Qualitative case examples illuminate patterns less visible in the quantitative data. We saw in the previous chapter, for example, that litigants sometimes amended, negotiated, and conceded their initial child custody claims. Such processes are more difficult to study quantitatively, and indeed are often altogether invisible in the written court decisions. However, quantitative analysis is the only means by which to paint a macroscopic picture of key patterns of judicial decision-making in huge and representative samples such as mine. Furthermore, some patterns emerge with clarity only from large-scale quantitative analyses. For example, the influence of the number and sex composition of children is difficult to tease out of a few qualitative examples but is striking in the full sample of child custody determinations.

Findings I present in this chapter confirm much of what we learned in the previous chapter. In particular, they reaffirm judges’ tendency to privilege defendants’ physical possession of children over plaintiffs’ claims of domestic violence. They also showcase additional extra-legal considerations that previous studies have found influence child custody determinations. Judges commonly split up siblings out of a sense of fairness to the parents and because fathers sometimes only pursue custody of their sons, which in turn allows or compels courts to grant custody of daughters to their mothers. Evidence of fathers’ “demand for sons” has also been found in the United States (Reference Dahl and MorettiDahl and Moretti 2008). When siblings do stick together, they are much more likely to go to the father than to the mother (Reference ZhangChen and Zhang 2015:28). Although it is an option available to judges, joint legal custody is exceedingly rare (Reference Hu and ShenHu and Shen 2016; Shandong Province Ji’nan Municipal Intermediate People’s Court Research Team 2018:184). Both siblings and only-children are often assigned to the parent matching their biological sex (Reference ZhangChen and Zhang 2015; Reference LangLang 2016; Reference DingZhao and Ding 2016). When granting custody of daughters to mothers, courts sometimes do so “in consideration of the biological characteristics of girls” (Reference DingZhao and Ding 2016:28–29). As one judge put it: “In a context of only-children and son preference, most families of fathers will not fight for custody of only-daughters, which causes courts to grant to mothers custody of 91.2% of only-daughters [in this particular sample of cases], which in turn causes them to feel abandoned” (Reference LangLang 2016:31). We will see that women’s child custody prospects were indeed determined to a large extent by both the number and sex composition of their children.

We will also see that these problems have been concentrated in rural areas, and that women’s outcomes in urban courts have been far better. According to one study, “in rural areas, the belief that men are superior to women is relatively influential, particularly among fathers and their families, who are therefore unenthusiastic about winning custody of daughters but regard winning custody of sons as relatively important” (Reference NiNi 2014:214). In another study, rural courts upheld the traditional norm of “sons stay with the husband’s family” (子留夫家; Reference ZhangChen and Zhang 2015:28–29).

Before turning to my empirical findings, let me briefly describe my analytical approach. I am ultimately interested in explaining variation with respect to one outcome: whether a court awarded child custody to a litigant. I will present findings from both descriptive bivariate analyses and multivariate regression analyses. In some analyses, I assess outcomes for plaintiffs and defendants simultaneously (taking as the unit of analysis the litigant, of which there are two per divorce case: one plaintiff and one defendant). However, given that both plaintiff and defendant in the same case can receive custody, most of the analyses take a plaintiff-oriented and then a defendant-oriented view (taking the case as the unit of analysis, from the perspective of either the plaintiff or the defendant only). In other words, I examine whether plaintiffs gained custody (which, in cases involving siblinged children, did not exclude defendants from also receiving custody) and then I examine whether defendants gained custody (again with plaintiffs also receiving custody in some cases). Differences between plaintiffs and defendants merit analysis. Moreover, mother–father differences emerge with greater clarity after looking at outcomes for plaintiffs and then for defendants. Assessing which parents received custody without attention to whether they were plaintiffs or defendants sometimes obscures differences between mothers and fathers simply because plaintiffs enjoyed an advantage over defendants and because mothers were overrepresented among plaintiffs.Footnote ¹

If all child custody orders were zero-sum (i.e., if custody of all children were granted to one side only), models for plaintiff outcomes would be inversely identical to models for defendant outcomes. Likewise, if the scope of analysis were limited to couples with only-children (custody of whom, in the absence of joint legal custody, was granted to one side only), models for plaintiff outcomes and models for defendant outcomes would also be inversely identical. Because the majority of couples in China have only one child, models for plaintiff outcomes and models for defendant outcomes are inversely similar. They are not inversely identical, however, because child custody is positive-sum for many couples with siblinged children, who are often split up with each parent receiving custody.

My multivariate analytical strategy is the same in this chapter as in Chapter 8. After estimating regression models, I use average marginal effects (AMEs) to assess the impact of litigant sex, domestic violence, and the physical possession of children on courts’ child custody orders, net of control variables included in the models. An AME can be interpreted as the effect of a change in a variable (e.g., a change in the composition of children from one girl to one boy) on the probability the outcome of interest occurs, holding all remaining variables at observed values (Reference Long and FreeseLong and Freese 2014:242–46; Reference MizeMize 2019:85–87). The regression models include interactions between litigant sex and all other explanatory and control variables. Models in this chapter contain the same control variables as models in Chapter 8. Whether a court is rural or urban is a good indicator of both the social origins of its litigants (Chapters 2 and 4) and the size and composition of its docket (Chapter 6). As before, therefore, I include fixed effects for the court that adjudicated the case in order to account for unobserved heterogeneity across contexts.

Fathers Enjoyed an Advantage in Rural Courts

From both the qualitative case examples and my review of previous research in Chapter 10, we have reason to expect that judges, when awarding child custody, cared more about which parent had physical possession than about domestic violence allegations. To the extent this is true, and insofar as patrilocal and patrilineal practices endemic to rural areas privilege fathers with respect to the physical possession standard, we also have reason to expect that fathers enjoyed an advantage in child custody determinations in rural courts. Let us now see whether these expectations hold up.

Variables of explanatory interest and their descriptive characteristics are shown in Table 11.1. I described in Chapter 4 my measurement methods for all variables.

Tables 11.2 and 11.3 contain AMEs calculated from logistic regression models. Table 11.2 contains models for rural courts and Table 11.3 contains models for urban courts. Each table contains models that take the litigant as the unit of analysis. Each model considers either plaintiffs or defendants, but not both simultaneously. Once again, models for plaintiffs’ outcomes are inversely similar to models for defendants’ outcomes because most cases are zero-sum: a win for the plaintiff is a loss for the defendant, and vice versa. They are not inversely identical, however, because some cases are positive-sum: when siblings are divided between mothers and fathers – which happens more often than not – a win for the plaintiff is also a win for the defendant.

An AME in these tables is interpreted as the effect of a change in a given explanatory variable on the probability a litigant will be awarded child custody. Since each variable in my models is categorical (as opposed to continuous), AMEs are interpreted as differences vis-à-vis omitted reference groups. For example, an AME for “female plaintiff” refers to the average difference between female plaintiffs and the omitted reference group of male plaintiffs, all else being equal. Similarly, an AME for “one girl only” refers to the average difference between plaintiffs with only-daughters and plaintiffs with only-sons (the omitted reference group), all else being equal.Footnote ² Let me illustrate how this works with rural Henan’s models for plaintiff outcomes in Table 11.2. According to Model 1 for plaintiff outcomes, the predicted probabilities of gaining child custody are .688 for mothers and .804 for fathers: .688 − .804 = −.116, which is the AME of −.12 for “female plaintiff.” According to Model 4 for plaintiff outcomes, which contains the full set of control variables, the predicted probabilities of gaining child custody are .729 for mothers and .731 for fathers: .729 − .731 = −.002, which is the AME for “female plaintiff.” Thus, by the time we reach Model 4, after incrementally adding control variables, the gap between female and male plaintiffs has vanished. The control variables added to intervening models level the playing field between mothers and fathers. In other words, mothers and fathers who had the same values of the newly added variables also had the same probability of receiving child custody. The mother–father gap in the probability of receiving child custody was associated with corresponding mother–father gaps in the intervening control variables. Mothers and fathers had unequal probabilities of receiving child custody because they were unequal in some other respect. I will focus on factors that account for mothers’ disadvantage in rural courts.

The same pattern of fathers’ large advantage in Model 1 shrinking with the addition of control variables in subsequent models – often to the point of disappearing entirely – persists across all analyses of rural courts (Table 11.2). This pattern does not extend to urban courts, however. In urban courts, by contrast, mothers enjoyed an advantage that persisted with the addition of control variables. Looking at Table 11.3, and taking plaintiffs in urban courts as an example, mothers enjoyed an advantage of .08 in Model 1 (the AME was calculated from predicted probabilities of .730 [for mothers] − .650 [for fathers] = .080). The extent of mothers’ advantage remained virtually unchanged with the addition of the full set of controls in Model 4. The same pattern of mothers’ advantage in Model 1 persisting with the addition of control variables in subsequent models persists across all analyses of urban courts – plaintiffs and defendants in both Henan and Zhejiang (Table 11.3).

Figure 11.1 depicts variation in the probability of gaining child custody by litigant sex (mothers versus fathers), litigant role (plaintiffs versus defendants), and urbanization without any controls. It therefore reflects the same mother–father gaps captured by all renditions of Model 1 in Tables 11.2 and 11.3. Three key takeaways emerge from Figure 11.1. First, in rural areas, where divorce trials were concentrated in both provinces, courts were much more likely to grant child custody to fathers than to mothers. Rural courts ruled in favor of men by a massive margin. The opposite is true in urban areas. Reflecting how the advantages of mothers and fathers flip between rural and urban courts, mothers’ and fathers’ lines (representing average probabilities of gaining child custody) cross over with urbanization in both provinces. Second, rural fathers’ advantage is larger in Henan than in Zhejiang, and urban mothers’ advantage is larger in Zhejiang than in Henan. Third, plaintiffs everywhere enjoyed an enormous advantage over defendants. Recall that, owing to differences between Henan and Zhejiang in how sub-provincial urbanization is measured, we can use urbanization to compare courts within but not between provinces.

Figure 11.1

Proportion of litigants (%) awarded child custody

Note: n = 18,216 for Henan and n = 2,529 for Zhejiang. With the exception of urban courts in Panel C, all sex differences are statistically significant (χ², P < .05). For more information on Henan’s scatterplot points, see the note under Figure 4.5. Panels B and D contain 177 scatterplot points each: 87 basic-level courts for female litigants and 90 basic-level courts for male litigants. Each panel contains best-fit lines for mothers and fathers.

Source: Author’s calculations from Henan and Zhejiang provincial high courts’ online decisions.

In what follows, I will focus on rural courts for two reasons. First, rural courts were more influential than urban courts in the overall landscape of child custody determinations. Second, the regression models identify clear reasons for fathers’ advantage in rural courts. We can identify factors that attenuated (and altogether eliminated) the mother–father gaps on the rural end of the urbanization spectrum in Figure 11.1. By contrast, mothers’ advantage persisted net of controls in urban courts. In urban courts, mothers did better than fathers regardless of the ways in which mothers and fathers otherwise differed. Nothing in the regression models attenuates mothers’ advantage in urban courts. The story of urban courts is therefore a relatively simple one. Moreover, because they handled far fewer child custody disputes than rural courts did, urban courts had a smaller overall impact in terms of numbers of divorce litigants and their children.

I will proceed by identifying and discussing the effects of various case characteristics on child custody outcomes in general and on mother–father gaps in child custody outcomes in particular. I will identify what matters and what does not matter to judges, including the impact of domestic violence allegations. In all analyses, plaintiffs’ domestic violence allegations hurt plaintiffs and rewarded defendants. For example, among plaintiffs in Henan’s rural courts, the probability of receiving child custody was .10 lower among those who made claims of domestic violence than among those who did not make such claims (Table 11.2, Model 1). Among defendants in Henan’s rural courts, by contrast, the probability of receiving child custody was .14 greater among those who were accused of abusing their spouses than among those who were not accused of such behavior (Table 11.2, Model 1). These effects shrink and sometimes disappear with the addition of control variables in subsequent models. The same patterns extend to urban courts (Table 11.3). After explaining why rural courts privileged fathers, I will then explain why, in Henan and Zhejiang alike, and in both rural and urban courts, plaintiffs (mostly mothers) were hurt by the domestic violence allegations they made and why defendants (mostly fathers) benefitted from abusing their spouses.

Fathers Enjoyed Advantages from Withholding Consent and Public Notice Trials

In Tables 11.2 and 11.3, Model 2 adds defendants’ consent and litigants’ requests for child custody. By comparing Models 1 and 2 in Table 11.2, we can see that controlling for these two variables dramatically shrinks inequality between mothers and fathers in Henan’s rural courts, and reverses it in Zhejiang’s rural courts. We can see in Table 11.2 that, by withholding consent to divorce, a defendant in a rural court in Henan reduced a plaintiff’s probability of gaining child custody by .05 and increased his own probability of gaining child custody by .07. This pattern extends to Henan’s urban courts (Table 11.3) but to neither type of court in Zhejiang (Tables 11.2 and 11.3).

Recall that, in order to avoid conflating a defendant’s affirmative consent with his failure to withhold consent, my measure of defendant consent to divorce includes whether the defendant participated in trial proceedings or was in absentia (Chapter 4). Unsurprisingly, judges tended to favor plaintiffs when defendants failed to participate in trial proceedings. Across both samples, and in rural and urban courts alike, public notice trials helped plaintiffs and hurt defendants (although the effects are not always statistically significant in the Zhejiang sample). Defendants’ failure to participate for other reasons exerts the same effect, but less strongly and only in the Henan sample. Moreover, fathers gained an advantage from this tendency of judges to favor plaintiffs when defendants were in absentia. Let me explain.

In rural courts, defendant consent to divorce accounts for a portion of – and thus partly mediates – fathers’ advantage in gaining child custody for two reasons. First, male defendants were far more likely than female defendants to withhold consent to divorce. As we know, withholding consent to divorce is an important source of leverage in child custody disputes, and men were more likely than women to exert it. As I reported in Chapters 4 and 8, female plaintiffs were far more likely than male plaintiffs to face defendant obstructionism of this nature. Second, in rural courts, female defendants were far more likely than male defendants to be served by public notice and thus to fail to participate in trial proceedings. Public notice trials (trials held without defendant participation after public notice service of process failures) were particularly common in Henan’s rural courts, where they were almost twice as likely among male plaintiffs (32%) as among female plaintiffs (17%). Even though public notice trials were less common in other contexts, male plaintiffs were more likely to have public notice trials in rural and urban courts in both provinces. The upshot is that, in rural courts, judges were more likely to award child custody to male plaintiffs than to female plaintiffs in part because public notice trials, which were more common for male plaintiffs, practically preclude the possibility of a contested child custody claim. A defendant who fails to participate in trial proceedings will not request child custody, greatly advantaging the plaintiff. As we will see next, “ask, and ye shall receive”: whether a litigant requests custody is a critical determinant of judges’ child custody orders.

Fathers Enjoyed an Advantage from Requesting Child Custody

Table 11.4 shows that mothers were far less likely to contest custody of sons than of daughters. As we have seen, daughters are less desirable to fathers in general and to fathers in rural areas in particular. In a context of entrenched son preference, many mothers appear to have abandoned hope of gaining custody of a son. Patterns of child custody requests in Table 11.4 reflect the importance of sons to the rural patriarchal family. In only-daughter couples, a request for child custody is, by definition, a request for custody of a daughter; in only-son couples, a request for child custody is, by definition, a request for custody of a son. By contrast, in couples with siblinged children, a child custody request could be for one or more of the children. Among couples with two children, for example, a child custody request could mean a request for custody of two children (two sons, two daughters, or one of each) or a request for one child (one of two sons, one of two daughters, the daughter of mixed-sex siblings, or the son of mixed-sex siblings). I am unable to differentiate between these various scenarios among couples with siblinged children owing to tremendous variation in how they are documented in the court decisions. Nonetheless, given that the majority of couples with more than one child less than 18 years old had at least one son (which we will see in greater detail below), and given the clarity of patterns of requests for custody of only-daughters and only-sons, Table 11.4 leaves no doubt that, in all contexts, fathers were more likely to request custody of sons than of daughters and that, also in all contexts, mothers were more likely to request custody of daughters than of sons. Compared to sons, daughters were less important to fathers.

One possibility is that son preference among fathers was mirrored by daughter preference among mothers. Another possibility is that mothers chose to request custody of daughters strategically, as the most realistic option. Perhaps mothers tried to avoid making requests that, in their estimation, could arouse the ire of their husbands, provoke contentious courtroom battles, and thus compromise their litigation goals. If a male defendant digs in his heels when a female plaintiff requests custody of a son, and if neither side is amenable to compromise, the court may be inclined to deny the entire divorce petition. A third and related possibility is that judges, motivated both to close cases quickly and to minimize the risk of violent retribution from angry husbands, applied pressure on mothers to give up their requests for custody of their children in general and of their sons in particular. A fourth possibility is that, to some mothers, particularly those in rural areas, the very notion of taking a son, especially an only-son, from his father is pure whimsy, culturally nonsensical, and a breach of prevailing patriarchal norms.

Table 11.4 also shows that, in rural courts, mothers were less likely than fathers to request child custody at all.Footnote ³ The opposite was true in urban courts, where mothers were more likely than fathers to request child custody. Mothers’ smaller incidence of requesting child custody in rural courts stems in part from their greater incidence of being in absentia as defendants. In order to be sure the effect of requesting child custody is not an artifact of the defendant’s failure to participate in trial proceedings, Model 2 (in Tables 11.2 and 11.3) controls for in absentia trials (as part of my measure for “defendant consent to divorce”).

The effects of requesting child custody are larger and more consistent than the effects of defendant consent to divorce. In rural courts, mothers were less likely than fathers to receive custody of a child of any sex in part because they were less likely than fathers to ask for it. Some rural women may worry that a child will reduce their remarriage prospects. In some divorce cases, women had already formed new partnerships. In some cases, women were under pressure to forfeit child custody in exchange for their freedom. Finally, to some rural women, challenging the patriarchal prerogatives of their husbands’ families would be counterproductive or altogether incomprehensible. In rural and urban courts alike, asking for child custody dramatically improved litigants’ chances of gaining it, and failing to do so had an equally dramatic effect in the opposite direction. In Table 11.2, the AMEs for female plaintiffs and defendants (which capture the extent and direction of inequality between mothers and fathers) shrink between Models 1 and 2 in part because mothers were less likely than fathers to request child custody in rural courts. Table 11.3 shows that in urban courts, by contrast, where mothers were more likely than fathers to request child custody, mothers’ advantage persisted.

Whether a litigant requested child custody was also a function of physical possession. Litigants whose children were already living with them were more likely than litigants whose children were living with their estranged spouses to request legal custody.

Fathers Enjoyed an Advantage from Physical Possession of Children

Model 3 (in Tables 11.2 and 11.3) adds physical possession, which almost completely explains away the gap between mothers and fathers in rural courts. In rural courts, fathers’ advantage in child custody orders shrinks or disappears among plaintiffs and defendants alike in both provinces (Table 11.2, Model 3). The reason is simple: fathers were significantly more likely than mothers to have physical possession of children. Rural patrilocality emerges in high relief from Figure 11.2, which depicts variation in the probability of physical possession of a child by litigant sex (mothers versus fathers), litigant role (plaintiffs versus defendants), and urbanization. In areas served by rural courts, children were more likely to be living with their fathers than with their mothers; in rural courts in both provinces, mothers were less likely than fathers to have physical possession of their children. The rural gap is particularly noteworthy in Henan; rural fathers’ advantage was smaller in Zhejiang than in Henan. And whereas in Henan the mother–father gap narrowed and disappeared with urbanization, in Zhejiang’s urban courts, mothers’ likelihood of having physical possession of a child surpassed that of fathers.

Figure 11.2

Proportion of litigants (%) with physical possession of a child

Note: n = 18,216 for Henan and n = 2,529 for Zhejiang. With the exception of urban courts in Panels B and C, all sex differences are statistically significant (χ², P < .05). For more information on Henan’s scatterplot points, see the note under Figure 4.5. Panels B and D contain 177 scatterplot points each: 87 basic-level courts for female litigants and 90 basic-level courts for male litigants. Each panel contains best-fit lines for mothers and fathers.

Source: Author’s calculations from Henan and Zhejiang provincial high courts’ online decisions.

Child custody outcomes (in Figure 11.1) clearly correspond to physical possession patterns (in Figure 11.2). While Figures 11.1 and 11.2 certainly suggest that physical possession influences child custody outcomes, multivariate regression analysis will help us assess more definitively the extent to which mother–father inequality in physical possession explains mother–father inequality in child custody determinations.

Many litigants without physical possession of their children challenged their estranged spouses by petitioning courts for child custody. Some, however, did not. Mothers were less likely to contest fathers’ physical possession of children than fathers were to contest mothers’ physical possession of children.Footnote ⁴ Litigants often asked courts to formalize the status quo. The current location of the child, which litigants often asked courts to preserve, reflects patrilocal norms. Table 11.5 shows that in both Henan and Zhejiang, in rural and urban courts alike, only-daughters were far more likely to be in the physical possession of their mothers. It also shows that the opposite was true for only-sons and siblings (most of whom include sons) in rural Henan. Although the same general pattern applies to rural Zhejiang, differences between mothers and fathers are not statistically significant. In urban courts, by contrast, physical possession of only-sons and multiple children was less gendered in both provinces, and even favored mothers in Zhejiang.

The limitations of my measure of requests for child custody also apply to this measure. Among couples with siblinged children, physical possession reflects one of a number of possibilities, including custody of two children (two sons, two daughters, or one of each) or of one child (one of two sons, one of two daughters, the daughter of mixed-sex siblings, or the son of mixed-sex siblings). I could not reliably or comprehensively distinguish between these various scenarios for the same reason I could not do so with my measure of requests for child custody. Nonetheless, Table 11.5 shows that, in all contexts, mothers were more likely to have physical possession of daughters than of sons and that, in rural areas, fathers were more likely to have physical possession of sons than of daughters.

An additional limitation of this measure is its high proportion of missing values (coded as “neither side or undisclosed”). Rates of physical possession in Figure 11.2 and Table 11.5 are therefore underestimates. Rates at which children were already split up by parents who had separated are also underestimates. Of all couples with siblinged children in the Henan and Zhejiang samples, I could identify 16% and 14%, respectively, who had split them apart prior to the divorce trial. Excluding decisions from which I could not ascertain the physical locations of children (often because they were undisclosed), the proportion of siblings split apart by their parents was in the range of 25–30% in both samples.Footnote ⁵

The introduction of physical possession in Model 3 (in Tables 11.2 and 11.3) greatly weakens the effect of requesting child custody in rural and urban courts in both provinces because, as we have seen, litigants who requested child custody often already had physical possession of the child. In rural areas, fathers were more likely than mothers to have physical possession of children, ergo they were more likely to request custody, ergo they were more likely to gain custody. The opposite was true in urban areas. In other words, rural women were less likely than rural men and urban women to request child custody in part because they were less likely to have physical possession of their children.Footnote ⁶

These patterns persist with the addition of control variables in Model 4. In Henan’s rural courts, the predicted probability a plaintiff with exclusive physical possession was awarded child custody was a whopping .96 compared to .78 for a plaintiff who shared physical possession of their child or children (the corresponding AME of .18 in Table 11.2, Model 4 is simply the difference between these predicted probabilities: .96 − .78 = .18). Meanwhile, when the defendant had sole physical possession of the child or children, a plaintiff’s predicted probability of gaining child custody was only .38 (.38 − .78 = −.40, which is the corresponding AME in Table 11.2, Model 4). In Zhejiang’s rural courts, a plaintiff with physical possession of a child enjoyed a similar advantage gaining child custody: a predicted probability of .91 compared to .63 for a plaintiff whose spouse also had physical possession of the child or children (.91 − .63 = .28, which is the corresponding AME in Table 11.2, Model 4). Finally, the predicted probability a plaintiff in rural Zhejiang gained child custody when the defendant had sole physical possession of the child or children was .37 (.37 − .63 = −.26, the value of the corresponding AME in Table 11.2, Model 4).

The story of urban courts stands in stark contrast. Nothing explains away mothers’ advantage in urban courts (Table 11.3).

Let us take stock of the empirical findings I have presented so far. Courts greatly favored fathers over mothers in rural areas, and mothers over fathers in urban areas. In rural courts, fathers benefitted from female defendants who were both more likely than male defendants to be in absentia and less likely than male defendants to withhold consent to divorce. Fathers in rural courts also gained some of their advantage from mothers’ lower rates of petitioning for child custody. Finally, fathers’ advantage gaining child custody from rural courts stemmed to a large extent from their correspondingly greater likelihood to have physical possession of their children.

Fathers Enjoyed an Advantage from Domestic Violence

In the samples of child custody decisions, the proportion of all plaintiffs who made claims of domestic violence was 26% and 32% in Henan and Zhejiang, respectively. Consistent with Figure 7.1, the incidence of plaintiffs’ allegations of domestic violence was slightly higher in urban areas than in rural areas. Most claims of domestic violence were made by women: 90% in Henan and 83% in Zhejiang. Thus, the proportion of female plaintiffs who made claims of domestic violence was 35% and 39% in the Henan and Zhejiang samples, respectively.

Plaintiffs who made domestic violence allegations had worse outcomes than those who did not, and defendants who were accused of domestic violence had better outcomes than those who were not. These effects are reflected in every version of Model 1 in Tables 11.2 and 11.3. Model 2 introduces defendant consent to divorce and requests for child custody. When these variables are added to the model, they attenuate the effect of plaintiff domestic violence allegations for plaintiffs and defendants alike in rural and urban courts in both provinces (Tables 11.2 and 11.3, Model 2). When defendants failed to participate in trial proceedings, plaintiffs were less likely to make claims of domestic violence, perhaps because there was less of a need to do so when the plaintiff’s request for child custody was uncontested. After all, as we can see from the AMEs in Model 2 (Tables 11.2 and 11.3), a defendant’s absence from trial proceedings greatly increased the likelihood that a court awarded child custody to the plaintiff. In short, the effect of domestic violence is partly mediated by the effects of defendant consent to divorce.

The effect of domestic violence is similarly mediated by the effects of requests for child custody. Put more plainly, plaintiffs who made allegations of domestic violence were less likely than those who did not make such allegations to request child custody. For this reason, what appears to be the effect of domestic violence is in part the effect of requesting child custody. Finally, as we have seen, an important reason why litigants did not request child custody is because they did not have physical possession of the child. In short, the regression models tell a story that can be summarized in the form of a path model: domestic violence→physical possession of child→request for child custody→child custody order. Let me elaborate.

Several of the qualitative examples I presented in Chapter 10 paint a grim picture in which domestic violence was the key reason why litigants did not have physical possession of their children in the first place. The quantitative data reveal the pervasiveness of this problem. In all contexts, domestic violence dramatically reduced the probability that women had physical possession of their children. Among plaintiffs in Henan who did and who did not make domestic violence claims, the proportion with physical possession of a child was 31% and 45%, respectively. Similarly, among plaintiffs in Zhejiang, the respective proportions were 35% and 44%. For this reason, when physical possession is added to Model 3 in Tables 11.2 and 11.3, the effect of domestic violence allegations shrinks yet again everywhere except in Zhejiang’s urban courts. We can infer from these findings that physical possession of children is an important reason why in prior models the effect of plaintiffs’ domestic violence allegations was negative for plaintiffs and positive for defendants. Thus, the effect of domestic violence is driven in part by inequality between abuse victims and abuse offenders with respect to the physical possession of children. When plaintiffs and defendants were equally likely to have physical possession of children (Model 3), the effect of domestic violence was much smaller or altogether absent. When their victims escaped, often without their children, perpetrators held an advantage with respect to child custody simply by having sole physical possession of their children. These patterns persist after adding all remaining control variables to Model 4.

Taken together, the regression models tell a story (depicted in the path model earlier) in which domestic violence reduced the probability of physical possession of children, which in turn reduced the probability of requesting child custody, which in turn reduced the probability of being awarded child custody. Leveling the playing field between abuse victims and abuse perpetrators with respect to physical possession washed out the negative effect of domestic violence. At the same time, however, plaintiffs’ domestic violence allegations failed to improve their custody outcomes. Among plaintiffs and defendants who were equally likely to have physical possession of children, were equally likely to request child custody, and were otherwise seemingly identical (Model 4), domestic violence still failed to increase victims’ or decrease perpetrators’ chances of winning custody. Simply put, domestic violence was of no importance and made no difference to judges.

Fathers Enjoyed an Advantage from Sons, Mothers Enjoyed an Advantage from Daughters

In contrast to the hit-or-miss nature of information about the physical possession of children and the motley assortment of ways it appears when courts do record it in their decisions, courts recorded child custody decisions in an almost uniform manner. I could therefore reliably and comprehensively identify and disaggregate siblings in courts’ child custody orders. When a court granted custody of at least one child among siblings to a litigant, I was able to distinguish the full array of possibilities for that particular litigant: custody of all children among same-sex siblings, custody of all children among mixed-sex siblings, custody of one daughter among same-sex siblings, custody of one daughter among mixed-sex siblings, custody of one son among same-sex siblings, custody of one son among mixed-sex siblings, and so on. Recall how rarely courts granted joint legal custody. For this reason, in no instances were both parents coded as simultaneously receiving custody of an only-child. We can likewise be confident that when both parents are coded as receiving custody of siblings, the codes reflect courts’ tendency to split siblings apart and do not imply joint legal custody.

Indeed, courts split siblings apart in over half of all child custody determinations involving siblings: 62% of the time in Henan and 51% of the time in Zhejiang. This happened more often in rural courts than in urban courts: 62% and 56%, respectively, in Henan, and 59% and 36%, respectively, in Zhejiang. Even twins were divided. In the handful of child custody determinations involving twins (about two dozen in the Henan sample and about a dozen in the Zhejiang sample), courts almost always split them apart. In one illustrative case, the twins were in the physical possession of the defendant’s parents. The twins’ mother, the plaintiff, made two requests: custody of one of the twins and the return of her dowry. In the course of the trial, she bargained away her dowry. This compromise – achieved when the plaintiff “voluntarily” forfeited her property rights, possibly in order to secure her right to child custody – spared the court the hassle of dealing with the marital estate. In its verdict, the court ordered custody of one twin to each parent and made no ruling on property division (Decision #224435, Linzhou Municipal People’s Court, Henan Province, October 19, 2009).Footnote ⁷ Of course, we have no way of knowing whether the defendant obeyed the court order. Given that courts so rarely enforce their decisions, the plaintiff would have ended up with nothing except her freedom in the not unlikely event the defendant failed to comply with the court order.

No law suggests – much less requires – that judges split up siblings. In one case, the judge invoked a generic “principle of fairness” (公平原则) as a rationale (Decision #1229092, Weishi County People’s Court, Henan Province, September 9, 2014).Footnote ⁸ In a similar case, when a male defendant withheld consent to divorce unless he was awarded custody of both children, the court granted the daughter to the plaintiff and the son to the defendant as a matter of fairness (Decision #1547183, Nanshao County People’s Court, Henan Province, December 26, 2013).Footnote ⁹ The female plaintiff in my final example, owing to her husband’s affair with another woman in the village, sought a divorce and ¥10,000 in damages for emotional distress. Her husband admitted to the affair and claimed he had broken it off. As an expression of his resolve to walk the straight and narrow, he had chopped off part of his finger – and the court affirmed both his act and motivation as factual. He then insisted on custody of both children as a condition of agreeing to divorce. Following the principle of fairness, the court gave custody of their son to him and custody of their daughter to the plaintiff (Decision #777658, Puyang County People’s Court, Henan Province, December 15, 2011).Footnote ¹⁰

Son preference emerges in high relief from Table 11.6. In both samples, only-son couples far outnumbered only-daughter couples. Indeed, in both provinces, couples with at least one son outnumbered couples with at least one daughter by a sizeable margin. In Henan, the proportions of couples with only-daughters and only-sons were 33% and 39%, respectively. In Zhejiang, the proportions were 40% and 48%, respectively. Child custody determinations, on average, involved a lot more children in Henan than in Zhejiang, since couples in Zhejiang were more likely to have only-children. The proportion of couples with more than one child less than 18 years old (“any siblings”) in Henan (29%) was more than double that in Zhejiang (13%).

In both provinces, son preference was endemic to rural areas more than to urban areas. In rural courts, as Table 11.6 shows, the overall proportion of couples with at least one son (64% in Henan and 57% in Zhejiang) was greater than the overall proportion of couples with at least one daughter (57% in Henan and 50% in Zhejiang). Table 11.6 also shows that the extent to which couples with sons outnumbered couples with daughters was greater in rural areas than in urban areas (although in the Zhejiang sample son preference was only slightly stronger in rural areas than in urban areas).

Before its abolishment in 2016, China’s so-called one-child policy allowed couples to have a second child under certain circumstances. For example, in support of rural patriarchal norms, rural couples were generally allowed to have a second child if the firstborn was a daughter. Even when they already had a son, two daughters, or one of each, many rural couples paid a fine after retroactively registering an “above-quota” or “out-of-plan” birth, or attempted to avoid a fine by hiding an unauthorized child (Reference Kennedy and ShiKennedy and Shi 2019; Reference Michelson, Oi, Rozelle and ZhouMichelson 2010).

Table 11.6 shows that, consistent with China’s family planning policies, couples were far more likely to limit fertility to one child after having a son than after having a daughter. Couples who started with a daughter were more likely than couples who started with a son to have a second child, often in efforts to bear a son. An overrepresentation of boys among only-children and relatively balanced sex ratios among siblings are telltale signs of a tendency to keep trying for a son after a firstborn girl. Sex ratios provide more direct evidence of this tendency. In both samples, the sex ratios of younger siblings were far more skewed than the sex ratios of older siblings. Sex ratios of boys to girls among the oldest children – including only-children – subject to child custody orders were 1.09:1 in Henan and 1.12:1 in Zhejiang. Younger siblings, by contrast, were much more likely to be boys than girls. Among younger siblings (higher-parity children), ratios of boys to girls were 1.64:1 in Henan and 1.34:1 in Zhejiang. The overall sex ratios and parity-specific sex ratios of children found in my samples of court decisions generally mirror those of young children in the Chinese population (Reference Michelson, Oi, Rozelle and ZhouMichelson 2010). Son preference in rural areas is responsible for such a high degree of overrepresentation of boys among younger siblings.

Some scholars, however, argue that China’s sex ratio imbalance is in large part a statistical artifact of an overrepresentation of girls among unregistered children (Reference Kennedy and ShiKennedy and Shi 2019). Of course, I am unable to assess the extent to which girls are “statistically missing” from (or “statistically invisible” in) court decisions. Perhaps some litigants did not disclose their unregistered children to courts. After all, a child custody claim should be supported by proof of parenthood in the form of a household registration booklet or birth certificate. I encountered only a small handful of court decisions in which children were described as “unregistered” (e.g., 未上户口). Even if it does inflate sex ratios, a failure to disclose unauthorized girls reinforces their liminal legal status, particularly if courts are complicit by looking the other way in their efforts to close cases, and is itself an indication of son preference.

Table 11.7 contains proportions of mothers and fathers receiving child custody by the number of children (only-children versus siblings) and urbanization (rural courts versus urban courts). It reaffirms two patterns we have already observed in both Figure 11.1 and the regression models in Tables 11.2 and 11.3, namely that mothers were less likely than fathers to be awarded child custody in rural courts and more likely than fathers to be awarded child custody in urban courts.Footnote ¹¹ It also shows a consequence of courts’ tendency to split siblings apart: litigants with siblinged children were vastly more likely than litigants with only-children to gain child custody. Custody of only-children is almost always a zero-sum game owing to the almost total irrelevance of joint legal custody. For this reason, the respective proportions of mothers and fathers receiving custody of only-children always sum to 100%. If siblings were never split up, the respective proportions of mothers and fathers receiving custody of siblinged children would likewise always sum to 100%. Because siblings were so commonly divided between parents, sums of proportions of mothers and fathers receiving custody of siblinged children were far in excess of 100% (and typically in excess of 150%). Table 11.7 also shows that, when siblings did stay together, they were more likely to go to fathers than to mothers in rural areas in both Henan and Zhejiang. Among couples with siblinged children divorcing in rural courts, the proportions of mothers and fathers granted custody of no children was 23% and 15% respectively in Henan and 25% and 17%, respectively, in Zhejiang. As striking as these patterns are, however, they obscure enormous variation according to the sex composition of children.

Table 11.8 contains litigants’ probabilities of receiving child custody by litigant sex, the number of children, and the sex composition of children. The sheer extent to which courts matched parent and child sex emerges with remarkable clarity. In Henan and Zhejiang, in both rural and urban courts, mothers were far more likely than fathers to receive custody of a daughter when there was only one daughter up for grabs. When there was more than one girl in the family, mothers and fathers were similarly likely to receive custody of a daughter because courts tended to split them up. Meanwhile, fathers were far more likely than mothers to receive custody of a son when there was only one available. An additional son greatly boosted mothers’ chances of receiving custody of a son because courts tended to split them up, although fathers remained advantaged in Henan’s rural courts even in cases involving more than one son.

Most striking of all in Tables 11.7 and 11.8 is how much more poorly mothers in rural courts fared in the absence of either a daughter or multiple children. Among only-son couples in rural courts, the proportion of mothers awarded child custody was 36% in Henan and 46% in Zhejiang (Table 11.8). In urban courts, mothers of only-sons did as well or better than fathers of only-sons (Table 11.8). Mothers of only-daughters did even better. Among only-daughter couples in rural courts, the proportion of mothers awarded child custody was 60% in Henan and 56% in Zhejiang (Table 11.8). In urban courts, mothers of only-daughters were awarded child custody 66% and 68% of the time, respectively (Table 11.8). Among couples with more than one child in rural courts (regardless of the children’s sex composition), the proportion of mothers awarded child custody was 77% in Henan and 75% in Zhejiang (Table 11.7).

The same patterns – the boost from daughters and multiple children to mothers’ chances of receiving child custody – emerge from the regression models in Tables 11.2 and 11.3. In Model 1, for both rural and urban courts, plaintiffs benefitted from only-daughters, and defendants were hurt by only-daughters because most plaintiffs were women and most defendants were men (Tables 11.2 and 11.3). In every model, plaintiffs and defendants alike benefitted tremendously from multiple children because courts tended to split them up. After adding requests for child custody to Model 2, the effects of only-daughters (positive for plaintiffs and negative for defendants) shrink across the board (in all version of Model 2 in Tables 11.2 and 11.3). Thus, the regression models tell another story summarized in the following path model: sex of child/sex of parent→physical possession of child→request for child custody→child custody order. I will elaborate.

The effects of the composition of children are partly mediated by the effects of requests for child custody, which, as we know, are partly mediated by the effects of the physical possession of children. In simpler terms, as previously discussed, mothers were more likely than fathers to request custody of daughters because they were more likely than fathers to have physical possession of daughters (this pattern extends to both rural and urban courts). By the same token, fathers were more likely than mothers to request custody of sons because they were more likely than mothers to have physical possession of sons (this pattern is limited to rural courts). Because the physical possession of children is so highly gendered, its addition to Model 3 further shrinks the effect of only-daughters (for plaintiffs and defendants alike everywhere except Zhejiang’s urban courts). In other words, Model 3, by equalizing litigants’ physical possession of children, shrinks the gap between those with only-daughters and only-sons. Thus, plaintiffs’ advantage securing custody of only-daughters can be attributed in part to plaintiffs’ greater likelihood to have physical possession of only-daughters (because plaintiffs were mostly women). Likewise, defendants’ advantage securing custody of only-sons can be attributed in part to defendants’ greater likelihood to have physical possession of only-sons (because defendants were mostly men).

In many, if not most, cases, courts simply preserved the status quo by applying the physical possession standard. As we saw, judges attached no importance whatsoever to plaintiffs’ domestic violence allegations. At the same time, however, the highly gendered nature of child custody orders – the degree to which courts matched parent and child sex – is not only a story about courts’ preserving the current situation established by the divorcing couple. Courts did more than formalize presorted gender matches made by the parents. Courts were also active agents of gender sorting. Courts themselves reproduced patriarchal norms by granting custody of daughters to mothers and of sons to fathers even when mothers and fathers were equally likely to request child custody and equally likely to have physical possession of children.

The regression models in Tables 11.2 and 11.3 show overall effects on litigants’ probabilities of receiving child custody – mothers and fathers taken together. They do not show whether the various effects they depict differ between mothers and fathers. In regression parlance, they do not show interaction effects. Table 11.9 contains predicted probabilities, calculated from Model 4 in Tables 11.2 and 11.3, that support tests of interactions between litigant sex and child sex. Note that differences between “one girl only” and “one boy only” in Table 11.9 correspond to AMEs for “one girl only” in Model 4 in Tables 11.2 and 11.3. For example, the AME of .05 for “one girl only” (vis-à-vis the omitted reference group of “one boy only”) among plaintiffs in Henan’s rural courts (Table 11.2, Model 4) corresponds to .675 − .626 = .049 in Table 11.9 (in the “all litigants” column).Footnote ¹² Likewise, the AME of −.05 for “one girl only” among defendants in Henan’s rural courts (Table 11.2, Model 4) corresponds to .36 − .41 = −.05 in Table 11.9 (in the “all litigants” column).

By showing whether the effects of the number and sex composition of children differed between mothers and fathers, Table 11.9 is similar to Table 11.8. Unlike Table 11.8, however, Table 11.9 allows us to assess whether the effects of the number and sex composition of children differed between mothers and fathers who were otherwise seemingly identical (i.e., after adding a full set of control variables).Footnote ¹³ Whereas Table 11.8 contains raw, descriptive probabilities, Table 11.9 contains postestimation predicted probabilities calculated from Model 4 in Tables 11.2 and 11.3.

Two patterns jump out of Table 11.9. First, control variables greatly shrink the effects of the number and sex composition of children. For example, in couples with only-daughters in Henan’s rural courts, mothers’ overall advantage was .20 (60% − 40% = 20% in Table 11.8). After adding controls, mothers’ advantage shrank to .07 among plaintiffs and .06 among defendants (Table 11.9). Had I calculated the contents of Table 11.9 according to Model 1 in Tables 11.2 and 11.3 (i.e., before controlling for requests for child custody and physical possession of children), mothers’ and fathers’ gender-specific advantages and disadvantages would have been far greater. Second, even net of controls, among mothers and fathers whose requests for child custody and physical possession of children were seemingly identical, courts favored mothers over fathers when determining the custody of only-girls and favored fathers over mothers when determining the custody of only-boys. Among plaintiffs and defendants in rural and urban courts in Henan and Zhejiang (i.e., in every analysis), mothers were advantaged by only-daughters. Mothers’ gender-specific advantage was statistically significant everywhere except Zhejiang’s rural courts. Net of controls, fathers were advantaged by only-sons only in Henan’s rural courts.

Differences between mothers and fathers (in the “mother–father differences” column in Table 11.9) are first differences, which are gaps between two groups in the probability the outcome of interest occurs. Differences between first differences are second differences. Mother–father differences did in fact differ between only-daughter couples and only-son couples to a statistically significant extent everywhere except Zhejiang’s rural courts. Tests of second differences in Table 11.9 tell us that, net of controls, gaps between mothers and fathers in the likelihood of gaining custody of only-daughters (which favored mothers) was, in most analyses, statistically significantly different from the gap between mothers and fathers in gaining custody of only-sons (which favored fathers).

Other Effects

In my presentation of results from analyses of child custody determinations, I have focused attention on what mattered most to judges: the physical possession of children, litigant sex, and the number and sex composition of children. I have also drawn attention to something unequivocally unimportant to judges: domestic violence allegations. My analyses include additional control variables I have not yet discussed. I simply note here that the effects of the other variables included in Model 4 paled in comparison to the effects I thoroughly discussed. For example, on the whole, the participation of female judges did not improve mothers’ chances of gaining child custody. Similarly, generally speaking, mothers with legal representation fared no better than mothers without legal representation. Other variables added to Model 4 had similarly inconsequential effects on child custody outcomes. After all, AMEs in Model 3 and Model 4 are virtually identical.

Summary and Conclusions

Chapter 8 demonstrated, on the basis of an analysis of over 60,000 court decisions in my Henan and Zhejiang samples, that, far more often than not, courts denied the first-attempt divorce petitions of domestic violence victims. A plaintiff’s claim of domestic violence, even when fully documented by admissible evidence, failed to increase even slightly the court’s probability of granting her divorce request. In Chapter 9, we saw that victims of marital abuse who were denied the divorces they requested were forced either to return home with their abusers or live elsewhere, often at the cost of separation from their children. My analyses of child custody determinations in both this chapter and Chapter 10 reveal that domestic violence allegations were similarly unimportant to Chinese judges when they made child custody determinations. Just as domestic violence claims did not increase the likelihood that courts granted women’s divorce requests (Chapter 8), they likewise did not increase the likelihood that courts granted child custody to marital abuse victims.

On the contrary, men were rewarded with child custody for beating their wives, because courts by and large preserved the status quo by granting custody to the parent who had physical possession of the child. When women left their abusive husbands, they were sometimes fleeing for their lives. With life and limb at risk, women sometimes left everything – including their children – behind. Owing to the importance judges attached to physical possession in their child custody orders, doing so perversely undermined their chances of winning child custody, and perversely put the health and safety of children at risk.

The over 20,000 child custody determinations I analyzed in this chapter bring into high relief judges’ impulse to preserve the status quo. In rural areas, which accounted for the majority of all child custody determinations, the status quo was a patriarchal one insofar as fathers were more likely than mothers to have physical possession of children in general and of sons in particular. Although a litigant’s request for child custody was another major source of influence on the court’s decision, many litigants requested child custody only when they already had physical possession of a child. Some litigants’ child custody requests were motivated by their desire to preserve and formalize such highly gendered de facto custody arrangements. Consistent with the logic of patriarchy and the title of this chapter, rural courts tended to grant custody of only-sons to fathers. The likelihood of receiving custody of an only-son from a rural court was far lower among mothers. In rural areas, mothers’ best chances for child custody came from multiple children and from only-daughters. In cases involving siblings, courts frequently split them up between the parents. In cases of mixed-sex siblings, courts typically granted custody of sons to fathers and custody of daughters to mothers.

To be sure, some litigants without physical possession of a child did contest the status quo by petitioning for custody of a child living with the other parent. Many, however, resigned themselves to – and thus did not contest – the living arrangements of their children at the time of the trial. Courts even pressured mothers to withdraw their requests for child custody, particularly when the child was already in the physical possession of the father. In rural areas, mothers desperate to escape miserable marriages not infrequently conceded custody of children, particularly of sons, in court-brokered compromises that favored fathers in terms of child custody and property division. Owing to both the cultural importance of patrilineality and pragmatic old-age security considerations in a context of patrilocality, rural fathers generally care more about sons than daughters. And owing to endemic son preference in rural China, when couples separate prior to divorcing, sons often stay with fathers, and daughters often follow mothers. In short, there are several mechanisms behind the strong association between litigants’ child custody requests and their physical possession of children. Much of the time courts passively affirmed and formalized the preexisting living arrangements of children. When parents had similar claims on children, courts were active agents of patriarchy by staunchly supporting the prerogatives of fathers, particularly in rural Henan.

In sum, judges, through their decision-making behavior, supported the very patriarchal norms which China’s family laws were designed to dismantle. More than serving to protect women from patriarchy, courts operated in the service of patriarchy. Courts did less to stand up to patriarchy and more to preserve it. On the whole, Chinese family law has significantly failed to penetrate the rural patriarchal order.

Fathers’ advantages in child custody determinations were limited to rural areas, which accounted for a sizeable majority of all child custody determinations. Urban courts, by contrast, favored mothers over fathers. Women’s better outcomes in urban courts may have less to do with courts per se – e.g., the social values of judges – and more to do with weaker patriarchal control over residential arrangements, to which courts tended to defer in their child custody orders. In rural areas, children were more likely to be in the physical possession of their fathers than of their mothers owing primarily to the norm of patrilocal marriage, whereby women marry into their husband’s families, and the entrenched tradition of patrilineality, whereby lines of descent are carried forward through sons. As neolocal residence becomes more common and newly married couples increasingly establish homes apart from their parents in rural areas, and as rural old-age pension support strengthens, the rural family’s patriarchal grip on sons should weaken and mothers’ prospects of gaining child custody in contested divorce cases should improve. Until either that happens or courts take the rights of women and the best interests of children more seriously, divorced mothers without daughters or siblinged children will continue to find themselves at a severe disadvantage in child custody determinations in China’s rural courts.

One limitation of my analysis is the absence of a measure for infant or nursing children. Information about the ages (or years of birth) of children is spotty at best. Decisions referring to children “under the age of two” (e.g., 未满两周岁, 不满两周岁, and 不到满两周岁) or “of nursing age” (e.g., 在哺乳期) were few in number, less than 5% in both samples. However, if judges tended to grant custody of infant and nursing children to their mothers, empirical results I have presented in this chapter indicating an overall disadvantage to mothers would be conservative. Imagine that, all else being equal, courts were equally likely to grant custody of children more than two years old to mothers and fathers. If this were true, and if judges tended to apply the infant standard, then mothers would have enjoyed an overall advantage in the probability of winning custody of a child of any age. For this reason, my empirical findings indicating an overall advantage to fathers likely understate the true extent of gender differences in child custody determinations.