Diverging destinies: changing trends of induced abortion in China

In the era of the second demographic transition, “diverging destinies” theory states that women with the most economic opportunities often obtain gains while women with the least economic opportunities suffer from losses. Using nationally representative data from the 2017 China Fertility Survey, we attempted to examine if diverging destinies theory can explain differences in the incidence of induced abortion among women with differing economic opportunities. We found that women with good economic opportunities (i.e., well-educated and holding urban hukou) were more likely to have induced abortions between 1980 and 2010 and less likely to do so between 2015 and 2017. Moreover, younger cohorts were far more likely to have induced abortions at a younger age and before marriage. These findings provide supportive evidence for diverging destinies in induced abortion scenarios in China. We thus advocate that sexual and reproductive health policy making shows an overarching concern for the interests of disadvantaged groups.


Introduction
As one of the most important facets of women's sexual and reproductive health, induced abortion has received intensive attention from both scholars and the general public. The World Health Organization reports that there are almost 121 million unintended pregnancies worldwide each year, and that 6 out of 10 unintended pregnancies end in induced abortion. 1 Notably, both the unintended pregnancy rate and abortion rate have remained at a high level during the past few decades. According to a recent work by Bearak et al. (2020), the worldwide unintended pregnancy rate for women aged 15-49 years was 79 per 1000 between 1990 and 1994 and 64 per 1000 between 2015 and 2019. Of these unintended pregnancies, 51% were aborted between 1990 and 1994 and 61% between 2015 and 2019.
Although increasing access to and the use of contraception has reduced the number of induced abortions, severe heterogeneity in abortions still exists across different regions, cultures, cohorts, and socio-economic status (SES). Prior research (e.g., Sedgh et al., 2014) has suggested that women who use contraception may unexpectedly become pregnant because their contraceptive methods fail to work or because they use them inconsistently or improperly. In developing countries, the failure to use contraceptives often leads to induced abortions. Notably, younger women are particularly vulnerable to peer pressures to have sex without contraception and thus are more likely to engage in unprotected sex and have abortions. McLanahan's (2004) "diverging destinies" framework is useful to study trends and heterogeneity of induced abortion. The theory is grounded in second demographic transition (SDT) theory and maintains that women with the most economic opportunities-such as women with higher levels of education-are more likely to enjoy advantages like delayed childbearing and increased employment opportunities in the SDT. An important question is whether women who have better economic opportunities may experience sexual and reproductive health benefits during the SDT. This study is interested in the following question: do women with better economic opportunities demonstrate lower rates of abortion? Theoretically, this question is twofold. On the one hand, well-educated women are more capable of acquiring sexual and reproductive knowledge and, accordingly, engaging in protective sex (e.g., using contraceptives), which prevents unintended pregnancies and therefore abortions. On the other hand, women with higher levels of education often have better employment, higher wages, and demanding professional roles, and are more likely to face difficult trade-offs between family and professional life. Compared with women with lower levels of education, well-educated women are more likely to intentionally delay childbearing and even terminate pregnancies to take advantage of professional opportunities (e.g., promotions) (Chen, 2005;Wang & Jiang, 2022). The answer to this study's research question, thus, has meaningful empirical and theoretical implications.

3
Diverging destinies: changing trends of induced abortion… China offers a unique, interesting, and important context in which to answer this question. Given China's huge population, induced abortion is a significant issue. Government statistics show that China's average number of abortions annually exceeded 10 million during the 1980s and early 1990s; furthermore, despite China's astonishing economic achievements over the past few decades, abortions per year still exceeded 9.5 million from 2014 to 2019 (National Health Commission of the People's Republic of China, 2020)-more than the total population of half of the countries in the world. Clearly, the Chinese context can provide useful insights into abortion patterns in large populations.
It is important to note that China's family planning policy may influence women's induced abortion patterns. To curb the country's soaring population growth rate, the Chinese government implemented the "one-child policy" in 1979, which required some families to have only one child. Accordingly, some pregnancies had to be aborted to avoid violating the one-child policy (Chen, 2007). This policy was adjusted in 1984 to satisfy the needs of families in some rural areas by allowing couples whose first child was a girl to have a second child. Starting in 2013, the family planning policy was relaxed nationally-families were allowed to have two children if at least one parent was from a one-child family. In 2016, a two-child policy was implemented nationally. Five years later, in 2021, China introduced a three-child policy, allowing pregnancies that would otherwise have been aborted to avoid violating the fertility policy.
China's case is also meaningful from a theoretical perspective. The diverging destinies phenomenon was first observed in Western industrialized societies and its applicability to East Asian societies remains uncertain. Thus, an empirical investigation of trends in the abortion-related diverging destinies of women with different economic opportunities in China is both theoretically necessary and practically feasible.

Diverging destinies framework
With rapid socioeconomic development since the 1970s, family formation and sexual behaviors have profoundly changed in Western societies. To understand these changes, the second demographic transition theory was first introduced in the 1980s to describe newly emerging living arrangements and sexual activities in Western industrialized societies that could not be explained by first demographic transition theories (van de Kaa, 1987;Lesthaeghe, 1995). These changes included delays in fertility and marriage, increases in non-marital cohabitation and births outside marriage, high levels of divorce, and increased rates of maternal employment. The SDT has occurred or is occurring in many, if not all, regions of the world in different ways (Lesthaeghe, 2022;Raymo, 2022;Yu & Xie, 2015. For instance, China has been exhibiting similar trends regarding marriage postponement, declines in fertility, and the rise of premarital cohabitation and sex. However, nonmarital births are very rare in China and marriage remains the main institution for procreation (Raymo et al., 2015;Yu & Xie, 2022). The coexistence of premarital sex and pregnancy and a low nonmarital birth rate in China during its SDT may cause shotgun marriages, i.e. marriage that is catalyzed by a pregnancy, and abortions (Che & Cleland, 2003;Raymo & Iwasawa, 2017).
Grounded in SDT theory, McLanahan (2004) developed the diverging destinies framework to conceptualize trends in the level of parental resources provided to children by mothers with different economic opportunities. Diverging destinies theory reasons that "changes in family behavior are increasing parental resources for children with well-educated mothers and reducing parental resources for children with less educated mothers" (McLanahan & Jacobsen, 2015). Evidence for the phenomenon of diverging destinies has been found in the United States, Japan, Australia, and countries in the European Union (McLanahan & Jacobsen, 2015). For example, in the United States, nonmarital births and single motherhood are more common among less-educated women due to less consistent and effective contraceptive use and less incentive to delay childbearing (England, 2016;England et al., 2012). Meanwhile, a recent study in China by Qian and Jin (2020) provides evidence of diverging destinies by suggesting that less-educated women are more likely to experience premarital pregnancies and that the educational gradient has widened across birth cohorts.
Given that sexual liberalization may have occurred for all women regardless of socioeconomic status (Pan, 2007) and that disadvantaged women may have limited access to reproductive health education and services, we reason that the likelihood that premarital sex will lead to premarital pregnancy and induced abortion differs by level of education. Therefore, we hypothesize that the diverging destinies framework is applicable to induced abortion in China: the abortion-related behaviors of women with the most economic opportunities (e.g., education) are likely to differ significantly from those of women with the least economic opportunities.

Induced abortion in China
China's high induced abortion rate is closely linked to strict enforcement of its family planning policy, rapid declines in fertility, and a rising preference for sons among Chinese couples (Chen, 2007). First, in 1979 China began requiring that each married couple have only one child. The one-child policy stipulated whether and when a couple could have a child and what contraceptive methods they should use (Chen, 2007). In 1984, China modified the one-child policy in response to the cultural preference for sons and allowed married couples in some rural areas whose first birth was a daughter to have a second child; this was called the one-child-and-a-half policy (Gu et al., 2007). Most families in urban areas remained subject to the one-child policy over time, while most families in rural areas were subject to the one-child-and-a-half policy. Thus, induced abortion became an important means of reducing unplanned, unwanted, or unauthorized births (Qiao, 2002). Second, the son preference in China was, to some extent, strengthened by the implementation of the strict family planning policy (Zeng et al., 1993). The belief that sons can ensure the continuity of the family line 1 3 Diverging destinies: changing trends of induced abortion… is deeply embedded in China's traditional agrarian culture, and many Chinese families want sons. As the increased availability and use of diagnostic technologies, such as B-mode medical ultrasound, made it possible to identify the sex of a fetus, sex-selective abortions became more common. Many women chose to have an induced abortion because they did not want to have a child of a particular sex, especially a girl (Yang, 2009;Zeng et al., 1993). Third, the desire to have children has gradually declined among women since the implementation of the family planning policy and in response to rapid economic development (Qiao, 2002). By regulating the maximum number of children in families, family planning policies have liberated women from traditional Chinese stereotypes that reduce their identities to motherhood. At the same time, economic development provides women with more opportunities like those men have to pursue careers. In this context, induced abortion has become a common way to deal with an unwanted pregnancy (Zeng et al., 1993). Finally, poor quality of sex education and reproductive health services are also important drivers of induced abortion, especially for younger women. In step with Confucian culture, Chinese parents seldom talk about sex and sexuality with their kids . Moreover, sex education at school and reproductive health services in society lag behind the needs created by actual sexual activities (Watts, 2004). This lag often leads to unexpected pregnancies and induced abortion.
After implementation of the one-child family planning policy in China, induced abortion demonstrated unique patterns related to the policy. Prior research has shown that women with a high SES, such as higher levels of education and urban hukou, were more likely to have an induced abortion (Qiao, 2002), in part because the policy's regulations varied across different groups. For instance, women with an urban hukou who worked in state-owned companies, government agencies, or public institutions could lose their jobs or opportunities for promotion if they violated the policy. Therefore, these women were much more likely to abort a pregnancy that might violate the policy than were women with a relatively low SES. Moreover, because women with a high SES are more able to focus on the quality of births rather than the quantity; when a strict family planning policy was in place, these women were more likely to abort a pregnancy that might violate the policy. In summary, previous research has established two things about induced abortion in China: (1) induced abortion is affected, if not largely driven by, family planning policies; thus, it is reasonable to conjecture that women's induced abortion behaviors may change if family policy changes and (2) women's induced abortion behaviors are highly heterogeneous across different levels of education, birth cohorts, and hukou statuses.
As noted above, after 2013 the one-child policy was adjusted and gradually relaxed to a two-child policy and finally three-child policy. Roughly concurrent with the introduction of the one-child policy, China began the economic reforms in 1978, and since that time the country has experienced a sexual revolution characterized by a rise in premarital sex (Parish et al., 2007), especially among lesseducated women (Qian & Jin, 2020). Furthermore, information about contraceptive methods and sexual and reproductive health have been widely disseminated among citizens over the past several decades and greater importance has been attached to sex education nationally. Together, these recent changes may be reshaping women's abortion behaviors.

Data
Data were obtained from the 2017 China Fertility Survey (CFS), a nationally representative sample of 249,946 women aged 15-60 years. The survey was organized by the National Health and Family Planning Commission of China 2 in 2017. These data were collected through a stratified three-stage probability-proportional-to-size sampling method, which was employed on a sampling frame of women aged 15-60 with Chinese nationality in mainland China from all 31 provinces (regions/municipalities) and the Xinjiang Production and Construction Corps (Zhuang et al., 2018).
The sampling units were townships (xiang, zhen, or jiedao) in the first stage, urban/rural communities (cunweihui or juweihui) in the second stage, and individual respondents in the third stage. The survey collected detailed information about women's pregnancy histories regardless of their marital statuses and other sociodemographic characteristics (Qian & Jin, 2020). The 2017 CFS asked about the year and month each pregnancy ended and whether it ended in a live birth, stillbirth, miscarriage, or induced abortion (hereafter simply "abortion"). Given that women born after 1990 were still too young to experience marriage or pregnancy, we dropped 37,440 women born between 1990 and 2002 (many of whom reported no history of pregnancy). We also excluded 244 women who became pregnant at 14 years of age or younger. We further dropped 5257 women who had never been pregnant and 358 women with missing data for the variables used in the analysis. Because very few of the women in our sample had been pregnant before 1980 and the numbers were too few to be representative of trends, we only kept records for pregnancies that occurred after 1980 and analyzed patterns between 1980 and 2017. Eventually, the sample we used for the main analyses consisted of 470,594 pregnancies from 206,402 women.

Empirical strategy
Although the 2017 CFS is a cross-sectional dataset, it asked women about their pregnancy histories, meaning that respondents could report multiple pregnancies up to 2017. By taking advantage of this unique feature, we reshaped the data into an "individual-pregnancy" panel-like dataset. Our dependent variable was whether a pregnancy ended in an abortion. We defined the value as 1 if a pregnancy had ended in abortion prior to the time when the survey was conducted and 0 for other outcomes (including live birth, stillbirth, miscarriage, etc.). The key 1 3 Diverging destinies: changing trends of induced abortion… independent variables included the date when a pregnancy ended (either as an abortion or other outcome), the women's birth cohort, her education level, her age and marital status when a pregnancy ended, and sex structure of the live birth(s).
The time period was measured using 5-year spans between 1980 and 2017, however, the most recent and final period had only three years (2015, 2016, and 2017). Birth cohort was classified into four cohorts : 1956-1959, 1960-1969, 1970-1979, and 1980-1989, and these were referred to as the "1950s cohort," "1960s cohort," "1970s cohort," and "1980s cohort." The 1950s cohort begins in 1956 because the oldest women in our sample were born in that year. For participants born after 1960, we used 10-year intervals to measure birth cohort. A 10-year birth cohort is a meaningful measure in the Chinese context because it speaks to popularly accepted identities based on the decade of birth (Qian & Jin, 2020;Sabet, 2011). Notably, birth cohorts are usually used to reflect changes and period effects over time. Education, an indicator commonly used to measure an individual's socio-economic status, was grouped into four categories: less than high school, high school, vocational college, and bachelor's degree (BA/BS) or above. In this study, education represents a woman's economic opportunities, and is necessary to situate her in the diverging destinies framework. Household registration type (i.e., urban or rural hukou) in China has important social significance, especially in terms of fertility-related behaviors. People with urban hukou generally have a higher SES than those with rural hukou. Also, the fertility policies to which married couples were subject differed by hukou type prior to the introduction of the universal two-child policy. Women's age when a pregnancy ended was a continuous variable in our analysis. To graphically display women's likelihood of having an abortion across different ages, we predicted the probability of abortion at ages 15, 25, and 35 years old. Premarital or marital pregnancy was identified using the date of first marriage and the date when a pregnancy ended. Pregnancies ending in abortion prior to marriage were labelled as "premarital abortion" and those ending in abortion within marriage as "marital abortion".
Our model further included women's ethnicity (Han or ethnic minority), number of siblings, birth order, work unit (danwei in Chinese), migrant status, and province of residence as control variables to account for potential endogeneity issues. Given the panel-like data structure at the individual-pregnancy level, we used logit regression models to predict the likelihood of abortion while clustering standard errors at the individual level.
Notably, we shifted our analysis unit to the individual level instead of the pregnancy level when we analyzed the association between son preference and abortion. First, we restricted our sample to women who had at least one live birth and a subsequent pregnancy. Then we examined how the sex of the prior live birth was linked with the outcome of the next pregnancy. We further restricted the sample to women who had at least two live births and a subsequent pregnancy and examined how the sex structure of prior live births affected the probability of abortion of the next pregnancy.
The demographic and socio-economic characteristics of the respondents are presented in Table 1.

Descriptive results
Table 2 presents the model-free evidence for women's abortions and the heterogeneity related to their SES (e.g., education, hukou) and cohorts. Three indicators were adopted to describe women's abortions: (1) the percentage of women who had experienced abortion(s) was used to describe the likelihood of having an abortion among women in China; (2) the average number of abortions was used to measure the overall level of abortion; and (3) the abortion ratio-defined as

3
Diverging destinies: changing trends of induced abortion… the total number of abortions over the total number of pregnancies-was used to represent the average percentage of abortions across all pregnancies for women in our sample.
Overall, across all four cohorts in our sample, 25.3% of women had experienced at least one abortion prior to the time of the survey and each woman had 0.36 abortions on average. The abortion ratio was 15.6 per 100 pregnancies. Trends of abortion varied across socio-demographic groups. Regarding birth cohort, women in the 1950s cohort were the least likely to have abortions in terms of all three indicators. Women in the younger cohorts were more likely to have abortions; 23.4% of women had abortions in the 1950s cohort, 24.4% in the 1960s cohort, and 27.6% in the 1970s cohort. The average number of abortions was 0.34 in the 1950s cohort, 0.35 in the 1960s cohort, and 0.40 in the 1970s cohort. Similar trends were observed for the abortion ratio across these cohorts. We did not interpret the descriptive evidence for the 1980s cohort because it fell out of the patterns described above due to truncation issues. That is, compared with the 1950s, 1960s and 1970s cohort, the 1980s cohort had the highest proportion of women who were still sexually active when surveyed and thus had the potential to have future abortions that would not be observed (i.e. were truncated).
Roughly, we observed an "inverted-U" shaped relationship between women's education and abortion. Women with the lowest levels of education (i.e., less than high school) and women with the highest levels of education (i.e., BA/BS or above) had a smaller likelihood of having an abortion and were likely to have a smaller number of abortions. In contrast, women with high school and vocational college education were more likely to have an abortion, a higher number of abortions, and a higher abortion ratio. Women with urban hukou were more likely to have an abortion than their rural counterparts. For example, 32.6% of women with urban hukou had an abortion, while only 22.6% of women with rural hukou had an abortion. Women with urban hukou also had a higher abortion ratio (25.3%) than women with rural hukou (13%). Similar patterns were observed for the average number of abortions.
When exploring the role of son preferences in abortions, we reshaped our sample by constructing two subsamples: women who had at least one living child (148,908 observations) and women who had at least two living children (34,598 observations). For both subsamples, we calculated women's likelihood of having an abortion across five conditions: one living boy, one living girl, two living boys, two living girls, and one living boy and one living girl (see Table 3). Among women with one living child, the probability of abortion (28.3%) was higher if the child was a boy than if the child was a girl (19.4%). If a woman's first child was a boy-that is, if the preference for a son had been satisfied-then she was equally likely to have a boy (33.5%) or a girl (33.6%) for her second child. In contrast, if the first child was a girl-that is, if the preference for a son had not been satisfied-then the woman was much more likely to have a boy (43.3%) than a girl (32.6%) for her second child. Women whose first two living children were girls-that is, if the preference for a son had not been satisfied after two attempts-demonstrated the lowest probability of having an abortion (13.2%) and the highest probability of having a boy for their third child (54.4%), indicating strong son preferences. In addition to son preferences, we observed a preference for a family composed of one son and one daughter; that is, women whose first two children were a boy and a girl were the least likely to have a third child and demonstrated the highest probability of having an abortion to terminate a third pregnancy (31.9%).

Regression results
This section reports results of logit regression models to identify changes in the patterns of abortions in China over time and across women's different economic 1 3 Diverging destinies: changing trends of induced abortion… opportunities, such as education and hukou status. As a commonly practiced method, the temporal perspective enables us to observe how abortion patterns have changed over time and enables comparisons with prior research in the West. Moreover, we analyzed the roles of the preference for a son, age, and marital status in abortion, all of which are commonly explored in prior abortion research. Because it is often difficult-if not impossible-to interpret the coefficients of the interaction terms in nonlinear models, we calculated predicted probabilities to interpret interaction terms in logit regression, following the practices recommended by Mize (2019).

Temporal trends
We charted the predicted probability of abortion from 1980 to 2017 to uncover the trends for abortion in China. As shown in Fig. 1, the trend for abortion in China roughly approximates an inverted-U shape, and includes three stages over the past four decades. In the first stage, which extends from 1980 to 1995, the predicted probability of abortion increased rapidly from 0.11 to 0.18. This rapid increase in abortion coincides with the implementation of the one-child family planning policy; many pregnancies were aborted to comply with this policy (Chen, 2007). In the second stage, which extends from 1995 to 2014, the predicted probability remained stable around 0.18; during this time, the one-child policy had become widely accepted, contraceptive methods were widely accessible, and women were likely able to adjust their sex and childbearing behaviors to comply with the policy and avoid abortion.
In the third stage, which extends from 2015 to 2017, the predicted probability of  Table 4 by time period abortion decreased rapidly to 0.15, largely because China relaxed its family planning policy.

Abortion patterns by time period and education
We now present the temporal trends of abortion by women's education (see Fig. 2) to demonstrate how the likelihood of abortion varies with women's economic opportunities. Overall, with the exception of women with the lowest level of education (less than high school), women who had completed high school, vocational college, or a BA/BS or above again showed an inverted U-shape for the predicted probability of abortion. The probability of abortion for women with the lowest level of education followed a slow increasing trend from 1985 to 1999, stabilized between 1999 and 2014, and sharply increased to a high level between 2015 and 2017. The diverging destinies framework assumes that women with better economic opportunities, such as higher levels of education, often enjoy advantages, such as a lower rate of abortion. However, this study found that the most educated women (i.e., women with a BA/BS or above) became increasingly likely to have an abortion between 1985 and 2004; that is, they demonstrated an increasing trend over time. In short, more educated women were more likely to have an abortion. However, it is important to note that the differences between the three groups of better educated women (high school, vocational college, BA/BS or above) were small while the differences between these women and the least educated women (less than high school) were large in all time periods. This pattern started to change between 2005 and 2009, when the probability of abortion declined dramatically among the most educated women. As a result, the educational differences between women who had completed high school and those with a BA/BS or above were no longer statistically significant. The educational pattern underwent further changes in the two most recent periods (2010-2014 and 2015-2017): women with a BA/BS or above rapidly became less likely to have an abortion and their differences with women with the lowest levels of education substantially decreased. The difference between the groups with high school and vocational college was insignificant. The probabilities of the four groups tended to converge between 2010 and 2014. Finally, the relationship between women's education and the likelihood of abortion completely reversed between 2015 and 2017, when a higher level of education became associated with a lower probability of having an abortion. During this period, the most-educated women had the lowest probability while the least-educated women had the highest probability of abortion.
In sum, during the 35-year period from 1980 to 2015 when China's family planning policy was strictly applied, women with higher levels of education were more likely to have abortions than their less-educated counterparts. However, after 2016, when the two-child policy came into effect, the relationship between women's education and abortion immediately reversed; that is, more educated women become less likely to have an abortion. This finding supports the assumptions of the diverging destinies framework.

Abortion patterns by time period and hukou
In China, hukou is another important indicator of women's economic opportunities. Women with urban hukou usually have access to better resources, such as better job opportunities, higher salaries, and easier access to high quality medical care. As shown in Fig. 3, women with urban hukou and women with higher levels of education demonstrated similar trends between 1980 and 2017. For both groups of women, the likelihood of abortion continued to rise and peaked between 1995 and 1999 and then began to rapidly decline. Correspondingly, women with rural hukou and women with lower levels of education also demonstrated similar trends; their likelihood of abortion reached its peak point between 1995 and 1999, then remained generally unchanged until around 2010-2014, after which it increased slightly between 2015 and 2017. Notably, for about 30 years before the 2010 to 2014 period, women with urban hukou were significantly more likely to have an abortion than their rural counterparts. The gap between these two groups started to narrow as the likelihood of abortion for women with urban hukou began to decline during the 2005 to 2009 period. By the 2015 to 2017 period, the gap had closed and the hukou-based abortion pattern had changed direction; that is, women with urban hukou were less likely to have an abortion than their rural counterparts.

Abortion patterns by birth cohort and education
Birth cohort is an important vantage point from which to understand the economic opportunities available to different generations. Figure 4 presents the predicted probabilities of abortion by birth cohort and education. Among the 1950s and 1960s cohorts, better-educated women,  Table 4 by time period and hukou type Fig. 4 The predicted probability of abortion by birth cohort and education. Note: the probability of a pregnancy ending in abortion is predicted based on Model 4 of Table 5 by cohort and education especially those with postsecondary education, were more likely than less-educated women to have an abortion. Meanwhile, the likelihood of abortion was higher in the 1960s cohort than in the 1950s cohort for all educational groups. Women with a BA/ BS or above (i.e., the most educated women) had the highest likelihood. The differences between the four educational groups were the greatest in the 1960s cohort. This pattern started to change in the 1970s cohort: the differences between the groups that had completed high school and postsecondary education lessened. The educational pattern underwent further changes among the 1980s cohort with the probability of abortion decreasing for women who had completed high school or vocational college, or had a BA/BS or above. For the 1970s cohort, women with less than a high school education were the least likely to have an abortion, while the difference in the likelihood of abortion between high school and postsecondary graduates declined. In the 1980s cohort, women with at least a bachelor's degree had the lowest likelihood of having an abortion among all educational groups, including the least-educated group. Notably, the likelihood of abortion increased among the least-educated women and strongly decreased among the most-educated women in the more recent cohorts. Figure 5 depicts variations in the effect of hukou status on abortion across birth cohorts. Consistent with prior findings, women from all four cohorts with urban hukou were more likely to have an abortion than their rural counterparts. However, the differences in the probability of abortion between urban and rural women decreased among younger cohorts due to two converging trends. First, the probability of having an abortion for women with Fig. 5 The predicted probability of abortion by birth cohort and hukou type. Note: the probability of a pregnancy ending in abortion is predicted based on Model 5 of Table 5 by cohort and hukou type rural hukou increased in younger cohorts, and second, the probability of having an abortion for women with urban hukou declined relatively rapidly in younger cohorts.

Abortion and its causes
Beyond the diverging destinies framework adopted in this study, prior research identified other important causes of abortion in China. To bring this study more directly into conversation with existing literature, we analyzed the prevalence of the three most common causes of abortion identified by existing studies in our sample; namely, age, premarital pregnancies, and the preference for a son. Figure 6 shows the probability of abortion by birth cohort and women's age when a pregnancy ended. In the 1950s and 1960s cohorts, older women were more likely to abort a pregnancy and differences between the two cohorts in the probability of abortion for different ages were small. Specifically, in the 1950s cohort, the predicted probabilities of aborting a pregnancy at ages 15, 25, and 35 were 0.13, 0.138, and 0.146, respectively. In the 1960s cohort, these probabilities were 0.142 at age 15, 0.145 at age 25, and 0.147 at age 35. However, in the 1970s and 1980s cohorts, younger women were more likely to abort a pregnancy and differences in the probability of abortion for different ages continuously increased for younger cohorts. Notably, in the 1950s cohort, the probability of abortion for 15-year-old women was less than 0.1. This probability rose to 0.206 in the 1970s cohort and 0.298 in the 1980s cohort. In contrast, in Fig. 6 The predicted probability of abortion by birth cohort and age at the end of the pregnancy. Note: the probability of a pregnancy ending in abortion is predicted based on Model 6 of Table 5 by cohort and age at the end of the pregnancy 1 3

Birth cohort and age
Diverging destinies: changing trends of induced abortion… the 1950s and 1960s cohorts, the probability of abortion for 35-year-old women was approximately 0.15. This probability dropped below 10% in the 1980s cohort. Figure 7 presents the likelihood of abortion by birth cohort and marital status when a pregnancy ended. Unmarried women were more likely to have an abortion; notably, this trend generally showed increases over time across all cohorts. The predicted probability of premarital abortion increased from 0.142 in the 1950s cohort to 0.365 in the 1980s cohort. In contrast, the likelihood of aborting a pregnancy when married stayed relatively stable across the three older cohorts and decreased significantly in the 1980s cohort. Specifically, the predicted probability of a marital abortion hovered around 0.15 in the 1950s, 1960s, and 1970s cohorts and decreased to 0.13 in the 1980s cohort. Figures 8a, b display how the effect of the sex structure of women's living child(ren) varies across cohorts. Women across all cohorts who had only one or two girls were much less likely to abort a subsequent pregnancy than those who had one or more boys, indicating the persistence of the preference for a son in fertility-related behaviors. However, the gap in the likelihood of women with one or two boys and women with one or two girls to have an abortion narrowed slightly among the 1980s cohort. This suggests that the preference for a son may be weakening. Fig. 7 The predicted probability of abortion by birth cohort and marital status at the end of the pregnancy. Note: the probability of a pregnancy ending in abortion is predicted based on Model 7 of Table 5 by cohort and marital status at the end of the pregnancy Fig. 8 a The predicted probability of abortion by birth cohort and sex of the first live child. Note: the probability of a pregnancy ending in abortion is predicted based on Model 8 of Table 6 by cohort and sex of the first child. The dependent variable is whether or not the subsequent pregnancy was aborted. b The predicted probability of abortion by birth cohort and sex composition of the first two living children. Note: the probability of a pregnancy ending in abortion is predicted based on Model 9 of Table 7 by cohort and sex of the first child. The dependent variable is whether or not the subsequent pregnancy was aborted 1 3

Birth cohort and son preferences
Diverging destinies: changing trends of induced abortion…

Discussion
In this study, we focused primarily on women's probability of abortion when graphically displaying the trends and discussing the key findings of this study. We also explored the heterogeneity of women's abortion behaviors from various perspectives such as women's education, hukou status, cohorts, age, marital status, etc. We selectively highlight three findings in the discussion section for a meaningful conversation with literature on women's abortion and the diverging destinies theory.
First, women's probability of having an abortion demonstrates an inverted-U shape over a 37-year time span from 1980 to 2017. The probability of abortion was low in 1980, then increased sharply between 1985 and 1995, remained stable at a high level for the twenty years between 1995 and 2015, and then decreased dramatically afterwards. Notably, the probability of abortion decreased dramatically right after the nationwide relaxation of the family planning policies. Thus, it is reasonable to conjecture that the sharp increase and high abortion rate in China before 2015 were partially, if not largely, attributed to the family planning programs especially the strict one-child policy. This study presents empirical evidence of an initial increase in the abortion rate followed by a lengthy period during which the rate remained high. The findings demonstrate, from a historical perspective, the necessity of making adjustments to family planning policies that have been introduced since 2015. It is worth noting that China has relaxed family planning policies nationwide to encourage childbearing behaviors. How the changes will affect women's abortion-related behaviors and how the effects may vary across cohorts, age groups and regions is worth close attention.
This study is also among the first using nationwide representative data to discover the inverted-U shape trend of women's abortion in China. As one of the most important facets of women's sexual and reproductive health, abortions have drawn intensive attention from academia (e.g., Chen, 2005Chen, , 2007Jiang et al., 2017;Wang & Jiang, 2022). Prior research commonly focuses on one unique perspective-such as rural China, women at certain age, and floating people-to disentangle women's abortions in China. For instance, Chen et al. (2020) focused on abortions of women aged 20-39 in rural China with data collected from five villages of five cities. Jiang et al. (2017) focused on post-abortion services with data from three provinces in China. Guo et al. (2020) focused on high school students between 15-19 years of age in rural areas of western China. Our study contributes to the abortion literature by documenting changing abortion trends at the national level in China from a longitudinal perspective.
Second, during the period between 1985 and 2014, women with good economic opportunities (e.g., women with high levels of education and urban hukou) were more likely to have an abortion than women with poor economic opportunities. This trend then reversed between 2015 and 2017-women with good economic opportunities became less likely to have an abortion than women with poor economic opportunities. Thus, this study contributes to the diverging destinies theory in the context of sexual and reproductive health in China from two aspects.
On one hand, our study shows that women with good economic opportunities do not necessarily always stay on the advantageous side. The core of diverging destinies theory states women with good economic opportunities say education tend to stay on the advantageous side in their social lives such as parenting resources (McLanahan & Jacobsen, 2015) or sexual and reproductive health status (England, 2016;England et al., 2012). This study evidences women with better economic opportunities were more likely to have abortions when China's family planning programs were in effect and the trend reversed right after the fertility policies relaxed. This effect likely occurred because fertility policy controls the fertility behaviors of women from privileged backgrounds by impacting their professional opportunities and their social welfare (Chen, 2005). Meanwhile, compared to women with poor economic opportunities, those with good economic opportunities desired smaller families (and thus were more motivated to regulate the size of their families) and had better access to abortion knowledge and services. As a result, women with good economic opportunities are more likely to have an abortion.
On the other, the reversed trend of abortion patterns that the economically disadvantaged groups are more likely to experience abortions during the most recent period also provides the first non-western evidence for the theory of diverging destinies during the SDT. As established above, both the theory of the SDT and the later theory of diverging destinies were developed in Western contexts. The evidence from the Chinese context offered by this study clarifies and extends these two theoretical perspectives. As a result of urbanization, modernization, and globalization, the SDT has been underway in China over the past few decades, characterized by a rapid decline in fertility; the postponement of marriage; and increases in premarital cohabitation, sexual activities, and divorces (Yu & Xie, 2015. The changes China is undergoing are very similar to those that continue to play out in Western societies. Nevertheless, the Chinese context is unique because the country's traditional culture continues to impact perceptions, beliefs, values, and behaviors in China today. As with fertility behaviors, marriage and childbearing are closely linked and nonmarital births are very rare in China; accordingly, recent increases in sexual activities and premarital pregnancies may be resulting in more abortions (Che & Cleland, 2003;Raymo & Iwasawa, 2017). Given the fact that economically disadvantaged groups have a higher risk of premarital pregnancy (Qian & Jin, 2020), they disproportionately bear the health consequences of abortion. This finding from China confirms the diverging destinies phenomenon and suggests that changes in the family as a result of the SDT may be a potential mechanism for the reproduction of social inequality in non-Western contexts (Qian & Jin, 2020).
Third, this study found that economically disadvantaged, younger and unmarried women were more likely to have abortions. Because China has some of the least restrictive abortion laws in the world (Huang, 2019) and because marriage and childbearing remain closely linked and nonmarital childbearing is illegal, stigmatized, and rare in China (Raymo et al., 2015), a higher likelihood of premarital pregnancy among less-educated women also means that this group is more likely to have more abortion in China (Qian & Jin, 2020;Wu et al., 1992). Thus, this may disproportionately harm less-educated women's reproductive health. Furthermore, young and unmarried people have limited access to the reproductive health education and contraceptive services and thus generally lack awareness of safe sex in China (UNE-SCO & UNFPA, 2018). According to a recent study by Luo et al. (2020), sexual activity outside marriage and unprotected sexual behaviors have increased among Chinese adolescents over the past decades, and their needs for reliable contraception remain largely unmet.
From a policy making perspective, our findings are a reminder that policies and related health services should address both pre-abortion and post-abortion needs. There is an urgent need to develop and implement effective policies that are friendly to economically disadvantaged women, younger women and unmarried women. For instance, it is common practice to organize lectures about sexual and reproductive health to be given in schools, community hospitals, or online, but this may not be an effective way to reach young women (Guo et al., 2020). Thus channels-such as live-streaming platforms, social network platforms like WeChat and short-form video platforms that align with the information acquisition habits of young women are better options for the provision of sexual and reproductive health information. Moreover, the content of such information should be targeted at helping women reduce unsafe sexual behaviors and preferences that may lead to unplanned pregnancies and abortions. Additionally, abortion before marriage, in Chinese traditional culture, is often interpreted as a shameful thing (Gu, 2021). Thus, it is important to enhance services of well-trained healthcare professionals who have expertise in working with younger and unmarried women (Zou et al., 2022).
This study had several limitations. First, when the China Fertility Survey (from which we drew our data) was conducted in 2017, the younger cohorts were still too young to marry or get pregnant. Our data showed that 46.4% of women born in the 1990s were married or had pregnancy experiences, which highlighted a truncation issue in the statistical analysis. Accordingly, the 1990s cohort was not included in our final analysis. However, the 1980s cohort, the youngest cohort included in our analyses, may have similar truncation issues, but to a lesser degree. Among the 1980s cohort, 89.2% were married and 92.7% had been pregnant. We also conducted robustness checks to minimize the effect of the truncation issue by restricting the cohort to older women born between 1980 and 1984, 1985, 1986, or 1987; and the results held. Future surveys and studies should include more young individuals to extend our findings. Second, because we used a cross-sectional dataset, we did not have detailed historical records of women's SES before pregnancy, such as their occupation, work unit, income, and wealth. We only used education and hukou-two socio-economic indicators that are usually established prior to pregnancy and that remain relatively stable over a woman's life course-to measure women's economic opportunities. Third, despite the fact that the reported pregnancy histories provided valuable information for our analysis of abortion experiences, we also realized that using retrospective information concerning activities that happened, in some cases, years in the past could be subject to recall bias. To strengthen the generalizability of our findings on socioeconomic differentials related to abortion and avoid recall bias, future studies should use longitudinal data to deepen our understanding of the underlying changes of family behaviors in China.

Conclusion
Using nationally representative data from the 2017 China Fertility Survey, we constructed a panel-like data to discover changing trends of abortion behavior in China over a 37-year time span from 1980 to 2017. This study found that women's probability of abortion manifested as an inverted-U shape. Meanwhile, women with good economic opportunities (e.g., women with high levels of education and urban hukou) were more likely to have an abortion when a strict fertility policy was in effect from 1985 to 2015. However, this trend reversed in 2015 to 2017 after the fertility policy relaxed nationwide, indicating a directional shift: in the most recent period covered by the study, disadvantaged women are more likely to have an abortion. Similarly, cohort heterogeneity exists in terms of women's likelihood of abortion with different economic opportunities, evidencing the emergence of new patterns. For instance, the likelihood of abortion increased among the least-educated women and decreased among the most-educated women in the most recent cohorts; specifically, the relationship between education and abortion incidence reversed in the 1980s cohort, the youngest one in our sample. We also observed that the relationship between age at the end of a pregnancy and the likelihood of abortion shifted across successive cohorts: in the older cohorts, older women were more likely to abort a pregnancy; among younger cohorts, younger women were more likely to abort a pregnancy. Meanwhile, the rate of premarital abortion has continuously increased while the rate of marital abortion has continuously declined over time. Additionally, our study on abortion patterns sheds light on changes in preferences for the sex of a child among women. In our sample, the preference for a son was weakest in the youngest cohort. However, it is worth noting that sex preferences persist today, including the preference for a balanced composition of sons and daughters.
We strongly advocate that sexual and reproductive health policy making show an overarching concern for the interests of disadvantaged groups. We also want to stress the importance of providing effective sexual and reproductive information and services to younger women and unmarried women. Efforts to target these women should take into consideration how these groups acquire information and address their needs for abortion-related procedures.