Unequal Duties and Unequal Retirement: Decomposing the Women's Labor Force Decline in Postreform China

Abstract

The gender gap in labor force participation (LFP) in China has grown over the last 30 years, despite substantial advances in women's education and economic development. Previous research has identified gender discrimination and work–family conflicts as two key explanations for the gap, both of which have risen since the start of China's economic reform in 1978. Using multiple waves of the national census and household panel data from China, this research shows that one overlooked mechanism widening the LFP gender gap lies in the institutional constraints that require women to retire earlier than men. This research also demonstrates how the impact of women's early retirement on the LFP gender gap has been exacerbated by two societal-level changes: (1) population aging, which increased the share of women who reached the retirement age; and (2) economic development, which increased the number of women entering nonfarming occupations and, hence, the gender-based retirement system. These findings suggest that without significant revisions to China's retirement system, the LFP gender gap will continue to expand as the population ages and economic development proceeds.

Introduction

The women's labor force participation (LFP) rate in China has declined dramatically in the past three decades, despite a rise in women's education and China's economic development. A substantial body of research has linked the gender gap in employment to escalating gender discrimination and family–work conflicts induced by the economic reform starting in 1978 (Ji et al. 2017; Wu 2019). The reform was marked by the state's retreat from social welfare provisions and a rise in private enterprises. As the burden of caring for family members postreform has transferred from the state to the family, the profit-driven economic system is increasingly disadvantaging women—particularly mothers—in the job market. Consistent with these themes, empirical studies have documented increasing gender inequality during the reform era, including the motherhood penalty in wages (Zhang et al. 2008), the widening gender gap in housework time (Tan et al. 2021; Zhang 2017), and the emerging workforce discrimination against women (Cao and Hu 2007; He and Wu 2017; Parish and Busse 2000).

Despite the considerable merit of studies along this line, not enough attention has been paid to women's employment patterns at older ages. Women retire at a younger age than men, because couples tend to retire together and women typically have an older partner (Moen et al. 2005). Elderly women are also in high demand to care for their grandchildren, especially in a society where the public support for parents is limited (Lumsdaine and Vermeer 2015). Moreover, women's early retirement has been institutionalized in China by a gender-based retirement system (Davis 1988): the mandated retirement age is 50 for nonprofessional women (e.g., blue collar workers, clerks) and 55 for professional women (e.g., senior government officials, professors, physicians), whereas the retirement age for men is 60 regardless of occupation. Although people may continue to work after retirement, opportunities are limited and typically offer only reduced salaries (Du and Dong 2009). Since its establishment in 1951, this retirement system has remained largely unchanged.

The impact of gender disparity in retirement age has likely been exacerbated by two societal-level changes: population aging and economic development.¹ China's population is aging rapidly as a result of decades-long, below-replacement-level fertility and advances in life expectancy (Cai et al. 2018; Chen and Liu 2009; Wang 2011). Policymakers and social scientists have expressed concern about the labor force shortage caused by aging, but little attention has been paid to how aging may affect men's and women's labor market prospects differently. Because women retire at a younger age than men, more women than men will leave the workforce as the population ages. Similarly, the radical industrialization and urbanization postreform have driven hundreds of millions of self-employed agricultural workers² into wage employment in the service and manufacturing sectors (Li et al. 2013), thus bringing more people into the gender-based retirement system.

This article investigates the impact of women's early retirement on the gender gap in LFP and how it is mechanically intensified by population aging and economic development. China provides a unique opportunity to empirically examine these potential mechanisms because of its rapid population aging and economic development over a relatively short period (Cai et al. 2018). Results from decomposition analyses show two main findings. First, both men's and women's LFP rates have fallen as the population ages, but women have been disproportionally affected. All else being equal, the change in age composition from 1990 to 2010 explains 27.8% of the labor force decline for women but only 7.8% of the decline for men. This is because female retirees increased faster than male retirees as a result of women's younger mandatory retirement age. Second, even controlling for the change in age composition, the retirement rate for women almost doubled from 1990 to 2010, accounting for 22.7% of the female labor force decline, partially explained by the expansion in service and manufacturing sectors. Women who would have joined the agricultural sector were drawn into paid employment and the gender-based retirement system. However, this change did not boost men's retirement rate, as men in both agriculture and paid employment maintained a high level of LFP in their 50s and even 60s. Other structural factors—such as increasing caregiving demands and massive layoffs induced by the economic restructuring—may also have contributed to the rise in women's early retirement.

This study makes two contributions to the literature and has important policy implications for China and other countries with a gender-based retirement system. The first contribution is empirical. This article joins a small number of studies that examine the role of retirement behind the puzzling decline in women's labor force in China. Although taking care of families was the major reason for women to quit the labor force, this study finds that the widening gender gap in LFP has been mainly driven by a rise in women's early retirement. The second contribution is both methodological and theoretical. The use of demographic decomposition on census data reveals that the population aging interacts with existing norms and institutional constraints to exacerbate gender inequality in the labor market at the population level. This mechanism can be easily overlooked in a survey-based study because the influence of population structure is difficult to account for.

The findings suggest that the gender gap in LFP will continue to grow as the population ages under a gender-based retirement system. Policies aimed at prolonging working life should avoid continuing or creating a gendered pathway. A supplementary analysis using individual-level data indicates that a significant proportion of women continue to work after the legal retirement age. Under the Chinese legislation, employees working beyond the retirement age are not protected by the labor law and thus are not entitled to such benefits as health insurance and a minimum wage. The phenomenon of women continuing to work after the legal retirement age implies that the static gender-based retirement system is not only unjust, but also unsuitable for many.

Prior Studies on the Gender Employment Gap

The apparent paradox between economic development and female labor force decline is well documented. Empirical studies suggest that female LFP often follows a U-shaped pattern during economic development (Boserup 1970/1989; Goldin 1994). According to Goldin (1994), the initial decline in female LFP is driven by a fall in demand for female manual labor in agriculture, coupled with social stigma against women working in manual labor jobs. As women become more educated, they increasingly enter the workforce as white collar employees, where the societal stigma is less prevalent, leading to a subsequent rise in female LFP. Similarly, recent studies have observed that the female labor force is stagnating in many developing countries despite educational expansion because of the lack of available jobs for educated women (Assaad et al. 2020; Chatterjee et al. 2018; Verick 2014). Yet, in the case of China, even though the share of women working in the service sector has increased significantly in the past four decades, the women's LFP rate declined continuously at the population level. To explain the decline, one line of research centers on the gender discrimination induced by the market transition in China since the 1978 economic reform. Before the reform, employment was described as “universal” for women in China (Bauer et al. 1992; Honig and Hershatter 1988). The state's deliberate promotion of equal employment under Marxist gender egalitarian ideology resulted in a remarkable increase in the proportion of women in paid labor. Postreform, however, the preference to hire men rather than women increased dramatically (Wu 2019).

Even women who already had a job were viewed as surplus workers and, instead, their domestic roles were emphasized (Ji et al. 2017). During the restructuring of state-owned enterprises in the late 1990s, women workers were disproportionally laid off (Dong and Pandey 2012; Giles et al. 2006). Furthermore, reemployment opportunities for women who had lost their jobs were limited. Women who have been dismissed or fired tend to have longer durations of unemployment than men (Du and Dong 2009) and many—especially mothers with young children—leave the labor force permanently after long-term unemployment (Appleton et al. 2002; Maurer-Fazio et al. 2011).

In recent years, a growing number of scholars have taken a feminist approach and have incorporated the role of family into the study of gender disparities in the labor market (Chen et al. 2014; Dong 2020; Zhang et al. 2008). They argued that an increase in work–family conflicts is the main factor that pulls women out of the labor market (Du and Dong 2013; Zhang et al. 2008). In the same vein, Du and Dong (2009) suggested that the availability of daycare in a local area is positively associated with the proportion of mothers participating in the labor force and with the number of hours per day that mothers worked. All things being equal, the presence of daycare in the communities increased mother's LFP by 10.5% and working hours by 5.3–6.7 hours per week. Their estimate also shows that the decline in daycare services accounted for 46% of the drop in mother's LFP between 1997 and 2000. In accordance with these studies, scholars consistently find that women in China spend more time on caregiving and household chores than men, and that the gender gap has been widening in the economic reform era (Tan et al. 2021; Zhang et al. 2008; Zhang 2017).

Gender Inequalities in Transition to Retirement

According to the retirement literature, women are more likely than men to be disadvantaged in retirement because they leave the labor force earlier (Dahl et al. 2003; Flippen and Tienda 2000), have less financial preparation for retirement (Noone et al. 2010), and receive less in pension benefits (Chen and Liu 2009; Giles et al. 2023; Shen et al. 2016). Often, women are more likely than men to retire in response to family caregiving responsibilities for both aging parents and young grandchildren (Feng and Zhang 2018; Pavalko and Artis 1997). In addition, married women often coordinate their retirement with their spouses. Because women are usually married to older men, they often retire at a younger age than their husbands (Moen et al. 2005; Ruhm 1996). Furthermore, women's comparatively low educational attainment and their concentration in low-skilled employment make their opportunity cost of exiting the labor market at old age lower than that of men. (Hare 2016; Radl 2013). In turn, early exit from the labor market leads to lower average pensions for women (Zhao and Zhao 2018). Moreover, at the societal level, early retirement is considered more socially acceptable for women than for men (Radl 2012; Settersten 2003). In fact, in many countries, women have a younger legal retirement age than men (World Bank 2021b).

In this article, I argue that the early retirement of women is an essential mechanism in explaining the labor force decline, a perspective that has not yet been fully explored. In recent years, a number of researchers have resorted to the theory of intersectionality to understand the “double jeopardy” that older women confront in the workplace (Duncan and Loretto 2004; Ford et al. 2021; Gee et al. 2007; Jyrkinen and McKie 2012; Moore 2009). Empirical studies suggest that women are perceived as “old” at relatively young ages and suffer greater ageism than men at older ages (Barrett 2022; Moore 2009). In addition, negative judgments of women's older appearance result in a devaluation of their competency and expertise (Ford et al. 2021; Jyrkinen and McKie 2012). These constraints have led to a disproportional preretirement exit for women in their mid-40s (Duncan and Loretto 2004). In the Chinese context, such age discrimination against women is institutionalized by a retirement system that compels women to retire earlier than men. Since 1951, China has mandated a retirement age for employees in the public sector, which was later extended to all sectors: age 60 for all men workers; 55 for women senior officials, professionals, and technicians (referred to as professionals); and 50 for all other women workers (referred to as nonprofessional). Evidence from survey data suggests that city employees generally follow mandatory retirement policies regardless of their education and occupation (Bauer et al. 1992; Hare 2018).

With few exceptions, scholars of gender inequality in China have paid little attention to age discrimination against women or the institutional constraint imposed by the retirement system. Using the 1987 1% population survey, Bauer et al. (1992) found that the compulsory retirement policy affected urban women of all educational levels. Another study conducted in the early 2000s on job disruption discovered that unemployed women in their 40s had lower reemployment rates than younger age groups (Du and Dong 2009). Du and Dong attributed the low reemployment rate to employers' cost–benefit analyses, which predicted that women approaching age 50 would soon leave the workplace because of the mandated retirement policy. Hare (2016, 2018) also identified age as the most important explanatory factor for the decline in women's LFP in China from 1991 to 2011; however, without comparing with a sample of male workers, Hare linked this finding to population aging but did not examine how aging affects men and women differently under the unequal retirement system.

Compositional Effects of Population Aging and Economic Development

The gender gap in LFP rate caused by women's early retirement may be further exacerbated by population aging. China's population is aging rapidly as a result of long-term low fertility and advances in life expectancy. Yet, the impact of population aging on women's LFP has not been fully addressed in gender stratification research. As the population ages, more women will reach retirement age because of the gendered retirement schedule.

The negative impact of the gendered retirement system may be further compounded by the rapid economic development following economic reform. Before 1978, less than 20% of the population were living in urban areas, but by 2020, the number had jumped to more than 60% (World Bank 2021c). Likewise, the decline in the agricultural sector was accompanied by the increased employment opportunities in the service and manufacturing sectors in urban areas (World Bank 2021a). Hundreds of millions of rural migrants were employed in cities and were thus integrated into the gender-based retirement system designed for wage employment³ (Li et al. 2013). Survey-based research often fails to account for the influence of population aging or changing occupational structure on the labor force decline at the population level. Given the scale and pace of population aging and economic development in China, ignoring the composition effects of changing age or occupational structure will offer an incomplete analysis.

The Current Study

Previous literature has identified gender discrimination in hiring and layoffs, as well as work–family conflicts, as explanations for the decline in the female labor force. Building on earlier research, this study argues that women's early exit before or near the legal retirement age is an important mechanism in explaining female labor force decline. In the Chinese context, this departure from the labor force is institutionalized, as the current retirement system requires women to retire 5–10 years earlier than men. I further suggest this mechanism gains even a greater significance as the population ages and economic development proceeds. The first process increases the proportion of women reaching the legal retirement age, while the second process increases the number of women who are subject to the retirement system. What effect do population aging and economic development have on LFP and how does it differ by gender? To what extent can the gap in LFP rate be attributed to women's early retirement? And to what extent do people comply with official retirement policies?

The answers to these questions are important. Similar to China, many countries employ a gender-based retirement system.⁴ Even in countries without such a system, women tend to retire earlier than men owing to age discrimination, poor health, or caregiving responsibilities, or to accommodate their husband's retirement (Moen and Flood 2013; Moen et al. 2005; Van Houtven et al. 2013; Warner et al. 2010). Therefore, this study not only provides suggestive evidence for evaluating the impact of early retirement on the gender gap in employment, but also offers insights into reforming the retirement system in such countries to promote gender equity in the context of global aging. Still, some might argue the entitlement to retire early is a benefit rather than a form of inequality. However, in most of the retirement plans, the pension received is tied to the number of years worked and the salary at the time of retirement. In addition, unemployment could also imply losses in health insurance and social networks. Therefore, early withdrawal from the labor market is a source of inequality—and in extreme cases, a cause of poverty among elderly women.

I explore these questions using Chinese data from 1990 to 2010. During this period, China witnessed not only rapid economic development but also a significant drop in women's LFP rate (Wu and Zhou 2015). Meanwhile, the Chinese population aged at an unprecedented pace from declines in fertility and mortality rates. Yet, the retirement system that requires women to retire earlier than men has remained intact. In this context, China offers an opportunity not only to evaluate the impact of gender-based retirement age on gender disparities in the labor market, but also to observe how the impact may be exacerbated by population aging over time. I further test the effectiveness of the retirement policy by examining the occupational differences in time of retirement and how the rising retirement rate in China may be linked to the change in occupational structure under the rapid economic development.

Data and Methods

Data

The analyses in this study rely on multiple data sources. To examine the influence of macro-level changes (i.e., population aging and economic development) on the LFP rate, I use long-form tabular data from the 1990, 2000, and 2010 National Population Census of China and from the 1995 and 2005 intercensal surveys. The decennial national census has been conducted to count every citizen living in China. In each census, 10% of the population is chosen at random to complete a longer questionnaire covering LFP status and reasons for nonparticipation. The intercensal survey (also known as the “mini census”) is a stratified, multistage cluster probability sample that is done in intercensal years. In the mini census, similar questions about LFP status were asked as in the decennial census. I focus on the period from 1990 to 2010 because it is when China's population and labor force structure underwent substantial change.

The census and mini census offer historical continuity and representativeness for examining trends in LFP cross-sectionally. Yet, they provide limited information about respondents' employment histories and other socioeconomic characteristics. Thus, I supplement the macro-level analysis with long-running household panel data from the China Health and Nutrition Survey (CHNS). The CHNS is a longitudinal survey that employs a multistage random cluster design to draw a stratified probability sample. Although the CHNS is not nationally representative, it covers rural, urban, and suburban areas in nine provinces and three autonomous megacities to ensure substantial variation in geography, economic development, and public resources. With a focus on examining how rapid social changes in China affect health and health behaviors, the survey also includes detailed information on socioeconomic status, occupation, and employment status.

I use CHNS data for two purposes: (1) to confirm the pattern found at the aggregate level showing that women retire earlier than men, controlling for a host of important sociodemographic factors, and (2) to examine whether the occupational differences in retirement correspond to national policy requirements. I use nine available waves, from 1989 to 2011, which cover the observational period in my macro-level analysis of census data. My final analytic sample from the CHNS data consists of 2,722 men and 2,473 women.

Measures and Methods

Demographic Analysis Using Census

In the Chinese census, the economically active population (those who participate in the labor market) is defined as respondents who had worked for pay for at least one hour in the week prior to interview; this includes respondents who were temporarily on leave, in training, or unemployed but looking for work. The remaining population is classified into mutually exclusive categories by reasons for not participating in the labor force: (1) enrollment in school, (2) caregiving, (3) retired, (4) disabled, and (5) other. The first group consists of people who were enrolled full-time in certified educational institutions at all degree levels; students who worked part-time or employees who received off-the-job training were classified as economically active. People classified as inactive due to caregiving activities include those who were undertaking unpaid care work for their own family and did not engage in any paid work. Retirement refers to the nonactive population who had formally retired, was receiving a pension, and had not engaged in any paid work since their retirement. People in the disabled group are those unable to participate in the labor market because of physical or cognitive disability. People who formally retired are counted in the retirement group regardless of current caregiving or disability status.

To examine the impact of age structure on LFP and how it varies by gender, I apply demographic decomposition to the rate change in labor force nonparticipation rate (hereafter referred to as LFNP) separately for both men and women (Das Gupta 1993; Kitagawa 1955). Specifically, I decompose the observed increase in the LFNP between 1990 and 2010 into the age composition effect and the rate effect. This analysis answers the questions of, cross-sectionally, how much of the increase in LFNP is caused by the change in population aging and how much by the increase in rate changes in age-specific LFNP. Using the same method, I decompose the gender gap in LFNP for each census year. This analysis evaluates the relative contribution to the gender gap in LFNP of each of the five reasons defined and how those contributing factors have changed in magnitude over time.

The LFNP is formally defined as the proportion of the economically inactive population among the total population aged 20–64.⁵ The crude LFNP can be written as the weighted average of the age-specific LFNP (Preston et al. 2001):

where C_i denotes the proportion of people in the age group i (i = 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64). $L$_i denotes the age-specific LFNP. It is easy to prove that the age-specific LFNP is a summation of mutually exclusive reason-specific LFNPs:

To simplify discussion, I reexpress Eq. (2) as

where $L$_ij denotes the reason- and age-specific LFNP ( j = in school, caregiving, retired, disabled, or other). To analyze the relative contribution of each factor to the differences in LFNP between year and between gender, I apply the demographic method of decomposition (Das Gupta 1993; Kitagawa 1955). The standard Kitagawa's decomposition can be expressed in a general form as

In Eq. (4), A and B refer to two different populations. For my first research question of how population aging affects the LFNP, A and B refer to the population in 1990 and 2010, respectively. For my second research question, which examines the main contributing factors to the widening gender gap in LFNP, A and B refer to the male and female populations, respectively. Δ denotes the difference of the observed LFNPs between populations A and B. The age composition effect, represented by the first component on the far right-hand side of Eq. (4), $∑iI(CiB−CiA)⋅[LiB+LiA2],$ measures the contribution of age composition differences to the Δ, holding constant the age-specific rate. The rate effect, represented by the second component, $∑iI(LiB−LiA)⋅[CiB+CiA2],$ measures the contribution of change in reason- and age-specific LFNP to the Δ, holding constant the age composition. Substituting Eq. (3) into Eq. (4), the decomposition can be extended to the following: the summation of the first component and the second component across all j adds up to the total difference in crude LFNPs between populations A and B, as shown in Eq. (5):

Equation (5) separates the age composition effect and rate effect from Eq. (4) into reason-specific age composition effects and reason-specific rate effects. It informs us of the relative contribution of each cause of labor force nonparticipation to the rate changes between populations A and B. This approach is mathematically equivalent to Das Gupta's (1993) 6-vector-factors decomposition.⁶ In the decomposition results, I present each reason-specific rate effect but show the age composition effect at the aggregate level.

Statistical Analysis Using CHNS

I use discrete-time logit regressions to measure the effects of occupation types on the hazard rate of retirement for men and women separately. This method allows estimations of the fixed and time-varying effects on retirement entrance without the proportionality assumption (Allison 1995). I extract two middle-aged cohorts from the CHNS data and follow them up to 11 years to investigate gender and occupational difference in retirement age. The first cohort consists of people aged 39–49 in the 1989 survey, and the second consists of people aged 39–49 in the 2000 survey. To increase the sample size, the two cohorts are merged; a dummy variable is added to indicate cohort. Because the focus of this article is on examining the gendered retirement schedule, the analytic sample is limited to men and women who were employed at their first interview (i.e., 1989 or 2000).

In this analysis, the dependent variable is a dichotomous indictor of whether the respondent had retired in the interval between two survey years. People who are categorized as retired must meet two criteria: they are not presently working and they report the cause of not working as retirement. Because the CHNS was not conducted on an annual basis, retirement status was determined using the most recent year available. Data were organized into person-year records so that one record represents a person-year of an individual at risk of retirement. A person at risk of retirement can leave the risk set by retiring or being censored; being censored is defined as leaving the labor force because of reasons other than retirement, being lost during follow-up, or death. Because the CHNS was not administered annually, the information in nonsurvey years was from the previous interview year.

The key independent variable is occupational type. I regroup this variable on the basis of the legal retirement age pertaining to women's occupation, which include professionals, nonprofessionals, and agricultural/self-employed individuals. I control for rural or urban residence, education level, marital status, ethnicity, cohort, and province. More details on the measures and descriptive statistics of the CHNS data sample are provided in the online appendix.

Results

Interrelationship Between Retirement System, Population Aging, and Economic Development

Panel A of Table 1 shows the LFNP rates and reasons for nonparticipation by gender and year. Between 1990 and 2010, both men's and women's LFNP rates increased, but women's rates increased faster, increasing the gender gap in LFP. Caregiving duties were the most common reason for women's nonparticipation, and their rate was approximately 16–21 times as high as among men, depending on the year studied. These patterns largely held after age standardization (using 1990 as the reference), as shown in panel B. From 1990 to 2010, the retirement rate for women more than doubled, rising from 2.2% to 5.1%; the pattern remained similar following age standardization, but to a lesser extent. The retirement rate for men increased slightly, from 2.4% to 3.1%; however, after age standardization, men's retirement rate stabilized at around 2.4%, aside from a spike in 2000.⁷

I further test the observations in Table 1 by decomposing the increase in LFNP between 1990 and 2010 into (1) the changes in age structure (composition effect) and (2) the change in age-specific nonparticipation rate (rate effect). The age composition effect captures the contribution of changing age structure on the increase in crude LFNP between 1990 and 2010, holding the age-specific LFPR rates constant. The rate effect captures the contribution of changing age-specific LFNP on the increase in crude LFNP between 1990 and 2010, assuming the age composition in these two time periods are the same.

Table 2 presents the decomposition results of the increase in LFNP between 1990 and 2010 by gender. Between 1990 and 2010, the LFNP for people aged 20–64 grew by 8.1% for women and by 5.6% for men. The rise in men's LFNP was explained primarily by an increase in school enrollment rate, which accounted for approximately half of the change. Similarly, from 1990 to 2010, school enrollment explained more than one third of the increase of LFNP among women.

The shift in age structure accounted for more than 27.8% of the rise in LFNP among women. In contrast, the aging effect accounted for just 7.8% of the rise in men's LFNP. The fact that women's LFP was more affected by the population aging than men's LFP can be explained by women's higher labor force dropout rate at older ages. Some may speculate that the population of women in China ages more quickly than that of men because of sex-selective abortion. Indeed, the sex ratio at birth in China witnessed a significant increase after 1990; however, my analytic sample mainly focuses on people born before 1990. The decomposition by gender (shown later) also reveals that differences in age structure between the two genders exist but explain little of the gender gap in LFNP.

For men, the contribution of retirement to the change in LFNP was only −1.7%. This negative estimate implies that, after adjusting for the age composition effect, men retired at a lower rate in 2010 than they did in 1990. In contrast, women had higher rates of retirement in 2010 than in 1990, accounting for 22.7% of the rise in LFNP from 1990 to 2010. The increase in women's retirement rate can be attributed to multiple factors. First, it might be caused by a rise in the volume of women employees subject to the gender-based retirement system because of economic development and changing occupational structure. This process also transferred a substantial number of men from the agricultural sector into other parts of the labor market, but because both men agricultural workers and paid employees maintained a high LFP rate through their 50s and beyond, this shift did not boost the retirement rate for men as it did for women. Secondly, the restructuring of state-owned enterprises from the late 1990s to the early 2000s may have led to more women than men nearing retirement age choosing to retire early. Lastly, the rising demand for caregiving due to the state's withdrawal from providing accessible family care services may have prompted older women to retire early and care for their grandchildren and elderly family members (Dong 2020; Giles et al. 2006).

Increasing Caregiving Rates at Younger Ages

According to Table 2, the caregiving rate contributed −1.6 to the change in LFNP among women, accounting for −19.4% of the total change. This implies that, if women in 2010 were engaged in caregiving at the same rate as in 1990, the increase in LFNP from 1990 to 2010 would be even larger, by a factor of 1.6. At first glance, this result appears to contradict previous research, which suggested that women were under more pressure to fulfill caregiving roles postreform than before the reform. In Figure 1, I present each age group's contribution to the overall differences in women's caregiving rates and retirement rates between 1990 and 2020, net of the change in age structure. The contribution of caregiving and retirement rates in all age groups adds up to −19.4% and 22.7%, respectively, as documented in Table 2.

Figure 1 reveals that the reduction in women's overall caregiving rate was mostly driven by declines among older age groups. However, in every age group younger than 45–49, women had higher caregiving rates in 2010 than they did in 1990, as indicated by the gray bars in the positive direction. After age 45, the contribution of caregiving rates turned negative, coinciding with increases in retirement rates in older age groups. It is important to note that reasons for nonparticipation are mutually exclusive by survey design. Therefore, women in their 40s and 50s may have cited retirement as the primary reason for nonparticipation even though they were also participating in caregiving duties. It is a common practice for women to retire early to look after their grandchildren. For example, a study by Wang and Zhang (2018) showed that older women were more likely to stop working if they had to take care of their grandchildren.

Factors Contributing to the Widening Gender Gap in LFNP

Findings from Table 2 support the notion that, cross-sectionally, women's LFNP is more sensitive to changes in age structure because women have had a higher labor force dropout rate at older ages. However, the results do not directly assess the magnitudes of each reason that contributes to the growing gender gap in LFPR over time. In this section, I address this question by decomposing gender difference in LFPR rates by year.

As shown in Table 3, the total gender gap in LFNP increased from 12.8% in 1990 to 15.2% in 2010. In each year, the gender gap in LFNP is largely attributed to women's higher caregiving rates than men's; however, over time, the contribution of women's higher retirement rates to the gender gap in LFNP has increased.

Gender and Occupational Differences in Retirement Schedules

One disadvantage of using cross-sectional census data is that it provides only an aggregate-level snapshot of the population. I supplement the aggregate-level analyses with individual-level data from the CHNS, which includes measures of employment status and comprehensive assessments of socioeconomic characteristics. Although Chinese legislation specifies the mandated retirement age on the basis of occupations and gender, local flexibilities in such policies and individual variation do exist. To examine the level of compliance with the national retirement policy, I utilize the CHNS to investigate the occupational differences in retirement. Recall that in my analysis, a retired person is someone who does not engage in economic activity and has declared the reason for not working as retirement.

Table 4 shows the estimated coefficients of the discrete-time hazards models for both men and women. The base models (Models 1 and 3) include only occupational type and its interaction with age. The interaction term is included to account for the fact that the effect of occupation on retirement may vary by age, with the effect potentially becoming stronger as individuals approach the legal retirement age. This adjustment relaxes the proportional hazards assumption, leading to a more accurate representation of the relationship between occupational type, age, and retirement. Models 2 and 4 introduce controls for education, marital status, ethnicity, and residence, and a set of dummy variables for province. Although interaction effects between age and occupation are significant, the magnitudes and significance of the coefficients carry little weight in the context of a nonlinear model (Ai and Norton 2003; Mize 2019). To facilitate interpretation, Figure 2 displays the average marginal effects of occupation type on the risk of retirement at varying ages based on Models 2 and 4. Working in agriculture and being self-employed—as opposed to having a professional job—reduce retirement likelihood for both men and women at all ages. While the effect of being a nonprofessional on retirement did not differ from that of being a professional for men, the effect of being a nonprofessional on women's retirement relative to that of being a professional was initially positive but became negative as women aged.

The patterns become even clearer in the corresponding predicted survival probability for men and women, shown in Figure 3 (based on Models 2 and 4). Three observations emerge from comparing the survival curves for men and women. First, men retired at a slower rate than women. By the age of 55, about 52% of professional and 43% of nonprofessional women remained in the labor force, compared with more than 80% of men, regardless of occupation. Second, the survival curves for professional and nonprofessional men are highly overlapped. In contrast, the survival curves for women show a clear disparity between professional and nonprofessional occupations. Third, women's survival curves between professional and nonprofessional occupation cross over at age 56. Women with professional occupations retire rapidly after 56, and by 60, nearly all (98%) have completed their retirement; however, around 14% of women in nonprofessional occupations remain in the workforce at the same age. This can be attributed to the fact that nonprofessional women receive smaller pensions, thus giving them a larger incentive to continue working after reaching the legal retirement age (Dong 2020, Giles et al. 2023; Hare 2018; Henry et al. 2018; Zhao and Zhao 2018); it may also reflect the scarcity of professional jobs for women at older ages. Nonetheless, the hypotheses that women retire earlier than men and that women have different retirement schedules depending on their occupational types largely hold.

These results lead to the macro-level question of whether the compositional change in occupation has accelerated women's retirement rates at the population level, as more women transition from agriculture to the service and manufacturing sectors. Figure 4 illustrates the occupational distribution of women aged 40–44, an age group that is still largely in the labor force but will be retiring in 10 years, according to the national retirement policies. The proportion of women working in agriculture dropped dramatically, from 66.0% in 1990 to 42.6% in 2010. Meanwhile, the proportion of women in the service and manufacturing sectors more than doubled, from 15.0% to 30.5%. The proportion of women in professional occupations, on the other hand, increased only slightly. Combining occupations in professional, service, and manufacturing jobs, approximately 40% of middle-aged women in 2010 were subject to the retirement system that required them to retire by age 55 or earlier.

In summary, between 1990 and 2010, the service and manufacturing sectors saw a significant increase in the influx of workers. Because men and women had different mandatory retirement ages, many of those women workers were expected to retire at 55 or earlier, whereas men workers could work until 60. However, it is worth noting that individuals who lack stable and formal employment positions—particularly migrant workers—may receive little or no pension after reaching retirement age; as a result, many are compelled to work beyond retirement age.

Discussion and Conclusion

Despite economic progress and rising rates of higher education enrollment among women, the gender gap in the LFP rate in China has widened during the last 30 years. Previous studies have not fully explored how demographic change interacts with traditional gender norms and institutional constraints to reinforce gender gaps in the labor market. I argue that China's rapid aging and economic development amplify the gender gap in the LFP rate at the population level by increasing the population of women who are subject to an ageist and sexist retirement system that requires women to retire early. The population of men has gone through a similar aging and economic development process; however, because men have a higher required retirement age across occupations, their LFP rate has been less affected alongside changes in the age structure and occupational structure. As a result, the effects of changing age structure and occupation structure play a smaller role in explaining the reduction in men's LFP. It should be noted that women's early exit from the labor market may be attributed to other structural factors, such as increasing caregiving demands and the large-scale layoffs in the late 1990s to early 2000s, which disproportionately affected older female workers and led to their early retirement (Dong 2020; Giles et al. 2006). However, these factors are not easily disentangled. The gender-based retirement system—while creating a direct obstacle for women to continue working at older ages—may also play an indirect role in discouraging their labor force participation by reinforcing women's reproductive roles and the traditional gendered division of labor.

This study also finds that a significant proportion of nonprofessional women remained in the labor market even after reaching the legal retirement age. This finding resonates with previous research, which found that a considerable number of older, low-skilled, urban female workers left their jobs in state-owned enterprises and transitioned to the informal service sector during periods of economic restructuring (Dong 2020). The phenomenon of continuing to work in the informal sector beyond the retirement age is even more pronounced among rural migrants, who have limited pension support and savings (Chen et al. 2017; Giles et al. 2023; Henry et al. 2018). While certain informal service jobs may offer better pay because of increasing demands, many of these jobs are precarious and lack insurance protection. Furthermore, women's decisions to stay are often contingent on their domestic caregiving needs (Dong 2020; Mao et al. 2018).

This article contributes to the literature in three ways. First, women's early retirement reveals an important mechanism in explaining the growing gender disparities in the labor market that have received little attention in previous research. This study implies that as the population ages, this mechanism will become much more prominent. Trends of women's early retirement and population aging are not unique to China. Many countries have similar policies that require women to retire early (World Bank 2021b). Even in countries without such institutional limitations, women typically retire earlier than men owing to gender discrimination, caregiving duties, and higher rates of morbidity and disability (Moen and Flood 2013; Warner et al. 2010). The institutionalized inequality in retirement age has other hidden social consequences as well. For example, because women's working life is often shorter than men's under the current retirement system, the maximum length of a commercial loan for women in China is typically five years less than that for men. In addition, the early retirement age for women may result in fewer women being promoted to higher positions in their careers (Cooke 2016). Moreover, because pensions and wages are closely related to the number of years worked, women are more likely to live in poverty and have fewer savings than men after retirement (Zhao and Zhao 2018). Future research on gender stratification should further investigate the opportunity costs and economic disparity created by women's early retirement, as well as how this is intensified by population aging.

Second, this article contributes to the literature on the paradoxical trend of women's rising educational attainment yet stagnant LFP by demonstrating that the paradox may be driven not only by limited job entry, but also by early exit in midlife. The paradox that increases in women's educational attainment do not translate into greater gender parity has been widely documented in developing countries (Assaad et al. 2020; Chatterjee et al. 2018; Verick 2014). Research has attributed the paradoxical trend to conservative gender norms and long-term gender segregation in the labor force that provide limited employment opportunities for educated women to enter the labor market. The case of China differs in that the job market for both men and women dramatically broadened postreform. However, as economic development proceeds and more women transitioned from agriculture to wage employment, the population subject to the gender-based retirement system increased, resulting in a larger gender disparity in LFP at the population level.

Third, although the role of early retirement becomes important as the female population ages, caregiving is still the biggest barrier preventing women from participating in the labor force. As shown in Figure 2, in 2010, many younger women were not in the labor force because of caregiving responsibilities. This finding is consistent with the extensive literature that indicates that women's labor supply is more negatively affected by childcare than men's, in China and elsewhere (Dong 2020; Feng and Zhang 2018; Lee and Yeung 2021; Lumsdaine and Vermeer 2015; Wang and Zhang 2018). In addition, the rising need for eldercare is still unfolding as population aging in China continues to accelerate (Liu et al. 2010; Luo and Chui 2019). While it remains unclear to what extent eldercare demands would reduce women's LFP, the institutionalized gender gap in retirement age might perpetuate the unequal gendered division of household labor.

The current retirement system not only disproportionately disadvantages women but is also unsuitable given the population realities in China (Cai et al. 2018; Feng et al. 2019). This study's findings highlight the need to incorporate a gender perspective into the ongoing discussions on retirement and pension reform. One important step could be to harmonize the retirement age and ensure equal labor rights for older workers before the consequences of population aging fully manifest. Policies aimed at preventing age discrimination should also be considered. An equal if not greater effort should be made to alleviate the caring duties borne disproportionately by women.

Acknowledgments

I am grateful to Emily Hannum, Pilar Gonalons-Pons, Xi Song, Michel Guillot, Annette Lareau, Rebecca Schut, Andrew Fenelon, Yuying Tong, three anonymous reviewers, and Demography editors and editorial assistants for their generous suggestions, comments, and assistance. I also thank the participants at the 2021 International Population Conference, the 2021 annual meeting of the Population Association of America, the 2021 annual meeting of the International Chinese Sociological Association, and the seminar on Mechanisms that Perpetuate or Reduce Inequality by Class, Race & Gender at UPenn Sociology for their insightful comments during the presentation of this research. Any errors are my own.

This research uses data from the China Health and Nutrition Survey (CHNS). I am grateful for research grant funding from the National Institutes of Health (NIH), the Eunice Kennedy Shriver National Institute of Child Health and Human Development (R01 HD30880 and R01 HD38700), the National Institute on Aging (R01 AG065357), the National Institute of Diabetes and Digestive and Kidney Diseases (R01 DK104371 and P30 DK056350), the National Heart, Lung, and Blood Institute (R01 HL108427), the NIH Fogarty grant (D43 TW009077), the Carolina Population Center (P2 CHD050924 and P30 AG066615) since 1989, the China–Japan Friendship Hospital, the Ministry of Health (CHNS 2009), the Chinese National Human Genome Center at Shanghai since 2009, and the Beijing Municipal Center for Disease Prevention and Control since 2011.

Notes

References