Socioeconomic and racial/ethnic disparities in sleep and cognitive and academic functioning across childhood and adolescence: a meta-analytic review

Abstract

Study Objectives

Youth from socioeconomically disadvantaged backgrounds and minoritized racial/ethnic groups face a greater risk for sleep disparities and poor cognitive/academic outcomes compared to their peers from higher socioeconomic status (SES) and majority racial/ethnic groups. This meta-analysis has two main objectives. First, it examined SES and racial/ethnic variables as moderators of the association between individual sleep parameters (i.e. objective duration, objective quality, subjective duration, subjective quality, schedule/chronobiology, and variability/consistency) and overall (i.e. combined assessment) cognitive/academic outcomes (e.g. memory, GPA). Second, it assessed sleep parameters and outcome domains as moderators of the association between sleep broadly (i.e. no distinction between various sleep parameters) and overall cognitive/academic outcomes to test whether the magnitude of the association varied across specific sleep parameters and outcome domains.

Methods

Thirty-three studies comprising 410 effect sizes and 52 854 participants were included. The meta-analysis was conducted per best practices.

Results

For sleep and cognitive functioning, the overall weighted effect was r = 0.11, 95% CI [0.08, 0.14], p < .001. Several significant differential associations between various sleep parameters and cognitive functioning emerged, showing that some parameters were stronger predictors. Academic performance yielded similar results, with an overall weighted effect size of r = 0.14, 95% CI [0.10, 0.17], p < .001. SES and race/ethnicity also modified relations between certain sleep parameters and cognitive/academic functioning.

Conclusions

The findings provide evidence for SES and race/ethnicity-related disparities, highlighting areas requiring further investigation. The discussion addresses methodological limitations in studying SES and race/ethnicity in relation to sleep and offers future research directions.

Graphical Abstract

Graphical Abstract.

Introduction

The role of sleep and its relation to cognitive and academic functioning in children and adolescents has been a subject of great interest and has been extensively examined in studies, systematic reviews, and meta-analyses [1–9]. Prior meta-analyses have found significant but modest links between sleep, particularly sleep duration, and multiple cognitive domains [2, 3, 6]. However, few studies have examined how other sleep parameters (e.g. quality, schedule) relate to cognitive and academic functioning, hereafter collectively referred to as “cognitive/academic functioning.” Additionally, understanding how sociocultural characteristics, such as socioeconomic status (SES) and race/ethnicity, may influence these associations has been an unexplored area of research. The aim of this meta-analysis is to address these gaps by investigating novel questions not examined in previous meta-analyses:

• (a) Do SES and race/ethnicity moderate the relationship between sleep and cognitive or academic functioning?

• (b) Are specific sleep parameters stronger correlates of cognitive/academic functioning than others?

• (c) Are certain sleep parameters more robustly linked to cognitive functioning than academic performance?

• (d) Does the association between sleep parameters and cognitive/academic functioning vary based on participant characteristics or methodological/study factors?

Cognitive functioning and academic performance are important aspects of child and adolescent development [10, 11]. Cognitive functioning is a broad construct that encompasses crystallized and fluid intelligence, executive functioning, reasoning, memory, attention, and processing speed [12]. These components may be examined individually or collectively to assess global cognitive functioning. Academic performance is assessed through a wide variety of measures, including objective metrics such as grade point averages (GPA) and standardized test scores, as well as subjective assessments like teacher reports of academic competence and achievement. Academic performance and cognitive functioning are separate but interrelated constructs. Academic performance is influenced by cognitive functioning as well as many noncognitive factors, including motivation, differences in quality of instruction, family and societal expectations, and mental health [13]. Less often acknowledged is that cognitive functioning may be influenced by noncognitive factors [14]. The development of cognitive functioning and academic performance in childhood and adolescence have bidirectional influences [15], suggesting that consideration of them both separately (and parsing specific measures within each) but also in combination is warranted.

Sleep is a multifaceted construct with several parameters, including duration, quality, timing (e.g. schedule or chronobiology), and consistency (variability or regularity), each crucial for understanding its relationship with cognitive/academic functioning and, therefore, the focus of the current review. In recent years, imaging studies have enhanced our understanding of physiological mechanisms that link sleep to the development of brain structure and function in children and adolescents [16–18]. Sleep measurement methods include objective assessments (i.e. polysomnography, actigraphy, and multiple sleep latency tests) and subjective reports (e.g. self-reports and sleep diaries), each offering distinct advantages and limitations [19]. Guided by established conceptualizations of sleep health [20, 21], we defined more optimal sleep as characterized by longer duration, better quality, earlier sleep timing, and consistency—all of which have been linked to enhanced cognitive/academic performance in youth across developmental stages [1, 22, 23].

Although individual studies provide evidence supporting associations between various sleep parameters and cognitive/academic functioning, findings from prior meta-analyses and systematic reviews remain sparse and inconclusive. These reviews have predominantly focused on sleep duration and its relation to cognitive functioning [1, 2, 6, 7] and academic performance [3, 4, 7], reporting associations ranging from small to medium in magnitude. Some studies have also indicated modest links between sleep quality and academic performance [2–4]. Musshafen and colleagues [4] concluded that the negligible associations between sleep parameters (duration, quality) and academic performance may be due to variations in definitions and assessment methods and suggested that unexamined mediating and moderating variables may obscure the discovery of direct relations.

Disparities in sleep and cognitive/academic functioning

Sociocultural factors such as SES and race/ethnicity play a critical role in disparities related to sleep, cognitive functioning, and academic performance in youth [24–27]. SES is a broad construct that encompasses multiple dimensions of financial well-being and hardship [28, 29] and is commonly assessed through objective metrics such as income, educational attainment, occupation/employment, and neighborhood characteristics. Subjective assessments of SES, which are less studied, examine perceived financial strain and well-being, offering insight on the effects of economic stress on individuals and families [30, 31]. The operationalization of SES across multiple studies is fraught with challenges due to variations in measurement, lack of standardization, and differences across geographic regions and historical periods. To mitigate these issues and maximize the retention of effect sizes, we categorized studies into low, middle, high, or mixed SES samples based on reported SES information.

Race and ethnicity are socially constructed categories that reflect shared social and cultural beliefs, values, and attitudes. Research on racial/ethnic disparities in sleep in the United States has primarily focused on European American/White and African American/Black individuals despite the fact that a quarter of the population belongs to other groups (e.g. Asian Americans, Latinx/Hispanic). Emerging research has begun to examine sleep disparities among other racial/ethnic groups in the United States, particularly Asian and Latinx/Hispanic populations [32–34]. Moreover, most research employs between-group comparison designs, often using European American/White groups as the reference, neglecting the importance of within-group analyses that could reveal greater heterogeneity. The growing population of individuals in the United States who identify as multiracial poses additional analytical challenges when examining sleep disparities. Although small sample sizes precluded testing within-group analytic effects, the current meta-analysis takes initial steps toward addressing these complexities by retaining racial/ethnic identity data from both between- and within-group studies. As a result, we applied a between-group analytic method, focusing on variations in effect sizes relative to the proportional representation of specific racial/ethnic groups.

Extensive research has documented disparities in sleep based on race/ethnicity [35] and SES [36]. However, few studies have examined whether SES and race/ethnicity might moderate the relationship between sleep and cognitive or academic functioning. An overview of research examining disparities in sleep and developmental outcomes among children indicates that youth from racially minoritized and socioeconomically disadvantaged groups derive greater benefits from more optimal sleep, which may be protective against adverse developmental outcomes [25]. Additionally, race/ethnicity has been shown to moderate cross-sectional and longitudinal associations between sleep quality and cognitive performance. For example, better sleep quality was more strongly associated with cognitive performance in African American/Black youth than in European American/White youth, both cross-sectionally in childhood [37] and longitudinally from childhood to adolescence [38].

Similar associations have been observed when SES serves as a moderator, where the relationship between sleep and cognitive/academic outcomes is more pronounced for youth from economically disadvantaged families [25]. In childhood, parent education has been identified as a significant moderator of cross-sectional and longitudinal linkages between multiple sleep parameters and various dimensions of cognitive ability [39]. Moreover, associations between sleepiness and executive functioning are stronger for adolescents from lower SES households compared to those from higher SES homes [40]. Likewise, early life SES has been shown to intensify the relationship between sleep problems and academic achievement during middle childhood, with the gap in achievement narrowing with improved sleep [41]. However, contrary findings exist. For instance, Philbrook and colleagues [38] reported that shorter sleep was associated with better cognitive performance among youth from lower-income households. In addition, some studies have not found SES to moderate links between sleep duration and teacher-reported academic performance [37]. These inconsistencies across studies highlight the need for a more nuanced understanding of SES as a potential moderator in the relationship between sleep and cognitive/academic outcomes.

Although few studies have assessed the role of SES and race/ethnicity in relation to sleep and cognitive/academic functioning, conducting a meta-analysis offers a unique opportunity to test how associations vary based on differences across the samples from which the bivariate data were drawn—such as the percentage of the sample identifying with a particular racial/ethnic group or the sample’s SES classification (low, middle, high, or mixed). The goal of this meta-analysis is to review and quantify the effect size of sleep on cognitive functioning and academic performance among school-aged children and adolescents. Addressing significant gaps in the literature, we also examine whether these associations vary by SES and race/ethnicity. Additionally, this analysis considers a range of individual (e.g. age, gender) and methodological (e.g. study design, reporter) characteristics as moderators to investigate potential sources of heterogeneity. These factors can influence the robustness of the weighted effect sizes. For example, the type of sleep assessment (objective, subjective), the methodology for assessing cognitive or academic outcomes (e.g. teacher reported, objective assessment), and the study design (cross-sectional, longitudinal) all play a role in shaping the observed relationships.

Of note, we examined similar questions in an independent meta-analysis focused on mental health outcomes (i.e. internalizing, externalizing). The large and distinct nature of the two literatures necessitated conducting separate analyses. Examining these questions in relation to youths' cognitive and academic functioning represents a distinct and pressing need.

Materials and Methods

The initial literature search was conducted in August 2022 and included the following databases: CINAHL, ProQuest, PsycINFO, PubMed, and Web of Science. Key word selection was based on operationalization rather than definition or terminology to ensure a comprehensive search across different literature (see Table 1). For a broad retrieval of records pertaining to the variables of interest, the keywords and phrases for our main predictor, sleep (“sleep duration” OR “sleep efficiency” OR “long wake episode” OR “long-wake episode” OR “wake minutes” OR “sleep onset latency” OR “subjective sleep quality” OR “sleep quality” OR “sleep wake problems” OR “satisfaction with sleep” OR “sleepiness” OR “sleep schedule” OR “circadian preference” OR “morningness” OR “eveningness” OR “sleep variability” OR “sleep minutes” OR “total sleep time” OR “sleep period” OR “time in bed” OR “sleep onset” OR “sleep midpoint” OR “sleep fragmentation” OR “wake after sleep onset” OR “sleep consistency” OR “sleep regularity” OR “sleep problems”) were cross-referenced with words and phrases for a broader search strategy that included mental health, cognitive, and academic outcomes¹ (“academic functioning” OR academic achievement OR “academic performance” OR “cognitive functioning” OR “cognitive performance” OR “cognitive skill^*” OR “cognitive ability^*” OR intelligence OR “executive function^*” OR “internalizing problems” OR “internalizing symptoms” OR depression OR anxiety OR “self-esteem” OR “self-worth” OR “social withdrawal” OR “emotional problem^*” OR “externalizing problems” OR “externalizing symptoms” OR “externalizing behavior^*” OR “problem behavior^*” OR “behavior^* difficult^*” OR anger OR aggression OR hostility OR “rule-breaking” OR “rule breaking” OR oppositional defiance OR conduct problems OR “well-being” OR “well being” OR happiness OR optimism OR “positive affect” OR “negative affect” OR “mental health” OR “mental illness” OR adjustment OR “adjustment problems” OR “emotion regulation”) as well as our moderators of interest, race/ethnicity and socioeconomic status (SES) (“socioeconomic^*” OR “socioeconomic status^*” OR “socioeconomic-status^*” OR “socioeconomic position^*” OR “social position^*” OR “socioeconomic adversity” OR “social class^*” OR “social rank^*” OR “sociodemographic^*” OR “ses” OR “neighborhood ses” OR “community ses” OR income OR education OR occupation OR employment OR “employment status^*” OR “economic deprivation” OR “disparit^*” OR race OR ethnicit^* OR “racial group^*” OR “ethnic group^*” OR “racially diverse” OR “ethnically diverse” OR “racial diversity” OR “ethnic diversity” OR “ethnic background” OR “ethnic origin^*” OR “national origin^*” OR “nationality” OR White OR Caucasian OR Caucasian American OR European American OR Black OR African American OR Asian OR Asian American OR Hispanic OR Latinx OR Latino OR Latina OR Latin American OR American Indians OR Alaska Natives OR Native Hawaiians OR Pacific Islanders OR “multiracial” OR “multi-racial”). Only cognitive and academic outcomes were included in the current review. Asterisks were used to capture variations in phrasing or wording (e.g. cognitive ability, cognitive abilities). Restrictions were applied to refine the initial results. Specifically, results were narrowed to exclude adult populations, manuscripts written in languages other than English (study inclusion was restricted to USA-based samples) and publications prior to the year of 1990. Searches included all available and accessible publications (e.g. journals, chapters) as well as unpublished dissertations/theses to reduce publication bias. See PRISMA flow diagram [42] (Figure 1).

Table 1.

Definitions of Sleep Parameters and Outcomes of Interest

Objective sleep duration	Amount of time spent asleep based on precise data provided by validated sleep monitoring tools such as actigraphy or polysomnography (e.g. duration, minutes)
Subjective sleep duration	Amount of time spent asleep based on an individual’s recall or perception gathered via surveys, sleep diaries, or interviews (e.g. “How many hours did you sleep last night?”)
Objective sleep quality	Measure of how well or effectively an individual sleeps and meets rest and recovery needs based on precise data provided by validated sleep monitoring tools such as actigraphy or polysomnography (e.g. sleep efficiency, minutes awake after sleep onset, activity)
Subjective sleep quality	Measure of how well or effectively an individual sleeps and meets rest and recovery needs based on an individual’s recall or perception gathered via surveys, sleep diaries, or interviews (e.g. “How rested did you feel upon waking?”)
Sleep schedule/chronobiology	The timing and patterning of sleep and wake cycles measured objectively or subjectively (e.g. wake time, bedtime, morningness-eveningness)
Sleep consistency/variability	The regularity and stability of the sleep–wake cycle measured objectively or subjectively (e.g. nightly fluctuations in duration, sleep onset, wake time)
Academic performance	Knowledge and achievement of academic subject matter using objective (e.g. GPA, subject test scores, subject grades) and subjective (e.g. teacher-reported academic achievement) measures
Cognitive functioning	Ability to rationalize, learn, and demonstrate abstract thinking indexed by crystallized and fluid intelligence, executive functioning, memory, attention, and processing measured using standardized tests such as the Woodcock-Johnson III (WJ-III)

Figure 1.

The PRISMA flow diagram of the original search.

Across all databases, the initial search resulted in 5042 records. The inclusion criteria for the meta-analysis were as follows:

• Records included empirical, quantitative data. Qualitative data, case studies, and reviews (e.g. systematic or narrative reviews) were excluded.

• Records were written in English. All other languages were excluded.

• Studies were conducted in the United States. Records using global samples were excluded, given interest in SES and race/ethnicity as moderators, which may vary greatly across countries.

• Records were dated after 1990.

• Records included data using child or adolescent samples (M_age = 5–18 years). Exceptions were made for high school samples wherein some students may be older than the age of 18 at the time of enrollment. Samples outside of this range (e.g. college samples, perinatal samples) were excluded.

• Sample sizes included within the record were n  28. Power analyses via pwr package [43] in R indicated that n = 28 was large enough to detect small correlations (r = 0.10) at  = .05 and a power level of .80.

• Records used community-based samples. Clinical samples were excluded to reduce confounding effects.

• Records used cross-sectional, longitudinal, or experimental designs. In the case of longitudinal designs, the outcome (e.g. cognitive functioning; academic performance) had to be antecedent to the predictor or moderator variable. In the case of experimental or intervention designs, effects at baseline assessments were used when available. Otherwise, the record was excluded.

• Only independent samples were considered. If multiple records were reported on the same sample, all unique effects were included (e.g. different assessments or informants).

• Records included at minimum one assessment of sleep measured objectively or subjectively (see Table 1). Exceptions were made for sleep measured using the Child Behavioral Checklist, which were excluded.

• Records included at minimum one assessment of cognitive and academic outcomes (see Table 1). Broader school performance variables (e.g. tardiness, peer relationships, teacher relationships, perseverance) were excluded.

• Records included measurements of the sample’s SES and/or racial/ethnic composition.

• Records included the zero-order correlation coefficients (r) for the association between sleep and cognitive and academic outcomes or provided sufficient details to compute an effect size. Authors were contacted for additional data if necessary.

Multiple tests were later conducted to assess the risk of selection and publication bias. Methodological quality or rigor was not part of the inclusion criteria.

After the removal of duplicates and the comparison of records against inclusion criteria, 3867 were then screened across titles and abstracts, resulting in 418 records. Of these 418 records, 26 were included following a full-text review for eligibility. This final set included 328 effects in the meta-analysis. Two coders were trained by the first author to review the titles and abstracts of all records. Screening reliability was assessed using Cohen’s kappa [44]. Coders achieved κ = .64 reliability across 31 percent of titles and abstracts. The coders then independently reviewed 100 percent of the full texts for screening. Reliability analyses were conducted to assess similarity in full-text screening for inclusion and exclusion. Moderate reliability was found between the two coders, κ = .66.

We conducted two updated literature searches following the same procedures as the initial search. The first update, conducted in September 2024, identified studies published after the original search and during manuscript preparation. This search yielded 460 records; after removing duplicates, 317 titles and abstracts were screened, 68 were selected for full-text review, and 5 met inclusion criteria (see Figure 2). The second update, conducted in March 2025, expanded the search to include the terms “sleep health” and “circadian misalignment” and identified additional studies published since September 2024. This search resulted in 6134 records. Following duplicate removal, including those already retrieved from the original search and first updated search, 274 titles and abstracts were screened, 32 full texts were reviewed, and 2 met inclusion criteria (see Figure 2).

Figure 2.

The PRISMA flow diagram of the updated search.

The first and second updates contributed 68 and 14 additional effects to the meta-analysis, respectively. Two coders, trained by the first author, reviewed all titles and abstracts. Interrater reliability was κ = .76 for the first update and κ = .88 for the second update, based on 100 percent of screened titles and abstracts. The first and second authors independently reviewed all full texts, with strong agreement (κ = .84 and .82 for the first and second updates, respectively).

Coding procedures

Articles that met inclusion criteria from the initial search were coded by the first author and a trained graduate research assistant. Both coders followed a detailed coding system to extract information on the participants, sample, methodology, and record characteristics of interest (see Table 2). All records were coded separately and independently by each coder. To assess the reliability of continuous variables, intraclass correlation coefficients (ICCs) were used, and reliability between coders was strong, ICCs = .93–1.00, M_ICC = .99. To assess the reliability of categorical variables, Cohen’s kappa [44] was used. Reliability between coders was moderate to strong (κs = .69–1.00, M_κ = .96). All discrepancies were discussed and addressed by both coders to create a set of consensus codes used in the study analyses [45].

Table 2.

Overview and Descriptive Statistics of Coded Variables

Variable	Code/description	k studies (%)*	M	SD	Mdn	Mode	Minimum	Maximum
Continuous Variables
Study details
Year	Year of publication/defense		2017	5.35	2019	2021	2007	2025
Sample size	N		1554.53	5645.11	275	302	31	32 980
Participant characteristics
Gender	Percentage of girls (%)		53.09†	11.84†	51.00	51.00	44.00	100.00
Race/ethnicity	Percentage of European American/White (%)		43.44†	16.15†	61.00	0.00	0.00	81.80
	Percentage of African American/Black (%)		8.12†	14.29†	22.07	0.00	0.00	76.00
	Percentage of Asian Americans (%)		1.71†	7.11†	0.00	0.00	0.00	87.00
	Percentage of Latinx/Hispanic (%)		5.90†	14.59†	2.30	0.00	0.00	100.00
	Percentage of other race/ethnicity or multi-racial/multi-ethnic (%)		40.66†	24.57†	0.00	0.00	0.00	58.20
Age	At sleep assessment		8.09†	3.54†	8.72	8.72	5.00	17.30
	At outcome assessment		10.23†	2.12†	10.80	10.80	5.00	17.30
Methodological characteristics
Time lag	Time elapsed between assessment of sleep and outcomes, in years		0.88	1.30	0.00	0.00	0.00	4.00
Categorical Variables
Study details
Record type	Published (peer-reviewed)	31 (93.94%)
	Unpublished (dissertation/thesis, author requested)	2 (6.06%)
Sample characteristics
Socioeconomic status	Low	11 (33.33%)
	Middle	4 (12.12%)
	High	3 (9.09%)
	Mixed	11 (33.33%)
	Not specified	4 (12.12%)
Methodological characteristics
Sleep parameter	Objective duration	14 (42.42%)
	Subjective duration	10 (30.30%)
	Objective quality	13 (39.39%)
	Subjective quality	18 (54.54%)
	Schedule/chronobiology	5 (15.15%)
	Consistency/variability	5 (15.15%)
Primary outcome domain	Academic	22 (66.67%)
	Cognitive	18 (54.54%)
Outcome reporter	Self	4 (12.12%)
	Parent	2 (6.06%)
	Teacher	4 (12.12%)
	Objective	28 (85.85%)
Study design	Cross-sectional	30 (90.91%)
	Longitudinal	11 (33.33%)

k = 33. Because of variations in time lags, ages, sample size, gender, and racial/ethnic makeups within a single study, descriptive statistics for sample size, gender, race/ethnicity, age, and time lag represent descriptive statistics across effect sizes when the entire data set was analyzed.

^* k _Studies and percentages may not add up to 33 and 100%, respectively, because many studies reported more than one effect and may have utilized different reporters, samples, methods, etc., for distinct effects within a single study.

^†Weighted by sample size.

The second round of coding—for articles included in the two updated searches—was conducted by the first and second authors using the same procedures as the initial round. To assess reliability, both authors independently coded 30 percent of the articles. The remaining articles were divided among the first and second authors. Interrater reliability for continuous variables was strong, ICCs = .99–1.00, M_ICC = .99. For categorical variables, reliability ranged from acceptable to strong (κs = .50–1.00, M_κ = .82). All discrepancies were reviewed and resolved by both coders to create a set of consensus codes used in the study analyses [45].

Data analysis

Calculating effect sizes

Pearson’s correlation coefficient (r), representing the association between sleep and cognitive and academic outcomes, was the effect size of interest. We also considered Cohen’s d, partial eta-squared (), and/or chi-square () for extraction and transformation into r. Cohen’s d was considered if group means, standard deviations, and sample sizes (i.e. number of participants) were reported. Cohen’s d is transformed into r using equations (1) and (2).

(1)

(2)

Within these equations, M_i and SD_i represent the mean and standard deviation for each group and were used to calculate a Cohen’s d effect size that was then transformed into Pearson’s r using equations (3) and (4).

(3)

(4)

To create the r equivalent of partial eta-squared (), the square root of the eta-squared value was taken using equation (5).

(5)

can be used to acquire Pearson’s r using equation (6).

(6)

Within each equation, n_i represents the sample size for each group, whereas n represents the total sample size. Consistent with established conceptualizations of sleep health [20, 21], effects were screened to ensure consistent directionality wherein higher scores reflect more optimal sleep (e.g. greater consistency, better quality) as well as greater cognitive and academic performance (e.g. WJ-III scores, GPA). Positive Pearson r values indicate that optimal sleep was associated with better cognitive and academic outcomes. No additional transformations were conducted. Effect size magnitudes were assessed using Cohen’s recommendations [46].

Dependency within data

Most of the included records reported multiple effect sizes, potentially introducing effect dependency within studies nested within samples. To address this, a hierarchical meta-analysis was conducted using an intermediary cluster term [47]. The analysis included three nested levels that accounted for variance among individual effect sizes (Level 1), variance among records (Level 2), and variance among samples (Level 3). We assessed heterogeneity within and between levels (e.g. comparing Level 1 variance to Level 2 variance).

Analysis plan

Hierarchical meta-analysis was conducted in R using the metafor package [48]. The current meta-analysis was structured to accomplish three aims across three different models. The first model accounts for the direct relationship between our predictor of interest, sleep, and cognitive and academic outcomes. The second model tests our moderators of interest, race/ethnicity and SES. The third model examines the heterogeneity within the associations between specific sleep parameters and specific outcomes of interest.

Model 1

The aim of Model 1 was to test for significant associations between sleep broadly (i.e. assessment of an overall sleep construct that does not distinguish between specific sleep parameters) and overall cognitive/academic outcomes. This was supported by three submodels.

Model 1.1

Broadest in conceptualization, the first model tested whether there was a significant association between sleep and cognitive and academic outcomes. Specifically, Model 1.1 tested whether the overall association between sleep and such outcomes was stronger (or weaker) for (1) any specific sleep parameter compared to another (Model 1.1A) or (2) for any specific outcome parameter compared to the other (Model 1.1B). The overall weighted effect size between sleep broadly and overall outcomes was calculated. Then, omnibus moderation tests of specific sleep parameters were conducted to assess whether this overall weighted effect size varied by specific sleep parameters (e.g. subjective quality, objective duration). Significant omnibus results were followed by dummy-coded pairwise comparisons across sleep parameters. To test Model 1.1B, the prior steps were repeated to account for different domains of cognitive and academic outcomes rather than specific sleep parameters.

Model 1.2

Model 1.2 was a narrowed iteration of Model 1.1 and accounted for whether a specific sleep parameter was more strongly (or weakly) associated with either cognitive or academic outcomes. In other words, Model 1.2 accounted for whether a specific sleep domain, like subjective quality, was more strongly associated with cognitive functioning (Model 1.2A) and academic performance (Model 1.2B) compared to another sleep domain, such as objective quality. Two overall weighted effect sizes were calculated: one between sleep broadly and cognitive functioning and the other between sleep broadly and academic performance. Omnibus moderation analyses followed, testing how weighted effects varied across sleep parameters. When omnibus results were significant, dummy-coded pairwise comparisons were conducted to examine the relative magnitude of the effect against all other sleep parameters on cognitive and academic outcomes.

Model 1.3

A variation of Model 1.2, Model 1.3 includes six separate models to account for the association between each sleep parameter of interest and overall outcomes. These models were designed to account for whether the association between each individual sleep parameter and cognitive functioning was stronger (or weaker) compared to academic performance. Using subjective quality as an example, the overall weighted effect between subjective quality and overall cognitive and academic outcomes was calculated. Then, cognitive functioning was tested as a potential moderator of this association compared to academic performance. The omnibus moderation effects were examined, and when significant, follow-up pairwise comparisons were conducted to assess the magnitude of the moderation effect for cognitive functioning compared to academic performance.

Model 2

As the primary aim of the study, we assessed whether the overall association between sleep and cognitive and academic outcomes would vary in magnitude by SES (Model 2.1) and race/ethnicity (Model 2.2). Of note, we tested moderation by SES and race/ethnicity following Model 1, to determine whether examining sleep parameters individually provides developmental and clinical value. The overall weighted effect between individual sleep parameters and cognitive and academic outcomes was calculated and then assessed for variation by SES and race/ethnicity separately. Any significant omnibus effects found for race/ethnicity were further examined using simple effects; significant omnibus effects for SES were further examined using pairwise comparisons.

Model 3

We also aimed to assess the heterogeneity found in the associations between individual sleep parameters and the overall outcome by participant or methodological characteristics. Significant variations in the association of interest by methodology would suggest an increased risk of bias based on the rigorousness of methodologies used across records. Therefore, we tested whether the association between each specific sleep parameter and overall cognitive and academic outcomes varied by participant and methodological characteristics, which were both continuous and categorical in nature.

In the case of continuous moderators (e.g. age, study lag), three studies at minimum were needed to calculate whether a moderation effect was present. Continuous effects were determined by the slope coefficient (b) in relation to the intercept, which was the value of the effect when the moderator was set to zero. A positive slope coefficient would suggest a strengthening effect or an increase, whereas a negative slope value would suggest a weakening effect or a decrease. Categorical effects were assessed when categorical moderators had at least three studies in any given category. The average weighted effect (r) of the reference group was calculated (i.e. intercept) and compared against the slope coefficient (b) representing the comparison category. Similar to the above, effects were recorded to account for positive values, indicating that the comparison category strengthens the effects compared to the reference category, and negative values weaken the effect.

Testing for bias in meta-analytic effect sizes

It is well-established that meta-analyses are prone to publication bias [49]. In attempts to identify the degree to which locating, selecting, and combining studies has biased our meta-analytic effect size(s), we have taken the following steps in alignment with established meta-analysis procedures [50]. First, we took steps to include gray literature by incorporating unpublished effects from dissertations and theses in our search strategy and by contacting authors of published studies when the bivariate data were unavailable. Second, we account for variations by publication status (i.e. published or unpublished effect) in our moderation analyses. Third, we conducted a multilevel meta-analysis of Egger’s regression test (MLMA) [47, 51, 52], which accounts for multiple levels (i.e. sources) of publication bias. Six MLMA Egger’s tests were conducted to account for each sleep parameter. Significant slope values indicated that the precision of the meta-analytic effect was affected by the publication status of the selected records. When the slope value was significant, the intercept of the MLMA Egger’s test model provided an adjusted weighted effect size. Finally, a commonsensical approach was used via analysis of contour-enhanced funnel plots [50]. The results of published versus unpublished effects were plotted and assessed for effect size convergence. Should only published records achieve statistical significance, then such findings may point to potential reporting bias.

Transparency and openness

This study was not pre-registered; however, data and coding materials are available on the Open Science Framework (https://osf.io/w278u/) [53].

Results

Descriptive statistics for all variables included in the meta-analysis can be found in Table 2. There were 410 effect sizes (M = 26.46, SD = 31.55, range = 1–123) taken from 33 studies comprising a total of 52 854 participants (range = 31–32 980). Across the 33 studies, 53.09 percent of participants identified as female, 43.44 percent as European American/White, 8.12 percent as African American/Black, 1.71 percent as Asian American, 5.90 percent as Latinx/Hispanic, and 40.66 percent as another race/ethnicity (i.e. “Other”; e.g. Indigenous American, Pacific Islander American, multiracial/ethnic). Across all effects, participants were, on average, 8.09 years of age at the time sleep was recorded and 10.23 years of age when cognitive and academic outcomes were recorded. Samples included were from a diverse array of socioeconomic backgrounds (SES), with 33.33 percent of studies classified as low SES, 12.12 percent as middle SES, and 9.09 percent as high SES. There were also 33.33 percent of studies classified as mixed SES (i.e. studies that included samples that did not represent a majority SES classification). Most studies were cross-sectional (90.91 percent). Only 33.33 percent of studies included longitudinal effects. The average lag between assessments was 0.88 years (11 months). Most effects were published (93.94 percent). A breakdown of study details by record is provided in Table 3.

Table 3.

Study Characteristics by Record

	Study	N	Record type	Study design	Lag (years)	Age at sleep (years)	Age at outcome (years)	Gender (% girls)	Race/ethnicity	SES	Sleep parameter	Outcome domain	Outcome reporter
1.	Becker et al. (2025) [54]	302	P	CS	0	13.2	13.2	45	82% EA, 5% AA, 5% AS, 8% O, 5% Lx	Hi	SD, SQ, S/C	Academic	O
2.	Bell and Juvonen (2020) [55]	2718	P	CS, L	0	13.0	13.0	100	20% EA, 12% AA, 15% AS, 18% O, 35% Lx	Md	SD	Academic	O
3.	Berger et al. (2019) [56]	103	P	CS	0	6.5	6.5	51	38% EA, 2% AA, 1% AS, 50% Lx, 8% O	Md	OQ	Cognitive	O
4.	Bub et al. (2011) [57]	250	P	CS, L	0, 1, 2	8.2	8.2, 9.3, 10.3	51	65% EA, 35% AA	Mx	SQ	Cognitive	O
5.	Buckhalt et al. (2007) [58]	166	P	CS	0	8.7	8.7	55	69% EA, 31% AA	Mx	OD, OQ, SQ, C/V	Cognitive	O
6.	Buckhalt et al. (2009) [39]	166	P	CS, L	0, 2	8.7, 10.8	8.7, 10.8	55	69% EA, 31% AA	Mx	OD, OQ, SQ, C/V	Academic, cognitive	O
7.	Cheng et al. (2021) [17]	4696	P	CS	0	9.9	9.9	48	75% EA, 21% AA	Md	SD	Cognitive	O
8.	Crichlow et al. (2024) [59]	176	P	CS	0	12.1	12.1	55	76% AA, 15% AS, 8% O	Lo	OQ	Cognitive	O
9.	Cusick et al. (2018) [60]	300	P	CS	0	13.0	13.0	45	82% EA, 5% AA, 5% AS, 5% Lx, 8% O	Md	SD, SQ	Academic, cognitive	O
10.	Dunbar et al. (2017) [61]	310	P	CS	0	14.5	14.5	64	23% EA, 11% AA, 38% AS, 24% Lx, 5% O	Md	SD, SQ	Academic	S, O
11.	El-Sheikh et al. (2007) [62]	111	P	CS	0	8.7	8.7	55	69% EA, 31% AA	Mx	OD, OQ	Academic	T
12.	El-Sheikh et al. (2019) [37]	199	P	CS	0	9.4	9.4	48	65% EA, 35% AA	Lo	OD, OQ	Academic	O
13.	El-Sheikh et al. (2020) [63]	272	P	CS	0	17.3	17.3	49	59% EA, 41% AA	Lo	OD, OQ, SQ	Cognitive	O
14.	El-Sheikh et al. (2014) [64]	252	P	CS	0	15.8	15.8	53	66% EA, 34% AA	Mx	OD, OQ	Cognitive	O
15.	Erath et al. (2015) [65]	280	P	CS	0	10.4	10.4	45	63% EA, 37% AA	Lo	OQ	Academic, cognitive	O
16.	Fredrick et al. (2023) [66]	302	P	L	0.5, 1, 2	13.2, 14.2	13.7, 15.2	45	82% EA, 5% AA, 5% AS	Hi	SD, SQ, S/C	Academic	O
17.	Keller et al. (2008) [67]	124	P	CS	0	8.7	8.7	54	77% EA, 23% AA	Mx	OD	Academic	O
18.	Lepore et al. (2013) [68]	402	P	CS, L	0, 0.5	12.8, 13.3	13.3	56	43% EA, 24% AA, 24% Lx, 9% O	Mx	OQ	Academic	O
19.	Lewin et al. (2017) [69]	32 980	P	CS	0	13.5	13.5	50	42% EA, 58% O	Mx	SD	Academic	S
20.	Lunsford-Avery et al. (2024) [70]	31	P	CS	0	15.4	15.4	61	81% EA, 3% AA, 6% AS, 9% O, 19% LX		SQ, S/C	Cognitive	S, P
21.	McHale et al. (2011) [71]	227, 191	U	CS	0	12.8, 15.7	12.8, 15.7	51, 50	100% Lx	Mx	SD, C/V	Academic	S
22.	Mitchell et al. (2024) [72]	278	P	CS, L	0, 1	9.3	9.3 10.3	49	49% Lx	Lo	SQ	Academic	T
23.	Orihuela et al. (2023) [73]	288	P	CS	0	12.0	12.0	54	37% EA, 48% AA, 5% O, 10% Lx	Mx	OD, SQ	Academic, cognitive	O, T
24.	Philbrook et al. (2018) [74]	282	P	CS, L	0, 2	9.4	9.4, 11.3	48	65% EA, 35% AA	Lo	OD, SQ	Academic, cognitive	O
25.	Philbrook et al. (2017) [38]	282	P	CS, L	0, 1, 2	9.4	9.4, 10.8, 11.3	48	65% EA, 35% AA	Lo	OD, OQ	Cognitive	O
26.	Rea-Sandin et al. (2022) [41]	707	P	CS	0	8.4	8.4	52	57% EA, 4% AA, 3% AS, 29% Lx, 8% O	Md	OD, SQ, S/C	Academic, cognitive	O, P
27.	Rubens et al. (2020) [75]	41	P	CS	0	11.7	11.7	44	100% Lc	Lo	SQ	Academic	O
28.	Rudd (2018) [76]	2330	U	CS, L	0, 4	5.0, 9.0	5.0, 9.0			Lo	SQ	Cognitive	O
29.	Saini et al. (2021) [77]	243	P	L	1	10.4	11.4	47	63% EA, 37% AA	Lo	OD, OQ	Cognitive	O
30.	Ursache et al. (2021) [78]	433	P	CS, L	0, 1	8.0, 6.8	8.0	52		Mx	SD, SQ	Academic	O
31.	Woods et al. (2024) [26]	3002	P	L	4	5.0	9.0		22% EA, 44% AA, 8% O, 27% Lx	Lo	S/C, SD, SQ	Academic, cognitive	O, T
32.	Xie et al. (2021) [79]	145	P	CS	0	14.3	14.3	65	87% AS, 18% Lx, 13% O	Mx	OD, SQ, C/V	Academic	O
33.	Yip et al. (2023) [80]	265	P	CS	0	15.3	15.3	71	21% AA, 42% AS, 37% Lx		C/V	Academic	O

Record type: P = published effect, U = unpublished effect. Study design: CS = cross-sectional, L = longitudinal. Race/ethnicity: EA = European American, AA = African American, AS = Asian American, O = other race/ethnicity and/or multiracial, Lx = Hispanic/Latinx. Socioeconomic status (SES): Lo = low, Md = middle, Hi = high, Mx = mixed. Sleep parameter: OD = objective duration, OQ = objective quality, SD = subjective duration, SQ = subjective quality, C/V = consistency/variability, S/C = schedule/chronobiology. Outcome Reporter: M = multi-informant, O = objective, P= parent, T = teacher, S = youth self-report.

Homogeneity was also assessed across clusters. Intraclass correlations (ICCs) were examined across all models (see Table 4). ICCs ranged from zero to large in size, though most models demonstrated modest ICCs (ICC_{Model 1.1} = .15; ICC_{Model 1.2} = .33 and .10 for cognitive functioning and academic performance, respectively; ICC_{Model 1.3} = .00–.95; M_{ICC Model 1.3} = .55; ICCs for Models 2 and 3 are the same as Model 1.3); therefore, the decision was made to continue with hierarchical meta-analysis to model the dependency in the data and reduce bias.

Table 4.

Homogeneity and Heterogeneity Statistics for Associations Between Sleep and Cognitive and Academic Functioning

Model	Q	df	Between sample variance		Between study variance within sample		Between effect variance within study		ICCs
			τ2	I 2	τ2	I 2	τ2	I 2
Model 1.1: Overall	6473.79***	409	.0000	.000001%	.0026	14.00%	.0142	77.21%	.15
Model 1.2A: Cognitive functioning	663.67***	216			.0015	23.04%	.0030	46.86%	.33
Model 1.2B: Academic performance	5477.32***	192			.0024	9.11%	.0224	85.55%	.10
Model 1.3A: Objective duration	115.46***	56			.0028	33.99%	.0011	12.87%	.73
Model 1.3B: Subjective duration	54.38*	33			.0004	31.85%	.0002	16.14%	.66
Model 1.3C: Objective quality	156.50***	72			.0088	60.56%	.0004	2.92%	.95
Model 1.3D: Subjective quality	5290.75***	148			.0036	10.69%	.0283	83.13%	.11
Model 1.3E: Schedule/chronobiology	194.57***	44			.0039	74.08%	.0008	14.95%	.83
Model 1.3F: Consistency/variability	119.09***	51			.0000	.0000002%	.0081	56.52%	.00

ICCs = intraclass correlation coefficients among effect sizes.

^* p < .05.

^*** p < .001. Significant at p = . 011

Model 1: Overall effect and variations by specific parameters

Model 1.1: Overall association between sleep and cognitive and academic outcomes

In first assessing the overall weighted effect size between sleep broadly and overall cognitive and academic outcomes, the overall weighted effect was small and significant, r = 0.11, 95% CI [0.08, 0.14], p < .001. There was also a significant degree of heterogeneity in the overall association across levels of analysis, as demonstrated by significant Q and I² values. Approximately .000001 percent of the variance was found between samples, 14.00 percent of the variance was found between studies within samples, and 77.21 percent of the variance was found between effects within samples. Given that variance accounted for the nesting of effects within studies within samples (i.e. Level 3) was negligible, we limited our hierarchical meta-analysis only to include effects between studies (i.e. Level 2) and effects within studies (i.e. Level 1).

Model 1.1A: Variations by Sleep Parameter

We determined whether the variance within the model could be attributed to the specificity of the sleep parameter. See Table 5 for all results of moderation for Models 1.1A and 1.1B. Most effects for individual sleep parameters except for sleep consistency were significant (i.e. different from zero) and of a small or medium magnitude (range = 0.08–0.14, all ps ≤ .001). The weighted effect of sleep consistency was nonsignificant (r = 0.04, p = .17). Omnibus moderation tests were significant, and pairwise comparisons were conducted among all individual sleep parameters. Subjective quality was statistically stronger compared to objective duration (b = .06, p = .02). Sleep consistency was statistically weaker compared to objective quality (b = −.07, p = .01), subjective duration (b = −.08, p = .02), subjective quality (b = −.10, p < .001), and schedule (b = −.07, p = .05).

Table 5.

Specificity in Associations Between Sleep and Cognitive and Academic Functioning

Model	Moderator	Level/category	k studies	k ES	Q between	r	SE	95% CI		Pairwise comparisons
								LL	UL
1.1A: Overall association	Sleep parameter		33	410	17.90**					Subjective quality > objective duration Consistency < subjective duration, objective quality, subjective quality, schedule
		Objective duration				0.08***	.02	0.04	0.13
		Subjective duration				0.12***	.03	0.06	0.17
		Objective quality				0.10***	.02	0.06	0.15
		Subjective quality				0.14***	.02	0.11	0.17
		Schedule/chronobiology				0.10***	.03	0.05	0.16
		Consistency/variability				0.04	.03	−0.02	0.09
1.1B: Overall association	Outcome domain		33	410	16.79***					Academic performance > Cognitive functioning
		Cognitive functioning				0.08***	.01	0.05	0.10
		Academic performance				0.14***	.01	0.11	0.17
1.2A: Cognitive	Sleep parameter		18	217	26.51***					Objective duration < subjective duration, subjective quality, schedule Subjective duration > objective duration, objective quality, subjective quality, schedule, consistency Consistency < subjective duration, objective quality, subjective quality, schedule
		Objective duration				0.04†	.02	−0.01	0.08
		Subjective duration				0.17***	.03	0.11	0.22
		Objective quality				0.07**	.02	0.03	0.11
		Subjective quality				0.09***	.02	0.05	0.12
		Schedule/chronobiology				0.10***	.03	0.05	0.15
		Consistency/variability				0.01	.03	−0.04	0.06
1.2B: Academic	Sleep parameter		22	193	11.66*					Consistency < objective quality, subjective quality
		Objective duration				0.14***	.04	0.07	0.22
		Subjective duration				0.11**	.03	0.04	0.17
		Objective quality				0.15***	.04	0.08	0.23
		Subjective quality				0.18***	.02	0.13	0.22
		Schedule/chronobiology				0.10**	.04	0.03	0.17
		Consistency/variability				0.06	.04	−0.02	0.14
1.3A: Objective duration	Outcome domain		14	57	18.87***					Academic performance > Cognitive functioning
		Cognitive functioning				0.04	.02	−0.004	0.08
		Academic performance				0.13***	.02	0.09	0.18
1.3B: Subjective duration	Outcome domain		10	34	2.21
		Cognitive functioning				0.08***	.01	0.06	0.11
		Academic performance				0.10***	.01	0.09	0.12
1.3C: Objective quality	Outcome domain		13	73	9.45**					Academic performance > Cognitive functioning
		Cognitive functioning				0.11***	.03	0.05	0.16
		Academic performance				0.18***	.03	0.12	0.24
1.3D: Subjective quality	Outcome domain		18	149	6.56*					Academic performance > Cognitive functioning
		Cognitive functioning				0.09**	.03	0.04	0.14
		Academic performance				0.18***	.03	0.12	0.23
1.3E: Schedule/chronobiology	Outcome domain		5	45	.08
		Cognitive functioning				0.10**	.03	0.03	0.16
		Academic performance				0.10**	.03	0.04	0.16
1.3F: Consistency/variability	Outcome domain		5	52	1.76
		Cognitive functioning				0.05*	.02	0.01	0.10
		Academic performance				0.09***	.02	0.05	0.14

^† p < .10.

^* p < .05.

^** p ≤ .01.

^*** p < .001.

Model 1.1B: Variations by Outcome

We assessed whether the overall weighted effect between sleep and cognitive and academic outcomes would vary by the type of outcome tested, cognitive functioning versus academic performance. The weighted effect size for cognitive functioning was small and significant (r = 0.08, p < .001), whereas the weighted effect size for academic performance was modest and significant (r = 0.14, p < .001). Omnibus moderation analyses were significant, suggesting that the overall weighted effect between sleep and cognitive and academic outcomes significantly varied by the type of outcome tested. Results from pairwise comparisons suggest that the weighted effect for academic performance was stronger than that of cognitive functioning (b = .06, p < .001).

Model 1.2: Further testing of sleep broadly and outcome type

To further understand how sleep and cognitive and academic outcomes are associated, the next tests examined whether each individual sleep parameter was associated with each type of outcome separately. Specifically, the next group of models tested for (1) the weighted effect size between sleep broadly and each outcome type separately and (2) whether individual sleep parameters moderated this weighted effect size. Results are separated by outcome type. All moderation outcomes can be found in Table 5.

Model 1.2A: Cognitive Functioning

When examining sleep broadly and only outcomes associated with cognitive functioning, the overall weighted effect size was modest, r = 0.08, 95% CI [0.05, 0.10], p < .001. More optimal sleep was associated with better cognitive functioning. Heterogeneity analyses indicated significant heterogeneity between levels, with 23.04 percent of the variance attributed to differences between studies and 46.86 percent of the variance attributed to differences between effects nested within studies. Omnibus moderation analyses were also significant, suggesting that there were variations in the overall weighted effect size of sleep and cognitive functioning by the specific sleep parameters. Weighted effect sizes for each sleep parameter were significant, except for objective duration (r = 0.04, p = .09) and consistency (r = 0.01, p = .71), and ranged from small to modest in size (r range = 0.07–0.17, all ps < .001). In further examining significant moderation effects, the effect of objective duration was weaker than that of subjective duration (r = −0.13, p < .001), subjective quality (r = −0.05, p = .01), and schedule (r = −0.06, p = .01). The effect of sleep consistency was weaker than that of objective quality (r = −0.06, p = .01), subjective duration (r = −0.16, p < .001), subjective quality (r = −0.08, p < .001), and schedule (r = −0.09, p = .002). Lastly, the effects of subjective duration were stronger compared to objective quality (r = 0.10, p = .002), subjective quality (r = 0.08, p = .003), and schedule (r = 0.07, p = .01).

Model 1.2B: Academic Performance

For the weighted effect size of sleep broadly and academic performance, the overall weighted effect size was modest, r = 0.14, 95% CI [0.10, 0.17], p < .001. More optimal sleep was associated with better academic performance. Significant heterogeneity in this association was found, with 9.11 percent of the variance attributed to differences between studies and 85.55 percent of the variance attributed to differences between effects within studies. Omnibus moderation tests were significant. The weighted effect for variability was near zero and nonsignificant, r = 0.06, p = .16. All other sleep parameters were modest in size and statistically significant (r range = 0.10–0.18, all ps < .01). Pairwise comparisons revealed that objective quality (b = .10, p = .05), and subjective quality (b = .12, p = .005) were significantly stronger effect sizes than consistency. No other pairwise comparisons were significant.

Model 1.3: Testing for each sleep parameter and overall cognitive and academic outcomes

Further testing of whether the association between specific sleep parameters and overall outcomes was stronger for the type of outcome (i.e. academic performance versus cognitive functioning) was then examined. Models were conducted to assess (1) weighted effect sizes between each individual sleep parameter and cognitive and academic outcomes and (2) whether each effect size was moderated by the type of outcome (i.e. academic performance versus cognitive functioning). Moderation results can be found in Table 5.

Model 1.3A: Objective Duration

For the association between objective duration and overall cognitive and academic outcomes, the weighted effect size was small and significant, r = 0.08, 95% CI [0.04, 0.12], p < .001. Longer sleep duration was associated with better cognitive and academic outcomes. Significant heterogeneity was found at the between study level (33.99 percent), and a smaller portion attributed to differences at the between effects level (12.88 percent). For objective duration, the omnibus test of moderation was significant. In other words, the weighted effect size of objective duration and cognitive and academic outcomes varied by type of outcome. The weighted effect size for cognitive functioning was not significantly different from zero (r = 0.04, p = .08). However, the weighted effect size for academic performance was statistically significant and modest in magnitude (r = 0.13, p < .001). Pairwise comparisons suggest that the weighted effect of academic performance was stronger than that of cognitive functioning (b = .09, p < .001).

Model 1.3B: Subjective Duration

The weighted effect size for subjective duration and cognitive and academic outcomes was small and significant, r = 0.10, 95% CI [0.08, 0.12], p < .001. Greater subjective sleep duration was associated with better cognitive and academic outcomes. Heterogeneity tests were significant. Modest variance was found at the between study level (31.85 percent) and between effects nested within studies (16.14 percent). The omnibus test of moderation was not significant, suggesting that the type of outcome did not significantly alter the magnitude of the weighted effect size of subjective duration and overall cognitive and academic outcomes. The weighted effects for cognitive functioning (r = 0.08, p < .001) and academic performance (r = 0.10 p < .001) were small and significant.

Model 1.3C: Objective Quality

For objective quality and cognitive and academic outcomes, the weighted effect size was medium and significant, r = 0.14, 95% CI [0.08, 0.19], p < .001. Greater objective quality was associated with cognitive and academic outcomes. Heterogeneity analyses were significant and suggested that the majority of the variance was attributed to differences between studies (60.56 percent), and a small portion of the variance was attributed to differences between effects (2.92 percent). Omnibus moderation testing was significant. The weighted effect size for cognitive functioning and academic performance were both statistically significant and moderate in magnitude (r = 0.11 and 0.18, respectively, ps ≤ .001). Corresponding pairwise comparisons indicated that the weighted effect size was stronger for academic performance compared to cognitive functioning (b = .08, p = .002).

Model 1.3D: Subjective Quality

The association between subjective quality and overall cognitive and academic outcomes had a significant weighted effect size of modest magnitude, r = 0.13, 95% CI [0.09, 0.18], p < .001. Greater subjective quality was associated with better cognitive and academic outcomes. Significant heterogeneity was found. Negligible variance could be attributed to differences between studies (10.69 percent) compared to that of the variance attributed to differences between effects (83.13 percent). An omnibus test of moderation was also significant. The main effects for cognitive functioning and academic performance were both modest in magnitude and significantly different from zero (r = 0.09, p = .0011 and r = 0.17, p < .001, respectively). Follow-up pairwise comparisons suggested that the weighted effect was statistically stronger for academic performance compared to cognitive functioning (b = .09, p = .01).

Model 1.3E: Schedule/Chronobiology

In assessing the association between schedule and overall cognitive and academic outcomes, the weighted effect size was modest and statistically significant, r = 0.10, 95% CI [0.04, 0.16], p = .002. Greater tendencies toward morningness were associated with better cognitive and academic outcomes. There was a significant amount of heterogeneity in the association, with most heterogeneity associated with differences at the between effects level (74.08 percent) and some variance found at the between study level (14.95 percent). The omnibus test of moderation was not significant. The weighted effect size for schedule/chronobiology and academic performance and cognitive functioning was r = 0.10 and 0.10, all ps < .003, respectively).

Model 1.3F: Consistency/Variability

The weighted effect size for variability, or consistency in sleep, was small and statistically significant, r = 0.07, 95% CI [0.04, 0.10], p < .001. Greater variability in sleep was associated with better cognitive and academic outcomes. Heterogeneity tests were significant. Over half of the variance was attributed to differences between effects (56.52 percent), and an almost undetectable amount of variance was attributed to differences between studies (.0000002 percent). The omnibus test of moderation was not statistically significant. The weighted effect for academic performance (r = 0.09, p < .001) was modest in size, and the weighted effect for cognitive functioning was slightly smaller in size (r = 0.05, p = .03).

Model 2: SES and race/ethnicity as moderators of the association among sleep parameters and overall cognitive and academic functioning

Model 2.1: Sample SES

We tested whether the weighted effect size of each individual sleep parameter and cognitive and academic outcomes was moderated by the sample’s SES (see Table 6). The overall weighted effects and heterogeneity breakdowns for each individual model are the same as those listed in Model 1.3. One significant moderation effect was observed. The weighted effect of schedule was weaker among low SES samples relative to middle (b = −.09, p < .001) and high (b = −.13, p < .001) SES samples (Figure 3).

Table 6.

Variation in the Association Between Each Sleep Parameter and Cognitive and Academic Outcomes by SES and Race/Ethnicity

Model	Moderator	Level/category	k studies	k ES	Q between	r	SE	95% CI		Pairwise comparisons
								LL	UL
Model 2.1A: Objective duration	Socioeconomic status		14	57	3.52
		Low				0.04	.03	−0.03	0.1
		Middle				0.05	.05	−0.04	0.15
		High				—	—	—	—
		Mixed				0.11***	.03	0.06	0.16
Model 2.1B: Subjective duration	Socioeconomic status		9	29	0.05
		Low				0.09*	.05	0.004	0.18
		Middle				0.10**	.04	0.03	0.18
		High				0.09**	.03	0.04	0.15
		Mixed				0.09**	.03	0.03	0.15
Model 2.1C: Objective quality	Socioeconomic status		11	70	0.2
		Low				0.10***	.03	0.05	0.16
		Middle				—	—	—	—
		High				—	—	—	—
		Mixed				0.12***	.03	0.06	0.19
Model 2.1D: Subjective quality	Socioeconomic status		16	131	0.11
		Low				0.13*	.05	0.03	0.23
		Middle				0.15	.09	−0.03	0.33
		High				0.16*	.08	0.003	0.31
		Mixed				0.13**	.04	0.05	0.21
Model 2.1E: Schedule/chronobiology	Socioeconomic status		4	41	58.63***
		Low				0.01	.01	−0.01	0.02
		Middle				0.10***	.02	0.06	0.15	Low < middle, and high
		High				0.14***	.02	0.1	0.17
		Mixed				—	—	—	—
Model 2.1F: Consistency/variability	Socioeconomic status		—	—	—
		Low				—	—	—	—
		Middle				—	—	—	—
		High				—	—	—	—
		Mixed				—	—	—	—
Model 2.2A: Objective duration	Race/ethnicity	% European American/White	14	57	.01	0.08	.08	−0.07	0.24	−.0001	.001	−0.003	0.002
		% African American/Black	14	57	.27	0.10*	.05	0.01	0.19	−.001	.002	−0.004	0.002
		% Asian American	10	46	.6	0.07**	.03	0.02	0.13	.001	.001	−0.002	0.003
		% Latinx/Hispanic	11	51	.05	0.08**	.03	0.03	0.14	−.0004	.002	−0.004	0.003
		% Other	11	51	.02	0.08**	.03	0.02	0.13	.001	.01	−0.01	0.01
Model 2.2B: Subjective duration	Race/ethnicity	% European American/White	8	31	.35	0.11***	.02	0.07	0.16	−.0002	.0004	−0.001	0.001
		% African American/Black	7	30	.11	0.10***	.02	0.06	0.14	−.0003	.001	−0.002	0.001
		% Asian American	5	21	.85	0.12***	.03	0.07	0.18	−.002	.002	−0.005	0.002
		% Latinx/Hispanic	6	20	1.56	0.11***	.03	0.05	0.18	−.001	.001	−0.003	0.001
		% Other	8	31	3.51†	0.09***	.01	0.07	0.1	.001†	.0004	−0.00	0.001
Model 2.2C: Objective quality	Race/ethnicity	% European American/White	12	71	2.53	0.29**	.09	0.11	0.47	−.002	.002	−0.005	0.001
		% African American/Black	13	73	4.16*	0.25***	.06	0.13	0.37	−.003*	.002	−0.007	−0.0001
		% Asian American	11	69	.16	0.11***	.02	0.06	0.15	.001	.001	−0.002	0.003
		% Latinx/Hispanic	11	69	1.61	0.14***	.03	0.08	0.19	.003	.002	−0.002	0.007
		% Other	12	71	.37	0.12**	.04	0.05	0.2	.005	.007	−0.01	0.02
Model 2.2D: Subjective quality	Race/ethnicity	% European American/White	15	132	2.10	0.05	.05	−0.05	0.16	.001	.001	−0.0004	0.003
		% African American/Black	15	132	.002	0.12**	.04	0.05	0.19	.0001	.001	−0.003	0.003
		% Asian American	12	99	.35	0.14***	.02	0.1	0.17	−.001	.001	−0.002	0.001
		% Latinx/Hispanic	14	108	.02	0.13***	.04	0.07	0.2	−.0002	.002	−0.003	0.003
		% Other	14	112	0.52	0.14***	.03	0.07	0.2	−.004	.005	−0.01	0.006
Model 2.2E: Schedule/chronobiology	Race/ethnicity	% European American/White	5	45	60.28***	−0.04***	.01	−0.06	−0.02	.002***	.0003	0.002	0.003
		% African American/Black	5	45	37.65***	0.14***	.02	0.11	0.17	−.003***	.001	−0.004	−0.002
		% Asian American	4	21	.82	0.01	.13	−0.24	0.26	.03	.03	−0.03	0.08
		% Latinx/Hispanic	4	35	1.61	0.19*	.09	0.02	0.35	−.005	.004	−0.01	0.003
		% Other	5	45	.57	−0.48	.77	−1.99	1.03	.07	.09	−0.11	0.26
Model 2.2F: Consistency/variability	Race/ethnicity	% European American/White	3	47	2.55	−0.12	.12	−0.35	0.11	.003	.002	−0.001	0.006
		% African American/Black	4	50	.68	−0.03	.14	−0.30	0.23	.004	.005	−0.006	0.01
		% Asian American	4	50	.29	0.09	.07	−0.05	0.23	−.001	.002	−0.005	0.003
		% Latinx/Hispanic	5	52	.07	0.07†	.04	−0.01	0.15	.0003	.001	−0.002	0.002
		% Other	3	47	2.47	0.07***	.02	0.03	0.1	−.01	.01	−0.03	0.004

^† p < .10.

^* p < .05.

^** p < .01.

^*** p < .001.

Figure 3.

Categorical moderators of weighted effects between sleep parameters and cognitive/academic functioning. Weighted effects of the moderator category are plotted with 95% CI for significant omnibus moderation analyses. A positive effect (r) denotes a positive association between sleep and cognitive/academic functioning (e.g., more consistent sleep is associated with greater cognitive and academic functioning).

Model 2.2: Racial/ethnic composition

The next models tested whether the weighted effect size between individual sleep parameters and overall cognitive and academic outcomes varied by sample racial/ethnic composition. For information on the heterogeneity and weighted effects of these models, see the results above in Model 1.3. Results delineating the effects of sample racial/ethnic composition can be found in Table 6. Omnibus tests of moderation were first conducted and assessed for statistical significance. Three significant moderation effects were found. The weighted effect of objective quality was moderated by whether participants identified racially/ethnically as African American/Black (b = −.003, p = .04). In other words, the association between objective quality and overall cognitive and academic outcomes decreased as the proportion of youth in the sample identifying as African American/Black increased (Figure 4, A). Whether participants identified as European American/White and African American/Black also moderated the association between sleep schedule and overall cognitive/academic outcomes. Specifically, the effects of sleep schedule on cognitive and academic outcomes strengthened as the proportion of youth in the sample identified as European American/White increased (b = .002, p < .001); see Figure 4, B. In contrast, the magnitude of this association weakened as the proportion of youth in the sample identifying as African American/Black increased (b = −.003, p < .001); see Figure 4, C. No other significant moderation effects were found.

Figure 4.

Variation in the weighted effect of sleep and cognitive/academic performance based on sample race/ethnicity. Panel (A) illustrates the decrease in the effect size between objective sleep quality and cognitive/academic performance as the proportion of African American/Black youth in the sample increased. Panel (B) shows the increase in the effect size between sleep schedule/chronobiology and cognitive/academic performance as the proportion of European American/White youth in the sample increased. Panel (C) presents the decrease in the effect size between sleep schedule/chronobiology and cognitive/academic performance as the proportion of African American/Black youth in the sample increased. Shaded regions represent the 95% CI. Point size and color represent weight in analyses (larger and darker indicate greater weight).

Supplemental qualitative review of published interaction effects

Although we found few moderating effects of SES and race/ethnicity in the meta-analytic association between sleep and cognitive/academic outcomes, prior research provides evidence of such effects. Given the limited number of studies available for quantitative synthesis, a brief supplemental qualitative review is an important step for contextualizing and extending the current findings. Eleven studies met inclusion criteria and examined interactions between sleep and SES (n = 10) and race/ethnicity (n = 5) in predicting cognitive (n = 11) and academic (n = 3) outcomes. Multiple studies tested the moderation of sleep duration, quality, and consistency, but no studies tested the moderation of sleep schedule.

Sleep Duration

Several studies identified SES as a moderator in associations between sleep duration and cognitive (n = 4) and academic (n = 3) outcomes [26, 38, 39, 41, 58, 63], though findings were mixed. Consistent with protective models in which sleep buffers against the negative effects of risk, youth from low SES backgrounds had lower cognitive/academic performance with short sleep but higher performance with longer sleep [39, 63]. Conversely, some found that longer sleep benefited cognitive functioning for youth in higher SES, but not lower SES, contexts [26, 38, 58]. Unexpectedly, among youth from low SES backgrounds, short sleep was more strongly associated with higher cognitive functioning than long sleep [38]. Longer sleep has also been linked with lower academic performance in high-SES contexts [41].

Race/ethnicity also moderated the relation between sleep duration and cognitive (n = 3) and academic (n = 1) outcomes [26, 38, 58]. Specifically, longer sleep was positively associated with cognitive functioning for African American/Black youth, whereas no such association was found for European American/White youth [58]. Likewise, latent growth models showed that African American/Black youth with longer sleep showed the greatest cognitive gains, and European American/White youth exhibited consistently high cognitive functioning regardless of sleep duration [38]. In contrast, one study found that longer sleep was associated with better cognitive and academic outcomes for European American/White youth but not African American/Black or Latinx/Hispanic youth [26].

Sleep Quality

Multiple studies reported SES as a moderator of the association between sleep quality and cognitive (n = 5) and academic (n = 2) outcomes [39–41, 58, 63]. Generally, poorer sleep quality was associated with lower cognitive/academic performance among youth from low SES backgrounds, whereas better sleep quality predicted higher cognitive/academic performance, narrowing SES-related disparities [39, 40, 58, 63]. In contrast, youth from high-SES backgrounds demonstrated high cognitive/academic performance regardless of sleep quality. However, alternative patterns also emerged: in some cases, poorer sleep quality was associated with worse cognitive/academic performance in high, but not low, SES contexts [39, 41]. Similarly, Buckhalt and colleagues [39] also found that in high-SES contexts, better sleep quality was linked to worse cognitive functioning.

There was also evidence of moderation by race/ethnicity in the association between sleep quality and cognitive (n = 3) and academic (n = 1) outcomes [37, 38, 57]. Latent growth models showed that among African American/Black youth, those with better sleep quality (i.e. better initial quality and improvements over time) had better cognitive functioning than those with poor and declining sleep quality [38, 57]. In contrast, European American/White youth displayed generally higher cognitive functioning regardless of sleep quality. Quadratic models also revealed that African American/Black youth had better cognitive performance at high levels of sleep quality compared to moderate or low levels [37]. No significant associations emerged for European American/White youth. Thus, racial/ethnic disparities in cognitive functioning were most pronounced at lower and moderate levels of sleep quality.

Sleep Consistency/Variability

One study reported SES as a moderator in the association between sleep consistency and cognitive (n = 1) and academic (n = 1) outcomes [39]. Among youth from low SES contexts, greater variability in wake time was linked to lower cognitive functioning, whereas in high-SES contexts, cognitive functioning was unrelated to variability in wake time. An alternative pattern was reported for variability in sleep onset—both youths from low and high SES backgrounds showed similar cognitive functioning when sleep variability was high. However, only high SES youth benefited from more consistent sleep, showing better cognitive functioning. Regarding academic outcomes, greater consistency in wake time was associated with similar academic performance across SES groups, but when variability was high, high-SES youth outperformed low-SES youth.

One study found that race/ethnicity moderated the link between sleep variability and cognitive functioning [58]. Greater variability was associated with worse cognitive outcomes for African American/Black youth but not European American/White youth. When sleep was more consistent, African American/Black and European American/White youth showed similar cognitive functioning. No race/ethnicity interactions were reported for academic outcomes.

Model 3: Accounting for heterogeneity among sleep parameters and cognitive/academic outcomes

We also considered the significance of participant and study characteristics in accounting for heterogeneity among our associations of interest. We conducted additional models to assess whether demographics such as age or gender, as well as study features such as design, outcome reporter (e.g. teacher or parent-reported), and time lag significantly accounted for the variability within our models. All results are available in Tables 7 and 8. Significant results are shared below.

Table 7.

Continuous Moderators of the Weighted Effect Between Each Sleep Parameter and Overall Cognitive and Academic Functioning

Model	Moderator	k studies	k ES	Q between	r	SE	95% CI		b	SE	95% CI
							LL	UL			LL	UL
Model 3A: Objective duration
	Gender (% girls)	14	57	.002	0.08	.17	−0.25	0.42	−.0001	.003	−0.007	0.006
	Age at sleep assessment	14	57	.62	0.14†	.08	−0.02	0.29	−.01	.007	−0.02	0.009
	Age at outcome assessment	14	57	.02	0.07	.08	−0.09	0.22	.001	.007	−0.01	0.02
	Time lag	14	57	3.40†	0.07**	.02	0.03	0.11	.03†	.02	−0.002	0.06
Model 3B: Subjective duration
	Gender (% girls)	9	26	2.05	0.05	.04	−0.03	0.12	.001	.001	−0.0003	0.002
	Age at sleep assessment	10	34	.09	0.09*	.04	0.01	0.16	.001	.003	−0.005	0.007
	Age at outcome assessment	10	34	.008	0.09	.06	−0.02	0.21	.0004	.005	−0.009	0.01
	Time lag	10	34	.26	0.10***	.01	0.08	0.13	−.004	.007	−0.02	0.01
Model 3C: Objective quality
	Gender (% girls)	13	73	.97	−0.03	.17	−0.37	0.30	.003	.003	−0.003	0.01
	Age at sleep assessment	13	73	.80	0.23*	.11	0.02	0.43	−.008	.009	−0.03	0.01
	Age at outcome assessment	13	73	.44	0.07	.10	−0.13	0.27	.006	.009	−0.01	0.02
	Time lag	13	73	3.19†	0.13***	.03	0.07	0.19	.02†	.01	−0.002	0.05
Model 3D: Subjective quality
	Gender (% girls)	16	138	.03	0.11	.20	−0.28	0.50	.001	.004	−0.007	0.008
	Age at sleep assessment	18	149	.26	0.09	.08	−0.06	0.25	.004	.007	−0.01	0.02
	Age at outcome assessment	18	149	.006	0.13	.10	−0.06	0.31	.001	.01	−0.02	0.02
	Time lag	18	149	.99	0.14***	.03	0.10	0.19	−.02	.02	−0.05	0.02
Model 3E: Schedule/chronobiology
	Gender (% girls)	4	21	.23	0.23	.21	−0.18	0.63	−.002	.004	−0.01	0.007
	Age at sleep assessment	5	45	21.36***	−0.06*	.03	−0.13	−0.001	.02***	.003	0.009	0.02
	Age at outcome assessment	5	45	.25	0.04	.12	−0.20	0.28	.01	.01	−0.02	0.03
	Time lag	5	45	13.72***	0.15***	.02	0.10	0.19	−.03***	.01	−0.05	−0.02
Model 3F: Consistency/variability
	Gender (% girls)	5	52	.63	−0.05	.15	−0.34	0.25	.002	.003	−0.003	0.01
	Age at sleep assessment	5	52	7.17**	−0.39*	.18	−0.74	−0.04	.04**	.01	0.01	0.06
	Age at outcome assessment	5	52	1.30	−0.03	.09	−0.21	0.15	.01	.009	−0.01	0.03
	Time lag	5	52	10.81**	0.08*	.04	0.01	0.16	−.06**	.02	−0.10	−0.03

^† p < .10.

^* p < .05.

^** p < .01.

^*** p < .001.

Table 8.

Categorical Moderators of the Weighted Effect Between Each Sleep Parameter and Overall Cognitive and Academic Functioning

Model	Moderator	Level/category	k studies	k ES	Q between	r	SE	95% CI		Pairwise comparisons
								LL	UL
Model 3A: Objective duration	Outcome reporter		—	—	—
		Self				—	—	—	—
		Parent				—	—	—	—
		Teacher				—	—	—	—
		Objective				—	—	—	—
		Multi-informant				—	—	—	—
	Study design		14	57	1.43
		Cross-sectional				0.07***	.02	0.03	0.11
		Longitudinal				0.10***	.03	0.04	0.16
	Document type		—	—	—
		Published				—	—	—	—
		Unpublished				—	—	—	—
Model 3B: Subjective duration	Outcome reporter		10	34	.23
		Self				0.09***	.03	0.04	0.14
		Parent				—	—	—	—
		Teacher				0.10***	.02	0.07	0.14
		Objective				0.10***	.01	0.07	0.12
		Multi-informant				—	—	—	—
	Study design		10	34	.01
		Cross-sectional				0.10***	.01	0.07	0.12
		Longitudinal				0.10***	.02	0.06	0.14
	Document type		—	—	—
		Published				—	—	—	—
		Unpublished				—	—	—	—
Model 3C: Objective quality	Outcome reporter		—	—	—
		Self				—	—	—	—
		Parent				—	—	—	—
		Teacher				—	—	—	—
		Objective				—	—	—	—
		Multi-informant				—	—	—	—
	Study design		13	73	1.70
		Cross-sectional				0.13***	.03	0.07	0.19
		Longitudinal				0.16***	.03	0.09	0.23
	Document type		—	—	—
		Published				—	—	—	—
		Unpublished				—	—	—	—
Model 3D: Subjective quality	Outcome reporter		18	149	1.51
		Self				0.19†	.10	−0.01	0.38
		Parent				0.09	.13	−0.16	0.35
		Teacher				0.19**	.06	0.07	0.32
		Objective				0.12***	.03	0.07	0.17
	Study design		18	149	.0001
		Cross-sectional				0.13***	.03	0.08	0.18
		Longitudinal				0.13***	.03	0.07	0.19
	Document type		18	149	.19
		Published				0.13***	.02	0.09	0.18
		Unpublished				0.08	.12	−0.15	0.31
Model 3E: Schedule/chronobiology	Outcome reporter		5	43	.19
		Self				0.05	.12	−0.19	0.29
		Parent				—	—	—	—
		Teacher				0.09*	.04	0.02	0.16
		Objective				0.09**	.03	0.03	0.16
		Multi-informant				—	—	—	—
	Study design		5	45	1.20
		Cross-sectional				0.13**	.04	0.05	0.22
		Longitudinal				0.07	.04	−0.02	0.15
	Document type		—	—	—
		Published				—	—	—	—
		Unpublished				—	—	—	—
Model 3F: Consistency/variability	Outcome reporter		—	—	—
		Self				—	—	—	—
		Parent				—	—	—	—
		Teacher				—	—	—	—
		Objective				—	—	—	—
		Multi-informant				—	—	—	—
	Study design		5	52	10.81**
		Cross-sectional				0.08*	.04	0.01	0.16
		Longitudinal				−0.04	.05	−0.14	0.06
	Document type		—	—	—
		Published				—	—	—	—
		Unpublished				—	—	—	—

^† p < .10.

^* p < .05.

^** p < .01.

^*** p < .001.

The weighted effect of all sleep parameters (i.e. Models 3A–3D) was robust across participant and study characteristics with the exception of sleep schedule and consistency.

Model 3.E: Schedule/chronobiology in sleep

There were two significant moderation effects found for schedule in association with cognitive/academic functioning, specifically for study lag and age in years at time of sleep measure. For study lag, shorter studies strengthened the effect size, b = −.03, p < .001 (Figure 5, A). For age, the weighted effect strengthened as the average age of the sample at the time sleep was measured increased, b = .02, p < .001, see Figure 5, B.

Figure 5.

Variation in the weighted effect of sleep schedule/chronobiology on cognitive/academic performance. Panel (A) presents the decrease in the effect size between sleep schedule/chronobiology and cognitive/academic performance as the lag between assessments increases. Panel (B) shows the increase in the effect size between sleep schedule/chronobiology and cognitive/academic performance as the age of participants at the time of sleep measure increased. Shaded regions represent the 95% CI. Point size and color represent weight in analyses (larger and darker indicate greater weight).

Model 3.F: Consistency/variability in sleep

There were three significant moderation effects found for sleep consistency and cognitive/academic functioning, specifically for study lag, study design, and age in years at time of sleep measure. For study lag, shorter studies strengthened the effect size, b = −.06, p = .001, see Figure 6, A. Likewise, the association was stronger for cross-sectional studies relative to longitudinal, b = .12, p = .001, see Figure 3. For age, the weighted effect strengthened as the average age of the sample at the sleep measurement occasion increased, b = .04, p = .01; see Figure 6, B.

Figure 6.

Variation in the weighted effect of sleep consistency/variability on cognitive/academic performance. Panel (A) presents the decrease in the effect size between sleep consistency/variability and cognitive/academic performance as the lag between assessments increases. Panel (B) shows the increase in the effect size between sleep consistency/variability and cognitive/academic performance as the age of participants at time of sleep measure increased. Shaded regions represent the 95% CI. Point size and color represent weight in analyses (larger and darker indicate greater weight).

Publication bias

We took a three-pronged approach to account for publication bias in the weighted effect between sleep and overall outcomes.

Document type as a moderator

We tested whether the effect was published versus unpublished as a potential moderator of the weighted effect between sleep and overall outcomes (Table 8). Due to cell size limitations for individual sleep parameters, we could only test this individually for subjective sleep quality. There was no significant difference in the weighted effect size for published versus unpublished effects for subjective sleep quality. Given cell size limitations, we also tested the weighted effect size of sleep broadly, which also did not vary across published and unpublished effect sizes (b = .06, p = .35).

MLMA Egger’s test

MLMA Egger’s tests were conducted for each individual sleep parameter and overall outcomes. Objective quality (b = −6.44, p < .001) and consistency (b = −8.12, p = .01) may have been subject to bias according to significant slope values, but the adjusted weighted effects for these parameters remained significant, all ps < .004. MLMA Egger’s slope results were not significant for objective duration, subjective duration, subjective quality, and schedule (bs ranged −.97 to .91, all ps ≥ .19), suggesting that a nonsignificant amount of reporting or selection bias was detected.

Commonsensical approach

Funnel plots with contour enhancement were also analyzed for reporting bias. No clear reporting bias was evident among sleep parameters and overall cognitive and academic outcomes (Figure 7). Few unpublished effects were included (subjective duration [n = 1], subjective quality [n = 3], consistency [n = 2]). Nevertheless, the significance level for records across all funnel plots ranged from p < .05 to p > .05. Therefore, concerns were limited.

Figure 7.

Contour-enhanced funnel plots depicting the level of significance for published (black circles) versus unpublished (open circles) studies. Few unpublished effects were included (n = 6). Nevertheless, the dispersion of published findings across significant and nonsignificant values can still provide insight into publication bias. No clear reporting bias in effects was found for objective duration (Panel A), subjective duration (Panel B), objective quality (Panel C), subjective quality (Panel D), schedule/chronobiology (Panel E), and consistency/variability (Panel F) and cognitive and academic outcomes.

Discussion

Prior reviews and meta-analyses have established direct links between SES, race/ethnicity, sleep, and cognitive or academic functioning, with the critical role of sleep for such outcomes being well documented. Building on this prior work, this meta-analysis is the first to quantify the interplay among these factors and examine whether SES and race/ethnicity moderate these relations. Additionally, we investigated whether specific sleep parameters are more strongly associated with particular outcomes, distinguishing between cognitive and academic performance. Our findings generally revealed that better (i.e. longer, higher quality, and earlier timing) sleep was associated with improved cognitive and academic functioning. Some evidence suggested that the strength of this association varied across SES and racial/ethnic groups. Sleep consistency was not correlated with academic performance, and neither objective duration nor sleep consistency was associated with cognitive functioning. All other associations between sleep parameters (i.e. subjective duration, objective and subjective quality, and schedule/chronobiology) and cognitive/academic outcomes were consistently significant.

SES- and racial/ethnic-related disparities in sleep and cognitive/academic functioning

The primary goal of the meta-analysis was to examine how SES and race/ethnicity moderated the associations between sleep and cognitive/academic performance. Unexpectedly, SES moderated only one association—the positive correlation between earlier sleep timing/morningness and cognitive/academic outcomes was stronger among middle and high SES samples relative to low-SES samples. Among low-SES samples, this association was small and nonsignificant. The moderating role of SES has yet to be tested in individual studies examining sleep schedules. Findings from the supplemental qualitative review of individual studies of other sleep parameters provide some insight. Consistent with the meta-analytic moderation results, some found that better sleep was associated with greater cognitive/academic functioning among youth from higher but not lower SES samples [26, 38, 39, 58]. However, others show that better (i.e. longer and more consistent) sleep was associated with higher cognitive/academic performance among youth from lower SES samples, suggesting a protective mechanism [39, 63]. Although collectively, these findings may indicate that youth from low-SES samples do not reap the same benefits from earlier schedules as youth from higher SES samples, the findings are complex and may not show the full picture.

Variability in how these factors are conceptualized and measured likely impedes the detection of moderating effects [81, 82]. For example, one meta-analysis of SES and cognitive functioning found that the overall effect size varied depending on the specific dimensions of SES and how SES was aggregated [83]. Traditional SES assessments primarily include income or education but do not account for the full scope of economic hardship [30]. For example, income-based poverty assessments do not fully capture the lived experience of economic strain [30]. Subjective assessments of financial hardship can offer unique insights into families’ perceived financial strain that are not captured by income-based measures [30, 31, 84]. Accordingly, it should be expected that the specific SES dimension measured would affect the degree to which SES modifies the sleep–cognitive/academic functioning association. However, despite considerable variability in SES conceptualizations [30, 83], many studies in this meta-analysis either lacked sufficient information to distinguish SES categories or did not include multiple SES measures. This hindered distinctions among SES dimensions, potentially masking important variance captured by different SES indicators.

Regarding race/ethnicity, our analysis yielded a few significant moderating effects. The beneficial effects of earlier sleep timing, or greater morningness, on cognitive and academic functioning strengthened as the proportion of youth identifying as European American/White increased. In contrast, the beneficial effects of objective sleep quality and sleep timing on cognitive/academic outcomes weakened as the proportion of youth identifying as African American/Black increased. Generally, these findings suggest that African American/Black youth do not reap the same benefits from good sleep.

In contrast to the meta-analytic race/ethnicity moderation results, findings from the qualitative review of individual studies examining the moderating role of race/ethnicity suggest that objective sleep quality operates as a protective factor that enhances cognitive functioning among African American/Black youth [37, 38]. The contradictory findings may be the result of differences in analytical models. Philbrook and colleagues [38] utilized latent growth models and examined how race/ethnicity moderated interindividual differences in initial levels of objective sleep quality (i.e. latent intercept) as well as interindividual changes in objective sleep quality over time (i.e. latent slope). El-Sheikh and colleagues [37] examined race/ethnicity as a moderator of quadratic effects of objective sleep quality. Although the current study addresses a similar question, examining moderation at a meta-analytic level is not analytically comparable to examining moderation of latent growth and quadratic parameters. Furthermore, we also found that associations between sleep schedules and cognitive/academic outcomes varied as a function of race/ethnicity. This has yet to be tested in individual studies and represents a critical gap in the literature.

Interpreting these conflicting findings is complex and underscores the importance of within-group designs. The meta-analytic moderation analyses by race/ethnicity rely on a comparative approach, in which an increase in the proportion of youth identifying as African American/Black youth necessarily corresponds to a decrease in the representation of youth identifying as European American/White, Asian American, Hispanic/Latinx, or “Other.” As a result, it is possible that, in comparison to other racial/ethnic groups, the positive effects of sleep quality and earlier sleep timing may appear dampened for African American/Black youth. However, this does not inherently imply that higher quality sleep and earlier sleep timing are not beneficial within African American/Black youth samples. Rather, comparative approaches may obscure these associations due to the systemic and institutional barriers that disproportionately affect this group [85–87]. When directly compared to groups not equally facing such barriers, the protective effects of sleep may be insufficient to offset broader structural inequities, thereby masking sleep’s potential benefits in between-group analyses [87, 88]. Findings from the supplemental qualitative review support this interpretation: among African American/Black youth, those who obtained better quality sleep showed higher cognitive and academic performance than those with poorer sleep quality [37, 38].

It is also important to consider the challenges posed by reporting standards for race/ethnicity in research. This work is often inter- and multidisciplinary—spanning fields such as psychology, education, public health, and sleep medicine—which adds to the complexity of establishing consistent reporting practices. Although some journals require the inclusion of racial/ethnic demographic data in manuscripts [89], there is no consistent standard for how this information should be reported, especially across disciplines. As a result, many studies either omit racial/ethnic demographic data or present it inconsistently, limiting the power to test race/ethnicity as a moderator [90]. Furthermore, existing reporting requirements often fail to account for the complex cultural contexts underlying racial/ethnic categorizations [91, 92]. For instance, cultural values and experiences of discrimination significantly influence sleep for youth from diverse backgrounds [27], yet broad racial/ethnic labels fail to capture the heterogeneity within these groups [93]. The use of broad categories such as “Other” can obscure important distinctions among groups, hindering the ability to examine differences at the meta-analytic level. These challenges in reporting standards contribute to an incomplete understanding of how associations between sleep and cognitive/academic outcomes vary among minoritized youth. However, it is acknowledged that it may not always be feasible to categorize multiethnic groups individually, especially when samples include many such multiethnic identifications. Collectively, our findings underscore the need for research on understudied racial/ethnic groups. Despite comprising over 25 percent of the US population [94], groups such as Asian Americans, Indigenous/Native Americans, Hispanic or Latinx individuals, and those of two or more races/ethnicities remain underrepresented in current research, though there are exceptions [95–98].

Although we observed few moderating effects, SES and race/ethnicity remain important in the sleep–cognitive/academic functioning association. Prior reviews have indicated the direct effects of SES and race/ethnicity on both sleep [35, 36] and cognitive/academic functioning [83]. Our findings on SES and race/ethnicity underscore the need for more specific reporting in studies. The way these socioecological constructs are conceptualized, defined, and measured across individual studies contributes considerable heterogeneity to effect sizes and increases the difficulty for meta-analyses to synthesize effects. Although we advocate for standard reporting of SES and racial/ethnic groups to foster a comprehensive understanding of health disparities, such grouping approaches are likely to become progressively challenging as complexities in SES and racial/ethnic groups grow. We recommend placing greater emphasis on within-group designs, which offer valuable insights beyond those provided by comparative between-group approaches. Between-group designs compare one SES or racial/ethnic group to another and fail to account for heterogeneity in sleep or cognitive/academic functioning within each group [88, 99]. This approach oversimplifies group membership by treating it as an inherent risk or protective factor, failing to capture average levels of functioning and variability within specific groups. In contrast, within-group designs are essential for understanding intragroup dynamics and the diverse experiences within a group and treating experiences rather than group membership as protective or risky. These designs can provide important insights into the associations between sleep and cognitive/academic functioning for specific groups, thereby enabling richer and more informative between-group comparisons. Additionally, measures of SES and race/ethnicity should be precisely defined and contextualized. Utilizing tools such as an SES decision tree [100] can help researchers select measures that better correspond to their theoretical aims. Specifically, this approach prompts greater consideration of the specific dimensions of SES or race/ethnicity believed to influence sleep and cognitive/academic functioning. For example, if researchers are interested in understanding how access to material resources promotes better sleep, their study should utilize measures related to material resources and not SES or financial distress more broadly. Alternatively, if the study is exploratory, then multiple targeted SES measures should be included to determine whether any specific dimension accounts for unique variance beyond the others.

Specificity in sleep parameters and cognitive/academic functioning

As a secondary goal, we examined the specificity of the association between sleep and outcome type across sleep parameters and cognitive/academic functioning. Findings underscore the importance of identifying which sleep parameters are most influential for cognitive and academic outcomes. Several key comparisons emerged. Subjective sleep quality was more strongly associated with overall cognitive/academic functioning, particularly when compared to objective duration. The stronger effect of subjective sleep quality is consistent with prior reviews [3] as well as some work in the adult literature. We also found that subjective sleep duration was more strongly associated with cognitive performance, especially in comparison to objective quality, subjective quality, and schedule. There are several plausible explanations. First, subjective sleep quality measures capture an individual’s satisfaction with their sleep or their feeling of restfulness, aspects that objective measures may overlook. Second, individual differences in perceived restfulness can exist among individuals with similar objective sleep quality [3, 101]. Third, subjective sleep duration is often overestimated in comparison to objective measures and may inflate effect sizes [102–104]. Fourth, subjective short and low-quality sleep may be stronger indicators of fatigue and daytime sleepiness [102], which may more proximally influence cognitive/academic performance by reducing daytime alertness and causing functional impairment (e.g. in school) [105, 106]. Finally, our broad assessment of subjective sleep quality, which encompasses self-reported satisfaction, sleep problems, and daytime sleepiness, may have amplified its effect. It is also worth noting that the larger number of studies and effects for subjective quality may have provided greater statistical power to detect significant differences.

In contrast, sleep consistency emerged as the weakest correlate of cognitive/academic functioning. Although sleep consistency is regarded as an important aspect of sleep–wake patterns, research on this parameter in youth remains limited [107]. Five studies in our review included measures of sleep consistency, which may reflect the emerging interest in this parameter. Systematic reviews have noted that associations between sleep consistency and cognitive/academic functioning remain inconclusive [108, 109]. Extending these narrative findings, we observed negligible to weak effects of sleep consistency when quantified across studies. However, these findings may reflect methodological challenges rather than a true absence of effects. For example, due to limited data and small cell sizes, we combined multiple aspects of sleep consistency (e.g. duration and schedule) to create an overall sleep consistency measure. It is possible that these different facets of sleep consistency have distinct or contrasting associations with cognitive/academic performance [39]. Additionally, studies use different methods to calculate sleep consistency (e.g. mean-centered coefficient of variance, standard deviation) [39, 79], which may further explain the lack of consistent findings. Thus, results related to sleep consistency must be interpreted with caution due to the limited number of studies assessing this parameter.

Likewise, we found a weaker association between objective sleep duration and cognitive functioning. Some research suggests that the effects of sleep duration may be nonlinear—specifically, that both short and long sleep durations can be detrimental [110, 111]. If the relationship follows a U-shaped pattern, the overall bivariate association may appear weaker or near zero, as the magnitude of the effect varies across the sleep duration spectrum (e.g. linear improvements in cognitive functioning from short to moderate sleep that plateau or decline with longer sleep). Notably, however, there is conflicting evidence for the nonlinear effects of sleep duration [37], and research overwhelmingly continues to identify short sleep duration as suboptimal, as highlighted in a comprehensive review [112].

Finally, sleep, particularly objective duration, objective quality, and subjective quality, was more strongly associated with youths’ academic performance than with their cognitive functioning. Academic performance, as operationalized in this meta-analysis, included measures of GPA and subject-specific performance. These measures may reflect a combination of socioemotional, behavioral, and cognitive factors. For example, GPA encompasses cognitive abilities as well as school-related factors such as positive behaviors and perseverance that contribute to school success [113]. Well-rested youth may excel in broader school-based domains, such as maintaining positive relationships with teachers and peers or pursuing extra credit opportunities that enhance their academic performance. These real-world benefits may not be fully captured by laboratory-based cognitive tests. Moreover, cognitive tests and standardized achievement tests typically demonstrate high reliability, whereas the inherent “impurity” of academic assessments such as GPA may introduce additional variability, potentially enhancing the ability to detect significant differences. Thus, it is not clear whether the findings have been influenced by such factors. This comparison does not imply that sleep has no effect on cognitive factors such as decision-making or reasoning. The current findings and prior research indicate that sleep is a modest correlate of cognitive functioning [6, 24]. Attention to the issues we have raised here by neuroscientists who study child and adolescent sleep with brain imaging has the potential to advance our understanding [16–18].

Other moderators and risk of bias

All sleep parameters, except for schedule and consistency, demonstrated robust effects across participant and study characteristics. The association between sleep schedule and consistency and cognitive/academic functioning was stronger in samples of older youth. During adolescence, sleep becomes more irregular [107] and preferences for eveningness increase [114], which, when coupled with structural and functional brain changes, may increase the risk for poor developmental outcomes. Youth who maintain earlier sleep timing and more consistent sleep during this critical period may be buffered against these risks and more likely to exhibit enhanced cognitive/academic performance. Additionally, the association between sleep consistency and cognitive/academic functioning was stronger in cross-sectional designs, suggesting potential bias from random measurement errors in concurrent assessments [115]. This effect was also more pronounced in studies with shorter time lags between measurements of sleep consistency and cognitive/academic performance, supporting revisionist models that propose a gradual attenuation of effects over time [116]. With the exception of sleep schedule, time lag did not moderate the effects of other sleep parameters, which supports enduring effects models and suggests that the effects of sleep duration and quality persist over time [116]. However, many correlational and experimental studies are designed to test cognitive/academic functioning in close proximity to sleep assessments because the effects are proposed to be stronger. Accordingly, findings related to study design and time lag do not necessarily indicate a design flaw.

Limitations

Several limitations warrant consideration. First, we adopted a US-centric focus, and the results may not be globally applicable due to differences in SES and the representation of minoritized groups worldwide. Second, the inclusion criteria requiring the reporting of SES or race/ethnicity demographic data may have reduced the number of studies included. Relatedly, aggregating SES and race/ethnicity data across studies is challenging due to missing data and the use of varying indicators and classifications. Third, variations in sleep terminology, operational definitions, and data coding procedures across studies may have influenced the results. Fourth, future meta-analyses could broaden their scope to include additional cognitive domains (e.g. neural activity, functional connectivity). Fifth, SES and race/ethnicity are intertwined [117], making it challenging to isolate their individual effects on sleep and cognitive/academic functioning [118, 119]. Finally, some analyses were restricted due to insufficient data, especially in regard to testing heterogeneity in the effects of sleep parameters individually.

Conclusions

This meta-analysis is the first to quantify the effect of sleep on youths’ cognitive and academic functioning across SES levels and racial/ethnic groups. A few significant moderation effects were found for SES and race/ethnicity. Additionally, all sleep parameters, except for sleep consistency, showed significant associations with cognitive/academic functioning. Subjective sleep quality and duration consistently emerged as the strongest predictors of cognitive/academic functioning, underscoring the importance of individual differences in restfulness and sleep satisfaction. These findings highlight the potential of sleep as a target for improving youths’ cognitive/academic performance and emphasize the need for more nuanced research approaches in this complex area.

Disclosure statement

Financial disclosure: None except funding from the National Heart, Lung, and Blood Institute (Grants R01-HL136752 and R01-HL093246) and the Eunice Kennedy Shriver National Institute of Child Health and Human Development (Grant R01-HD046795) awarded to Mona El-Sheikh.

Non-financial disclosure: The authors have declared no non-financial conflicts of interest.