5.1)	22 (31.4)	23 (12.7)	21 (17.5)	1 (1.1)
35–44 years	235 (11.2)	100 (27.9)	-	99 (17.9)	6 (1.0)	21 (30.0)	2 (1.1)	7 (5.8)	-
45–54 years	157 (7.5)	63 (17.5)	-	82 (14.9)	4 (0.6)	5 (7.1)	1 (0.6)	2 (1.7)	-
55–64 years	116 (5.6)	34 (9.5)	-	78 (14.1)	1 (0.2)	1 (1.4)	-	2 (1.7)	-
≥65 years	158 (7.6)	-	-	158 (28.6)	-	-	-	-	-
Female, n (%)	1,655 (69.7)	331 (91.9)	73 (59.3)	266 (48.2)	653 (74.6)	45 (64.3)	121 (66.5)	92 (76.7)	74 (79.6)
Race, n (%)
White	1,457 (67.3)	290 (81.7)	78 (63.4)	401 (72.9)	606 (67.6)	60 (88.2)	43 (23.9)	37 (30.8)	8 (8.6)
Black	292 (13.5)	19 (5.4)	13 (10.6)	142 (25.8)	115 (12.8)	3 (4.4)	-	-	2 (2.2)
Asian	237 (10.9)	34 (9.6)	4 (3.3)	7 (1.3)	47 (5.2)	3 (4.4)	121 (67.2)	78 (65.0)	79 (84.9)
Other/multiracial	179 (8.3)	12 (3.4)	28 (22.8)	-	128 (14.3)	2 (2.9)	16 (8.9)	5 (4.2)	4 (4.3)
Hispanic/Latinx, n (%)	178 (8.9)	41 (11.5)	38 (31.4)	1 (0.2)	92 (10.2)	6 (8.7)	-	-	-
College education, n (%)	2,044 (93.6)	360 (100.0)	123 (100.0)	332 (70.6)	904 (100.0)	68 (97.1)	164 (100.0)	-	93 (100.0)
Employed, n (%)	617 (67.0)	360 (100.0)	51 (41.5)	-	-	60 (85.7)	67 (43.2)	59 (49.2)	20 (21.5)
Relationship status, n (%)
Single	1,308 (71.4)	96 (26.7)	118 (96.7)	-	827 (93.3)	13 (18.6)	118 (65.2)	68 (56.7)	68 (73.1)
Married	349 (19.1)	229 (63.6)	3 (2.5)	-	51 (5.8)	31 (44.3)	15 (8.3)	17 (14.2)	3 (3.2)
Other	175 (9.6)	35 (9.7)	1 (0.8)	-	8 (0.9)	26 (37.1)	48 (26.5)	35 (29.2)	22 (23.7)
Children, n, mean (SD)	0.40 (0.91)	1.34 (1.25)	0.02 (0.18)	-	0.07 (0.38)	-	-	-	-
BMI, kg/m2, mean (SD)	24.82 (5.39)	26.97 (5.62)	24.66 (5.64)	26.36 (5.23)	23.72 (4.97)	29.39 (7.75)	22.31 (3.63)	22.76 (3.55)	21.99 (3.56)
Southern hemisphere, n (%)	395 (16.4)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	182 (100.0)	120 (100.0)	93 (100.0)
Completed diary days, mean (SD)	10.17 (3.51)	13.89 (1.87)	6.91 (0.46)	13.93 (0.49)	6.97 (0.24)	6.91 (0.37)	10.95 (1.72)	6.86 (1.27)	14.02 (2.25)
Completed actigraphy days, mean (SD)	8.87 (2.31)	12.64 (2.53)	6.92 (0.54)	-	-	6.69 (1.43)	10.49 (2.08)	3.69 (3.27)	12.78 (4.01)
Completed EEG days, mean (SD)	8.48 (4.64)	-	-	-	-	5.93 (1.61)	-	-	11.03 (3.84)

Abbreviations: BMI, body mass index; BVM, Bayesian Variability Model; CV, coefficient of variation; EEG, electroencephalography; iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences.

Columns indicate study ID. n and % are used to represent number of participants within the group and percentage of participants out of the total sample within the group.

Table 2 shows the mean and IIV in sleep duration and efficiency by each measurement type and IIV metric. Across sleep measurement types (diary, actigraphy, EEG), sleep duration fluctuated from night-to-night within-people (by 85.80 [diary], 77.41 [actigraphy], and 67.04 [EEG] min for iSD; by 118.91, 108.89, and 91.93 min for RMSSD; by 19.19%, 19.11%, and 18.57% for CV; and by 60.60, 58.20, and 48.60 min for BVM). Sleep efficiency also fluctuated within-people across sleep measurement types (by 5.18% [diary], 5.22% [actigraphy], and 6.46% [EEG] for iSD; by 7.01%, 7.08%, and 8.44% for RMSSD; by 5.80%, 6.27%, and 8.14% for CV; and by 3.40%, 3.58%, and 4.16% for BVM). Significant group differences by study (p < 0.05) were detected for all sleep parameters except for EEG-determined mean sleep duration and most variability measures (see Table S2 for Tukey’s HSD group comparisons). Therefore, all demographic comparison models were run as one-way ANCOVAs including categorical study as a covariate to control for between-study heterogeneity. For descriptive purposes, Tables S3 and S4 display correlations (Pearson r) examining relationships between sleep duration (Table S3) and sleep efficiency (Table S4), as measured by diary, actigraphy, and EEG, and summarised by mean, iSD, RMSSD, CV, and BVM.

TABLE 2.

Descriptive statistics of mean and intraindividual variability in sleep parameters for overall sample and stratified by study

	Overall
Study	Mean (SD)	p
Diary DUR, n	2,404
Mean	451.99 (48.58)	<0.001
iSD	85.80 (41.03)	<0.001
RMSSD	118.91 (63.35)	<0.001
CV	19.19 (4.44)	<0.001
BVM	60.60 (25.20)	<0.001
Actigraphy DUR, n	885
Mean	412.98 (50.30)	<0.001
iSD	77.41 (33.74)	<0.001
RMSSD	108.89 (52.96)	<0.001
CV	19.11 (8.94)	<0.001
BVM	58.20 (21.60)	<0.001
EEG DUR, n	159
Mean	375.05(47.09)	0.627
iSD	67.04(28.74)	0.163
RMSSD	91.93(43.94)	0.053
CV	18.57(10.24)	0.287
BVM	48.60(18.00)	0.023
Diary EFF (n)	2404
Mean	91.71(4.44)	<0.001
iSD	5.18(3.87)	<0.001
RMSSD	7.01(5.48)	<0.001
CV	5.80(4.59)	<0.001
BVM	3.40(2.19)	<0.001
Actigraphy EFF (n)	885
Mean	85.81(5.91)	<0.001
iSD	5.22(3.35)	0.020
RMSSD	7.08(4.48)	0.004
CV	6.27(4.55)	<0.001
BVM	3.58(1.40)	0.337
EEG EFF (n)	159
Mean	82.60(6.71)	0.001
iSD	6.46(3.92)	0.953
RMSSD	8.44(5.35)	0.858
CV	8.14(5.87)	0.480
BVM	4.16(2.04)	0.753

Abbreviations: BVM, Bayesian Variability Model; CV, coefficient of variation; DUR, sleep duration; EEG, electroencephalography; EFF, sleep efficiency; iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences.

Descriptive sleep variables for whole sample and then stratified by study. The significance test and partial eta squared are the results of one-way analysis of variance (ANOVA) models examining mean differences in sleep parameter (row) by study (column). ANOVA models assessing intraindividual variability sleep parameters were fit to a gamma residual distribution with a log link. ANOVA models assessing mean sleep parameters were fit to a Gaussian error distribution. Bolded p values represent significance at p < 0.05.

Age differences in IIV in sleep duration and efficiency

Sleep duration

There were no significant age differences in EEG-determined sleep duration iSD, RMSSD, CV, or BVM after controlling for study (Table 3). There were significant age differences in diary-determined sleep duration iSD, RMSSD, CV, and BVM (Table 3; see Figure 2 for an unadjusted comparison). Individuals aged 18–24 years had significantly greater sleep duration iSD than all other age groups (Cohen’s d ranging 0.28–0.99). Individuals aged 25–34 years also had significantly greater sleep duration iSD than individuals aged 55–64 years (Cohen’s d = 0.44) and ages 65+ (Cohen’s d = 0.71). Individuals ages 35–44 (Cohen’s d = 0.78) and 45–54 (Cohen’s d = 0.78) had significantly greater sleep duration iSD than individuals aged ≥65 years. Post hoc comparisons for the RMSSD, CV, and BVM models matched the significant group differences, directionality, and effect sizes of the results for iSD (Table 3). When adjusting for gender, race, and study we found similar results (Table S5).

TABLE 3.

Estimated marginal mean differences in sleep parameters by age

Study	Age brackets 18–24 yearsa EMM (SE)	25–34 yearsb EMM (SE)	35–44 yearsc EMM (SE)	45–54 yearsd EMM (SE)	55–64 yearse EMM (SE)	≥65 yearsf EMM (SE)	p
Diary DUR, n	1,101	322	235	157	116	158
Mean	456.00 (1.92)c	447.00 (3.06)	442.00 (3.72)f	438.00 (4.39)f	439.00 (4.99)f	461.00 (4.80)	<0.001
iSD	87.30 (1.57)b,c,d,e,f	76.20 (2.18)e,f	71.70 (2.51)f	71.20 (2.93)f	61.40 (2.88)	53.90 (2.43)	<0.001
RMSSD	121.60 (2.43)b,c,d,e,f	104.60 (3.33)e,f	97.60 (3.78)f	97.80 (4.47)f	85.90 (4.47)	74.50 (3.72)	<0.001
CV	19.50 (0.36)b,c,d,e,f	17.20 (0.51)e,f	16.50 (0.60)f	16.50 (0.70)f	14.20 (0.69)	11.90 (0.56)	<0.001
BVM	62.70 (1.06)c,d,e,f	56.60 (1.51)e,f	52.90 (1.74)f	52.80 (2.01)f	47.60 (2.06)	41.90 (1.75)	<0.001
Actigraphy DUR, n	424	190	123	70	37	0
Mean	411.00 (2.66)	417.00 (4.49)	416.00 (5.51)	416.00 (6.77)	406.00 (8.64)	-	0.636
iSD	76.90 (1.79)b,c,d,e	65.70 (2.59)	57.80 (2.79)	57.90 (3.43)	52.90 (4.01)	-	<0.001
RMSSD	109.20 (2.85)b,c,d,e	90.40 (4.00)	78.20 (4.23)	80.00 (5.31)	68.10 (5.77)	-	<0.001
CV	19.20 (0.48)b,c,d,e	15.90 (0.67)	14.00 (0.72)	13.90 (0.88)	13.20 (1.07)	-	<0.001
BVM	56.70 (1.19)	52.40 (1.85)	50.90 (2.19)	52.00 (2.73)	46.40 (3.09)	-	0.069
EEG DUR, n	111	22	19	5	1	0
Mean	372.00 (5.68)	391.00 (11.30)	374.00 (12.21)	-	-	-	0.418
iSD	70.80 (3.64)	59.60 (6.08)	54.90 (6.06)	-	-	-	0.319
RMSSD	98.30 (5.53)	72.40 (8.07)	77.80 (9.38)	-	-	-	0.205
CV	20.00 (1.30)	15.40 (1.99)	14.70 (2.05)	-	-	-	0.270
BVM	47.60 (2.48)	55.70 (5.53)	40.70 (4.25)	-	-	-	0.164
Diary EFF, n	1101	322	235	157	116	158
Mean	92.40 (0.18)e	91.99 (0.28)e	92.44 (0.34)e	91.86 (0.40)	91.84 (0.46)	90.57 (0.44)	0.003
iSD	5.13 (0.16)	5.15 (0.26)	4.94 (0.30)	5.47 (0.39)	5.38 (0.44)	5.12 (0.40)	0.826
RMSSD	6.91 (0.23)	6.84 (0.36)	6.61 (0.42)	7.42 (0.56)	7.45 (0.64)	6.90 (0.57)	0.698
CV	5.71 (0.19)	5.74 (0.30)	5.48 (0.35)	6.12 (0.46)	6.00 (0.52)	5.84 (0.48)	0.828
BVM	3.30 (0.09)	3.28 (0.14)	3.17 (0.16)	3.47 (0.21)	3.46 (0.24)	3.36 (0.22)	0.756
Actigraphy EFF, n	424	190	123	70	37	0
Mean	84.80 (0.28)d	85.60 (0.47)	86.90 (0.57)	87.60 (0.70)	86.30 (0.90)	-	0.002
iSD	5.35 (0.18)	5.01 (0.29)	4.25 (0.30)	4.00 (0.35)	4.57 (0.50)	-	0.009
RMSSD	7.30 (0.24)c	6.73 (0.38)	5.48 (0.38	5.70 (0.48)	5.59 (0.60)	-	0.002
CV	6.44 (0.24)	5.98 (0.38)	5.01 (0.40)	4.63 (0.45)	5.67 (0.70)	-	0.010
BVM	3.70 (0.08)	3.48 (0.13)	3.35 (0.16)	3.08 (0.17)	3.28 (0.23)	-	0.067
EEG EFF, n	111	22	19	5	1	0
Mean	81.70 (0.78)	85.00 (1.55)	83.30 (1.67)	-	-	-	0.162
iSD	6.62 (0.48)	5.97 (0.87)	5.60 (0.88)	-	-	-	0.538
RMSSD	9.41 (0.69)	5.92 (0.86)	6.41 (1.01)	-	-	-	0.031
CV	8.59 (0.73)	7.22 (1.22)	6.70 (1.22)	-	-	-	0.472
BVM	4.19 (0.29)	3.98 (0.54)	4.29 (0.61)	-	-	-	0.950

Abbreviations: BVM, Bayesian Variability Model; CV, coefficient of variation; DUR, sleep duration; EEG, electroencephalography; EFF, sleep efficiency; EMM, estimated marginal means (controlling for study and reported in the original scale); iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences; SE, standard error.

Bold values represent significant values (p < 0.01) after adjusting for false discovery rate using Benjamini–Hochberg procedure. Superscripts represent significant p < 0.01 group pairwise comparisons. Analysis of covariance (ANCOVA) models assessing intraindividual variability sleep parameters were fit to a gamma residual distribution with a log link. ANCOVA models assessing mean sleep parameters were fit to a Gaussian error distribution. Type III. Controlled by Study. Post hoc comparisons were not tested for age groups with <10 people.

FIGURE 2.

Unadjusted notched boxplots of diary- and actigraphy-determined intraindividual variability in sleep duration by age. Acti, actigraphy; BVM, Bayesian Variability Model; iSD, intraindividual standard deviation; RMSSD, root mean squared successive differences; TST, total sleep time represents sleep duration. The line/notch in each boxplot represents the median value. The box represents the interquartile range (quartile groups 2 and 3). The lower and uppers whiskers represent quartile groups 1 (from minimum value observed) and 4 (to maximum value observed), respectively. Distributions are not adjusted for study.

There were also significant age differences in actigraphy-determined sleep duration iSD, RMSSD, and CV, but not BVM, after controlling for study (Table 3; see Figure 2 for an unadjusted comparison). Post hoc comparisons revealed that individuals aged 18–24 years had significantly greater sleep duration iSD than individuals aged 25–34 years (Cohen’s d = 0.38), individuals aged 35–44 years (Cohen’s d = 0.69), individuals aged 45–54 years (Cohen’s d = 0.69), and individuals aged 55–64 years (Cohen’s d = 0.90). Post hoc comparisons for the RMSSD and CV models matched the significant group differences, directionality, and effect sizes of the results for iSD (Table 3). When adjusting for gender, race, and study we found a similar pattern of results (Table S5).

Sleep efficiency

There were no significant age differences in diary- or EEG-determined sleep efficiency iSD, RMSSD, CV, or BVM after controlling for study (Table 3). There were significant age differences in actigraphy-determined sleep efficiency RMSSD, but not iSD, CV, or BVM (Table 3). Individuals aged 18–24 years had significantly more sleep efficiency RMSSD than individuals aged 35–44 years (Cohen’s d = 0.46; Table 3). When adjusting for gender, race, and study we found a similar, but less robust pattern of results (Table S6).

Gender differences in IIV in sleep duration and efficiency

Sleep duration

There were no significant gender differences in diary-, actigraphy-, or EEG-determined sleep duration iSD, RMSSD, CV, or BVM after controlling for study (Table 4). When adjusting for age, race, and study we found a similar pattern of results (Table S5).

TABLE 4.

Estimated marginal mean differences in sleep parameters by gender

Gender	Malea EMM (SE)	Femaleb EMM (SE)	p
Diary DUR, n	720	1.655
Mean	450.00 (2.06)	453.00 (1.51)	0.219
iSD	79.50 (1.58)	82.00 (1.19)	0.149
RMSSD	109.00 (2.39)	114.00 (1.83)	0.067
CV	17.80 (0.36)	18.40 (0.27)	0.186
BVM	57.40 (1.03)	59.30 (0.78)	0.080
Actigraphy DUR, n	194	691
Mean	398.00 (3.43)a	417.00 (2.10)	<0.001
iSD	68.90 (2.17)	72.70 (1.39)	0.146
RMSSD	96.70 (3.40)	101.40 (2.18)	0.243
CV	17.60 (0.59)	17.80 (0.36)	0.796
BVM	51.80 (1.44)	55.90 (0.95)	0.017
EEG DUR, n	42	117
Mean	378.00 (7.32)	374.00 (4.49)	0.631
iSD	60.50 (4.03)	68.70 (2.77)	0.109
RMSSD	80.60 (6.04)	94.50 (4.26)	0.076
CV	16.30 (1.38)	19.20 (0.98)	0.105
BVM	45.70 (2.97)	48.20 (1.87)	0.506
Diary EFF, n	720	1655
Mean	92.79 (0.18)a	91.98 (0.14)	<0.001
iSD	4.62 (0.15)a	5.40 (0.13)	<0.001
RMSSD	6.20 (0.21)a	7.26 (0.18)	<0.001
CV	5.12 (0.18)a	6.03 (0.15)	<0.001
BVM	3.10 (0.09)a	3.39 (0.07)	0.002
Actigraphy EFF, n	194	691
Mean	83.70 (0.37)a	85.70 (0.23)	<0.001
iSD	5.34 (0.25)	5.07 (0.14)	0.342
RMSSD	7.18 (0.33)	6.86 (0.19)	0.380
CV	6.50 (0.34)	6.08 (0.20)	0.255
BVM	3.70 (0.11)	3.53 (0.06)	0.174
EEG EFF (n)	42	117
Mean	84.10 (1.00)	81.70 (0.61)	0.038
iSD	4.95 (0.44)a	7.04 (0.38)	0.002
RMSSD	6.67 (0.64)	9.15 (0.53)	0.008
CV	5.97 (0.61)a	9.02 (0.56)	0.001
BVM	3.63 (0.30)	4.36 (0.22)	0.070

Abbreviations: BVM, Bayesian Variability Model; CV, coefficient of variation; DUR, sleep duration; EEG, electroencephalography; EFF, sleep efficiency; EMM, estimated marginal means (controlling for study and reported in the original scale); iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences; SE, standard error.

Bold values represent significant values (p < 0.01) after adjusting for false discovery rate using Benjamini–Hochberg procedure. Superscripts represent significant p < 0.01 group pairwise comparisons. Analysis of covariance (ANCOVA) models assessing intraindividual variability sleep parameters were fit to a gamma residual distribution with a log link. ANCOVA models assessing Mean sleep parameters were fit to a Gaussian error distribution. Type III. Controlled by Study.

Sleep efficiency

There were no significant gender differences in actigraphy-determined sleep efficiency iSD, RMSSD, CV, or BVM after controlling for study (Table 4). However, there were significant gender differences in diary-determined sleep efficiency iSD, RMSSD, CV, and BVM (Table 4). Post hoc comparisons revealed that men had significantly less diary sleep efficiency iSD, RMSSD, CV, and BVM than women (Cohen’s d ranging −0.14 to −0.21; Table 4). There were also significant gender differences in EEG-determined sleep efficiency iSD and CV, but not for RMSSD or BVM (Table 4). Post hoc comparisons revealed that men had significantly less sleep efficiency iSD (Cohen’s d = −0.63) and sleep efficiency CV (Cohen’s d = −0.67) than women. When adjusting for age, race, and study we found a similar pattern of results (Table S6).

Racial differences in IIV in sleep duration and efficiency

Sleep duration

There were no significant racial differences in EEG-determined sleep duration iSD, RMSSD, CV, or BVM after controlling for study (Table 5). However, there were significant racial differences in diary-determined sleep duration iSD and CV, but not RMSSD or BVM (Table 5). White adults had significantly shorter sleep duration iSD than Black individuals (Cohen’s d = −0.22). Post hoc comparisons for the CV model mirrored the significant group differences for iSD between White adults and Black adults. Post hoc comparisons for CV also revealed an additional significant group pairwise comparison between individuals where White adults had significantly less sleep duration CV than Asian individuals (Cohen’s d = −0.25; Table 5). There were also significant racial differences in actigraphy-determined sleep duration CV, but not for iSD, RMSSD, or BVM (Table 5). White adults had significantly shorter sleep duration CV than Asian individuals (Cohen’s d = −0.40; Table 5). When adjusting for age, gender, and study we found a similar, but less robust pattern of results (Table S5).

TABLE 5.

Estimated marginal mean differences in sleep parameters by race

Race	Whitea EMM (SE)	Blackb EMM (SE)	Asianc EMM (SE)	Multiracial/Otherd EMM (SE)	p
Diary DUR, n	1,523	294	373	195
Mean	457.00 (1.84)c	456.00 (3.27)	443.00 (2.89)	455.00 (3.68)	0.002
iSD	77.80 (1.36)b	86.60 (2.68)	85.50 (2.35)	85.40 (2.98)	<0.001
RMSSD	109.00 (2.11)	118.00 (4.06)	117.00 (3.55)	120.00 (4.65)	0.004
CV	17.30 (0.31)b,c	19.10 (0.60)	19.60 (0.55)	19.00 (0.68)	<0.001
BVM	57.40 (0.92)	62.00 (1.74)	60.70 (1.52)	58.00 (1.85)	0.012
Actigraphy DUR, n	496	37	279	64
Mean	427.00 (2.69)c	403.00 (7.77)	394.00 (3.29)	410.00 (5.93)	<0.001
iSD	68.70 (1.70)	73.60 (5.26)	75.10 (2.28)	75.20 (4.26)	0.113
RMSSD	96.80 (2.69)	100.00 (8.00)	104.20 (3.54)	104.50 (6.38)	0.368
CV	16.30 (0.43)c	18.40 (1.40)	19.40 (0.62)	18.60 (1.08)	<0.001
BVM	55.10(1.21)	57.60 (3.58)	54.40 (1.46)	54.90 (2.65)	0.884
EEG DUR, n	65	5	81	6
Mean	389.00 (7.64)	-	363.00 (8.03)	-	0.043
iSD	59.70 (4.19)	-	75.60 (5.59)	-	0.291
RMSSD	85.50 (6.73)	-	97.50 (8.12)	-	0.783
CV	15.80 (1.42)	-	21.80 (2.07)	-	0.294
BVM	46.00 (2.94)	-	50.20 (3.36)	-	0.764
Diary EFF, n	1523	294	373	195
Mean	92.03 (0.16)b	91.07 (0.29)c	92.91 (0.26)	92.01 (0.33)	<0.001
iSD	5.17 (0.15)	5.61 (0.29)	4.99 (0.23)	5.35 (0.31)	0.307
RMSSD	6.92 (0.21)	7.40 (0.40)	6.81 (0.33)	7.20 (0.44)	0.536
CV	5.77 (0.18)	6.29 (0.35)	5.53 (0.27)	5.99 (0.37)	0.276
BVM	3.30 (0.08)	3.53 (0.16)	3.23 (0.13)	3.35 (0.17)	0.398
Actigraphy EFF, n	496	37	279	64
Mean	85.70 (0.30)	83.60 (0.86)	84.80 (0.36)	84.60 (0.66)	0.024
iSD	4.78 (0.17)d	5.06 (0.52)	5.40 (0.24)	6.45 (0.51)	0.002
RMSSD	6.43 (0.23)	7.09 (0.72)	7.39 (0.32)	8.29 (0.64)	0.004
CV	5.70 (0.23)d	6.13 (0.71)	6.54 (0.32)	7.93 (0.70)	0.002
BVM	3.47 (0.08)	3.73 (0.25)	3.68 (0.10)	3.72 (0.19)	0.280
EEG EFF, n	65	5	81	6
Mean	82.20 (1.08)	-	82.40 (1.13)	-	0.880
iSD	6.60 (0.66)	-	6.41 (0.67)	-	0.434
RMSSD	9.46 (0.99)	-	7.68 (0.85)	-	0.332
CV	8.40 (0.98)	-	8.13 (0.99)	-	0.468
BVM	4.12 (0.36)	-	4.24 (0.38)	-	0.863

Abbreviations: BVM, Bayesian Variability Model; CV, coefficient of variation; DUR, sleep duration; EEG, electroencephalography; EFF, sleep efficiency; EMM, estimated marginal means (controlling for study and reported in the original scale); iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences; SE, standard error.

Bold values represent significant values (p < 0.01) after adjusting for false discovery rate using Benjamini–Hochberg procedure. Superscripts represent significant p < 0.01 group pairwise comparisons. Analysis of covariance (ANCOVA) models assessing intraindividual variability sleep parameters were fit to a gamma residual distribution with a log link. ANCOVA models assessing mean sleep parameters were fit to a Gaussian error distribution. Type III. Controlled by Study. Post hoc comparisons were not tested for age groups with <10 people.

Sleep efficiency

There were no significant racial differences in diary- or EEG-determined sleep efficiency iSD, RMSSD, CV, or BVM after controlling for study (Table 5). There were also no significant racial differences in actigraphy-determined sleep efficiency RMSSD or BVM. However, there were significant racial differences in actigraphy-determined sleep efficiency iSD and CV. White adults had significantly less actigraphy sleep efficiency iSD and CV than Multiracial/Other individuals (Cohen’s d ranging from −0.56 to −0.57; Table 5). When adjusting for age, gender, and study we found a similar, but more robust pattern of results (Table S6).

Ethnic differences in IIV in sleep duration and efficiency

Sleep duration

There were no significant ethnic differences in diary-, actigraphy-, or EEG-determined sleep efficiency iSD, RMSSD, CV, or BVM after controlling for study (Table 6).

TABLE 6.

Estimated marginal mean differences in sleep parameters by ethnicity

Ethnicity	Not Hispanic/Latinx EMM (SE)	Hispanic/Latinx EMM (SE)	p
Diary DUR, n	1,822	178
Mean	444.00 (1.65)	442.00 (3.69)	0.593
iSD	80.90 (1.31)	86.80 (3.13)	0.055
RMSSD	112.00 (1.99)	122.00 (4.82)	0.043
CV	18.40 (0.30)	19.80 (0.72)	0.048
BVM	58.70 (0.86)	60.60 (1.97)	0.351
Actigraphy DUR, n	456	84
Mean	395.00 (2.53)	383.00 (5.00)	0.029
iSD	71.30 (1.75)	80.60 (3.91)	0.017
RMSSD	101.00 (2.71)	115.00 (6.11)	0.016
CV	18.30 (0.48)	21.40 (1.10)	0.004
BVM	53.80 (1.12)	58.60 (2.44)	0.052
EEG DUR, n	60	6
Mean	377.00 (4.82)	-	0.013
iSD	62.40 (3.61)	-	0.348
RMSSD	82.80 (5.16)	-	0.412
CV	17.10 (1.19)	-	0.207
BVM	53.80 (1.12)	-	0.052
Diary EFF, n	1822	178
Mean	91.14 (0.15)	91.35 (0.34)	0.550
iSD	5.61 (0.14)	5.11 (0.29)	0.127
RMSSD	7.60 (0.20)	6.99 (0.42)	0.184
CV	6.31 (0.17)	5.76 (0.35)	0.159
BVM	3.56 (0.08)	3.46 (0.17)	0.568
Actigraphy EFF, n	456	84
Mean	82.60 (0.30)	82.50 (0.59)	0.769
iSD	5.50 (0.22)	6.07 (0.48)	0.241
RMSSD	7.52 (0.29)	8.18 (0.63)	0.291
CV	6.87 (0.31)	7.58 (0.68)	0.296
BVM	3.65 (0.09)	3.81 (0.19)	0.407
EEG EFF, n	60	6
Mean	80.70 (0.97)	-	0.461
iSD	6.39 (0.57)	-	0.679
RMSSD	8.35 (0.69)	-	0.517
CV	8.42 (0.94)	-	0.702
BVM	3.65 (0.09)	-	0.407

Abbreviations: BVM, Bayesian Variability Model; CV, coefficient of variation; DUR, sleep duration; EEG, electroencephalography; EFF, represents sleep efficiency; EMM, estimated marginal means (controlling for study and reported in the original scale); iSD, intraindividual standard deviation; RMSSD, root mean square of successive differences; SE, standard error.

Bold values represent significant values (p < 0.01) after adjusting for false discovery rate using Benjamini–Hochberg procedure. Superscripts represent significant p < 0.01 group pairwise comparisons. Analysis of covariance (ANCOVA) models assessing intraindividual variability sleep parameters were fit to a gamma residual distribution with a log link. ANCOVA models assessing mean sleep parameters were fit to a Gaussian error distribution. Type III. Controlled by Study. Post hoc comparisons were not tested for age groups with <10 people.

Sleep efficiency

There were no significant ethnic differences in diary-, actigraphy-, or EEG-determined sleep efficiency iSD, RMSSD, CV, or BVM after controlling for study (Table 6).

DISCUSSION

We pooled eight datasets to compute multiple metrics of IIV in sleep duration and sleep efficiency across diary, actigraphy, and EEG sleep measures. We also examined differences in IIV in sleep across age, gender, and racial/ethnic identities. Across all sleep measurement types, sleep duration and efficiency varied from night-to-night within-people. Demographic comparisons revealed: (i) younger adults had more IIV in diary and actigraphy sleep duration; (ii) gender differences in sleep IIV were inconsistent, and (iii) White and non-Hispanic/Latinx adults had less IIV in sleep duration and efficiency compared to racial/ethnic minority groups. These results suggest that even among relatively healthy sleepers, sleep patterns are not consistent from night-to-night.

Our findings generally align with previous studies, which report mean iSD values ranging from 72–98 min for sleep duration and 6%–7% for sleep efficiency (Jonasdottir et al., 2020; Slavish et al., 2019; Slavish et al., 2020). We added to this work by incorporating a Bayesian metric of sleep IIV. Compared to the iSD, CV, and RMSSD, the key advantage of BVM is that it takes into account the measurement error of IIV, which produces relatively unbiased estimates for the relations between IIV and an outcome, in contrast to an iSD where measurement error is ignored (Bei et al., 2017).

Additionally, there were significant mean differences in metrics of IIV in sleep duration and/or sleep efficiency by gender, age, and racial/ethnic identity after adjusting for study. Individuals in the youngest age bracket (18–24 years) consistently had greater IIV in diary and actigraphy sleep duration across all IIV metrics compared to all other age groups. There were no significant age differences in IIV in EEG sleep. These findings corroborate previous studies showing high IIV in diary- and actigraphy-determined sleep duration in younger age groups (Bailey & Silver, 2014; Bei et al., 2019; Diem et al., 2016; Jonasdottir et al., 2020; Merklinger-Gruchala et al., 2008; Slavish et al., 2020; Xu et al., 2018). In general, younger adults appear to have more variability in sleep, which may be due to a combination of biological factors (e.g., changes to circadian rhythms across young adulthood), as well as more dynamic social, work, and/or school schedules (e.g., social jetlag) that interfere with the ability to maintain a consistent sleep schedule compared to older adults (Logan & McClung, 2019).

Age differences in IIV in sleep efficiency were generally not supported. In the present study, participants with mean diary sleep efficiency <80% were excluded to enhance internal validity and capture a more precise sample of healthy sleepers. This may have restricted the range of IIV in sleep efficiency, suppressing the ability to detect possible age differences. In general, individuals with lower average sleep efficiency tend to have greater variability in sleep efficiency (Dietch & Taylor, 2021). Future studies should compare IIV in sleep efficiency in a more age-diverse sample of healthy sleepers and in individuals with clinical levels of insomnia and other sleep disorders.

There were no gender differences in IIV in sleep duration across all sleep measurement types and IIV metrics. These findings add to a growing body of evidence suggesting there are not robust gender differences in IIV in sleep duration (Bei et al., 2019; Kaufmann et al., 2016; Minors et al., 1998; Ogilvie et al., 2016; Xu et al., 2018). However, it is possible that contextual factors such as age, racial/ethnic identify, or weekly obligations moderate the association between gender and IIV in sleep duration, leading to conflicting results (Grandner et al., 2016). For example, Jonasdottir et al. (2020) found small significant gender differences in actigraphy-determined IIV in sleep duration when results were stratified by weekday versus week-end. Examining all days simultaneously in the present study may have obscured these gender differences in IIV in sleep duration. In addition to, day of the week, gender differences in IIV in sleep duration may vary widely as a function of other contextual factors we did not measure (e.g., caregiving responsibilities, daily stress). When examining IIV in sleep efficiency, men had less IIV in diary- and EEG-determined sleep efficiency. To our knowledge, no other studies have examined gender differences in IIV in sleep efficiency. Findings preliminarily suggest women may have greater night-to-night variability in sleep efficiency than men. These findings map onto previous research with mean sleep efficiency (Mallampalli & Carter, 2014).

White adults had the lowest diary-determined IIV in sleep duration compared to Black, Asian, or Multiracial/Other individuals. This disparity may be attributed to White adults experiencing fewer daily stressors than people from minority racial/ethnic groups (Grandner et al., 2016; Ogilvie et al., 2016). Stress is robustly associated with greater IIV in sleep (Kalmbach et al., 2018). People from minority racial/ethnic groups are also disproportionately engaged in shift work and more likely to have unpredictable work schedules, which may interfere with consistent sleep (Grandner et al., 2016). Further, there are vast racial and ethnic disparities in healthcare coverage and self-reported health among Black and Hispanic/Latinx adults (Centers for Disease Control and Prevention, 2020). These inequities may relate to increases in medical symptoms, disability, pain, and/or medication use that interfere with the ability to maintain consistent sleep. White individuals also had the lowest actigraphy-determined iSD and CV IIV in sleep efficiency compared to Black, Asian, or Multiracial/Other individuals. Previous studies examining racial differences in IIV in sleep efficiency have primarily been conducted using actigraphy to measure sleep, and corroborate our results (Knutson et al., 2007; Lunsford-Avery et al., 2018; Mezick et al., 2009; Ogilvie et al., 2016).

Contrary to two previous studies that found Hispanic/Latinx individuals had greater actigraphy-determined iSD IIV in sleep duration and sleep efficiency (Lunsford-Avery et al., 2018; Ogilvie et al., 2016), we found no significant ethnic differences in any measures of IIV in sleep duration or efficiency. It is possible that our study was underpowered to detect significant mean differences in IIV measures of sleep duration and efficiency due to fewer Hispanic/Latinx individuals (n = 178) across all eight studies.

Limitations and future directions

This study was the first to assess descriptive characteristics of multiple metrics of IIV in two facets of sleep (i.e. duration and efficiency) and across three different sleep measurement types (i.e., diary, actigraphy, and EEG) in a pooled sample of >2,000 participants and >20,000 days of sleep data. However, it is important to note several limitations of this approach. First, study heterogeneity (e.g., design, location, year of collection) between studies cannot be ignored, even when relatively standardised measurements of sleep duration and efficiency were used across studies. By controlling for study, a significant portion of shared variance may have been removed from the model, which may explain the smaller than expected effect sizes for demographic variables. Furthermore, due to differences in sample sizes across studies, it is possible that studies with larger samples drove the main effects found in the overall results. Additionally, not all studies used the same sleep measurement types or included all demographic groups. Therefore, different analyses resulted in different subsamples of studies. Although our analyses capitalised on pooling results across all available data, larger studies with more diverse samples and standardied sleep measures are needed to derive more direct comparisons of IIV across sleep measurement types and demographic groups.

Second, while pooling datasets provided a larger and more diverse sample, most participants were still young adults (aged 18–24 years; 49.1%), non-Hispanic/Latinx (91.1%), White (67.3%), and college-educated (93.1%). Several demographic groups were under-represented within the pooled datasets, such as Hispanic/Latinx individuals and older adults (aged ≥65 years). Underrepresentation of these demographic groups particularly impacted the generalisability of results for both actigraphy- and EEG-determined sleep. Additionally, due to lack of standard information on clinically diagnosed sleep disturbances and sleep medication usage across studies, we resorted to using mean diary-determined quantitative criterion to exclude non-healthy sleepers and to increase the validity of the sample used in the study. However, this may have led to a restriction in range in IIV in sleep, and underpowered results for several demographic groups. Finally, we did not examine potential mediators of associations, such as socioeconomic status, shift work, caregiving, medical symptoms, stress, or discrimination. Future studies should identify mediating mechanisms between IIV in sleep and demographics to better understand the source of IIV in sleep between and within demographic groups.

Third, although at least 12 observed nights are needed as the “gold standard” to achieve temporal stability in mean sleep parameters (Wohlgemuth et al., 1999), this has not yet been assessed for IIV metrics of sleep. In the present study, we chose to exclude participants with <3 days of sleep data, as prior work suggests including of more nights of sleep data may increase the reliability of IIV metrics of sleep parameters (Bei et al., 2016). However, future research is needed to establish a standard for the field. Finally, future studies should consider extending this work by examining and comparing properties of other metrics of IIV in sleep (e.g., sleep regularity indices; Fischer et al., 2021), and examining IIV in other unique dimensions of sleep, such as sleep timing (Buysse, 2014).

Overall, along with findings from other studies, our work helps characterise what is a typical amount of within-person variability in sleep among healthy sleepers. Given the association between IIV and physical and psychological outcomes (Slavish et al., 2019), such information may be useful to collect (in addition to mean sleep parameters) to inform clinical guidelines and treatment initiatives, as well as help identify who is most at risk of impaired sleep. Additionally, there appear to be some important demographic differences in IIV in sleep duration and efficiency. This research underscores the importance of considering demographic differences when analysing the effects of IIV in sleep. Findings may help identify who is in greatest need of interventions to help stabilise sleep patterns and improve health outcomes.

DATA AVAILABILITY STATEMENT

The data underlying this article were provided by co-authors and their corresponding labs by permission. Data will be shared on request to the corresponding author with permission of the co-authors.