Sleep duration and subject-specific academic performance among adolescents in China

Abstract

This study draws on data from 54,102 eighth-grade students across 717 middle schools in Shanghai, China, to examine the relationship between sleep duration and academic performance across subjects. Sleep duration was self-reported as average nightly sleep on school days, and academic performance was assessed through standardized subject tests administered in the same semester. Using OLS regression, threshold models, and Shapley value decomposition, we found that around 8 h of sleep was most strongly associated with higher academic performance, particularly in mathematics and science. The positive association was more evident among lower-achieving students. Girls generally required slightly longer sleep than boys to achieve their best performance in most subjects, except mathematics. Additionally, homework time and electronic device use were significantly linked to reduced sleep duration. These findings underscore the importance of adequate sleep in supporting adolescent learning and suggest that interventions should address sleep habits alongside academic demands in high-pressure educational contexts.

Introduction

In recent years, there has been a growing global concern about the physical and mental health of adolescents. As a core component of well-being, sleep has increasingly drawn attention from both the academic community and policymakers. Educational and health authorities in several countries have issued reports highlighting the widespread issue of insufficient sleep among adolescents^1,2. Despite its vital importance, research indicates that students across all educational levels are experiencing reduced sleep duration^3,4. Particularly in the context of increasing academic pressure, reduced sleep duration has become an important factor affecting adolescents’ physical and mental health and academic performance^5–8. Sleep is essential for maintaining cognitive function, emotional regulation, and physical health. Sufficient sleep not only facilitates brain processes that consolidate and integrate memory but is also crucial for executive functions such as attention, working memory, and problem-solving abilities^9–11. This is especially important for the middle school population, which is under a lot of academic pressure¹². However, the average amount of sleep for adolescents is decreasing globally as the burden of schooling increases^13,14. Particularly in Asian countries, influenced by Confucian culture, academic achievement is seen as an important measure of individual and family success^15,16, which result in students in countries such as China investing more time in academic competition, compressing necessary sleep time^15,16.

Given the established consensus that adolescents should sleep between 8 to 10 per night,this study does not aim to determine a universal optimal sleep duration. Instead, it seeks to provide a more nuanced understanding of how sleep duration relates to academic performance across different subjects and among different student groups. Using large-scale, standardized data from eighth-grade students in Shanghai, this study investigates the non-linear associations between sleep duration and subject-specific academic outcomes. It further explores how these relationships vary across academic performance levels and gender, and examines how homework time and electronic device use influence students’ sleep duration. By combining objective test scores, self-reported behavior, and demographic background, this research aims to offer subject-specific, evidence-based insights to improve adolescent sleep health and academic achievement within a high-pressure education system.

Sleep duration is defined as the number of hours of sleep per night⁵, and it has been widely recognized as a crucial factor influencing adolescents’ physical and mental health, cognitive development, and emotional stability⁶. According to the National Sleep Foundation’s recommendations, children aged 6−13 years should ideally sleep 9–11 h and adolescents aged 14−17 years should ideally sleep 8−10 h¹⁷. The American Academy of Sleep Medicine recommends that adolescents get 8 to 10 h of sleep per night, but nearly 40% of adolescents in the United States report getting <7 h of sleep most nights¹³.

Research has shown that adolescent sleep duration is influenced by a combination of biological, environmental, and social factors. Biologically, the commonly observed “phase delay” during puberty is driven by a shift in the timing of melatonin secretion, resulting in changes to circadian rhythms. This leads to delayed sleep onset and increased difficulty waking up in the morning^18,19. This physiological change and the socially imposed daily routine, especially early school mornings, create a sleep time mismatch that often leads to sleep deprivation among adolescents²⁰. Environmental factors also play a crucial role in influencing adolescent sleep. For example, increased evening use of electronic devices is closely associated with reduced sleep duration and poorer sleep quality^21,22. The blue light emitted from these devices suppresses melatonin production, delaying sleep onset²³. Additionally, high academic pressure compresses adolescents’ sleep time, as students spend significant hours after school completing homework or attending evening tutoring sessions—especially prevalent in countries where academic achievement is highly emphasized^24–26. Social influences are also can not be neglected, such as parental control over bedtime and peer interactions. Studies indicate that adolescents lacking parental supervision over bedtime tend to go to bed later and have shorter overall sleep duration^27,28. Furthermore, social media use and evening interactions with peers further disrupt sleep patterns among adolescents^14,29. These intertwined factors create a complex network affecting adolescents’ sleep duration, leading to insufficient sleep, which has negative impacts on their academic performance, emotional regulation, and overall health.

A growing body of recent research has confirmed that sleep significantly influences adolescents’ academic performance, with multiple contextual factors shaping this relationship. For example, early school start times are associated with chronic sleep deprivation and systematically lower academic outcomes³⁰. Even when students obtain sufficient sleep, a mismatch between their biological sleep preferences and school schedules can negatively affect their academic performance³¹. Additionally, assessments administered at suboptimal times of day can impair performance, especially in cognitively demanding subjects like mathematics³². These findings highlight the complex interplay between sleep patterns and learning outcomes, underscoring the need to examine this relationship within specific cultural and educational contexts.

The impact of sleep duration on adolescents’ academic performance has garnered widespread attention^8,33, especially the varying sensitivity of different subjects to sleep duration, which has become a focal point of recent academic research. Overall, insufficient sleep impairs adolescents’ cognitive abilities⁵, particularly in areas such as attention, memory processing, learning motivation, and executive function^9,10,34, all of which are critical factors for academic success³⁵. Short sleep duration and poor sleep quality were the sleep metrics most strongly associated with a higher likelihood of lower GPA⁶. The inverted U‑shaped relationship between sleep duration and academic performance remained significant even after controlling for academic‑related confounders such as socioeconomic status and prior academic performance³⁶. Additionally, from a physiological perspective, shortened or disrupted sleep increases adolescents’ stress response, physical pain, emotional disturbances, and memory deficits—issues that further affect their academic performance³⁷.

Differences in the short-term and long-term effects of sleep duration on adolescent academic performance exist, with short-term sleep deprivation typically occurring during test preparation and peak study periods, especially before final exams, when many adolescents reduce their sleep hours in response to academic stress. However, research suggests that short-term sleep deprivation, while negatively affecting immediate academic performance, has relatively small consequences for long-term academic development, and that sleep deprivation during middle and high school adolescence has a greater impact on future academic performance relative to sleep deprivation at the college level³⁸.

Prior research has primarily examined the link between sleep duration and overall academic performance, often operationalized as cumulative GPA or mean test scores³⁶. However, academic performance is not a uniform construct; different subjects engage distinct cognitive skills and mental processes. The impact of sleep duration on adolescents’ academic performance varies across different subjects. Subjects such as mathematics and science, which require high levels of cognitive processing, are especially dependent on adequate sleep. Studies have shown that sleep deprivation significantly reduces adolescents’ math performance, as sleep directly affects problem-solving, logical reasoning, and inferential skills^39,40. Additionally, subjects like math and science involve high levels of abstraction, requiring students to engage in complex deduction and reasoning at advanced cognitive levels, abilities that are severely compromised by insufficient sleep^41,42. Compared to math and science, language-based subjects (such as English), while also reliant on cognitive abilities, are somewhat less sensitive to sleep deprivation. Although insufficient sleep can impair memory, especially in processing vocabulary and grammatical rules, language and arts subjects generally rely more on long-term accumulation and mastery. Therefore, short-term sleep deprivation has a more moderate impact on performance in these areas⁴². However, prolonged sleep insufficiency still leads to declines in language comprehension and expression abilities, particularly in analyzing complex texts and applying language structures⁴³.

In China, sleep problems among adolescents are becoming increasingly severe, especially amid rapid urbanization, where academic pressure and social expectations collectively impact their sleep quality and duration. Studies show that the average sleep duration of Chinese adolescents is generally below international standards, particularly during middle and high school years, with many students sleeping less than 7 h per night^15,44,45. This phenomenon is even more pronounced in large Chinese cities like Shanghai and Beijing, where the pressures of extracurricular tutoring and homework further reduce students’ sleep time^46,47.

A randomized cross-sectional survey of 4801 Chinese adolescents aged 11−20 found that 51.0% of adolescents did not achieve optimal sleep duration on weekdays (defined as <8.0 h per day)⁴⁸. Similarly, a cross-sectional study of children and adolescents aged 8-16 in Shanghai found that over half of the participants (57%) did not meet the sleep duration recommended by the National Sleep Foundation³⁵. Likewise, a substantial number of adolescents experience either insufficient sleep or poor sleep quality. This situation not only affects their daily learning and life but may also have long-term impacts on their physical and mental health. Studies have found a significant association between sleep deprivation and mental health issues among adolescents, such as anxiety and depression^25,49.

The causes of sleep deprivation among Chinese adolescents can be attributed to multiple factors, including not only biological aspects but also social and cultural influences unique to China. Socially, high parental expectations for academic performance⁵⁰ and school schedules—particularly early school start times⁴⁴—further exacerbate adolescents’ sleep deprivation⁵¹. In East Asian societies with a Confucian cultural background, academic achievement is highly valued, leading families and society to hold exceptionally high expectations for academic success. This cultural context often drives adolescents to dedicate significant time to their studies, making sleep deprivation particularly severe⁴⁶. Many students and parents view sacrificing sleep as necessary to achieve academic goals. Cross-cultural studies show that Asian adolescents tend to go to bed later and have shorter sleep durations than their Western counterparts^3,52.

Additionally, many parents encourage their children to attend extracurricular tutoring classes to gain an edge in the highly competitive academic environment. Tutoring has become a common social phenomenon in China, especially under the examination-oriented education system. To prepare for entrance exams and various standardized tests, many students participate in after-school tutoring programs. While this high-intensity learning approach aims to improve academic performance, it often takes up students’ after-school time and extends into the evening, placing adolescents in a dilemma between studying and resting, further reducing their sleep time²⁹. The Chinese government has implemented a series of policies to safeguard students’ sleep duration, such as the “Standards for Management of Compulsory Education Schools,” which explicitly mandates at least 10 h of sleep per night for primary school students, 9 h for middle school students, and 8 h for high school students. However, due to prevailing public perceptions and cultural constraints, these policies have not gained sufficient attention from schools and parents, and adolescent sleep health has not significantly improved⁵³. Overall, the sleep issues among Chinese adolescents warrant serious attention, especially in raising public awareness and enhancing policy interventions.

Academic research on the relationship between sleep duration and academic performance has made some progress, yet several key issues remain to be addressed. First, the generalisability of the study is limited. There is a lack of large-scale studies specifically examining sleep duration among Chinese adolescents, particularly within the Confucian cultural context. A few earlier studies with small sample sizes indicated severe sleep duration issues among Chinese adolescents. Most studies use self-reported or parent-reported school performance rather than standardized academic tests, which may result in larger effect size³⁵.

Second, empirical findings regarding the relationship between sleep duration and academic achievement remain mixed⁵⁴. Some studies report inverted U-shaped patterns, while others find null or even linear associations^55,56. Third, prior research often concentrates on overall academic performance, overlooking potential variation across different subjects. However, academic domains such as mathematics, language, and science differ in cognitive demands and developmental trajectories. As such, a subject-specific approach may reveal patterns that are obscured when only average performance is analyzed. Finally, the influence of sleep duration may differ across student subgroups—for instance, by academic ability or gender—yet such heterogeneity is underexplored.

Despite the growing attention to the relationship between adolescent sleep and academic performance, several research gaps remain. Previous studies have largely relied on small-scale samples or self-reported academic outcomes, with limited exploration of how sleep duration differently affects various academic subjects or students at different academic levels. This study contributes to the literature in four key ways. First, it adopts a subject-specific perspective by distinguishing the sensitivity of various academic domains to changes in sleep duration. Second, it employs nonlinear modeling techniques, including quadratic regression and threshold analysis, to identify subject-specific optimal sleep ranges and to summarize the overall optimal sleep duration across subjects. Third, it applies Shapley decomposition to quantify the relative contribution of sleep duration to academic outcomes across subjects, offering deeper insights into its role beyond structural factors such as socioeconomic status. Fourth, the study examines heterogeneity across students of different academic performance levels and genders. Notably, this research draws on a large-scale dataset of standardized test scores from 86,127 middle school students—a dataset that is exceptionally rare in terms of both size and richness, incorporating multi-subject objective academic outcomes, detailed behavioral measures, and background information. Especially within the context of East Asia’s high-pressure educational environment, the scale, objectivity, and multidimensionality of this dataset provide an invaluable empirical foundation for uncovering the complex relationship between sleep and academic performance, and for informing more precise and actionable educational interventions.

Results

Based on the valid sample of 54,102 students who provided both self-reported sleep duration and academic performance data, ~13,549 students (25%) reported an average sleep duration of no more than 7 h per night during school days. About 33,098 students (61%) slept for 8 h or less, while 5144 students (around 9%) reported sleeping for >9 h.

For subsequent analysis, mathematics performance is used as the representative academic outcome variable. Figure 1 illustrates both the academic performance and the number of students corresponding to each sleep duration category. A clear inverted U-shaped trend is observed: when students slept fewer than 6 h, their average math score was 458, the lowest across all groups. As sleep duration increased, scores gradually improved. Students who slept 7–8 h scored an average of 506, and those in the 8–9 h group performed best, with an average score of 509. Beyond this range, performance began to decline: students sleeping 9–10 h scored 505, and those sleeping >10 h dropped to 482 on average.

Fig. 1. Student numbers and academic performance corresponding to different sleep duration.

This figure illustrates the distribution of sleep duration and mathematics scores within the mathematics sample. Sleep duration is divided into six self-reported categories: <6 h, 6–7 h, 7–8 h, 8–9 h, 9–10 h, and >10 h. The left y-axis represents mathematics scores standardized to a mean of 500 and a standard deviation of 100, while the right y-axis indicates the number of students in each sleep category.

Table 1 presents the differences in sleep duration and academic performance across student background characteristics, including gender, urban-rural classification, local residency status, and school type. Analyses for other subjects (Chinese, English, science, and arts) are provided in Supplementary Information.

Table 1.

Differences in sleep duration and academic performance among students from different backgrounds

Variable		Gender		Urban or rural		Household registration		School type
		Female	Male	Rural	Urban	Non-local	Local	Private	Public
Sleep duration	Mean	3.145	3.262	3.163	3.304	3.252	3.191	3.105	3.225
	SD	0.994	1.035	1.008	1.029	1.010	1.019	1.032	1.013
	T	5.616***		14.914***		6.025***		26.461***
	Cohen’s d	0.115		0.139		0.059		0.118
Math score	Mean	496.134	502.213	493.523	514.447	466.975	510.843	532.066	494.078
	SD	95.657	103.808	102.525	92.508	111.025	93.599	81.146	102.009
	T	13.283***		18.485***		36.653***		9.854***
	Cohen’s d	0.060		0.210		0.446		0.383

Note: *p < 0.05, **p < 0.01, ***p < 0.001, Sleep duration values are category codes (1–6), not actual hours of sleep. Same notation applies in the following tables.

In terms of gender, boys reported significantly longer sleep duration than girls (T = 5.616, p < 0.001, Cohen’s d = 0.115), and also outperformed girls in math (boys: M = 502.213; girls: M = 496.134; T = 13.283, p < 0.001). However, the effect size was small (Cohen’s d = 0.060), indicating that although gender differences exist in both sleep and academic performance, the magnitude is limited.

For urban-rural differences, urban students had significantly longer sleep duration than their rural counterparts (T = 14.914, p < 0.001, Cohen’s d = 0.139) and also performed substantially better in math (urban: M = 514.447; rural: M = 493.523; T = 18.485, p < 0.001, Cohen’s d = 0.210). This suggests that both sleep and academic outcomes are affected by residential background, with a stronger impact observed on academic performance.

Regarding household registration (hukou) status, non-local students slept slightly longer than local students (T = 6.025, p < 0.001, Cohen’s d = 0.059), and the difference in math performance was particularly pronounced (local: M = 510.843; non-local: M = 466.975; T = 36.653, p < 0.001), with a large effect size (Cohen’s d = 0.446), indicating a strong academic disparity associated with residency status.

In terms of school type, students in public schools reported slightly longer sleep duration than those in private schools (T = 26.461, p < 0.001, Cohen’s d = 0.118). However, their math scores were lower (public: M = 494.078; private: M = 532.066; T = 9.854, p < 0.001, Cohen’s d = 0.383), indicating notable differences in academic performance by school type.

In summary, student background factors—including gender, urban-rural location, household registration, and school type—are significantly associated with both sleep duration and academic outcomes. Among these, residential location, registration status, and school type show larger effect sizes on academic performance, suggesting that they may play a more prominent role in shaping students’ learning outcomes.

Table 2 presents the results of the OLS regression analysis examining the relationship between sleep duration and students’ academic performance across subjects. The findings show that sleep duration is significantly and positively associated with academic achievement in all subjects (p < 0.001), while the squared term of sleep duration is significantly negative (p < 0.001), indicating an inverted U-shaped relationship. This suggests that, within a certain range, longer sleep is associated with better academic outcomes, but excessive sleep beyond a critical point is linked to lower performance.

Table 2.

OLS regression results on the impact of sleep duration on academic performance

		Model 1	Model 2	Model 3	Model 4	Model 5
		Chinese	Math	English	Science	Art
Intercept		347.8***	292.5***	287.1***	295.4***	411.5***
		(6.530)	(5.033)	(5.393)	(6.471)	(7.872)
CV	Gender	-32.87***	8.501***	-19.57***	3.058**	-52.94***
		(1.312)	(1.134)	(1.080)	(1.300)	(1.587)
	Urban or rural	1.701	1.31	2.688**	5.143***	7.208***
		(1.467)	(1.262)	(1.200)	(1.454)	(1.759)
	Household registration	4.040**	13.09***	18.38***	8.262***	-0.0886
		(1.607)	(1.381)	(1.326)	(1.592)	(1.87)
	School type	-6.036***	-11.68***	-27.41***	-9.721***	-4.606**
		(1.879)	(1.612)	(1.536)	(1.862)	(2.264)
	Family Socioeconomic and Cultural Status	1.760***	2.567***	2.809***	2.279***	1.270***
		(0.042)	(0.037)	(0.035)	(0.042)	(0.051)
	Homework completion duration	3.525***	4.048***	3.700***	2.556***	0.879
		(0.723)	(0.627)	(0.597)	(0.717)	(0.889)
	Electronic device usage duration	-9.643***	-10.74***	-8.310***	-8.193***	-7.997***
		(0.490)	(0.428)	(0.407)	(0.485)	(0.590)
IV	Sleep	46.81***	32.28***	41.50***	45.18***	35.64***
	duration	(3.253)	(2.203)	(2.701)	(3.225)	(3.987)
	(Sleep)	-6.430***	-3.764***	-5.511***	-5.868***	-4.722***
	(Duration)2	(0.487)	(0.322)	(0.403)	(0.483)	(0.595)
Model fit	R2	0.173	0.274	0.34	0.208	0.174
	F	439.28	953.76	1299.06	552.62	278.60
Optimal sleep duration coefficient		3.639	4.287	3.765	3.850	3.774
Optimal sleep duration		7–8 h	8–9 h	7–8 h	7–8 h	7–8 h

Note: CV represents control variables, and IV represents independent variables. *p < 0.05, **p < 0.01, ***p < 0.001. Values in the table represent unstandardized regression coefficients. Numbers in parentheses are robust standard errors.

To identify the subject-specific optimal sleep duration—that is, the amount of sleep statistically associated with the highest predicted academic performance—we included both the linear and quadratic terms of sleep duration in the regression model. This allows us to capture the curvilinear relationship between sleep and academic outcomes. Given the inclusion of a quadratic term in the regression model, the relationship between sleep duration and academic performance is modeled as a parabolic function. The turning point of the quadratic function, calculated using the formula −b/2a, where b and a are the estimated coefficients for sleep duration and its squared term, respectively, represents the peak of the curve, corresponding to the range of sleep duration most strongly associated with higher academic performance. Since sleep duration is treated as an ordinal variable (ranging from 1 = “<6 h” to 6 = “>10 h”), the computed vertex value falls within a specific interval. This method enables us to quantitatively estimate the amount of sleep most conducive to academic achievement in each subject. The results suggest that for Chinese, English, science, and the arts, the optimal duration is approximately between 7 and 8 h, while for mathematics, it lies between 8 and 9 h.

Regarding the control variables, several patterns emerge consistently across subjects. Female students tend to perform significantly better in Chinese, English, and the arts, while boys have an advantage in mathematics, aligning with prior research on gender differences in subject-specific academic strengths. Urban students and those with formal household registration (hukou) in urban areas generally score higher, reflecting potential advantages in educational resources and access. Students attending higher-tier schools, however, often have slightly lower scores, which may reflect the more competitive peer environment or stricter grading standards in such institutions. Family socioeconomic and cultural status is positively and significantly associated with academic performance in all subjects, underscoring the role of household background in shaping educational outcomes. Furthermore, more time spent completing homework is positively correlated with higher scores, particularly in core subjects, while longer durations of electronic device use are negatively associated with academic performance, suggesting potential distraction effects. All variables in the model passed multicollinearity diagnostics, with variance inflation factor (VIF) values below 5, indicating no serious multicollinearity concerns.

Figure 2 provides a visual representation of this inverted U-shaped relationship. The horizontal axis shows the average daily sleep duration, while the vertical axis displays academic performance by subject. The solid blue lines represent the fitted regression curves, and the dashed red lines indicate the estimated optimal sleep duration for each subject. As illustrated, student performance improves with longer sleep until reaching the optimal point, after which further increases in sleep duration are associated with a decline in scores.

Fig. 2. Inverted U-shaped relationship between sleep duration and academic performance.

This figure depicts the inverted U-shaped relationship between sleep duration and academic performance across five subjects: Chinese, Mathematics, English, Science, and Arts. The x-axis represents sleep duration coded as an ordinal variable (1 = “<6 h” to 6 = “>10 h”), while the y-axis shows standardized academic performance scores (mean = 500, SD = 100). The solid blue curves represent the fitted quadratic regression lines, illustrating the curvilinear association between sleep and academic outcomes. The red dashed lines mark the estimated turning points calculated using the quadratic vertex formula (−β₁/2β₂), indicating the sleep durations statistically associated with the highest predicted scores for each subject. These results demonstrate that academic performance improves with increasing sleep up to a moderate range, followed by a slight decline beyond the peak.

These results indicate a clear nonlinear effect of sleep on academic performance. Although the degree of sensitivity varies slightly by subject, the overall findings suggest that maintaining 7–9 h of sleep per night may represent an optimal range for enhancing middle school students’ academic outcomes. This provides empirical support for future policy interventions and recommendations regarding adolescent sleep habits.

Table 3 presents the relative contribution of various influencing factors to academic performance across subjects, based on the Shapley value decomposition method. The results show that sleep duration (including both the linear and quadratic terms) contributes substantially to explaining performance in Chinese, mathematics, and science, with total contributions of 6.94%, 5.11%, and 6.24%. In contrast, the contribution of sleep duration is lower for foreign language and arts, accounting for 2.95% and 4.36%. These findings suggest that although sleep is nonlinearly associated with performance across all subjects, its impact is more pronounced in cognitively demanding subjects such as mathematics and science.

Table 3.

Shapley Decomposition Results

Variable	Chinese		Math		English		Science		Art
	Value	%	Value	%	Value	%	Value	%	Value	%
Gender	0.029	16.77%	0.001	0.36%	0.011	3.24%	0.000	0.08%	0.08	46.31%
Urban or rural	0.001	0.58%	0.004	1.46%	0.005	1.47%	0.004	1.92%	0.003	1.74%
Household registration	0.005	2.89%	0.018	6.56%	0.027	7.95%	0.011	5.28%	0.003	1.74%
School type	0.004	2.31%	0.010	3.65%	0.025	7.36%	0.007	3.36%	0.003	1.74%
Family Socioeconomic and Cultural Status	0.091	52.62%	0.193	70.37%	0.237	69.78%	0.150	71.99%	0.056	32.42%
Homework completion duration	0.002	1.16%	0.002	0.73%	0.002	0.59%	0.002	0.96%	0.000	0.18%
Electronic device usage duration	0.027	15.61%	0.032	11.67%	0.023	6.77%	0.021	10.08%	0.020	11.58%
Sleep duration	0.007	4.05%	0.008	2.92%	0.006	1.77%	0.008	3.84%	0.005	2.89%
(Sleep Duration)2	0.005	2.89%	0.006	2.19%	0.004	1.18%	0.005	2.40%	0.003	1.74%
Variance explained	6.94%		5.11%		2.95%		6.24%		4.63%

In addition, the gender variable shows high explanatory power in Chinese and arts, with contributions of 16.77% and 46.31%, indicating that gender differences should not be overlooked in academic performance related to language and arts. Other background factors, including household registration (hukou), school type, and urban-rural classification, exhibit relatively modest explanatory power overall, though they show some influence in subjects such as mathematics and foreign language.

While sleep duration and its squared term contribute meaningfully to academic performance, they are not the dominant explanatory factors in any subject. In all cases, their contribution is lower than that of students’ family socioeconomic status, measured by the ESCS (Economic, Social, and Cultural Status) index, which highlights the central role of social background in shaping academic outcomes.

Table 4 reports the results of the threshold regression models examining the nonlinear effects of sleep duration on academic performance. Both single-threshold and double-threshold models were employed to identify specific ranges where the impact of sleep duration on achievement varies significantly.

Table 4.

Threshold Regression Estimation Results of Sleep Duration on Academic Performance

Threshold variable		Single threshold		Double threshold
		Qi ≤ φ	Qiå φ	Qi ≤ φ1	φ1 <Qi ≤ φ2	Qiå φ2
Chinese	Sleep duration threshold	23.165*** (1.388)	-10.312*** (2.137)	23.163*** (1.388)	104.180*** (0.879)	-37.324*** (7.082)
	Intercept	346.010*** (4.815)	459.001*** (9.757)	345.734*** (4.8147)	—	598.736*** (36.266)
	Threshold value φ	7–8 h		7–8 h, 8–9 h
Math	Sleep duration threshold	22.370*** (1.212)	-6.958*** (1.860)	22.370*** (1.212)	86.533*** (0.761)	-29.240*** (6.046)
	Intercept	272.799*** (4.182)	374.628*** (8.485)	272.750*** (4.181)	—	490.420*** (31.077)
	Threshold value φ	7–8 h		7–8 h, 8–9 h
English	Sleep duration threshold	21.774*** (1.159)	-7.247*** (1.752)	31.548*** (3.177)	117.192*** (0.951)	-7.244*** (1.751)
	Intercept	286.963*** (4.001)	384.322*** (8.020)	269.741*** (6.574)	—	384.287*** (8.018)
	Threshold value φ	7–8 h		6–7 h, 7–8 h
Science	Sleep duration threshold	23.343*** (1.372)	-8.381*** (2.109)	23.3428*** (1.372)	92.0718*** (0.867)	-24.568*** 9(6.994)
	Intercept	294.219*** (4.759)	402.418*** (9.632)	294.0648*** (4.759)	—	486.1678*** (35.814)
	Threshold value φ	7–8 h		7–8 h, 8–9 h
Art	Sleep duration threshold	30.260*** (4.708)	0.378 (1.259)	30.277*** (4.704)	2.907 (1.959)	-38.548*** (8.668)
	Intercept	381.171*** (9.775)	453.729*** (6.198)	381.299*** (9.767)	445.225*** (7.947)	651.665*** (44.454)
	Threshold value φ	6–7 h		6–7 h, 8–9 h
Control variables		YES		YES

Note: *p < 0.05, **p < 0.01, ***p < 0.001. Values in the table represent unstandardized regression coefficients. Numbers in parentheses are robust standard errors.

For Chinese, mathematics, and science subjects, the model detected two statistically significant thresholds, located between 7–8 h and 8–9 h. When sleep duration was below the first threshold (7–8 h), the regression coefficients were significantly positive, indicating that additional sleep was associated with improved academic performance. In the intermediate range (between the two thresholds, i.e., 7–9 h), the coefficients increased markedly, suggesting that the effect of sleep on academic performance was strongest in this window. However, once sleep exceeded the second threshold (8-9 h), the coefficients became significantly negative, implying that excessive sleep duration may be associated with lower academic outcomes. These findings suggest that the optimal sleep range for these three subjects is between 8 and 9 h.

In the case of foreign language subject, the two detected thresholds were 6-7 h and 7-8 h. When sleep was below 6-7 h, the coefficient was 31.548 (p < 0.001); in the 7–8 h range, it rose to 117.192 (p < 0.001). Beyond 8 h, however, the coefficient turned negative (-7.244, p < 0.001). This indicates that 7-8 h of sleep may be most beneficial for performance in foreign language subjects, while both insufficient and excessive sleep may be detrimental to language-based cognitive functioning

For arts subject, the two thresholds were 6–7 h and 8–9 h. Below 6–7 h, sleep duration had a positive effect on performance (30.277, p < 0.001). In the intermediate range (7–9 h), the coefficient remained positive (2.907) but was not statistically significant. Beyond 8–9 h, the coefficient turned significantly negative (-38.548, p < 0.001), indicating that excessive sleep may impair performance in arts-related subjects. These results suggest that a sleep duration of 7−9 h is generally optimal for students in the arts, with diminished returns beyond this range.

Taken together, the threshold regression results corroborate the inverted U-shaped relationship identified in the earlier OLS analysis. They provide a clearer specification of the subject-specific optimal sleep intervals, which generally fall within the 7−9 h range. Around 8 h appears to be the overall optimal point for maximizing academic performance across subjects.

Figure 3 presents the results of the quantile regression analysis, which examines whether the effect of sleep duration varies among students with different levels of academic performance. Unlike traditional regression that estimates the “average effect,” quantile regression captures the varying impacts of sleep duration across different percentiles of the academic achievement distribution. The results show that, in mathematics and foreign language, the positive effect of sleep duration is most pronounced among students in the 20th to 50th percentiles, suggesting that students with average or below-average performance are more likely to benefit from adequate sleep.

Fig. 3. Quantile regression results by academic performance level.

This figure illustrates how the association between sleep duration and academic performance varies across different points (quantiles) in the distribution of academic performance. Quantiles divide the sample into equal portions. The green solid line indicates the estimated coefficients of sleep duration at different quantiles of academic performance, capturing the marginal effects from quantile regression. The gray shaded area represents the 95% confidence interval. The black dotted line shows the average effect from OLS regression, and the thinner black dotted lines indicate its 95% confidence bounds.

In science and arts, the strongest sleep effects appear among students in the lower deciles (10th-20th percentiles), indicating that students with the lowest academic performance levels are more dependent on adequate sleep in these areas. This pattern may reflect a marginal compensation mechanism of basic cognitive resources: for students with weaker foundational skills, sufficient sleep plays a more direct role in enhancing attention, information processing, and emotional regulation.

Overall, the marginal effect of sleep duration on academic performance tends to decrease as performance levels rise. This finding highlights that the academic benefits of sleep are greater for lower-achieving students. For this group, ensuring proper sleep schedules may not merely be a matter of lifestyle adjustment, but a key intervention point for improving learning efficiency and academic outcomes.

To examine the heterogeneity in the relationship between sleep duration and academic performance across genders, we conducted separate quadratic OLS regression models for male and female students, including both linear and squared terms of sleep duration and controlling for individual-level covariates. The results are reported in Table 5. Overall, both male and female students exhibit a statistically significant inverted U-shaped relationship between sleep duration and academic performance across all subjects. However, there are observable differences in the estimated optimal sleep durations. For most subjects, the estimated optimal sleep duration is slightly longer among girls than boys. In contrast, for mathematics, boys show a higher estimated optimal sleep duration than girls. These findings indicate that the association between sleep and academic performance varies by gender and subject, suggesting potential heterogeneity in students’ sleep needs and learning patterns.

Table 5.

Gender-specific regression results on the impact of sleep duration on academic performance

Variable		Model 1 Chinese	Model 2 Math	Model 3 Foreign language	Model 4 Science	Model 5 Art
Male sample
Intercept		260.052*** (9.020)	241.289*** (7.275)	217.688*** (7.128)	257.293*** (8.635)	298.282*** (11.345)
IV	Sleep duration	63.574*** (4.748)	51.034*** (3.872)	54.366*** (3.783)	54.947*** (4.547)	48.997*** (6.082)
	(Sleep Duration)2	-8.396*** (0.710)	-6.063*** (0.576)	-6.840*** (0.561)	-6.923*** (0.679)	-6.249*** (0.904)
Control variables		Yes	Yes	Yes	Yes	Yes
Model fit	R2	0.121	0.257	0.329	0.197	0.086
	F	228.579	678.537	957.388	387.678	97.343
Optimal sleep duration coefficient		3.786	4.209	3.974	3.968	3.920
Female sample
Intercept		336.805 (7.761)	254.693 (7.486)	282.988 (6.909)	277.069 (8.095)	416.82 (8.664)
IV	Sleep duration	43.833 (4.332)	48.610 (4.179)	43.396 (3.860)	48.069 (4.517)	26.348 (4.856)
	(Sleep Duration)2	-5.616 (0.661)	-6.115 (0.641)	-5.601 (0.591)	-5.989 (0.689)	-3.26 (0.740)
Control variables		Yes	Yes	Yes	Yes	Yes
Model fit	R2	0.145	0.248	0.310	0.206	0.890
	F	254.593	602.507	821.936	388.945	92.537
Optimal sleep duration coefficient		3.903	3.975	3.874	4.013	4.041

Note: *p < 0.05, **p < 0.01, ***p < 0.001. Values in the table represent unstandardized regression coefficients. Numbers in parentheses are robust standard errors.

A regression analysis was conducted to examine the effects of homework time and electronic device use on students’ sleep duration, with the results presented in Table 6. Both homework time and screen use were found to have significant negative effects on sleep duration. Specifically, the coefficient for homework time was -0.335 (p < 0.001), indicating that each additional hour spent on homework was associated with a reduction of ~0.335 h in average sleep duration. The coefficient for electronic device use was -0.118 (p < 0.001), also showing a statistically significant negative association with sleep.In Model 3, where both variables were included simultaneously, the regression coefficients remained robust: -0.332 for homework time (p < 0.001) and -0.114 for electronic device use (p < 0.001). In terms of model fit, Model 1 and Model 2 explained 10.8% (R² = 0.108) and 3.5% (R² = 0.035) of the variance in sleep duration, respectively. When both predictors were included in Model 3, the explanatory power increased to 13.2% (R² = 0.132), indicating a notable improvement in model performance.

Table 6.

OLS regression results of homework time and electronic device usage time on sleep duration

Variable		Model 1	Model 2	Model 3
Intercept		4.016*** (0.024)	3.473*** (0.024)	4.348*** (0.025)
CV	Gender	0.097*** (0.008)	0.131*** (0.009)	0.112*** (0.008)
	Urban or rural	0.070*** (0.009)	0.183*** (0.010)	0.091*** (0.009)
	Household registration	-0.044*** (0.010)	-0.071*** (0.011)	-0.044*** (0.010)
	Shool type	0.062*** (0.012)	0.099*** (0.012)	0.049*** (0.012)
	Family Socioeconomic and Cultural Status	-0.001*** (0.001)	-0.002*** (0.001)	-0.002*** (0.001)
IV	Homework completion duration	-0.335*** (0.004)		-0.332*** (0.004)
	Electronic device usage duration		-0.118*** (0.003)	-0.114*** (0.003)
Model fit	R2	0.108	0.035	0.132
	F	1088.951	329.640	1161.79

Note: *p < 0.05, **p < 0.01, ***p < 0.001. Values in the table represent unstandardized regression coefficients. Numbers in parentheses are robust standard errors.

These findings suggest that while screen time does contribute to reduced sleep, the academic workload—specifically time spent on homework—may play a more substantial role in shaping adolescents’ sleep habits. Therefore, alongside concerns about digital media use, greater attention should be paid to academic scheduling, particularly in balancing homework demands with sufficient sleep opportunities.

Discussion

This study provides new empirical evidence on the relationship between adolescents’ sleep duration and academic performance across various subjects. Results show that maintaining ~8 h of sleep is most beneficial for students’ academic performance. Related studies have also found that 7–9 h of sleep is optimal¹³. The impact of sleep duration is particularly significant in subjects with high cognitive demands, such as Mathematics, Science, and Chinese. This finding aligns with existing literature, indicating that cognitively demanding subjects, like Mathematics and Science, are more dependent on adequate sleep. Sleep deprivation negatively impacts performance in these subjects, as they require strong skills in logical reasoning, problem-solving, and abstract thinking, which are vulnerable to the effects of insufficient sleep^39,41,43. Improvements in Mathematics and Science performance are closely linked to adequate sleep duration, consistent with existing literature on the role of sleep in cognitive function, attention, and memory consolidation⁹.

In contrast, subjects such as Arts and Foreign Languages appear to be less sensitive to variations in sleep duration. This may be attributed to the nature of these disciplines, which rely more heavily on knowledge accumulation, long-term memory, or creativity - capacities that are not entirely dependent on short-term sleep states^42,44. Therefore, while adequate sleep is essential for overall cognitive performance, different subjects have varying sleep needs. This finding provides important insights for education policymakers, suggesting that learning and rest schedules should consider the unique characteristics of each subject to better balance study and rest times.

Across the overall sample, nearly one-quarter of students (13,549 students, 25%) reported sleeping no more than 7 h per night, while ~10% (5144 students, 9%) reported sleeping >9 h. Both shorter and longer sleep durations are associated with lower academic performance, forming an inverted U-shaped relationship. The high prevalence of short sleep may reflect the academic pressure commonly faced by adolescents, particularly in competitive educational contexts where students often reduce sleep time to manage heavy workloads. In contrast, longer sleep duration among adolescents may not necessarily indicate a causal risk factor for lower achievement. Rather, it could be compensatory—for instance, due to poor sleep quality—or reflect underlying health issues, fragmented sleep, or reduced time available for academic activities. Therefore, the observed associations between longer sleep and lower academic performance should be interpreted with caution, as they may be influenced by unmeasured confounders rather than indicating a direct detrimental effect of extended sleep duration.

This study found that the relationship between sleep duration and academic performance is more significant for students whose academic performance is in the lower 20%-50% range. In subjects such as Mathematics and English, lower-performing students need adequate sleep even more. For these students, lack of sleep exacerbates cognitive fatigue and reduces motivation to learn, leading to further declines in academic performance⁵⁷. Conversely, higher-achieving students show greater adaptability, maintaining relatively stable academic performance even with slightly less sleep. This aligns with the findings of Diekelmann & Born (2010)³⁴, suggesting that high-ability students may possess stronger self-regulation skills, allowing them to rely on acquired knowledge and skills to handle exams or learning tasks even under short-term sleep deprivation.

Thus, for students with weaker academic performance, ensuring sufficient sleep is a key strategy for improving their academic outcomes. This finding also provides important intervention suggestions for schools and families. Rather than increasing academic pressure on lower-performing students, priority should be given to improving their sleep quality. This approach not only enhances learning efficiency but also improves emotional well-being, reducing anxiety and depression symptoms related to academic pressure⁵³. For these students, improving sleep quality may be more effective than extending study time. Especially when facing complex subjects and high-intensity learning tasks, adequate sleep enables better cognitive processing and knowledge integration.

In addition, the results revealed significant gender heterogeneity. Overall, girls tended to have slightly longer optimal sleep durations across subjects compared to boys, suggesting that girls may rely more on the restorative effects of sleep to maintain academic performance. This may be due to the fact that eighth-grade girls are in the middle of puberty, experiencing greater fluctuations in circadian rhythms and higher emotional sensitivity, making them less tolerant of sleep deprivation and thus requiring longer sleep durations to sustain cognitive functioning and emotional stability⁵⁸. However, in mathematics, boys exhibited a longer optimal sleep duration than girls, possibly reflecting their greater reliance on spatial reasoning and strategic processing in mathematical learning⁵⁹. These findings suggest that the impact of sleep on academic performance is influenced not only by students’ academic levels but also by gender and subject-specific characteristics. Therefore, educational practice should consider the characteristics of different student groups and develop differentiated schedules and intervention strategies accordingly.

This study also reveals the dual impact of homework time and electronic device usage on adolescents’ sleep health. In China, excessive homework time is one of the primary reasons for students’ sleep deprivation. In pursuit of high academic achievement, parents and schools often assign a large amount of homework, forcing students to extend their study hours at night, thereby reducing time that should be allocated for rest^47,59. This excessive academic pressure not only leads to insufficient sleep but may also cause a range of health issues, such as anxiety, depression, and other psychological problems²⁵.

Electronic device usage is also a significant factor affecting sleep duration. With the proliferation of smartphones and electronic devices, many students use them for online games or study before bedtime. When using electronic devices, students are easily drawn into social media or gaming, further reducing the time allocated for rest²¹. A survey study in Guangzhou, China, indicates that the addiction rate to online games among Chinese adolescents is on the rise, with a general addiction prevalence of 26.50%⁶⁰. Additionally, the blue light emitted by electronic devices suppresses melatonin secretion, an essential hormone for regulating sleep. Reduced melatonin secretion can lead to difficulty falling asleep⁶¹. Therefore, managing electronic device usage time, especially before bedtime, should be a key focus for parents and schools.

In conclusion, this study, based on a large-scale dataset of 86,127 adolescents in Shanghai, China, examined the relationship between sleep duration and academic performance across multiple subjects, with particular attention to heterogeneity across performance levels and gender. The findings suggest that maintaining around 8 h of sleep per night is associated with better academic outcomes, especially in cognitively demanding subjects such as mathematics and science. Notably, the beneficial effects of sleep duration are most evident among lower-achieving students, and gender differences emerge in optimal sleep durations across subjects, with girls generally benefiting from slightly longer sleep except in mathematics.

While the results show an inverted U-shaped relationship between sleep duration and academic performance, this should not be interpreted as a causal link in both directions. Short sleep duration is well-supported as a plausible contributor to lower academic outcomes, given its documented effects on attention, memory, and emotional regulation. However, longer sleep durations should not be interpreted as “excessive” or harmful without qualification, as they may reflect compensatory behavior (e.g., due to poor sleep quality), underlying health conditions, or reduced study time. In observational studies like this, such associations are likely confounded by unmeasured variables, and reducing long sleep does not imply improved academic performance. These nuances should be carefully considered when interpreting the U-shaped patterns.

The study also finds that time spent on homework and screen exposure are negatively associated with sleep health, suggesting that policies or school practices that encroach on sleep time—for example, extending evening study hours—may be counterproductive for learning outcomes.

Despite its strengths in sample size and analytical rigor, this study has several limitations. First, sleep duration was self-reported via a single multiple-choice item, rather than allowing students to provide continuous or averaged responses over several days. This limits measurement precision and introduces potential recall or reporting biases. Second, sleep was measured as a general pattern during school days, whereas academic performance was captured via assessments during specific periods. It is plausible that sleep immediately before exams has stronger effects on performance than average sleep, and this study cannot disentangle those temporal dynamics. Third, the study has limitations in the measurement dimensions of sleep. While it focused solely on sleep duration, it did not account for other important sleep-related variables such as chronotype, sleep quality, regularity of sleep-wake cycles, or daytime sleepiness—all of which are known to affect academic performance. Additionally, the lack of information on exam timing and students’ biological rhythms limits the ability to interpret how sleep interacts with test schedules. These omissions restrict a deeper understanding of the mechanisms linking sleep and academic achievement. Future research should incorporate more comprehensive and multidimensional sleep indicators to better capture the complexity of sleep-related influences on student outcomes. Fourth, sleep duration categories were binned rather than continuous, constraining the ability to precisely estimate nonlinear thresholds. Finally, potential mediators such as mental health, academic stress, or family socioeconomic status were not fully accounted for, which may explain part of the observed associations.

Future studies should aim to incorporate objective and multidimensional sleep measures (e.g., via actigraphy or sleep diaries), collect repeated academic assessments, and account for individual sleep preferences, school timing, and contextual stressors. A more nuanced understanding of how sleep behaviors interact with environmental and psychological factors is essential for designing effective interventions to support adolescent learning and well-being.

Methods

The research data were drawn from the 2024 large-scale survey on basic education quality and subject assessments conducted in Shanghai, China, covering junior secondary school students from 717 schools across 17 districts. Eighth-grade students were selected as the target population for this study. Because this group typically represents the mid-adolescence stage, during which biological sleep rhythms, academic pressure, and workload are undergoing dynamic transitions—making them developmentally representative. Compared to seventh-grade students, who are still in the early stages of school adjustment, eighth graders have generally stabilized in their learning routines. At the same time, unlike ninth graders who are under intense pressure from upcoming high school entrance exams, eighth graders’ academic performance is less affected by high-stakes testing. Therefore, eighth-grade students offer a more natural context in which to observe regular patterns of sleep and academic performance, enabling a more precise estimation of sleep duration’s impact on academic achievement. Based on these considerations, this study focuses on eighth-grade students currently enrolled in public and private junior secondary schools in Shanghai, excluding those attending special education schools, special education classes within regular schools, and students in inclusive education settings.

When selecting schools within districts, a stratified sampling method was used based on school type and educational structure, followed by a Probability Proportional to Size sampling method relative to the number of students in each school. In cases where schools had particularly small or large student populations, adjustments were made: in schools with fewer than 90 eighth-grade students, all students participated in the survey and assessments; in schools with over 361 eighth-grade students, 240 students were sampled. For schools of intermediate size, a proportional sampling approach was used to ensure the scientific rigor of the sampling process. This method included all school types across the region, resulting in a total sample of 54,012 eighth-grade students. Due to item-level nonresponse, the final matched sample sizes vary across subjects: 36,176 students for mathematics, 36,139 for English, 31,744 for Chinese, 31,714 for science, and 18,088 for arts. In terms of sample distribution, the gender breakdown includes 44,737 male students (51.9%) and 41,390 female students (48.1%). Regarding geographic location, 59,645 students (69.3%) are from suburban areas, while 26,482 students (30.7%) are from urban areas. Concerning household registration status, 64,678 students (75.1%) have local registration, while 21,449 students (24.9%) do not. As for school type, 13,335 students (15.5%) attend private schools, and 72,792 students (84.5%) attend public schools.

The research process consisted of two components: a student questionnaire and subject assessments. The questionnaire was administered through a computerized system and completed by students in their school computer labs under a unified arrangement. It covered a range of topics including basic demographic information, family background, daily routines (such as sleep, homework time, and electronic device usage), and academic experiences. All procedures were conducted via a government-supported standardized platform, ensuring uniform testing interfaces and protocols to maintain data consistency and comparability.

The subject assessments were organized and implemented by the Shanghai Educational Quality Monitoring Center. All test items were developed by certified professional item writers. Students were randomly selected by class to participate in different subject tests, and were randomly assigned to examination rooms to avoid clustering by class. Each test room included students from various classes to minimize potential testing bias. All assessments were conducted in standard classrooms that met specific environmental criteria—safety, cleanliness, quietness, adequate ventilation, and lighting. Each test room was equipped with 30 single-person desks, and test center signs were posted at the entrance of each room indicating the site name and room code. Student ID numbers were placed on the upper left corner of each desk, and each room was supervised by trained proctors to ensure standardized and scientifically sound test administration.

This large-scale quality monitoring survey was approved by the Center for Basic Education Quality Monitoring of the Shanghai Municipal Education Commission, a government-led public service institution aimed at providing a basis for policy-making and supporting school improvement. The survey and testing in question fall within the scope of regular educational quality monitoring and are not human experiments. Relevant arrangements were fully informed to participants (and the parents or legal guardians of participants under the age of 16), and all participants took part voluntarily. All procedures complied with the principles of educational ethics, data privacy protection, and the Declaration of Helsinki. Data access for this study was formally authorized, and all data used in the analysis were de-identified to ensure compliance with the standards of anonymity, non-interference, and ethical research conduct. The study was reviewed and approved by the Center for Basic Education Quality Monitoring of the Shanghai Municipal Education Commission.

Academic performance

The subject assessments in this study included Chinese, Mathematics, English, Science, and Arts. These standardized eighth-grade test papers were developed over a 4-month period by a team of professional curriculum experts and outstanding frontline teachers under the coordination of the Shanghai Educational Quality Monitoring Center. Each test had a full score of 100 points. To facilitate comparison across subjects and enhance interpretability, raw scores were rescaled to a 500-point scale using a PISA-like standardization approach, in which the mean was set to 500 and the standard deviation to 100. The Chinese language test lasted 120 min and assessed students’ abilities in character recognition and writing, reading and appreciation, expression and communication, as well as information sorting and inquiry. The Mathematics test, with a duration of 100 min, evaluated students’ competencies in numbers and algebra, shapes and geometry, basic functions and probability/statistics, and their integrated and temporal reasoning abilities. The English, Science, and Arts assessments each lasted 90 min. The English test focused on students’ language skills, cultural awareness, cognitive thinking, and learning ability. The Science assessment covered knowledge areas in physical science, life science, earth science, and technological/engineering literacy. The Arts test assessed students’ aesthetic perception, artistic expression, creative practice, and cultural understanding.

Sleep duration

Following Yao & Chen (2023)⁵³ and Titova et al.⁵⁶, adolescent sleep duration was measured through self-reported habitual sleep duration with the question: “On average, how many hours do you sleep per night during school days this semester?” Response options included: less than 6 h, 6–7 h (excluding 7 h), 7–8 h (excluding 8 h), 8–9 h(excluding 9 h), 9−10 (excluding 10 h) hours, and >10 h. This ordinal categorical variable was coded as 1−6 to represent increasing sleep duration.

Homework time

It was measured based on Liu et al.⁶² using self-reported daily time spent on homework: “On average, how many hours do you spend on teacher-assigned written homework at home during school days this semester?” Response options included: within 1 h, 1–2 h (excluding 2 h), 2–3 h (excluding 3 h), 3–4 h (excluding 4 h), and more than 4 h. Results were also converted to continuous variables to calculate the duration of homework completion.

Electronic device usage

Students were asked, “Approximately how much time do you spend using electronic devices (such as smartphones, tablets, computers) after school each day this semester?” Response options included: within 10 min, 11−30 min, 31−60 min, 61−90 min, 91−120 min, 121−150 min, 151−180 min, and >181 min. These responses were likewise converted into continuous variables to measure the duration of electronic device usage.

Control variables

Studies indicate^15,62 that family background factors, such as socioeconomic, cultural, and social status, influence student performance. Thus, this study includes a socioeconomic and cultural background index as a control variable. Following the PISA assessment methodology⁶³, this variable is constructed using several indicators: parental education level based on the International Standard Classification of Education, parental occupational status based on the International Standard Classification of Occupations, family material resources (e.g., the number of books, electronic devices, and other study-related resources), and family economic conditions (such as travel history). These dimensions are synthesized into a single socioeconomic and cultural background index through factor analysis to capture the comprehensive socioeconomic and cultural background of the student’s family. Additionally, common control variables affecting academic performance include student gender, urban-rural classification, household registration, and school type^49,62,64. These are binary variables, treated as dummy variables in the analysis: gender (0 = female, 1 = male), urban-rural classification (0 = suburban, 1 = urban), household registration (0 = non-local, 1 = local), and school type (0 = private, 1 = public).

This study analyzes the data through the following steps: (1) Descriptive Analysis and Independent Sample T-Test: Present the distribution of sleep duration among surveyed adolescents, and examine differences in sleep duration and academic performance across students from different backgrounds. (2) Multiple OLS Regression Models: Analyze the effects of homework duration and electronic device usage on sleep duration using a multiple OLS regression model. Then, construct another multiple OLS regression model to analyze the relationship between sleep duration (including a quadratic term for sleep duration) and academic performance, identifying the sleep duration most conducive to academic achievement. (3) Shapley Decomposition Model: Compare the explanatory power of sleep duration on academic performance across different subjects. (4) Threshold Regression Model: Further verify the robustness of the OLS regression results using a threshold regression model. (5) Quantile Regression Model: Examine the heterogeneity of sleep duration’s effects on students of different genders and academic abilities using a quantile regression model. The stata17 software was used in the data analysis process.

The study employs an Ordinary Least Squares (OLS) regression model, with the basic econometric model specified as follows:

$$Yij={\beta }_0+{\beta }_1{I}_{ij}+{\beta }_2{F}_{ij}+{\beta }_{3}Tim{e}_{ij}+{\beta }_{4}Tim{e_{ij}}^2+{\epsilon }_{ij}$$ 1

Yij represents the academic performance of student i in subject j. Iij includes individual background characteristics as control variables, such as gender, household registration, urban-rural classification, and school type. Fij denotes the family socioeconomic and cultural status index for student i in subject j. Timeij is the average daily sleep duration for student i in subject j. Timeij2 represents the quadratic term of sleep duration to capture any nonlinear effects on academic performance. eij is the error term accounting for unobserved factors affecting academic performance. This model allows for the estimation of the effects of both linear and quadratic sleep duration on academic performance while controlling for individual and family background characteristics.

The Shapley decomposition model is used to measure the relative contribution of each explanatory variable (independent variable) to the dependent variable by distributing the marginal contribution of each explanatory variable to the model’s predicted value⁶⁵. In this study, it is primarily used to decompose the explanatory power of sleep duration across models for different subjects. The purpose of Shapley decomposition is to break down the total explanatory power of the regression (R²) into contributions from each independent variable.

To conduct a robustness test on the OLS regression results and examine the nonlinear threshold effects of sleep duration on students’ academic performance, this study applies Hansen’s non-dynamic panel threshold regression model. This model addresses the limitations of group tests or interaction terms in accurately estimating threshold values within the model. It not only allows for the estimation of the threshold value but also enables significance testing for the existence and validity of the threshold. The basic structure of Hansen’s panel threshold model is as follows:

$$Yij={\beta }_0+{{\beta }_1}^{\prime }{x}_{ij}I(q_{it}\le \gamma )+{{\beta }_2}^{\prime }{x}_{ij}I(q_{it}>\gamma )++e_{ij}$$ 2

i represents the student sample. $q_{\mathit{it}}$ is the threshold variable, used to determine the cutoff point in the model. $\gamma$ is the unknown threshold value that needs to be estimated within the model. $e_{\mathit{it}}$ is the random disturbance term. $I$ is an indicator function, which takes a value of 1 when the condition in the parentheses is true, and 0 otherwise.

To estimate the heterogeneous effects of sleep duration on students at different academic ability levels, a quantile regression model is used. Students can be grouped by academic performance, and different quantiles can be selected to estimate the effect of sleep duration on students with varying academic performance levels. The quantile regression model can capture the heterogeneity among students across different academic ability levels, while traditional OLS regression can only estimate the mean effect, overlooking heterogeneity at different distribution positions. Quantile regression is less sensitive to outliers and is suitable for cases where the dependent variable has an asymmetric distribution. The quantile regression equation is expressed as follows:

$$Q_{\theta }\left(y|X\right)=Min{\beta }_q\sum _{i:yi\ge xi\beta q}^{n}q\left|yi-xi\beta q\right|+\sum ^{n}i:yi\le xi\beta q(1-q)\left|yi-xi\beta q\right|$$ 3

Author contributions

Conceptualization and resources, M.W.; writing—original draft preparation, M.W., Z.C., N.H., H.Y.; writing—review and editing, Z.C.; visualization and supervision, H.Y.

Data availability

The data of this study belongs to Shanghai Municipal Government and is confidential.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41539-025-00361-y.