Gender differences in sleep behaviors under China’s universal two-child policy

Abstract

Since 2016, China has progressively relaxed family planning policies to stimulate birth rates. This paper examines the behavioral health repercussions of China’s 2016 universal two-child policy (UTCP) by analyzing sleep pattern data from China Family Panel Studies. Napping is a composite indicator that denotes health outcomes, job quality, and personal well-being. It reveals work conditions and environments to some extent. UTCP may lead to heightened social expectations regarding pregnancy likelihood, and changes in social expectations within the workplace may make work environments less equitable and more stressful for females. Leveraging a difference-in-difference model, this paper explores how napping behaviors among the working-age cohort have responded to the policy shifts. Our analysis reveals a gender discrepancy in response to the policy, specifically, females exhibit a discernible reduction in the likelihood of napping, as well as in the duration of both daytime naps and daily sleep. Conversely, such effects are not significant among males. These results suggest policy consequences extend beyond individuals directly impacted by childbirth or contemplating parenthood. Hence, while promoting fertility is still the government’s goal, policymakers are encouraged to consider the broader challenges the female population faces from social and workplace environment factors.

Introduction

The growing prevalence of aging societies worldwide has prompted policymakers and economists to prioritize social programs aimed at increasing birth rates. In response to its aging population, China’s policymakers have gradually eased the country’s stringent birth control policies, with the introduction of the universal two-child policy (UTCP) in 2016, followed by the implementation of the three-child policy in 2021. Despite an initial uptick in birth rates following the 2016 policy change, the increased birth rate proved to be short-lived, with birth rates eventually continuing their long-term decline (see Fig. 1). The effectiveness of UTCP appears limited, moreover, there are unintended negative consequences impacting the workforce. Our study seeks to investigate the impacts of UTCP through the lens of health behavior.

Figure 1.

Birth rate and population growth rate in China. It shows the birth rate and the natural growth rate of the population in China. Data comes from the China Statistical Yearbook 2022.

At the core of our investigation are sleep patterns, particularly daytime napping, which usually occurs right or shortly after lunch. Sleep patterns, including both nighttime sleep and daytime naps, are known predictors of health^1,2, productivity, psychological well-being, and cognitive function³. Napping also reflects job flexibility, a critical non-wage job characteristic. By examining the policy’s impact, we introduce an alternative framework for understanding labor market outcomes. Despite the cultural acceptance of napping in China, changes in napping behavior may signal disruptions in personal well-being, influenced by workplace dynamics⁴ and peer influence⁵. Thus, napping serves as a comprehensive indicator of health outcomes, labor market dynamics, and subjective well-being.

The UTCP makes having two children legal for all couples, regardless of ethnic group, location, and whether they have siblings. Exploiting the fact that the UTCP does not affect the fertility decision of women with two or more children and that the public also perceives this, our study employs a difference-in-difference specification using data from the China family panel studies (CFPS) to examine the behavioral health implications of UTCP on sleep patterns in China. Our analysis reveals a significant gender difference in response to the UTCP, with females showing reductions in both the likelihood and duration of napping, as well as changes in daily sleep duration. These effects exist among working-age females regardless of whether they have a child or their intention of having another child. We contend that the UTCP renders napping more costly for women, primarily due to gender inequality in the labor market.

Our study contributes to existing literature in several ways. First, we add empirical evidence to the literature on the relationship of fertility-labor market outcomes. Earlier studies rely on the composition of children’s sex or twins to deal with the endogenous issue of fertility and labor supply^6–8. We utilize the UTCP as an exogenous policy shock like other studies in the Chinese context^9–12, which allows us to assess the effects on the broader population since our study does not need to be conditional on women’s birth history. We reach consistent results suggesting that fertility policy affects labor market outcomes. While previous studies focus on labor force participation, working hours, and wages, our research seeks to enrich existing academic thought by examining the influence of family planning policies on labor market outcomes through non-wage job characteristics such as job welfare and flexibility.

Second, we highlight that the impact of the family planning policy extends beyond individuals directly involved in childbirth or those planning to have children. Existing studies often assume family planning policies influence outcomes primarily through changes in fertility rates, birth timing, and birth space^13,14. The decrease in fertility rates and postponement of motherhood contribute to women’s educational attainment and human capital accumulation, thereby enhancing female labor market outcomes^15,16. On the contrary, we show that the relaxation of the family planning policy affects women’s outcomes, not necessarily through fertility decisions. Thus, the effects apply to all working-age females, not restricted to those who change their fertility decisions due to the UTCP.

Third, we present evidence of a gender disparity in sleep patterns stemming from changes in the workplace environment in China. The shift in family planning policies from permitting to actively encouraging childbirth may lead to heightened social expectations regarding pregnancy likelihood, irrespective of women’s actual fertility intentions¹⁷. Changes in social expectations within the workplace may impact working conditions and environments, and force women to adjust their behaviors during working time. In contrast to previous research that discusses the gender gap in sleep duration resulting from childcare responsibilities^18,19, we demonstrate that the shift in social expectations about childbearing could be a potential mechanism driving differential changes in sleep patterns, particularly daytime napping behaviors.

This paper offers opinions for China and other nations that changes in family planning policies have extensive effects in terms of objects and scopes, and there is a need to take action to keep gender equality in the labor market while conducting these policies. Although the Law of the People’s Republic of China on the Protection of Women’s Rights and Interests (2022 Revision) explicitly states that an employer shall not determine recruitment (hiring) on marriage and childbearing restrictions, or marital and childbearing status, the government should still strengthen the implementation of the law and consider the broader challenges faced by the female population from social and workplace environment factors.

The paper is structured as follows: Sect. "Institutional background" provides a brief overview of the institutional background of family planning policy in China. Section "Theoretical framework" presents the theoretical framework. Section "Methods" outlines the data and methodology employed in the study. Section "Results" discusses the primary results and robustness checks, focusing on the impact of UTCP on napping and sleep behaviors. In Sect. "Mechanism", we delve into the mechanisms. Finally, Sect. "Conclusion" offers concluding remarks.

Institutional background

The inception of modern family planning policies in China can be traced back to the late 1970s. Policy in 1978 was initially structured by the principles of “Late, Sparse, and Few,” this framework gradually evolved into a comprehensive population policy by the end of 1980. This policy advocated for “late marriage,” “late childbearing,” “fewer children,” and “eugenics.” A significant milestone occurred in 1982 when the central government issued an official document endorsing the famous “One Child Policy” (OCP). The implementation of this policy was ratified in the Constitution, emphasizing the state’s promotion of family planning to align population growth with economic and social development plans. The key objective was to reduce the natural population growth rate to 10 per thousand by promoting the birth of only one child per couple. Violators of the OCP faced penalties, including fines and potential job loss, with additional repercussions such as issues concerning the legal status of “unregistered resident” if penalties were not paid. However, this rigorous family planning had exceptions. In China, there are 56 ethnic groups, among which 55 ethnic minority groups make up 9% of the total population, and the remaining 91% is the Han ethnic group. Ethnic minorities were generally allowed to have two or three children.

Despite its effectiveness in achieving its primary goals, the rigorous implementation of the family planning policy also led to various conflicts. Notably, the policy’s uniform application resulted in labor shortages in the agricultural sector and exerted pressure on social security systems. Responding to these challenges, the Central Government convened a family planning committee in 1984, emphasizing the need for flexibility in rural areas based on local conditions. This led to the introduction of the “One-and-a-half Child Policy,” allowing rural couples with a firstborn daughter to have an additional child.

Subsequent adjustments to the family planning policy were made to address demographic challenges. In 2002, the government introduced the “Selective Two-child Policy” (STCP), permitting some families to have two children under specific conditions. This policy was further relaxed in 2013, encouraging families where one member is an only child to have a second child.

A significant policy shift occurred in October 2015, when the Fifth Plenary Session of the 18th Communist Party of China (CPC) Central Committee announced the “Universal Two-child Policy,” marking the end of the OCP after 35 years. Subsequently, in 2021, the government introduced the “Three-child Policy” to promote balanced population development. Various supporting measures, such as tax deductions for parents with young children, were also implemented. For example, parents with children under the age of three are eligible for the 12,000 RMB (approximately 1700 US dollars) deduction of personal income tax per child per year, which increased to 24,000 RMB (approximately 3400 US dollars) in 2023.

However, despite these policy changes, fertility rates in China exhibited only temporary increases after 2013 and 2016, followed by a sustained decline. The natural growth rate of the population (0.34% in 2021) turned negative in 2022 (refer to Fig. 1), highlighting concerns about low fertility levels. The seventh national population census conducted in 2020 revealed that the fertility rate of women of childbearing age in China was 1.3, below the 1.5 threshold generally perceived as the risk of falling into the low fertility trap. Consequently, there is an expectation for increased pronatalist policies in the future.

Our study encourages policymakers to be concerned with the side effects and unintended consequences of the pronatalist policies beyond the scope of the traditional paradox of fertility and labor supply.

Theoretical framework

Sleep patterns, integral to health considerations^1,2, are also recognized for their potential impact on labor supply and productivity^18,20,21. However, distinctions between naps and nighttime sleep merit attention. While increased nighttime sleep duration is associated with modest declines in labor supply without discernible effects on cognitive functions or overall well-being, short afternoon naps are linked to reduced work time alongside significant enhancements in productivity, psychological well-being, and cognition³. Consequently, the reduction in napping, whether from extensive or intensive margins, is posited to negatively influence both health outcomes and productivity.

Beyond conventional metrics such as productivity and income, recent scholarship underscores the significance of non-wage facets in assessing labor market outcomes. These encompass dimensions such as work autonomy, job security, learning opportunities, career advancement prospects, social standing, societal contribution, and work-life balance²². We contend that napping behavior not only correlates with productivity but also serves as a barometer of workplace quality. In China, napping is entrenched as a cultural practice widely accepted across occupational strata, transcending socioeconomic divides. As seen in Table 1, approximately half of the individuals in our sample take naps. Naps are not exclusive to the white-collar class; blue-collar workers also take naps²³. Therefore, naps reflect job flexibility rather than socioeconomic status in China.

Table 1.

Descriptive data.

Variables	Male					Female
	Treatment		Control		t-test	Treatment		Control		t-test
	Mean	S.E.	Mean	S.E.		Mean	S.E.	Mean	S.E.
Nap=1	0.415	(0.013)	0.481	(0.020)	0.006	0.503	(0.013)	0.506	(0.017)	0.897
Napping time in minutes	28.989	(1.138)	36.143	(1.872)	0.001	35.685	(1.193)	36.317	(1.565)	0.748
Time to sleep	1380.01	(4.027)	1357.21	(5.544)	0.001	1361.04	(3.706)	1345.14	(4.746)	0.009
Sleep hours	7.930	(0.032)	7.878	(0.056)	0.389	8.124	(0.032)	8.018	(0.050)	0.062
Age	30.084	(0.160)	35.145	(0.183)	<0.001	30.052	(0.158)	34.442	(0.175)	<0.001
Education
Less than compulsory education	0.195	(0.010)	0.417	(0.020)	<0.001	0.212	(0.011)	0.499	(0.017)	<0.001
Compulsory education	0.337	(0.012)	0.395	(0.020)	0.012	0.319	(0.012)	0.394	(0.017)	<0.001
High school	0.230	(0.011)	0.137	(0.014)	<0.001	0.215	(0.011)	0.083	(0.009)	<0.001
College and above	0.239	(0.011)	0.051	(0.009)	<0.001	0.254	(0.011)	0.024	(0.005)	<0.001
Urban=1	0.573	(0.013)	0.361	(0.019)	<0.001	0.584	(0.013)	0.370	(0.016)	<0.001
Family income per person	17,518.8	(691.51)	8361.7	(320.94)	<0.001	16,090.9	(654.60)	8593.2	(289.34)	<0.001
Marital status
Single	0.292	(0.012)	0.002	(0.002)	<0.001	0.156	(0.009)	0.002	(0.002)	<0.001
Married	0.677	(0.012)	0.984	(0.005)	<0.001	0.822	(0.010)	0.990	(0.003)	<0.001
Cohabited	0.009	(0.003)	0.005	(0.003)	0.307	0.006	(0.001)	0.001	(0.001)	0.079
Divorced	0.022	(0.004)	0.008	(0.004)	0.024	0.013	(0.003)	0.005	(0.002)	0.053
Widowed	–	–	0.002	(0.002)	0.131	0.003	(0.001)	0.002	(0.002)	0.862
Employment status
Unemployed	0.017	(0.003)	0.010	(0.004)	0.210	0.023	(0.004)	0.010	(0.003)	0.030
Employed	0.948	(0.006)	0.946	(0.009)	0.866	0.753	(0.011)	0.771	(0.014)	0.317
Out of labor force	0.036	(0.005)	0.045	(0.008)	0.328	0.225	(0.011)	0.219	(0.014)	0.742
Working hours per week	45.318	(0.622)	44.533	(1.064)	0.504	33.590	(0.677)	31.702	(0.931)	0.097
Housework hours	0.984	(0.035)	1.461	(0.074)	<0.001	1.895	(0.040)	2.748	(0.063)	<0.001
Exercise hours per week	2.121	(0.146)	1.797	(0.255)	0.244	1.829	(0.130)	1.551	(0.177)	0.201
Commute time in minutes	18.548	(0.612)	10.927	0.693	<0.001	14.754	(0.521)	6.081	(0.495)	<0.001
TV hours per week	10.436	(0.024)	10.295	0.321	0.731	11.050	(0.260)	9.752	(0.303)	0.002
Exercise frequency	1.457	(0.061)	1.123	(0.088)	0.002	1.408	(0.064)	0.988	(0.076)	<0.001
Drink = 1	0.223	(0.011)	0.290	(0.018)	0.001	0.017	(0.003)	0.010	(0.003)	0.173
# of cigarettes smoked/day	7.403	(0.240)	9.717	(0.427)	<0.001	0.108	(0.030)	0.058	(0.029)	0.266
Poor health = 1	0.768	(0.011)	0.791	(0.016)	0.248	0.824	(0.010)	0.843	(0.012)	0.242
Poor memory = 1	0.626	(0.013)	0.689	(0.018)	0.006	0.670	(0.012)	0.766	(0.014)	<0.001
# of individuals	1433		628			1487		864

Family income per person is reported in Chinese Yuan and is inflation-adjusted to the 2010 base as provided by the survey. Time to sleep is calculated based on a 24-hour timing system that converts the expression of 9 pm to 21:00. If the individual went to sleep at 22:05, we set the value at $22\times 60+5$, and if the individual went to sleep after 24:00, for example, 1:05 am in the next day, we set the value at $(24+1)\times 60+5$. Therefore, higher values of this variable indicate the individual went to sleep at a later time. For time to sleep and sleep hours, we keep the full sample in this table.

However, the decision to nap and its duration remain personal choices influenced by social cognitive factors within the work environment²⁴. Previous studies suggest napping behaviors are shaped by workplace dynamics⁴ and peer influences⁵. To explain this phenomenon, we propose a simple economic model delineating the demand and supply of naps. We posit that individuals, positioned as consumers of naps, must give up working time to gain longer naps, given that napping typically occurs during working hours. The elasticity of nap demand becomes a relevant question, particularly in response to price changes resulting from policy interventions such as the UTCP.

A spectrum of health behaviors, spanning dietary choices, physical activity levels, tobacco and substance use, and adherence to medical guidelines, may be targeted by public policies to promote public health²⁴. Sleep patterns, both nighttime sleep and daytime naps, have emerged as indicators of health status^1,2. Altering sleep patterns presents challenges other than those of health behaviors such as smoking, as evidenced by social-cognitive theories²⁵. Thus, from an economic standpoint, we posit the demand for naps exhibits elasticities in response to price fluctuations, and different relative prices of naps between males and females result in differential responses to the policy change.

Next, the relative price of naps to work after the UTCP is reviewed. We contend that UTCP renders napping more costly for women, primarily due to discriminatory hiring practices¹⁷. In China, women enjoy more extensive legal parental leave entitlements compared to men, thereby imposing additional costs on employers. Moreover, societal norms presuppose parenthood as a normative progression following marriage, thereby disincentivizing the recruitment of married women without children. However, UTCP’s shift from permissive to incentivizing second births for all exacerbates this situation. For women who have one child, their current situation has turned out to be the same as that of married women without children under the OCP, and then employers may anticipate female employees having two children, irrespective of their actual fertility intentions¹⁷. Thus, married women with one child are encountering workplace discrimination. Women with no child are also affected since they are expected to give birth eventually. Further, women may encounter diminished promotional prospects.

As women face increasing challenges in the workplace, taking naps becomes more burdensome for them. Gender discrimination creates a less supportive work environment for women, leading to heightened pressure. Additionally, disparities in workplace treatment may compel women to exert greater effort to maintain their positions or advance in their careers. Consequently, women may feel compelled to reduce their nap time in order to dedicate more hours to work. Given these dynamics, women are more likely to respond to the policy change, fearing it may exacerbate gender disparities in the labor market¹⁷. We hypothesize that females reduce their naps at extensive and intensive margins to respond to the inequality in the workplace induced by the UTCP, and we investigate this hypothesis in the subsequent empirical sections.

Methods

Data and sample selection

We utilized data from the China Family Panel Studies spanning 2014 to 2018. The CFPS is a nationally representative longitudinal survey conducted biennially by the Institute of Social Science Survey at Peking University in China. It captures information on individuals, families, and communities across contemporary China, covering diverse topics such as economic activities, educational outcomes, family dynamics, migration patterns, and health indicators. Besides, since the CFPS tracks individuals for multiple waves, it allows us to include the individual fixed effect to control for unobserved heterogeneity among individuals.

Although CFPS began in 2010, our analysis excludes data in 2010 and 2012 for two reasons. First, there is limited information on the number of children in 2012. Time to sleep and daily sleep hours are not recorded in these two years, thus only the effects on napping behaviors could be examined. Second, by doing so, we avoid the potential confounding effects resulting from the STCP implemented in 2013. For the Han ethnic group, whether they were allowed to have two or more children before the UTCP depends on the couples’ sibling status. The change made in 2013 is that as long as one side, instead of both sides, of the couple is the only child of their parents, the couple is eligible to have a second child. Unfortunately, we do not have proper data to identify whether the couple is eligible for the STCP, making it difficult to identify the policy effect of STCP. Therefore, we exclude the data before 2013 to ease the threat of confounding effects. However, data from 2010 and 2012 are included when we conduct the parallel trend test in Sect. "Validity of DD specification".

Additionally, we eliminate the most recent 2020 data due to the disruptive influence of the COVID-19 pandemic on sleep patterns, attributable to shifts in employment statuses and changes in time allocation, particularly with the proliferation of remote work arrangements. For instance, remote work may alter parental caregiving dynamics and individual engagement in physical activities^26–28. Instead, we conduct a robustness check by adding 2020 data in Sect. “Extend the study period”.

Further, our analysis excludes minority groups, as they were subject to a more lenient policy program. Moreover, Han ethnicity and minority groups exhibit divergent patterns across various domains, including lifestyle, cultural practices, educational attainment, decision-making processes, and government subsidies, and are eligible for different educational, political, and economic policies.

Our study focuses specifically on individuals within the childbearing age range of 18 to 45, as they represent the demographic most directly impacted by the implementation of UTCP.

Outcome variables

Our study focuses on a comprehensive set of sleep behavior outcomes, which encompass daytime napping, total daily sleep duration, and the timing of sleep onset. To assess daytime napping, respondents were queried regarding their habitual engagement in afternoon naps and the corresponding duration in minutes. Instances where individuals did not partake in napping were recorded as a duration of 0 minutes. Subsequently, our analysis was refined to include only those individuals who reported engaging in napping for analysis at the intensive margin.

Respondents were prompted to provide information on their sleep duration, stratified by weekdays and weekends. For individuals not engaged in employment, a general inquiry was made into their daily sleep duration, while those employed were asked to specify their sleep duration on both weekdays and weekends. To derive daily sleep duration, we computed the average of weekday and weekend sleep hours for the employed. Extreme outliers, defined as those sleeping less than 2.55 hours or more than 15 hours, were excluded from the analysis to mitigate the influence of aberrant cases, which accounted for approximately 0.05% of observations.

Respondents were also queried on their usual bedtime, aiding in delineating the timing of night sleep onset. To ensure data integrity, samples were restricted to individuals whose reported bedtime fell within the range of 9 pm to 4 am, thereby excluding extreme cases that accounted for approximately 17% of observations.

We used a 24-hour based timing system that converts the expression of 9 pm to 21:00. If the individual went to sleep at 22:05, we set the value at $22\times 60+5$, and if the individual went to sleep after 24:00, for example, 1:05 am in the next day, we set the value at $(24+1)\ast 60+5$. Therefore, higher values of this variable indicate the individual went to sleep at a later time.

Table 1 shows the descriptive statistics of control and treatment groups for males and females separately. We observe females sleep more than males during daytime naptime and total sleep hours, consistent with previous findings²⁹. Also, females sleep earlier than males at night. Individuals in the treatment group tend to be younger, more highly educated, single, living in urban areas, and have higher family incomes. Napping behaviors do not vary across employment status, which is consistent with our assumption that napping is a culture-shaped behavior in China.

Treatment identification

The section presents the treatment identification. Under the UTCP, each couple is allowed to have two children. The fertility decisions of couples with two or more children and their situations in the workplace are not affected by the UTCP, and thus their related outcomes should not be affected either. Accordingly, we define those who have two or more children as the control group using the number of children in 2014. If individuals had two or more children in 2014, they would not have been affected by the UTCP.

If individuals had only one child in 2014, they were affected by the UTCP if they were not eligible to have a second child. The individuals of childbearing age in 2014 were born in the years that OCP was executed rigorously. Although we are not able to distinguish whether an individual is really treated, individuals having one child are most likely to be affected by the UTCP.

For those who do not have children, they are still affected. For example, they may change their fertility plans after UTCP. If they prefer to have two children, they may plan earlier parenthood when they are allowed to. The change in the fertility plan may change their labor supply decisions. Most important of all, we argue that the effect of UTCP is not necessarily through giving birth and maybe through unfriendly working environments; females with no child are also expected to have a child sooner or have multiple pregnancies.

In addition, to ensure the validity of the identification strategy, we use the number of children before UTCP to define the treatment group to avoid endogeneity. That is to say, the number of children may change after UTCP. We deal with this issue in the robustness check Sect. “Sample selection”.

Specifications

A difference-in-difference (DD) model is specified in Eq. (1).

$$Y_{\mathit{it}}={\alpha }_i+{\beta }_1{\mathit{treatment}}_i\ast {Year16}_t+{\beta }_2{\mathit{treatment}}_i\ast {Year18}_t+{\gamma }_t+{{\mathbf{X}}^{\prime }}_{\mathbf{it}}\mathrm{\Delta }+{\epsilon }_{\mathit{it}}.$$ 1

$Y_{\mathit{it}}$ denotes the outcome variables, including sleep behaviors, time usage, other health behaviors, and health status. ${\mathit{treatment}}_i$ is defined as one if an individual $i$ has no children or one child in 2014 and is set to be zero otherwise. We use the number of children prior to UTCP to define the treatment group to avoid endogeneity. Regarding the concern that some individuals may have had additional children between 2014 and 2018, we conduct a robustness check in Sect. “Sample selection” by excluding those who gave birth after UTCP. ${Year16}_t$ takes the value of one in the year 2016 and zero in the years 2014 and 2018. ${Year18}_t$ takes the value of one in the year 2018 and zero in the years 2014 and 2016. ${{\mathbf{X}}^{\prime }}_{\mathbf{it}}$ is a vector of individual characteristics, including age, education, marital status, family income per person (inflation-adjusted to the year 2010), whether living in urban areas, and provincial locations. ${\alpha }_i$ and ${\gamma }_t$ denote individual and year fixed effects, respectively. An assumption of DD specification is that the treatment group and control group had the same trend before the intervention, and the two groups would still be parallel in the absence of the policy change. Since we only have one round of data before the policy implementation, a permutation test is employed as a robustness check in Sect. "Validity of DD specification". To ease the interpretation of the regression results, we use the linear probability model with fixed effects for the binary outcome variables to estimate the policy effects. Our main interest is the estimation of ${\beta }_1$ and ${\beta }_2$ indicating the family planning policy’s effect on the outcome variables. We conduct the sub-sample analysis of females and males because we expect a greater impact of the family planning policies amongst females, as discussed in Sect. "Theoretical framework".

Results

Napping and sleep behaviors

Table 2 shows the baseline estimations of the dynamic effects of UTCP. The left panels show that the UTCP has little effect on sleep behaviors except for the time to sleep for males. The right panel reports the estimates for the females that the probability of having a nap was reduced by 6.1% in 2016 (column 6), and the length of napping time was reduced by 5.687 minutes (column 7). Thus, we find a gender gap in adjusting the sleep behaviors to the family planning policies. However, when we focus on those who take a nap, the negative effect on napping length is only significant in 2018 (column 8). Based on these regression results, we infer that among those who usually take naps and still choose to take naps, the length of napping time is not affected in the short term, indicating their demand for naps is inelastic in the short term, but they may still reduce the length of napping time in the long term.

Table 2.

Baseline estimations.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Treatment*year16	− 0.013	1.175	0.410	5.170*	− 0.068	− 0.061**	− 5.687***	− 4.413	− 3.507	− 0.123*
	(0.027)	(2.338)	(3.397)	(2.983)	(0.071)	(0.024)	(2.167)	(2.944)	(2.532)	(0.066)
Treatment*year18	0.005	0.838	0.965	6.476**	− 0.089	− 0.028	− 3.858*	− 4.817*	− 4.112	− 0.003
	(0.027)	(2.395)	(3.414)	(3.087)	(0.073)	(0.026)	(2.221)	(2.915)	(2.597)	(0.068)
Age	− 0.039*	− 3.069*	− 0.234	− 6.995**	− 1.491	− 0.134***	− 9.224***	− 4.382*	0.177***	0.142*
	(0.024)	(1.848)	(3.151)	(2.717)	(2.993)	(0.032)	(2.305)	(2.660)	(0.059)	(0.074)
Education
Less than compulsory education	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Compulsory education	0.076*	9.974***	12.653**	2.371	0.840	0.006	0.678	3.035	− 0.205**	− 0.027
	(0.041)	(3.833)	(5.012)	(4.647)	(3.749)	(0.036)	(3.125)	(4.351)	(0.098)	(0.103)
High school	0.080	10.577*	15.854**	7.000	6.401	− 0.009	3.934	16.713**	− 0.044	0.050
	(0.062)	(5.583)	(7.443)	(6.766)	(6.030)	(0.061)	(5.312)	(7.124)	(0.140)	(0.166)
College and above	0.164**	16.870***	15.464	10.239	5.631	0.020	6.029	15.593*	− 0.015	0.129
	(0.074)	(6.246)	(9.586)	(7.913)	(7.638)	(0.079)	(6.465)	(8.172)	(0.167)	(0.195)
Marital status
Single	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Married	− 0.072*	− 5.437	− 1.929	− 5.217	− 10.309**	0.020	2.674	3.173	− 0.093	− 0.033
	(0.039)	(3.358)	(5.847)	(4.542)	(4.490)	(0.047)	(3.743)	(4.779)	(0.111)	(0.104)
Cohabited	− 0.004	0.759	− 25.883*	− 1.547	− 14.802	0.133	16.758**	30.136**	− 0.073	0.440
	(0.108)	(7.795)	(13.719)	(10.197)	(9.290)	(0.099)	(8.132)	(12.505)	(0.275)	(0.302)
Divorced	− 0.185***	− 9.652*	− 5.855	− 11.169	− 8.103	− 0.103	− 9.884	4.896	− 0.070	− 0.198
	(0.068)	(5.466)	(8.362)	(7.334)	(7.766)	(0.108)	(7.247)	(9.082)	(0.184)	(0.253)
Widowed	0.093	10.302	7.638	123.218***	9.340	0.051	3.695	2.094	− 0.744	− 0.185
	(0.380)	(23.143)	(6.671)	(14.512)	(16.314)	(0.193)	(11.566)	(12.416)	(1.095)	(0.248)
Log family income	− 0.006	− 0.307	− 0.513	− 0.223	1.074*	0.003	0.262	0.430	− 0.019	0.032**
	(0.006)	(0.511)	(0.790)	(0.650)	(0.548)	(0.006)	(0.452)	(0.694)	(0.015)	(0.015)
Urban=1	− 0.014	1.287	5.513	3.280	5.143	0.046	0.674	− 6.830*	0.050	0.014
	(0.031)	(2.657)	(5.273)	(3.822)	(3.569)	(0.036)	(2.712)	(3.868)	(0.092)	(0.082)
Employment status
Unemployed	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Employed	− 0.003	− 4.280	4.026	2.890	10.824*	− 0.032	− 2.442	2.025	− 0.120	0.090
	(0.058)	(4.662)	(8.653)	(6.393)	(5.963)	(0.048)	(4.321)	(5.740)	(0.177)	(0.134)
Out of labor force	− 0.030	− 6.897	3.435	− 3.977	9.406	0.046	4.917	7.546	− 0.234	0.166
	(0.068)	(5.626)	(11.148)	(7.829)	(6.094)	(0.049)	(4.409)	(5.846)	(0.205)	(0.137)
# of individuals	2061	2061	1405	1981	2061	2351	2351	1763	2257	2351
# of observations	6183	6183	2810	5173	6167	7053	7053	3622	5742	7038

This table is constructed based on Eq. (1). Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. We also do the regressions without deleting the extreme values for both the sleep hours and the time to go to sleep. The results are not significant, indicating that the policy affects those who have ordinary sleep behaviors and does not change the sleep behaviors of those who have unusual sleep patterns. All regressions include individual fixed effect, province fixed effect, and the year fixed effect. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

The negative impact on the likelihood of taking a nap suggests that some people switch from taking naps to not taking naps. One possible explanation is that some females adjust their sleep behaviors to overcome the implicit stress in the work environment after UTCP. A less-equitable working environment may not only make females work harder but also push them to make their efforts “visible.” Females have a higher probability of incurring sleep problems than males³⁰. Most of the gender gap in sleep behaviors could be explained by work and family responsibilities and varies across life course stages²⁹. We provide evidence that gender differences in sleep behaviors could be affected by fertility policies. Nevertheless, it does not mean that females reduce their sleep time; they may prolong their nighttime sleep to substitute for their naps. To examine this statement, we estimate the policy effects on total sleep hours and time to sleep. While we see a decline in total sleep hours of 12.3% (column 10) in females, we do not find a significant effect on males. However, males tend to sleep later, and their time to sleep is postponed by 5.17 minutes (column 4). Combined with the results of time allocation shown in Table 13, we infer that males go to sleep later due to doing more housework. Besides, males tend to have shorter sleep latency³¹, so they may stick less to the time to go to bed.

Table 13.

Mechanism-time usage, other health behaviors, and health outcomes.

Panel A: time allocations	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Working hours	Commute time (in minutes)	Housework hours	Exercises hours	TV hours	Working hours	Commute time (in minutes)	Housework hours	Exercise hours	TV hours
Treatment*year16	1.910	− 1.148	0.197**	− 0.153	− 0.065	0.139	− 0.313	0.109	0.018	− 1.127**
	(1.693)	(1.159)	(0.089)	(0.471)	(0.512)	(1.464)	(0.595)	(0.083)	(0.316)	(0.474)
Treatment*year18	− 0.721	1.156	0.146	− 0.367	0.612	− 2.507**	0.150	0.151*	− 0.857**	0.011
	(1.419)	(1.041)	(0.092)	(0.505)	(0.499)	(1.270)	(0.746)	(0.083)	(0.388)	(0.491)
# of individuals	2061	2061	2061	2061	2061	2351	2351	2351	2351	2351
# of observations	6183	6183	6183	6183	6183	7053	7053	7053	7053	7053

Panel B: other health behaviors and health status	Exercise frequency	Drinking alcohol =1	# of cigarettes smoked/day	Poor health=1	Poor memory =1	Exercise frequency	Drinking alcohol =1	# of cigarettes smoked/day	Poor health=1	Poor memory=1
Treatment*year16	0.104	− 0.042*	0.179	0.016	0.003	− 0.109	− 0.006	− 0.007	0.029	0.014
	(0.138)	(0.023)	(0.391)	(0.021)	(0.026)	(0.130)	(0.006)	(0.042)	(0.019)	(0.022)
Treatment*year18	0.019	− 0.007	0.833**	0.039*	− 0.008	− 0.320**	− 0.001	− 0.076	0.039**	0.069***
	(0.163)	(0.024)	(0.384)	(0.023)	(0.026)	(0.138)	(0.008)	(0.054)	(0.019)	(0.022)
# of individuals	2061	2061	2061	2061	2061	2351	2351	2351	2351	2351
# of observations	6183	6183	6183	6183	6183	7053	7053	7053	7053	7053

This table is constructed based on Eq. (1) with different outcome variables. In Panel A, columns 1 and 6 show the policy effect on working hours for males and females, respectively. Columns 2 and 7 show the policy effect on commute time between home and workplace in minutes for males and females, respectively. Columns 3 and 8 show the policy effect on housework hours for males and females, respectively. Columns 4 and 9 show the policy effect on exercise hours for males and females, respectively. Columns 5 and 10 show the policy effect on TV hours for males and females, respectively. In Panel B, columns 1 and 6 show the policy effect on exercise frequency for males and females, respectively. Columns 2 and 7 show the policy effect on the probability of drinking alcohol for males and females, respectively. Columns 3 and 8 show the policy effect on the number of cigarettes smoked per day for males and females, respectively. Columns 4 and 9 show the policy effect on the probability of having poor health conditions for males and females, respectively. Columns 5 and 10 show the policy effect on the probability of having a poor memory for males and females, respectively. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Next, we examine the effects observed in 2018. The impact of UTCP on the likelihood of taking naps is statistically insignificant. However, the policy does have a negative effect on the duration of naps among all females and at the intensive margin. These findings suggest that the UTCP prompts females who typically nap to stop doing so, and those who continue to nap experience a reduction in nap duration. Males, on the other hand, adjust their sleep patterns primarily by delaying their bedtime.

In summary, while both males and females respond differently to the relaxation of family planning policies, females experience more substantial negative consequences in terms of their sleep behaviors.

Robustness checks

Validity of DD specification

Permutation test

The DD method relies on the parallel trend assumption, which requires that the average outcomes for treated and controls would have followed parallel paths over time in absence of the treatment. However, there is only one period of data before the policy, so we are not able to conduct the parallel trend test. We perform the permutation test, which looks at how the policy change affects a post-treatment variable known to be unaffected by the policy change³².

Permutation tests are based on the concept of scrambling – that is, permuting – the order of a variable in all possible ways to determine whether the estimated results are statistically significant or just because of random chance, particularly if the sample size is small. Additionally, the permutation test makes no presumptions on the underlying distribution of the data. In the permutation test, we test the null hypothesis that there is no effect of the UTCP on the odds of outcome variables³³. Under the null hypothesis, the estimated coefficient from the actual data can be considered a random sample from the permutation distribution. We can produce the permutation distribution of the estimated coefficients and use it for statistical inference. Specifically, we have a balanced dataset of 4412 individuals each year, and 2920 of them are in the treatment group. Thus, we randomly assign the 2920 treatment indicators to all individuals and construct placebo treatment status for the same sample of individuals as in the baseline analysis for males and females. We plot the cumulative distribution function of the distribution of placebo treatment effects from 1000 random assignments in Fig. 2 for males and females in Panel A and B, respectively. The dashed vertical line shows the estimated treatment effect from the baseline analysis. The p-value of each permutation placebo test is the proportion of placebo estimates equal to or larger in absolute value than the corresponding estimate from the baseline analysis³³. We have five outcome variables for each gender group for two years. For the significant policy effects from the baseline analysis, the p-values in Fig. 2 should be close to zero to reject the null hypothesis of no policy effect. The test results shown in Fig. 2 are consistent with the baseline results, providing further support for our identification strategy.

Figure 2.

Permutation test. We plot the cumulative distribution function of the distribution of placebo treatment effects from 1000 random assignments for males and females in Panel (A) and (B), respectively. The dashed vertical line shows the estimated treatment effect from the baseline analysis. The p-value of each permutation placebo test is the proportion of placebo estimates equal to or larger in absolute value than the corresponding estimate from the baseline analysis. For the significant policy effects from the baseline analysis, the p-values should be close to zero to reject the null hypothesis of no policy effect.

Parallel trend

Even though we conduct the permutation test, the parallel trend is still a concern. Thus, we use those observed from 2010 to 2018 to conduct the parallel trend analysis. In 2012, the number of children was not asked, and we use the status of 2010 and 2014 to determine the status. For example, in 2010 and 2014, the person did not have a child, and in 2012, he/she had no child. Consider another case, the person had a child in 2010, and he/she had two children in 2014, then the number of children in 2012 is not clear. If the 2012 status is hard to determine, we set the 2012 treatment indicator as missing. The sample size is relatively small, and the STCP in 2013 has a profound effect. Thus, we provide it here as a robustness check. Figure 3 shows the parallel trend for females and males. We find that the parallel trend assumption is satisfied.

Figure 3.

Parallel trend. The parallel trend test needs to be carried out to verify the causal effect. We utilize those who were observed from 2010 to 2018 to conduct the parallel trend analysis. In 2012, the number of children was not asked, and we use the status of 2010 and 2014 to determine the status. If the 2012 status is hard to determine, we set the 2012 treatment indicator as missing. Figure 3 shows that the parallel trend assumption is satisfied for both males and females.

Control variables

First, we examine the robustness of the baseline results without control variables to eliminate the perfect multicollinearity between covariates (see Table 3 Panel A). Control variables in the baseline regressions, such as education and provincial locations, are time-variant but generally do not change substantially over time. Potential collinearity could emerge as we attempt to assess the impact of variations in the explanatory variables on the outcome variable. If an explanatory variable does show limited variation in our dataset, it will be difficult to estimate its effect³⁴. Even though we do not include any control variables in Table 3 Panel A, time-invariant controls are captured by the individual fixed effect. We find that the coefficients do not change a lot compared to those in Table 2. Second, there may be some omitted variables that cannot be observed. For example, different areas may have different prevalence rates of napping behaviors due to local culture, and it may change with time, but we do not have relevant information. We incorporate the province-year cross fixed effect to seize the temporal shocks and variables we cannot observe within each province and year. We find similar results in Table 3 Panel B, indicating that the policy effect still exists after eliminating the macroeconomic disturbances.

Table 3.

Robustness checks-control variables.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: no control variables
Treatment*year16	− 0.016	1.059	0.631	4.471	− 0.077	− 0.063***	− 5.862***	− 4.042	− 3.874	− 0.116*
	(0.027)	(2.333)	(3.362)	(2.950)	(0.071)	(0.024)	(2.140)	(2.911)	(2.506)	(0.065)
Treatment*year18	− 0.005	0.125	0.882	5.658*	− 0.109	− 0.020	− 3.146	− 4.159	− 4.802*	− 0.002
	(0.027)	(2.371)	(3.397)	(3.066)	(0.071)	(0.025)	(2.179)	(2.833)	(2.539)	(0.066)
# of individuals	2061	2061	1405	1981	2061	2351	2351	1763	2257	2351
# of observations	6183	6183	2810	5173	6167	7053	7053	3622	5742	7038
Panel B: include province-year fixed effect
Treatment*year16	− 0.012	0.790	1.039	6.183*	− 0.063	− 0.052**	− 5.722**	− 5.325*	− 2.796	− 0.149**
	(0.028)	(2.464)	(3.637)	(3.161)	(0.074)	(0.026)	(2.326)	(3.086)	(2.705)	(0.072)
Treatment*year18	0.013	1.405	2.407	7.629**	− 0.091	− 0.016	− 3.502	− 5.722*	− 2.877	− 0.044
	(0.029)	(2.529)	(3.510)	(3.245)	(0.076)	(0.027)	(2.363)	(3.001)	(2.800)	(0.071)
# of individuals	2061	2061	1405	1981	2061	2351	2351	1763	2257	2351
# of observations	6183	6183	2810	5173	6167	7053	7053	3622	5742	7038

Panel A is constructed based on Eq. (1) but without control variables. In Panel A, all regressions include individual fixed effect, province fixed effect, and the year fixed effect. Panel B is constructed based on Eq. (1), including controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Inverse probability of treatment weighting estimation

The parallel trend assumption of DD method may be implausible if pre-treatment characteristics supposed to be associated with the dynamics of the outcome variable are unbalanced between the treated and the untreated group³⁵. To reduce the effects of confounding arising from the characteristics of the subjects that frequently occur in treatment selection, statistical methods based on propensity scores (PS) are commonly adopted, such as covariate adjustment, stratification or subclassification, matching, and inverse probability of treatment weighting (IPTW)^36,37. The PS is defined as a subject’s probability of treatment assignment, conditional on observed baseline covariates³⁶. To keep all of the sample units in the analysis, we use the IPTW method to make the adjustment before the DD estimation. A synthetic sample is created by assigning weights to subjects based on the inverse probability of the treatment received, and then the treatment assignment is not influenced by the measured baseline covariates³⁶. Those less likely to be treated are given more weight to rebalance the data. Then, with the weight adjustment, the policy effect is measured as its average treatment effects, which shows the average difference in potential outcomes between treated and control groups across the sample.

We carry out the IPTW method as follows. First, we use the logit model to get the PS, which is the probability of being treated (i.e. had no child or only one child in 2014) conditional on personal characteristics, such as gender, education, family income, living in urban or rural areas, and employment status. Then, we assign the weights of 1/PS to the treatment group and 1/(1–PS) to the control group to balance the two groups. After assigning different weights to treatment and control groups, we check the balance of PS and plot the estimated densities of the probability in Fig. 4, which can be used to check whether the overlap assumption is violated. We also conduct the overidentification test, and we cannot reject the null (with Prob > chi2 = 0.7010) that covariates are balanced. Finally, we perform the DD estimations with the propensity score weights. The results are presented in Table 4, which are consistent with those in Table 2.

Figure 4.

Density plot of the propensity scores. To ease the confounding issue and improve the satisfaction of the parallel trend assumption, we employ the inverse probability of weighting method to make the adjustment before the DD estimation. After estimating the probability of being treated in the first step, we assign different weights to the treatment group and the control group to balance the two groups. Figure 4 plots the estimated densities of the probability being treated, which can be used to check whether the overlap assumption is violated.

Table 4.

PSW-DD estimations.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Treatment*year16	− 0.010	1.988	0.216	4.267	0.037	− 0.061**	− 6.415***	− 4.416	− 4.263	− 0.139**
	(0.031)	(2.552)	(3.633)	(3.169)	(0.077)	(0.028)	(2.239)	(3.225)	(2.968)	(0.070)
Treatment*year18	0.013	1.903	− 0.300	3.370	− 0.036	− 0.034	− 4.828**	− 5.724*	− 5.573**	− 0.042
	(0.031)	(2.594)	(3.544)	(4.114)	(0.085)	(0.031)	(2.317)	(2.990)	(2.834)	(0.076)
# of individuals	2061	2061	1405	1981	2061	2351	2351	1763	2257	2351
# of observations	6183	6183	2810	5173	6167	7053	7053	3622	5742	7038

This table is constructed from the propensity score weighting and DD specification. First, we use a logit model to calculate the probability of being treated based on control variables, such as gender, education, family income, living in urban or rural areas, and employment status. The probability is the propensity score, recorded as PS. Second, we assign the weights of 1/PS to the treated and 1/(1-PS) to the control units and then run the regressions using Eq. (1), including individual fixed effect, province fixed effect, and the year fixed effect. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Sample selections

Another assumption for the DD approach to be valid is that the composition of the two groups must remain constant over the period. In the primary analysis, we define the treatment and control groups by the number of children in 2014. Because individuals may have more children between 2014 and 2018, we drop out those who did give birth after UTCP to eliminate the confounding effect. The results are consistent with our main results, as shown in Table 5 Panel A.

Table 5.

Robustness checks-sample selections.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: eliminate those who gave birth in 2016 and 2018
Treatment*year16	− 0.013	1.104	0.057	5.184*	− 0.068	− 0.062**	− 5.749***	− 4.355	− 3.345	− 0.124*
	(0.027)	(2.356)	(3.436)	(2.967)	(0.072)	(0.024)	(2.190)	(2.962)	(2.556)	(0.067)
Treatment*year18	0.001	0.389	0.490	6.444**	− 0.087	− 0.027	− 3.880*	− 4.831*	− 3.665	− 0.007
	(0.027)	(2.406)	(3.415)	(3.096)	(0.073)	(0.026)	(2.240)	(2.929)	(2.597)	(0.068)
# of individuals	2050	2050	1396	1971	2050	2336	2336	1754	2242	2336
# of observations	6150	6150	2791	5147	6134	7008	7008	3606	5705	6993
Panel B: consider living in a community that was allowed to have a second child before the UTCP
Treatment*year16	0.007	3.216	0.660	4.108	− 0.029	− 0.071**	− 6.104**	− 4.338	− 5.422*	− 0.104
	(0.031)	(2.767)	(4.031)	(3.593)	(0.083)	(0.029)	(2.622)	(3.556)	(3.069)	(0.078)
Treatment*year18	− 0.005	0.521	2.263	7.499**	− 0.129	− 0.029	− 3.001	− 5.067	− 4.052	0.021
	(0.033)	(2.884)	(4.033)	(3.769)	(0.085)	(0.031)	(2.662)	(3.418)	(3.166)	(0.080)
# of individuals	1193	1,193	816	1142	1193	1474	1474	1118	1408	1474
# of observations	3579	3579	1662	2964	3557	4422	4422	2342	3534	4398
Panel C: draw back the minorities
Treatment*year16	− 0.003	0.873	− 0.658	3.639	− 0.024	− 0.054**	− 5.084**	− 4.340	− 2.859	− 0.094
	(0.024)	(2.094)	(3.220)	(2.738)	(0.064)	(0.022)	(2.028)	(2.866)	(2.346)	(0.059)
Treatment*year18	0.007	− 0.025	0.566	3.884	− 0.102	− 0.033	− 4.499**	− 4.642	− 3.710	0.042
	(0.025)	(2.152)	(3.235)	(2.893)	(0.065)	(0.024)	(2.097)	(2.860)	(2.425)	(0.060)
# of individuals	2243	2243	1503	2157	2243	2565	2565	1897	2455	2565
# of observations	6729	6729	2988	5637	6697	7695	7695	3870	6250	7653

This table is constructed based on Eq. (1) with different samples. Panel A eliminates those who gave birth in 2016 and 2018 from our baseline samples. Panel B considers living in a community that was allowed to have a second child before the UTCP. The observations who had no child or only one child but lived in a community that was allowed to have a second child were dropped, and the treatment group is defined as those who had no child or only one child and lived in a community that was not allowed to have a second child before the UTCP. Those who have 2 or more children are still in the control group. Panel C draws back the minorities into the original samples. All minorities are in the control group, no matter how many children they have. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Table 5 Panel B shows another adjustment for the sample selection. Despite the strict implementation of the OCP, certain groups in rural areas were allowed to have more than one child prior to UTCP, as discussed in Sect. "Institutional background". The 2014 CFPS community survey is helpful in identifying special conditions for allowing a second child, from the survey question: “Currently, couples in your village are allowed to have two children under which of the following circumstances?” The potential answers could be:

1. The husband and wife, or one of them is an ethnic minority.

2. The first child is a girl.

3. Both of the husband and wife are the only child of their original family.

4. Either husband or wife is the only child of their original family.

5. The husband and wife, or one of them engaged in a special occupation.

6. The first child has disabilities.

7. Special circumstances.

8. Having two children is not allowed.

We do not track who gave birth from the survey question, but we see how the family planning policy was carried out locally. If a couple lives in a community where the second child was permitted previously, the UTCP has little effect on them. Thus, we grouped “not applicable,” “not allowed,” and “both spouses or one of them is an ethnic minority,” which are regarded as “the living environment is not allowed to have a second child before the policy takes place.”

We treat the answer of “both spouses or one of them is an ethnic minority” into the “not allowed” category because minorities are an implicit exception under the OCP, and the answer reveals that the second child for the Han ethnic group was not allowed prior to UTCP.

We re-define the treatment group as those who had no child or only one child and lived in a community that was not allowed to have a second child before the UTCP. Those who have two or more children are still in the control group. We drop the observations of those who had no child or only one child but lived in a community that was allowed to have a second child to gain more clear treatment effects. We obtain the consistent results in Table 5 Panel B.

Table 5 Panel C draws back the minorities into the sample. So, the control group is extended to those of the Han ethnicity group with two or more children and those of the minorities, no matter how many children they have. The results in Table 5 Panel C are similar to those in the baseline analysis in Table 2.

Placebo test

First, as discussed in Sect. “Sample selection”, if a couple resides in a community where a second child was previously permitted, the UTCP has minimal impact on them because their social environment remains largely unchanged from the period before UTCP. Thus, we estimate the baseline specification based on the samples who lived in areas that were allowed to have a second child before the UTCP. Results are shown in Table 6 Panel A, and there is no significant effect of the UTCP for the placebo sample.

Table 6.

Placebo test-alternative samples.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: those living in a community that was allowed to have a second child before the UTCP
Treatment*year16	0.064	4.927	− 3.755	6.761	− 0.128	− 0.068	− 6.187	− 4.458	− 3.870	− 0.032
	(0.043)	(3.952)	(5.671)	(5.019)	(0.122)	(0.041)	(3.910)	(5.000)	(4.070)	(0.102)
Treatment*year18	0.036	0.491	− 6.706	8.182	− 0.083	− 0.054	− 4.726	− 0.531	− 3.416	− 0.074
	(0.044)	(3.982)	(5.635)	(5.191)	(0.123)	(0.044)	(3.941)	(4.850)	(4.506)	(0.101)
# of individuals	811	811	563	772	811	942	942	743	897	1181
# of observations	2433	2433	1145	1996	2427	2826	2826	1527	2248	3537
Panel B: for age 60 and older
Treatment*year16	− 0.009	− 1.800	− 2.006	− 5.183	− 0.138	− 0.013	− 2.720	− 1.252	− 1.000	0.044
	(0.028)	(2.672)	(2.894)	(3.200)	(0.091)	(0.035)	(2.815)	(4.199)	(3.663)	(0.112)
Treatment*year18	− 0.001	− 0.703	− 1.200	− 7.061**	− 0.177**	− 0.033	− 2.196	2.827	2.003	− 0.031
	(0.029)	(2.828)	(2.919)	(3.319)	(0.088)	(0.035)	(2.987)	(4.113)	(4.219)	(0.111)
# of individuals	1777	1777	1486	1368	1777	1470	1470	1179	1120	1468
# of observations	5331	5331	3535	2980	5299	4410	4410	2677	2458	4371

This table is constructed based on Eq. (1) with different samples. Panel A is based on those living in a community that was allowed to have a second child before the UTCP. Panel B is based on those who are age 60 and older. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. Standard errors are clustered at individual level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Second, the UTCP is designed to impact individuals of childbearing age. We posit that the UTCP influences the working environment, subsequently affecting women’s health behaviors. Therefore, it should not impact individuals who are retired and beyond childbearing age. To verify this, we conduct a placebo test using samples of individuals aged 60 and older. As shown in Table 6 Panel B, the UTCP does not significantly affect this older population, supporting our analysis.

Third, we conduct the other test by randomly dropping 10% of all observations in the sample systematically and perform Eq. (1) 1000 times in order to rule out the possibility that certain control units drive effects of interest. Table 7 shows the results from 1000 sub-samples. We test the equality of the mean of the 1000 coefficients and the baseline coefficients in Table 2. We find that all p-values exceed 0.1, indicating that the null hypothesis that the means of the 1000 coefficients and the baseline coefficients in Table 2 are equal cannot be rejected. In addition, we report the 95% confidence interval (CI) of all treatment effects in Table 2 and find that all 1000 coefficients lie in the CI, indicating that each of 1000 sub-samples can represent the total sample. In other words, the results of the sub-samples do not deviate from the baseline sample, which suggests that our primary results are not driven by certain control units.

Table 7.

Placebo test-randomly drop observations.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Drop 10% of all units
Treatment*year16
Average coefficients	− 0.013	1.200	0.400	5.188	− 0.069	− 0.061	− 5.679	− 4.370	− 3.473	− 0.122
Equality test	0.492	0.319	0.794	0.556	0.178	0.209	0.733	0.180	0.195	0.262
Min	− 0.043	− 1.437	− 3.818	1.882	− 0.143	− 0.087	− 8.227	− 7.786	− 7.029	− 0.193
Max	0.014	3.634	3.891	8.218	0.008	− 0.037	− 3.176	− 1.231	− 0.927	− 0.047
95% CI of baseline coefficients
Lower	− 0.066	− 3.410	− 6.253	− 0.680	− 0.208	− 0.108	− 9.936	− 10.188	− 8.472	− 0.253
Upper	0.039	5.761	7.074	11.020	0.072	0.013	− 1.437	1.362	1.459	0.007
Treatment*year18
Average coefficients	0.005	0.856	0.985	6.480	− 0.087	− 0.028	− 3.879	− 4.790	− 4.143	− 0.002
Equality test	0.751	0.465	0.586	0.910	0.122	0.713	0.371	0.386	0.239	0.177
Min	− 0.023	− 1.873	− 2.864	3.030	− 0.158	− 0.055	− 6.061	− 7.710	− 7.417	− 0.087
Max	0.028	3.575	4.739	9.316	− 0.010	− 0.0002	− 1.583	− 1.331	− 1.005	0.077
95% CI of baseline coefficients
Lower	− 0.048	− 3.859	− 5.732	0.423	− 0.231	− 0.078	− 8.215	− 10.535	− 9.204	− 0.136
Upper	0.058	5.536	7.662	12.530	0.054	0.022	0.498	0.901	0.980	0.130

This table is constructed by randomly dropping 10% of all units to perform Eq. (1) 1000 times. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. We report the average value of the 1000 coefficients under the label “Average coefficients”. We expect that the mean value of the 1000 coefficients is statistically no different from the baseline coefficient in Table 2, eliminating the possibility that certain control units may be driving effects. We test the equality of the mean value of the 1000 coefficients and the baseline coefficient in Table 2 under the label “Equality test”. We find that all p-values exceed 0.1, indicating that the null hypothesis that the means of the 1000 coefficients and the baseline coefficients in Table 2 are equal cannot be rejected. We report the minimum and maximum values of all the 1000 coefficients under the labels “Min” and “Max”. We report the 95% confidence interval (CI) of baseline coefficients estimated in Table 2 under the labels “Upper” and “Lower”. We find that all 1000 coefficients lie in the CIs, indicating that each of the 1000 sub-samples can represent the total sample. In other words, the results of the sub-samples do not deviate from the baseline sample, which means that the elimination of some units will not make the sub-samples different from the original sample. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours.

Extend the study period

The most recent wave of the CFPS is 2020. We exclude the 2020 data in the baseline estimation because we expect a significant lifestyle change after the spread of COVID-19. Individuals may choose to work from home, so their working style and time allocation are more flexible^26–28. During the pandemic, some individuals experienced job loss, resulting in significant changes to their employment status, which subsequently affected their sleep patterns and time management. Table 8 presents consistent findings from 2016 and 2018, reflecting the dynamic patterns observed each year following the implementation of UTCP.

Table 8.

Include 2020 data after experiencing the COVID-19 shocks.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Treatment*year16	− 0.019	1.690	2.716	2.300	− 0.006	− 0.046	− 5.039*	− 6.099*	− 5.210*	0.017
	(0.035)	(3.031)	(4.093)	(3.949)	(0.086)	(0.032)	(2.875)	(3.617)	(3.162)	(0.082)
Treatment*year18	− 0.014	− 0.574	− 1.245	2.724	− 0.104	− 0.007	− 2.317	− 2.520	− 7.927**	0.060
	(0.037)	(3.147)	(4.028)	(3.860)	(0.091)	(0.034)	(3.068)	(3.553)	(3.185)	(0.085)
Treatment*year20	− 0.036	1.224	4.374	3.511	− 0.036	− 0.018	− 2.603	− 1.380	− 4.720	0.074
	(0.038)	(3.183)	(4.053)	(3.986)	(0.093)	(0.034)	(3.033)	(3.479)	(3.321)	(0.086)
# of individuals	1213	1213	910	1184	1213	1433	1433	1182	1397	1433
# of observations	4852	4852	2307	4115	4832	5732	5732	3057	4770	5708

This table is constructed based on Eq. (1) with an extended year and year dummy variables. Specifically, $Y_{\mathit{it}}={\alpha }_i+{\beta }_1{\mathit{treatment}}_i\ast {Year16}_t+{\beta }_2{\mathit{treatment}}_i\ast {Year18}_t+{\beta }_3{\mathit{treatment}}_i\ast {Year20}_t+{\gamma }_t+{{\mathbf{X}}^{\prime }}_{\mathbf{it}}\mathrm{\Delta }+{\epsilon }_{\mathbf{it}}$ . ${Year20}_t$ takes the value 1 if in year 2020, and takes value 0 otherwise. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. We also exclude those who sleep more than 15 hours, and less than 2.55 hours. Standard errors are clustered at individual level in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Heterogeneous treatment effects

In the baseline estimation, we include individuals with no children and those with one child in the treatment group. However, we are concerned about heterogeneous treatment effects, as lifestyle differences between those with no children and those with one child may exist, despite both being affected by the UTCP. Intuitively, the UTCP should have a more significant impact on those with one child, as they are the primary target population. First, we use the traditional DD approach outlined in Eq. (2) to demonstrate the heterogeneous effects on sleep patterns. ${No\_child}_i$ takes the value of one if individual $i$ has no children in 2014, and takes the value or zero otherwise. ${One\_child}_i$ takes the value of one if individual $i$ has one child in 2014, and takes the value or zero otherwise. Other variables are the same as in the Eq. (1). There is no markedly different between these two groups, as shown in Table 9 Panel A. Individuals with no children are also impacted by the UTCP, adjusting their napping behaviors in response to the policy change. This finding suggests that the family planning policy may influence women who have the potential to give birth, regardless of whether they had a child.

$$Y_{\mathit{it}}={\alpha }_i+{\beta }_1{No\_child}_i\ast {Year16}_t+{\beta }_2{One\_child}_i\ast {Year16}_t+{\beta }_3{No\_child}_i\ast {Year18}_t+{\beta }_4{One\_child}_i\ast {Year18}_t+{\gamma }_t+{{\mathbf{X}}^{\prime }}_{\mathbf{it}}\mathrm{\Delta }+{\epsilon }_{\mathbf{it}}.$$ 2

Table 9.

Treatment intensity.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: traditional DD
No child*year16	− 0.005	1.104	− 1.085	4.425	− 0.064	− 0.084**	− 8.294**	− 1.936	− 8.168**	− 0.111
	(0.032)	(2.891)	(4.596)	(3.982)	(0.088)	(0.036)	(3.221)	(4.686)	(3.854)	(0.099)
No child*year18	0.023	4.126	6.424	9.650**	− 0.074	− 0.025	− 4.389	− 3.225	− 5.849	− 0.049
	(0.034)	(3.118)	(4.628)	(4.306)	(0.094)	(0.039)	(3.350)	(4.433)	(4.138)	(0.099)
One child*year16	− 0.017	1.464	1.553	5.977*	− 0.070	− 0.053**	− 4.816**	− 5.110*	− 1.766	− 0.130*
	(0.029)	(2.509)	(3.653)	(3.162)	(0.076)	(0.026)	(2.301)	(3.094)	(2.638)	(0.069)
One child*year18	− 0.005	− 0.911	− 1.692	4.730	− 0.096	− 0.030	− 3.739	− 5.208*	− 3.561	0.010
	(0.029)	(2.526)	(3.680)	(3.229)	(0.076)	(0.027)	(2.330)	(3.049)	(2.686)	(0.071)
# of individuals	2061	2061	1405	1981	2061	2351	2351	1763	2257	2351
# of observations	6183	6183	2810	5173	6167	7053	7053	3622	5742	7038
Panel B: negative weights in TWFE
# of neg. weight	44	44	113	123	46	40	40	135	159	42
# of pos. weight	2742	2742	893	2195	2737	2864	2864	1129	2226	2860
Neg. weight ratio	1.58%	1.58%	11.2%	5.31%	1.65%	1.38%	1.38%	10.68%	6.67%	1.45%
min σ(Δ)	0.052	1.320	1.156	10.043	0.390	0.306	29.391	4.922	8.198	0.374

Panel A is constructed based on Eq. (2). Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1. Panel B is constructed from the twowayfeweight stata package. TWFE denotes for two-way fixed effect. The number of negative weights and min σ(Δ) compatible with β from the TWFE and Δ_TR = 0 are reported. Δ_TR denotes the average treatment effect across all treated units.

In the traditional two-way fixed effect estimations, as in our baseline model, negative weights arise because the two-way fixed effect estimator is a weighted sum of several average treatment effects (ATEs) in various groups at multiple periods³⁸. The negative weights are an issue when the ATEs are heterogeneous across groups or periods. Next, we estimate the weights attached to our baseline two-way fixed effect estimators ($\widehat{\beta }$). In Table 9 Panel B column 2, a small number of weights are negative. In addition, we could assess the robustness of the two-way fixed effect coefficient by checking the ratio of the absolute value of the expectation of $\widehat{\beta }$ to the standard deviation of the weights. The ratio is the minimal value of the standard deviation of the ATEs across the treated in group $g$ and period $t$. If the ratio is not close to zero, the treatment effect heterogeneity would not be a serious concern for the validity of that coefficient. In our study, the traditional DD approach is valid.

Mechanism

In this section, we provide a mechanism analysis to explain why UTCP affects female health behaviors but not male health behaviors. As discussed in Sect. "Theoretical framework", being treated differently in the workplace may compel women to exert greater effort to advance in their careers^17,39. The following sub-sections are designed to expound on the circumstances under which women would undertake changes in their sleep behaviors and deliberate whether they make changes in other behaviors corresponding to the policy change.

Workplace promotion

Females who anticipate career promotions may adjust their work behaviors and become more sensitive to their working environment. To investigate, we use a survey question asking, “What kind of promotion do you anticipate having?” with potential responses being “management job promotion,” “technical job promotion,” “both”, or “neither.” We construct an indicator, “expect promotion,” which equals one if respondents chose any of the first three options and zero if they chose “neither.” Table 10 shows that females who expect to receive a promotion tend to reduce their napping time after the implementation of UTCP. In contrast, males do not show a significant response to the UTCP. This indicates that the policy’s impact on health behaviors is more pronounced among females motivated by career advancement opportunities.

Table 10.

Mechanism-workplace promotion.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Treatment*year16*expect promotion	0.059	− 14.895	− 44.621	6.842	0.116	− 0.245	− 27.459*	− 9.515	− 20.962	− 0.576
	(0.213)	(15.799)	(31.233)	(22.377)	(0.516)	(0.214)	(14.675)	(25.280)	(20.133)	(0.417)
Treatment*year18*expect promotion	− 0.026	− 8.681	− 18.241	− 5.631	0.122	− 0.250	− 22.282**	− 15.119	− 18.040	− 0.586*
	(0.177)	(13.536)	(16.742)	(16.588)	(0.467)	(0.171)	(10.893)	(17.969)	(16.940)	(0.351)
# of individuals	1317	1317	725	1202	1315	1296	1296	784	1173	1295
# of observations	2205	2205	1003	1939	2201	2096	2096	1074	1827	2095

This table is constructed based on Eq. (1) but with an additional interaction term of treatment effect with the expect promotion indicator. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Working sectors

There is considerable heterogeneity in job security across occupations. Job security could be satisfied by the broader institutional environment⁴⁰, for example, public sectors are relatively lenient with their employees, and their employees will be slack⁴¹. Also, the rates of absenteeism among public sector employees are significantly higher compared to those of private sector employees⁴². Those who work in Chinese state-owned enterprises and various levels of the public sector have relatively low work pressure and good work environments⁴³. We consider that public-sector workers have higher job security and stability and are thus more likely to take naps regardless of UTCP. Consequently, we expect UTCP to have a greater impact on private-sector workers compared to public-sector workers. The estimated coefficients presented in Table 11 align with our expectations.

Table 11.

Mechanism-working sectors.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: public sector
Treatment*year16	− 0.091	− 15.399	− 17.797	− 16.472	− 0.529*	0.216	27.234	− 71.649***	− 59.149***	0.865
	(0.134)	(10.617)	(15.640)	(15.678)	(0.300)	(0.287)	(32.444)	(14.559)	(22.292)	(0.786)
Treatment*year18	− 0.042	− 7.165	− 11.128	19.008	− 0.211	0.179	− 7.649	−	− 31.362	1.007
	(0.133)	(11.228)	(15.124)	(19.153)	(0.316)	(0.400)	(48.848)	−	(50.475)	(0.824)
# of individuals	339	339	186	307	339	222	222	119	199	222
# of observations	476	476	239	422	476	276	276	130	239	276
Panel B: private sector
Treatment*year16	0.005	2.515	− 0.352	7.434**	− 0.021	− 0.072**	− 6.130**	− 3.826	− 4.369	− 0.128
	(0.029)	(2.551)	(3.782)	(3.255)	(0.079)	(0.029)	(2.567)	(3.462)	(2.967)	(0.081)
Treatment*year18	0.007	− 0.008	− 2.475	5.882*	− 0.061	− 0.044	− 3.161	− 2.279	− 4.731	− 0.005
	(0.030)	(2.680)	(3.746)	(3.346)	(0.081)	(0.031)	(2.652)	(3.486)	(3.122)	(0.081)
# of individuals	1961	1961	1293	1864	1961	1863	1863	1302	1778	1863
# of observations	5315	5315	2417	4422	5303	4822	4822	2492	3973	4814

We focus on the employed observations, and estimations are based on Eq. (1). Panel A is based on those who work in the public sector, and the number of observations is small. Panel B is based on those who work in the private sector. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Job choices

Napping has significant health implications. For individuals who cannot get enough sleep or experience poor sleep quality at night, naps can serve as a beneficial supplement to improve their overall health. Conversely, if napping is a habitual practice, disruptions to this routine can negatively impact their lifestyle and well-being⁴⁴. In China, napping is culturally ingrained and often becomes a habit from a young age. For such individuals, the demand for naps is relatively inelastic.

In an extreme scenario where the demand for naps is perfectly inelastic, individuals may be unwilling to adjust their napping habits to accommodate changes in job flexibility and work environment brought about by the UTCP. While this scenario is not likely to be common, we cannot exclude the possibility that some individuals might quit or change jobs if their naps cannot be promised. If this occurs, our baseline results may underestimate the negative effects of UTCP.

To address this issue, we categorize our sample based on whether individuals changed jobs. Although we do not have detailed information on the reasons for job changes, if napping habits prompted these changes, we would expect napping behaviors to remain unaffected after the job transition. Our findings indicate a negative effect on the probability of napping for females who left their main jobs, although this effect is only significant at the 10% level. This suggests that quitting due to napping habits is an isolated case. Those who did not change jobs are more likely to adjust their napping behaviors in response to the UTCP, reflecting an unfriendly work environment.

Table 12 shows significant effects on the probability of napping and nap durations for females who retained their main jobs. These results support previous findings that job flexibility has minimal impact on employment choices but may significantly affect wage distribution⁴⁵.

Table 12.

Mechanism-job choices.

	Male					Female
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours	Nap=1	Napping time in minutes	Napping time (intensive)	Time to sleep	Sleep hours
Panel A: stopped the main job
Treatment*year16	0.050	3.486	− 2.102	2.171	0.016	− 0.064*	− 3.431	− 0.649	− 4.147	− 0.076
	(0.035)	(3.174)	(4.474)	(3.626)	(0.097)	(0.036)	(3.396)	(4.347)	(3.508)	(0.101)
Treatment*year18	0.061*	2.533	− 1.577	3.241	− 0.014	− 0.042	− 2.049	0.016	− 7.156*	− 0.013
	(0.036)	(3.255)	(4.523)	(3.806)	(0.095)	(0.038)	(3.385)	(4.175)	(3.662)	(0.103)
# of individuals	2020	2020	1275	1878	2019	2130	2130	1430	1972	2130
# of observations	4690	4690	2214	3922	4680	4650	4650	2465	3830	4643
Panel B: maintained the main job
Treatment*year16	− 0.131	− 4.708	9.963	2.628	− 0.360	− 0.117**	− 11.510**	− 15.039*	8.971	− 0.276*
	(0.119)	(9.390)	(24.205)	(12.446)	(0.276)	(0.054)	(5.071)	(8.595)	(5.961)	(0.156)
Treatment*year18	− 0.027	6.425	11.874	− 16.477	0.188	− 0.136*	− 15.313**	− 18.541*	1.707	− 0.128
	(0.145)	(12.487)	(17.266)	(19.676)	(0.353)	(0.075)	(6.799)	(10.458)	(7.434)	(0.205)
# of individuals	1182	1182	511	1005	1180	1577	1577	863	1340	1576
# of observations	1427	1427	569	1191	1422	2310	2310	1099	1841	2305

This table is constructed from the subsample analysis based on Eq. (1). Panel A is based on the sample who stopped the main job. Panel B is based on those who maintained the main job. Columns 1 and 6 show the policy effect on the probability of having the nap for males and females, respectively. Columns 2 and 7 show the policy effect on nap time in minutes for males and females, respectively. In columns 3 and 8, samples are restricted to those who have naps, and these two columns show the policy effect at the intensive margin for males and females, respectively. Columns 4 and 9 show the policy effect on time to sleep for males and females, respectively. We restrict samples to those who go to sleep between 9 pm and 4 am to avoid extreme situations. Columns 5 and 10 show the policy effect on the daily sleep hours in total for males and females, respectively. All regressions include controls for age, education, marital status, family income per person, whether living in urban areas, provincial locations, individual fixed effect, province fixed effect, and the year fixed effect, as in Table 2. Standard errors are clustered at individual levels in parentheses. ***p<0.01, **p<0.05, *p<0.1.

Time usage

In the baseline model, we observe females respond to the UTCP by adjusting their sleep behaviors, including napping time and total sleep hours. In this section, we explore whether these adjustments extend to the allocation of time to other activities.

The most crucial activity to examine is working hours, as it directly indicates labor supply and measures the intensive margin of labor supply⁴⁶. For individuals who do not work, working hours are set to zero. Our finding in column (6) in Table 13 Panel A exhibits that the UTCP has a negative impact on the employment of female workers and the supply of labor in China, consistent with Li and Wang⁴⁷. Their study shows that the effect on the quality of employment is much more profound than that on the employment status itself⁴⁷, indicating that females tend to sacrifice the quality of employment rather than stop working. Our regression results echo their findings. Combined with the negative effect on napping time, we infer that females do not immediately reduce labor supply when feeling work-related stress; instead, they adjust their napping behaviors to adapt to environmental changes. In the long term, the exacerbating work environment may lead to a change in labor supply and even drive one out of the workforce⁴⁸.

Beyond working hours, survey questions address time spent on housework (per week), exercise (per week), daily commute (in minutes), and watching TV (per week). We examine changes in time allocation for these activities to assess the welfare effect of relaxing the family planning policy, with results shown in Table 13, Panel A. Column (7) shows no change in daily commute hours. Columns (3) and (8) reveal that both males and females spend more time on housework, with males responding more promptly. One possible explanation is that over 30% of males did not engage in housework before UTCP. Specifically, our data shows that in 2014, 69.6% (1434 out of 2061) of males did housework, while 93.7% (2203 out of 2351) of females did housework. Another explanation is that men’s and women’s fertility intentions may be misaligned, causing conflicts between spouses and extended families³⁹. The participation of husbands in housework is positively related to wives’ fertility willingness^49,50. Thus, males may increase their housework time to alleviate the burden on females to encourage females’ fertility intention, though this effect does not persist. Females tend to sacrifice leisure activities, such as watching TV and exercising. The time spent on these activities decreased by 1.127 hours in 2016 and by 0.857 hours in 2018, respectively (see columns (9) and (10) in Table 13, Panel A).

Other health behaviors and health outcomes

Napping is a health-related behavior. Given that our study suggests negative effects on napping, other health behaviors may respond to the UTCP or changes in napping. Napping could supplement night sleep to benefit people’s health⁵¹. Insufficient sleep may lead to poor memory⁵². We may consider napping as a personal habit that may be physically and mentally challenging to change. Therefore, it is essential to examine the policy effects in the health domain. In this section, we include outcome variables related to other health behaviors and health status. Health behaviors encompass exercise frequency, drinking, and smoking habits in addition to sleep behaviors. Specifically, we introduce a set of dummy variables to indicate health behaviors based on survey questions: “How many times have you worked out in the past week, including all indoor exercises and outdoor activities?”, “Have you consumed alcohol three times or more in a week during the last month?”, and “How many cigarettes do you smoke per day?”.

While the less equitable workplace induced by the UTCP changes an individual’s napping behaviors, it may also lead to changes in other health behaviors. Here we examine whether the UTCP is associated with changes in health behaviors, including exercise, alcohol consumption, and smoking, and also the health status. Health status is measured by two indicators: self-reported health condition and memory ability. Self-reported health status is derived from the question, “How would you rate your health status?” with responses ranging from 1 to 5, denoting very healthy, healthy, fair, poor, and very poor conditions, respectively. For analysis, we group responses 2–5 to indicate poor health. The memory indicator is based on the question, “Are you able to remember important events that occurred within the past week?”. Responses range from 1 to 5, indicating situations where individuals remember completely, most, half, only a few, and only a little bit, respectively. We group responses 2-5 and set a dummy variable, poor_memory, equal to 1 to denote poor memory, and set poor_memory to 0 when the memory ability is rated as 1. Outcome variables regarding health status and other health behaviors are shown in Table 13 Panel B. The negative effect of UTCP on female exercise frequency is consistent with the above time spent on exercises. For males, the pressure from UTCP changes their drinking and smoking behaviors, even though they are in opposite directions. For females, drinking and smoking behaviors have no significant changes, which may be due to the low drinking and smoking prevalence (only 1.84% and 1.29% of females drink or smoke, respectively). Both males and females exhibit a decline in the likelihood of poor health, but only females express poor memory after the UTCP.

Conclusion

The Chinese government aims to increase the birth rate by progressively relaxing its family planning policies. While numerous studies have explored female labor supply, the majority have concentrated on labor force participation and working hours, with limited attention given to job quality.

Napping behaviors, which bear significance for both the labor market and health outcomes, have been affected by the implementation of UTCP. The observed gender disparity in sleep behaviors suggests that the relaxed family planning policy exerts a broader impact on the working-age female population, extending beyond those who have given birth or intend to do so.

Furthermore, if napping constitutes a personal habit, women may need to modify their routines to cope with the unintended social stress generated by UTCP, potentially compromising their well-being. Although Chinese laws guarantee that women enjoy equal rights with men to work and social security, it is not enough to ensure women to be treated equally in the labor market. Consequently, policymakers should adopt a comprehensive approach that considers the entire female population and addresses the potential side effects of changes in family planning policies within social and work environments.

Author contributions

Yuan Fang: Methodology, Software, Writing - Review & Editing. Shih-Ting Huang: Writing—Original draft, Writing - Review & Editing.

Data availability

The data are from China Family Panel Studies (CFPS), funded by Peking University and the National Natural Science Foundation of China. The data are publicly accessible at https://www.isss.pku.edu.cn/cfps/en/data/public/index.htm. The CFPS is maintained by the Institute of Social Science Survey of Peking University.

Competing interests

The authors declare no competing interests.