Can Risk Perception Alter Son Preference? Evidence from Gender-Imbalanced Rural China

Abstract

It is well established that son preference is the crucial driver for sex ratio imbalance, and that there are risks stemming from such imbalance. Whether risks associated with a gender-imbalanced society may alter son preference will be explored in this study, which has so far received scant scholarly attention. Using data from the Consequences of Gender Imbalance Survey conducted in 2018, exploiting structural equation modelling, this paper shows that risk perception of gender imbalance has a significant and negative effect on stated son preference; however, entrenched traditional norms underpinning the institution of son preference, measured as gender role attitudes and the value of sons, are reinforced by risk perception. The effect of risk perception on weakening stated son preference is suppressed by gender role attitudes and the value of sons which are still upholding stated son preference. Overall, the effects of risk perception and social norms are additive, influencing stated son preference simultaneously, but traditional norms act as counteracting forces. This study makes an important step toward shedding light on both continuity and change in son preference in gender-imbalanced rural China, and offers new perspectives for future research.

Keywords:

• Risk perception
• son preference
• social norms
• gender imbalance
• rural China

1. Introduction

Son preference has been widely documented in East and South Asia, and is reflected in skewed sex ratios in favour of males. In China, the sex ratio imbalance has been a cause of concern for demographers, economists, sociologists, and policy-makers during the last few decades. The emphasis of previous studies has been on the impact of son preference on gender imbalance, while the focus of this study is on the inverse of this relationship. The most recent China census data of 2020 indicate that the sex ratio at birth (SRB) was 111.3, a decrease of 6.8 compared with 2010. The fact that SRB is decreasing combined with the fact that it remains well above the normal level of around 105 call for more knowledge about what underpins and what erodes son preference.

Research about the determinants of son preference has proliferated during the last three to four decades. Intertwined explanations of socioeconomic, sociocultural, household, and individual characteristics have evolved, among which social norms are decisive factors underpinning son preference (Chung & Gupta, 2007; Gupta et al., 2003; Murphy, Tao, & Lu, 2011; Pande & Astone, 2007). Social norms are rules or expectations of behaviours for certain populations (Bicchieri, 2010). In terms of gender relations, social norms are constructed based on accepted ideas about what women and men should be and do, what constitutes appropriate masculinity and femininity, and – of relevance for this study – parents’ different expectations on sons and daughters (Yilmaz, 2018). Essentially, son preference is a gendered institution that builds upon conservative gender role attitudes and the value of sons.

China is the country with the longest duration, the widest geographical range, and the highest degree of severity of gender imbalance in the world (Bongaarts & Guilmoto, 2015). Gender imbalance is typically portrayed as generative of social crises (Eklund, 2018; Driessen & Sier, 2021), manifested in missing girls, involuntary bachelors, and related social problems. These include escalating bride prices, economic pressure, and vulnerability at old age for those who never marry, and rising crime rates related to violence and trafficking in women (Jiang, Sánchez-Barricarte, Li, & Feldman, 2011; Jiang, Zhang, & Sánchez-Barricarte, 2015; Ling, 2017). These risks are likely to raise concerns among parents about opting for sons, as suggested by an increasing number of recent studies (Li, Chan, Spencer, & Yang, 2016; Ling, 2017; Porter, 2017; Robitaille, 2020). It has been suggested that: ‘An increase in the relative scarcity of females would offset the preference for sons’ (Li et al., 2016, p.1065). Yet, no study to date has made a quantitative investigation into the degree to which underlying preferences for sons may persist in a gender-imbalanced society, i.e. how son preference is reshaped by gender imbalance. This research seeks to shed light on this issue, in a rural context.

The study aims to explore the impact of gender imbalance on the institution of son preference. Accordingly, we use micro-level data from The Consequences of Gender Imbalance Survey conducted in 2018 in rural China in order to further the understanding of how son preference relates to social norms in a gender imbalanced risk society. Deploying structural equation modelling (SEM), this paper makes theoretical and empirical contributions to the existing literature in several ways. First, due to the unique property of the dataset, it analyses perceived risks of gender imbalance at the individual level. Second, it estimates the impact of risk perception on reported son preference in gender imbalanced rural China, using a rigorous approach through SEM. Finally, to the best of our knowledge, it is the first theoretical probe for the causal mechanism explaining the influence of gender imbalance on the institution of son preference.

2. Theoretical framework

2.1. The institution of son preference

2.1.1. The value of sons

Many studies have used the value of children (VOC) to parents approach to understand fertility preferences and behavioural intentions (e.g. Kagitcibasi & Ataca, 2005; Eklund, 2016). Initially, the basic values identified in the VOC approach contain nine categories. Through further refinements, subsequent research has focused on two dimensions, namely, utilitarian value and social value. Utilitarian value comprises economic returns and child labour in housework, i.e. providing old age support, financial support, and day-to-day care. Social value entails the social acceptance parents gain from having children, and one of the main features is the need for children to carry on the family lineage and act as a source of prestige.

Son preference is rooted in patrilineal societies, where patriarchal institutions and norms value sons higher than daughters (Croll, 2000; Den Boer & Hudson, 2017; Pande & Astone, 2007; Tafuro, 2020). Elderly parents receiving more long-term care from sons is the principal cause of son preference both in China and India (Kadoya & Khan, 2017; Kadoya & Yin, 2014), and sons have been found to provide financial assistance more frequently than daughters in rural China (Lei, 2013). In China, most studies support that sons and daughters-in-law serve as primary day-to-day caregivers for parents (Cong & Silverstein, 2008; Kadoya & Yin, 2014), although some studies show that daughters provide more or equal instrumental support, i.e. help with household chores or caretaking (Lei, 2013; Zeng, Brasher, Gu, & Vaupel, 2016). In addition, sons are wanted for their ability to maintain the family line in countries like China, where Confucianism dictates that the lineage is traced solely through male descendants (Chung & Gupta, 2007). In sum, ‘a strong son preference is a result of the high perceived worth of males and low perceived worth of females’ (Fuse, 2013, p. 1029).

2.1.2. Gender role attitudes

Sons’ value in providing old-age support, financial support, day-to-day care, and continuing the family line generally rests on the idea that gender determines unequal abilities for sons and daughters to contribute to the family over the life course, manifested in gendered divisions of labour where husbands are breadwinners and wives homemakers (Jergins, 2021). Gender role attitudes capture the extent to which individuals support traditional gender divisions of paid and unpaid labour. Stereotypical gender roles are at the root of gender differences in both investment and return in economic and emotional terms, something which causes parents to obtain higher value from sons than from daughters (Choi & Hwang, 2020).

A growing literature has examined whether and how gender preferences are shaped by gender role attitudes. Gender preferences for children are found to be latent manifestations of the level of gender equity within a particular society (Mills & Begall, 2010). Mills and Begall (2010) note stronger son preference in societies where there is low gender equality, and Andersson, Hank, Rønsen, and Vikat (2006) claim stronger daughter preference in relatively gender equal societies (see also Miranda, Dahlberg, & Andersson, 2018). A longitudinal study in China finds that adherence to traditional norms on gender roles is negatively associated with daughter preference (Kim & Fong, 2014). Similarly, in South Korea, the probability of preferring daughters over sons increases among those less vested in the traditional gender role attitudes; conversely, the odds of preferring sons are higher among those holding more conservative views on gender roles (Chun & Gupta, 2022).

2.2. Risks of gender imbalance

Due to China’s large population, the highly skewed SRB translates into substantial numbers of ‘excess men’ and ‘missing women’ at the macro level, resulting in various kinds of pressures on families and individuals. Marriage is a mainstream practice in China, and social acceptance and status largely depend on being married (Attané, Zhang, Li, Yang, & Guilmoto, 2013; Hesketh, Li, & Zhu, 2011). The shortage of marriageable women tends to leave marginalised men (poor, uneducated, disabled, etc.) with few marriage prospects (Eklund & Attané, 2017). Guilmoto (2012) estimates that the proportion of men who are likely to remain unmarried at age 50 is expected to rise to 15% by 2055 in China even if SRB would have returned to normalcy by 2020. Previous studies claim that the marginalisation of these surplus men will lead to increased levels of social problems, and eventually threaten societal stability and security, something which may even spill over to regional and global society (Hudson & Den Boer, 2002; Jiang et al., 2011).

To improve the marriageability of a son by way of providing a higher bride price, marriage expenses have been increasing rapidly in China, and families with sons face heavy financial burdens (Jiang & Sánchez-Barricarte, 2012). A recent study suggests that marriage market pressures on men are associated with a higher tendency to commit financially rewarding crimes (Cameron, Meng, & Zhang, 2019). Edlund, Li, Yi, and Zhang (2013) estimate that one-seventh of the rise in violence and property crime in China between 1998 and 2004 is attributable to masculine sex ratios and adverse marriage market conditions. A not-so-surprising yet important finding is that the need for care of the unsupported bachelors poses a social challenge (Larsen & Kaur, 2013). Involuntary bachelors who cannot establish their own family will live without economic and old-age support usually provided by the wife and children (Jiang, Guo, Li, & Feldman, 2013).

Some scholars predict that marriage-squeezed men stimulate the occurrence of crimes, such as rape and abductions (Banister, 2004), which threaten the safety of women, families, and communities. The clustering of involuntary bachelors and their involvement in the violation of community security has been reported by Jin, Liu, Li, Feldman, and Li (2013), though a recent study found that social discrimination is an important mechanism that triggers marriage-squeezed men to engage in risky behaviours that threaten self- and community safety, while being unmarried is in itself not associated with risk for community safety (Yang, Wang, & Eklund, 2020).

Others argue that meeting unmarried men’s sexual needs will lead to a large-scale expansion of the sex industry, even including coercion and trafficking in women (Hesketh & Zhu, 2006). Prakash and Vadlamannati (2019), using panel data from twenty-nine Indian states, find that a 100-unit increase in the child sex ratio would increase 0.635 per cent of girl trafficking. Zhao (2003) takes a multi-disciplinary approach to explore the fundamental causes of trafficking in women in China, and reports that the rising sex ratio drives the prevalence of trafficking of women for marriage. Xiong (2022) notes that every woman in a gender-imbalanced society is a potential victim of sexual assault and sexual abuse, thus, female personal safety gets compromised with an excess of males.

As noted above, risks of gender imbalance broadly fall into two dimensions, one is related to the male marriage squeeze and the other is related to female safety issues (Jin et al., 2013).

2.3. Conceptualising risk perception, social norms, and stated son preference

Bélanger (2015) argues that ‘son preference has been reshaped and insufflated with new meaning over time’ (p. 211). Exploring how son preference is reconfigured in a patrilineal society experiencing gender imbalance is of special interest. Li et al. (2016) estimate that an increase of 1% in the number of the sex ratio in the marriage market (marriageable males relative to females) would decrease the probability of having a male birth by 0.02 percentage points. Porter (2017) documents that where women are scarce, married women are better able to bargain for more time devoted to caring for their own parents, suggesting that highly imbalanced sex ratios eventually may increase the value of daughters and thereby contribute to a shift in parental preferences towards daughters. Robitaille (2020) finds that marriage expenditures reduce the desire for sons in India, either as measured by stated son preference or revealed son preference.

This study extends our understanding of the relationship between risks of gender imbalance and son preference at a micro level by revealing underlying processes. Drawing inspiration from exposure-based theories (cf. Rimal & Real, 2003), we depart from the understanding that when exposed to a high sex ratio context, perceived risk associated with that context holds the potentiality of triggering attitudinal change, in this case changes in son preference. Yet, the relationship between perceived risk associated with sex ratio imbalance and son preference is not understood as linear, but as embedded in complex social systems. Indeed, while a growing body of literature has argued that risk perception influences intentions and behaviour, mechanisms may be more complicated (Veflen, Scholderer, & Langsrud, 2020). Sheeran, Harris, and Epton (2014) find that risk perception has only a small main effect on people’s intentions and behaviours, yet it has a substantially larger effect when it also triggers anticipatory emotions. This result implies that mediation effects are important to take into consideration when investigating the mechanism of various behavioural intentions.

Son preference is conceptualised as an institution grounded in norms that carry extrinsic and intrinsic rewards and benefits, which hold the potential of impeding the influence of gender imbalance. Hence, changes in son preference are seen as the result of a complex trade-off between gender imbalance and sticky social norms. Investigating how risk perception influences sets of norms and practices that are known for upholding notions of son preference, i.e. gender role attitudes and the value of sons, is one way of understanding how the institution of son preference becomes renegotiated with gender imbalance. We hypothesise that gender role attitudes and the value of sons act as mediators in the relationship between risk perception and stated son preference for three reasons. First, gender imbalance alters the relative status of men and women on the marriage market, which in turn exerts an effect on the intra-household gender division of labour (Grosjean & Khattar, 2019; Guttentag & Secord, 1981). Second, long-run consequences of male-biased sex ratios, such as involuntary bachelors and high marriage expenses, increase the costs of raising sons over daughters and affect expectations of sons’ value and demands for sons (Dong, Alhaj-Yaseen, Jiao, & Zhong, 2021; Eklund, 2016). Third, the association between gender role attitudes and the value of sons signify their serial-mediating roles in upholding the institution of son preference. Elucidating the mechanism of how institutionalised son preference changes can provide vital insights for actual fertility decisions in gender imbalanced rural China.

3. Data

The data in this paper are derived from the Consequences of Gender Imbalance Survey conducted in Hubei Province, in which individuals’ behaviours and perceptions about gender imbalance were included. The two-week survey was launched by the National Health Commission, implemented in August 2018. Hubei Province in central China is the focus of national and local policies to address the high SRB and ameliorate the imbalance. The census data showed that the SRB in Hubei was 128.02 in 2000 and 123.94 in 2010. At the end of 2015, the SRB was still high at 114.6. Since becoming one of the three pilot provinces in the United Nations Population Fund (UNFPA) – Government of China Eighth Country Programme (2016–2018), Hubei has begun to promote gender mainstreaming, change social norms, and transform customs of son preference (United Nations Population Fund & China Population & Development Research Center, 2018).

Data were collected by multistage and stratified random sampling method. In the first stage, counties were selected based on the population size. In total, 103 counties were included in the sampling frame. H County with a population over one million and J County with a population below one million were randomly selected as sampling sites. In the second stage, 12 of the 16 towns were randomly selected from H County, and one village/community was randomly selected in each town. In J County, three of the eight towns were randomly selected, and two villages/communities were randomly selected from one town and three villages/communities from the other two towns in accordance with the population distribution. In the third stage, 25 rural households were randomly selected in each village/community. In each household, two individuals were invited to answer the questionnaire.

Interviewers met the respondents in public venues, government institutions, or private homes. All respondents were explained the purpose of the survey, that they could discontinue the survey at any time without any consequences, and that they could remain anonymous. Before participating, each respondent provided informed consent. No information that can lead to identifying an individual was collected. The research project was approved by the ethics committee of Xi’an Jiaotong University.

A total of 1032 valid questionnaires were distributed among 20 villages/communities. The overall response rate was 100 per cent. The average age of the respondents was 42.38 years (SD = 9.29), with an average education of 9.38 years (SD = 3.17). 53.88% of the total respondents identified as women. As to marital status, 93.02% of the respondents were married.

The four key variables of interest in the analysis are stated son preference, the value of sons, gender role attitudes, and risk perception, consisting of 16 items; each is ranked on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).

Stated son preference is obtained by asking respondents about the extent to which they agree or disagree with the statement ‘I think that it is necessary to have a son’. A higher score indicates a stronger stated son preference.

The value of sons is captured from the original VOC study by asking the respondents to rate the importance they attributed to reasons for wanting a son (Kagitcibasi & Ataca, 2005), consisting of four items: ‘provide old age support’, ‘offer financial support’, ‘render day-to-day care’, and ‘carry on the family lineage and prestige’.

To capture gender role attitudes, the indicators are created by one subscale of four items adopted from the Third Wave Survey on the Social Status of Women in China: ‘The domain for men is in public and the domain for women is within household’; ‘Earning money to support the family should be done by men’; ‘The development of the husband is more important than the development of the wife’; ‘For women, having a good marriage is better than a career’. Higher values signify more traditional gender role attitudes.

Risk perception is the ‘process of collecting, selecting, and interpreting signals about uncertain impacts of events or activities’ (Wachinger, Renn, Begg, & Kuhlicke, 2013, p.1049). Respondents were asked seven questions about gender imbalance. These questions cover seven aspects of risks, that is male marital difficulties, exorbitant marriage expenses, forced bachelor families face enormous economic pressure, forced bachelor families face problems of old age support, trafficking in women and children, female personal safety, and increasing crime rate due to more unmarried men. Out of the seven items, the first four items are related to male marriage squeeze, while the remaining three are related to female safety issues.

Following prior research, a set of personal characteristics were designated as covariates. Thus, gender, educational attainment, age, and annual income were controlled in the analyses. Covariates were added as exogenous variables (Kline, 2016), each of which has a direct influence on stated son preference, the value of sons, and gender role attitudes.

4. Methodology

In order to analyse the relationship between risk perception and stated son preference, covariance-based structural equation modelling is applied. SEM is a comprehensive and flexible statistical technique that allows conducting a precise analysis by combining factor analysis with multiple regression analysis (Kline, 2016). It is more versatile than most other multivariate statistical tools, particularly valuable in analysing models encompassing multi-item constructs. The first overriding advantage of SEM is its ability to correct for measurement errors, which allows for specifying the estimated error between the theoretical constructs (latent variables) and the observed data (manifest variables). The second benefit of SEM is the provision of more straightforward tests of mediation, in which all mediation paths (multiple mediators) are estimated simultaneously rather than being restricted to an additive model, thus it is considered to be more rigorous than typical stepwise regression (Garson, 2013).

SEM contains two related sub-models: measurement model and structural model. The former shows the association between latent variables and their observed indicators, and the latter serves to estimate causal hypotheses between latent variables of interest. Figure 1 visualises the path diagram of the measurement model (dashed rectangles) and the structural model (solid rectangle). Prior to validating the structural model, confirmatory factor analysis (CFA) was employed to evaluate the reliability and validity of the measurement model. The maximum likelihood estimation was used to analyse the structural model under the assumption that data are missing at random. The bootstrap test, repeated sampling 1,000 times, was used to assess the mediating effects.

Figure 1. Path diagram of the model.

Notes: The measurement models are the dashed rectangles; the structural model is the solid rectangle. Latent variables are shown as ellipses; observed variables are presented by rectangles. The arrows represent hypothesised direct effect.

The structural equation model is represented by the following equations: (1) $$\mathrm{\chi }={\Lambda }_{\chi }\xi +\delta$$(1) (2) $$\mathrm{y}={\Lambda }_y\eta +\epsilon$$(2) (3) $$\mathrm{\eta }=\mathrm{B}\mathrm{\eta }+\mathrm{Г}\xi +\zeta$$(3)

Equations (1)(1) $$\mathrm{\chi }={\Lambda }_{\chi }\xi +\delta$$(1) and (2)(2) $$\mathrm{y}={\Lambda }_y\eta +\epsilon$$(2) correspond to the measurement models. $\mathrm{\chi }$ and $\mathrm{y}$ are vectors of indicators for exogenous latent variables $\xi$ (i.e. risk perception) and endogenous latent variables $\eta$ (i.e. gender role attitudes, the value of sons and stated son preference). ${\Lambda }_{\chi }$ and ${\Lambda }_y$ are matrices of factor loadings. $\delta$ and $\epsilon$ are vectors of measurement errors for indicators. Equation (3)(3) $$\mathrm{\eta }=\mathrm{B}\mathrm{\eta }+\mathrm{Г}\xi +\zeta$$(3) denotes the structural model, where $\mathrm{B}$ is a coefficient matrix for the relations among the endogenous latent variables, and $\mathrm{Г}$ is a coefficient matrix for the effects of the exogenous latent variables on the endogenous latent variables. $\zeta$ is the error term.

Conventional recommended and used Goodness of Fit Statistics are the CFI (comparative fit index); the TLI (Tucker–Lewis index); the RMSEA (root mean square error of approximation); the SRMR (standardised root mean square residual). CFI and TLI use cut-off scores, requiring values equal to or higher than 0.90 (Hu & Bentler, 1995). Following Browne and Cudeck (1992), the value of RMSEA between 0.05 and 0.08 indicates an adequate fit. As a rule of thumb, the SRMR should be lower than 0.05 (Hu & Bentler, 1995). Analyses were conducted using Stata 14.0 and Mplus 7.0.

5. Results

5.1. Measurement model

Results of the measurement model for latent variables are summarised in Table 1. The factor loadings represent the extent to which the latent variables are related to the observed indicators, showing how well the indicator (item) discriminates between individuals. Each of the factor loadings on their respective latent factor is statistically significant and positive. Cronbach’s alpha coefficients are used to evaluate the internal consistency of the multi-item scales. All alpha coefficients are above the threshold of 0.6, indicating adequate reliability for the survey instrument.

Table 1. Constructs, survey items, and standardised factor loadings

Constructs	Survey items	Factor loadings
Value of sons (VOS) α = 0.866	Old age support	0.798***
	Financial support	0.789***
	Day-to-day care	0.821***
	Carrying on the family lineage and prestige	0.740***
Gender role attitudes (GRA) α = 0.659	The domain for men is in public and the domain for women is within household	0.612***
	Earning money to support the family should be done by men	0.669***
	The development of the husband is more important than the development of the wife	0.659***
	For women, having a good marriage is better than a career	0.359***
Risk perception (RP) α = 0.758	Male marital difficulties	0.506***
	Exorbitant marriage expenses	0.454***
	Forced bachelor families face enormous economic pressure	0.705***
	Forced bachelor families face problems of old age support	0.689***
	Trafficking in women and children	0.273***
	Female personal safety	0.504***
	Increasing crime rate due to more unmarried men	0.553***

Notes: N = 1032; RMSEA = 0.059; CFI = 0.924; TLI = 0.908; SRMR = 0.045; ***p < 0.01.

According to the results, day-to-day care is the best indicator of sons’ value, whereas carrying on the family lineage and prestige is the weakest. In terms of gender role attitudes, ‘Earning money to support the family should be done by men’ loads the highest, while ‘For women, having a good marriage is better than a career’ is the weakest among the indicators. As for risk perception, indicators showing agreement on ‘Forced bachelor families face enormous economic pressure’ and ‘Forced bachelor families face problems of old age support’ have greater factor loadings, while concern for ‘Trafficking in women and children’ is the weakest indicator.

5.2. Structural model

The results of the structural model are shown in Table 2, revealing the hypothesised relationships between risk perception, social norms, and stated son preference. The goodness of fit is acceptable according to a CFI value of 0.908, TLI 0.888, RMSEA 0.051 (90% confidence interval: 0.047, 0.056), and SRMR 0.044, though the TLI, which penalises models that estimate many parameters, is barely sufficient.

Table 2. Estimation results of the structural model

Direct effect	Coefficient	Std. Err	Standardised coefficient
Dependent variable: stated son preference
Value of sons	0.267***	0.037	0.287
Gender role attitudes	0.087**	0.043	0.086
Risk perception	−0.097**	0.038	−0.091
Gender^a	−0.162**	0.069	−0.075
Education level	−0.018	0.012	−0.052
Age	0.006	0.004	0.055
Income level	0.004	0.042	0.003
Dependent variable: value of sons
Gender role attitudes	0.399***	0.048	0.367
Risk perception	0.207***	0.043	0.179
Gender^a	−0.351***	0.077	−0.152
Education level	−0.050***	0.014	−0.134
Age	−0.004	0.004	−0.030
Income level	−0.087*	0.047	−0.063
Dependent variable: gender role attitudes
Risk perception	0.166***	0.045	0.157
Gender^a	−0.133	0.083	−0.062
Education level	−0.074***	0.015	−0.219
Age	0.015***	0.005	0.129
Income level	−0.006	0.051	−0.004

Notes: N = 1032; RMSEA = 0.051 (90% confidence interval: 0.047, 0.056); CFI = 0.908; TLI = 0.888; SRMR = 0.044; ^abase category: male; ***p < 0.01; **p < 0.05; *p < 0.10.

Women have a much lower stated son preference and the individual attributes of being female, more educated, and having a high income are all related to weaker expectations of sons’ value. Any education works to diminish traditional gender role attitudes. Younger individuals hold more egalitarian gender role attitudes than older ones.

Stated son preference is found to be positively and significantly affected by stereotypical gender role attitudes and the value of sons. In a similar vein, the strength of stereotypical gender role attitudes positively affects the value of sons. These findings are in line with previous studies and not surprising since specific gender reasoning (Croll, 2000) and gender division of labour underpin gender traditionalism, underlining the superior value of sons.

The key findings regarding the role of risk perception are also reported in Table 2. It can be noted that risk perception of gender imbalance has a significant and negative effect on stated son preference, however, it is also found to be positively correlated with more traditional gender role attitudes and the value of sons. Taken as a whole, these results illuminate the complexity surrounding shifting notions about the need and importance of having a son in rural China today.

Yet, persistent stated son preference in rural China is attenuated by risk perception of gender imbalance. In other words, the stronger the perceived risk, the lower the stated preference for sons. This is in line with Ling (2017) who argues that in light of the hurdles faced by young men in China’s marriage market today, son preference is encountering its own shadow. Yet, risk perception is having a significant positive impact on the value of sons, which indicates that the value of sons remains a resilient norm to some extent in rural China. As the risk perception increases, the expectation attached to sons’ value rises. More interestingly, risk perception positively influences gender role attitudes in a more traditional direction. This speaks against the idea, put forward by Driessen and Sier (2021), that male marriage squeezes boost gender egalitarianism, as traditional gender role attitudes continue to prevail in gender imbalanced rural China. These findings support the hypothesis that stated son preference is the result of a trade-off involving perceived risk and social norms.

The results of indirect effects are displayed in Table 3. According to Preacher and Hayes (2008), the significance of the total indirect effect is not a prerequisite for identifying significant specific indirect effects in a multiple mediator model. Thus, we explain the specific mediating effects. The value of sons mediates the relationship between risk perception and stated son preference, while the mediating effect of gender role attitudes is non-significant. Further, the serial-mediating effect from risk perception to stated son preference through gender role attitudes and the value of sons is significant. Specifically, while the direct effect of risk perception on stated son preference is negative (as shown in Table 2), the indirect effect mediated by gender role attitudes and the value of sons is positive. The opposite signs illustrate an inconsistent mediation effect, that is, gender role attitudes and the value of sons operate as suppressors. Additionally, the indirect effect of risk perception on the value of sons through gender role attitudes is significant, accounting for 24% of the total effect.

Table 3. Bootstrapping indirect effects of gender role attitudes and value of sons

Indirect effect						Bootstrapping
						90% CI
				Standardised coefficient	Std. Err	LLCI	ULCI	p-Value
RP→	GRA→	SSP		0.013	0.009	−0.001	0.120	0.028
RP→	VOS→	SSP		0.051***	0.015	0.027	0.000	0.075
RP→	GRA→	VOS→	SSP	0.016***	0.005	0.008	0.002	0.025
RP→	GRA→	VOS		0.057***	0.017	0.029	0.001	0.086

Notes: RP: risk perception; SSP: stated son preference; GRA: gender role attitudes; VOS: value of sons; ***p < 0.01; **p < 0.05; *p < 0.10.

5.3. Robustness checks

For robustness check, this study further estimates the effects of risk perception related to male marriage squeeze and risk perception related to female safety issues separately, which are presented in Appendix Tables B1–B4. Results of interest remained virtually unchanged, but a more detailed illustration is provided. The direct effect of risk perception related to male marriage squeeze on stated son preference is negative and statistically significant (Table B1). Yet the indirect effect, mediated through gender role attitudes and the value of sons, is positive (Table B2). The direct effect of risk perception related to female safety issues on stated son preference is also significant and negative (Table B3), which is larger than the effect of risk perception related to male marriage squeeze. Its indirect effect on stated son preference is only partially mediated through the value of sons (Table B4).

In addition, following the procedure developed by Oster (2019), we test the robustness of the results to omitted variable bias. The method relies on coefficient and R-squared movements after the inclusion of controls, which links the bias explicitly to coefficient stability. Results are reported in Table B5 and imply that selection on unobservables would have to be five times as important as selection on observables to entirely attribute the results to the omission of unobservables, i.e. explain away the influence of risk perception on stated son preference, a value well beyond the recommended threshold of one. Thus, the statistics rule out the concern that omitted variable bias may be driving our results.

6. Discussion

This study, using an original dataset, puts forward new understandings of changing notions of son preference in rural China by introducing a previously unexplored variable, i.e. risk perception of gender imbalance. We find signs of changes in son preference, but also shed light on mechanisms that delineate the resilience of son preference in a male-biased society.

The results elucidate that traditional gender role attitudes and the value of sons still hold considerable ground in contemporary rural China, underpinning the persistence of son preference. The magnitude of sons’ value is greater than gender role attitudes, which resonates with the finding of Tafuro (2020) that son preference is related to a strong reliance on son-based intergenerational support instead of insufficient advancement of women in the economy, although they interrelate to some extent. Complementary evidence about the relationships between gender role attitudes and stated son preference is provided by the analysis of sons’ value. Traditional gender role attitudes are important for sustaining the value of sons, highlighting the tenacious nature of traditional norms. Specific gender reasoning and the gender division of labour underpin traditional gender role attitudes, underlining the gender division model of men belonging to the public sphere and women belonging to the private sphere (Chen, Peng, Xu, & Yin, 2020). In patriarchal societies, gender reasoning implies that husbands are responsible, authoritarian, strong, and dominant, and by extension, men are natural providers for their families (Yilmaz, 2018). Traditional gender role attitudes place high expectations on men, and sons are therefore preferred where such attitudes still prevail.

Prior research suggests that the consequences of the skewed sex ratio ‘may unravel over very long periods. People may be so inured to them that they go unremarked’ (Dyson, 2012, p.453). This is largely uncharted territory as virtually no attention has been devoted to understanding the role of risk perception. The data, while not exhaustive, provide a unique opportunity to examine the role of risk perception quantitatively. There is compelling evidence that the extent of stated son preference decreases with the increase in risk perception, concurring with the view of Dong et al. (2021) that greater sex imbalance, creating greater pressure in mating competition, leads to lower son preference. This is a sign that the perceived risk of gender imbalance may potentially reverse the current preference for sons in rural China, illustrating a potential path of ‘high risk perception -- weak son preference -- rebalance of the sex ratio’, i.e. a self-correcting of the gender imbalance. The results highlight the importance of risk perception in understanding both continuity and change in son preference in rural China.

The results reveal strong evidence that traditional gender role attitudes increase with the level of risk perception related to male marriage squeeze, though not to female safety issues. This suggests that in gender-imbalanced rural China, individuals perceiving a higher level of male marriage squeeze are more likely to confine women to the domestic sphere. This finding is supported by Driessen and Sier (2021), who argue that in the context of gender imbalance, family members attempt to rescue the dignity of their sons and brothers through gender labour and ‘giving gender’ to bolster unmarried men’s gender authenticity and thereby strengthen their marriageability. The finding also echoes the evidence that economic obstacles of marriage are positively associated with traditional views about gender roles among marriage-squeezed men (Attané, Eklund, & Zhang, 2018). Similarly, establishing a link between gender imbalance and gender role attitudes, another study demonstrates that in areas with more male-biased sex ratios, both men and women held more conservative views about gender roles (Grosjean & Khattar, 2019). These studies, as well as our results showing that high-risk perception of gender imbalance strengthens traditional gender role attitudes, can be understood in light of the work by Guttentag and Secord (1981), who based on historical data proposed that in high sex ratio societies, women and men would adopt complementary roles, women would be valued for their beauty and their role as mothers, and the institution of marriage would become stronger. This perspective offers important cues as to the mechanisms that mediate the impact of risk perception on stated son preference, and can also explain the finding that there is a tendency to place more importance on the value of sons as a consequence of perceived risk related to gender imbalance. The results remain unchanged in further analyses of the two dimensions of risk perception, which signals that sons’ value of filial piety and morality, as well as symbolic and ritual practices, does not necessarily abate with socioeconomic development but may even be strengthened by perceived risk in a gender imbalanced society. Another explanation pertains to safety concerns, and that an ardent expectation of sons is related to concerns about raising a daughter in the context of a heightened risk of crimes against women and girls (Eklund, 2016; Theerthaana & Sheik Manzoor, 2019). The unique finding of the mediating effect in this study provides further evidence that risk perception augments the value of sons by reinforcing traditional gender role attitudes. It must be noted that the value of sons is mainly measured by perceived benefits. Risk perception, however, logically links with the costs of sons following the relatively simple economic intuition. Thus, the absence of costs in measuring sons’ value may be one reason behind this strong relationship.

The impact of risk perception on stated son preference is significantly mediated by the strength of sons’ value. In addition, gender role attitudes and the value of sons play a serial mediating role between risk perception and stated son preference. Risk perception of gender imbalance highlights traditional gender role attitudes, which leads to higher expectations of sons’ value in turn, which finally increases stated son preference. Hence, although holding the potential to undermine son preference, risk perception may also contribute to the resilience of the institution of son preference by strengthening traditional gender role attitudes and the value of sons.

7. Conclusion

The negative consequences and risks of long-term gender imbalance which have now piled up over three decades in China deserve much attention. No previous studies have investigated how risks of gender imbalance influence the institution of son preference. This study attempts to fill this conspicuous gap in the literature by incorporating risk perception of gender imbalance into the empirical research model. The interlinkages between risk perception, gender role attitudes, and the value of sons are important to take into account in order to get a more nuanced understanding of how son preference evolves.

This study suggests that, first, when individuals perceive risks of gender imbalance, they are likely to reduce their stated preference for sons. However, entrenched traditional norms underpinning the institution of son preference, measured as gender role attitudes and the value of sons, are reinforced by risk perception. Second, gender imbalance is a double-edged sword in altering the institution of son preference. While risk perception has a negative impact on stated son preference, traditional norms have a counteracting positive effect of a stronger magnitude. Last but not least, the weakening effect of risk perception on stated son preference is suppressed or outweighed, to some extent, by gender role attitudes and the value of sons. Consequently, these findings contribute to understanding why rural China has continued to be gender imbalanced in spite of the growing manifested risks.

Several limitations regarding this study should be mentioned. The primary limitation pertains to the data design. First, this study was a cross-sectional survey, limiting both determinations of causality and observations of how son preference changed over time. Second, in the measurements of certain variables, mature scales were not used but only self-developed questions. Finally, the self-reported nature of the survey may have led to the under-reporting of son preference and risk perception. Furthermore, there are content limitations in this study. One limitation is the possibility that variables may not be capturing all of the relevant dimensions. For example, the value of sons was construed as the sum of costs and benefits but mainly measured by the perceived benefits.

Therefore, follow-up studies should expand the dimensions of latent variables. Future research may also wish to explore and determine whether the predictions of risk perception would change in a context of more aggravated gender imbalance, which China is projected to experience in the near decades to come.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data and codes are available upon request.

Additional information

Funding

This work was supported by the National Social Science Fund of China under Grant (number 16CRK022).

Appendix A

Table A1. Descriptive statistics (N = 1032)

(strongly disagree = 1, strongly agree = 5)	Mean	SD
Stated son preference	2.30	1.08
Latent variable: value of sons	2.56	0.94
Old age support	2.60	1.11
Financial support	2.53	1.10
Day-to-day care	2.48	1.09
Carrying on the family lineage and prestige	2.65	1.17
Latent variable: gender role attitudes	3.14	0.87
The domain for men is in public and the domain for women is within household	3.28	1.21
Earning money to support the family should be done by men	3.23	1.24
The development of the husband is more important than the development of the wife	3.28	1.26
For women, having a good marriage is better than a career	2.78	1.25
Latent variable: risk perception	3.40	0.67
Male marital difficulties	3.70	0.93
Exorbitant marriage expenses	3.86	0.92
Forced bachelor families face enormous economic pressure	3.41	1.01
Forced bachelor families face problems of old age support	3.32	1.03
Trafficking in women and children	3.11	1.22
Female personal safety	3.46	1.07
Increasing crime rate due to more unmarried men	2.97	1.12

Note: The underlined values are the mean values and standard deviations of the latent variables. There is no statistical significance.

Appendix B:

Robustness checks

Table B1. Estimation results of the structural model: Risk perception related to male marriage squeeze

Direct effect	Coefficient	Std. Err	Standardised coefficient
Dependent variable: stated son preference
Value of sons	0.263***	0.037	0.283
Gender role attitudes	0.087**	0.043	0.086
Risk perception related to male marriage squeeze	−0.080**	0.039	−0.075
Dependent variable: value of sons
Gender role attitudes	0.394***	0.048	0.363
Risk perception related to male marriage squeeze	0.212***	0.044	0.184
Dependent variable: gender role attitudes
Risk perception related to male marriage squeeze	0.183***	0.047	0.172

Notes: N = 1032; RMSEA = 0.049 (90% confidence interval: 0.043, 0.055); CFI = 0.935; TLI = 0.916; SRMR = 0.037; ***p < 0.01; **p < 0.05; *p < 0.10. Covariates are similar as in Table 2 and are omitted.

Table B2. Bootstrapping indirect effects of gender role attitudes and value of sons: risk perception related to male marriage squeeze

Indirect effect						Bootstrapping
						90% CI		p-Value
				Standardised coefficient	Std. Err	LLCI	ULCI
RPM→	GRA→	SSP		0.015	0.010	−0.001	0.031	0.125
RPM→	VOS→	SSP		0.052***	0.015	0.027	0.077	0.001
RPM→	GRA→	VOS→	SSP	0.018***	0.006	0.008	0.027	0.002
RPM→	GRA→	VOS		0.063***	0.018	0.032	0.093	0.001

Notes: RPM: risk perception related to male marriage squeeze; SSP: stated son preference; GRA: gender role attitudes; VOS: value of sons; ***p < 0.01; **p < 0.05; *p < 0.10.

Table B3. Estimation results of the structural model: Risk perception related to female safety issues

Direct effect	Coefficient	Std. Err	Standardised coefficient
Dependent variable: stated son preference
Value of sons	0.267***	0.037	0.285
Gender role attitudes	0.080*	0.044	0.078
Risk perception related to female safety issues	−0.116***	0.041	−0.108
Dependent variable: value of sons
Gender role attitudes	0.424***	0.048	0.388
Risk perception related to female safety issues	0.145***	0.046	0.127
Dependent variable: gender role attitudes
Risk perception related to female safety issues	0.062	0.051	0.059

Notes: N = 1032; RMSEA = 0.043 (90% confidence interval: 0.037, 0.049); CFI = 0.951; TLI = 0.934; SRMR = 0.037; ***p < 0.01; **p < 0.05; *p < 0.10. Covariates are similar as in Table 2 and are omitted.

Table B4. Bootstrapping indirect effects of gender role attitudes and value of sons: risk perception related to female safety issues

Indirect effect						Bootstrapping
						90% CI		p-Value
				Standardised coefficient	Std. Err	LLCI	ULCI
RPF→	GRA→	SSP		0.005	0.006	−0.005	0.014	0.426
RPF→	VOS→	SSP		0.036***	0.014	0.013	0.059	0.009
RPF→	GRA→	VOS→	SSP	0.007	0.006	−0.003	0.017	0.281
RPF→	GRA→	VOS		0.023	0.021	−0.011	0.058	0.278

Notes: RPF: risk perception related to female safety issues; SSP: stated son preference; GRA: gender role attitudes; VOS: value of sons. ***p < 0.01; **p < 0.05; *p < 0.10.

Table B5. Selection on unobservables

	Stated son preference
	Uncontrolled	Controlled
Coefficient	−0.006	−0.014
R-squared	0.001	0.124
δ(β = 0)		−5.068
β(δ = 1)		−0.017

Notes: δ designates the ratio of selection on unobservables to selection on observables that would be required to fully attribute our results to omitted variable bias (β= 0); β is derived from imposing the reasonable assumption that selection on unobservables is equally as important as selection on observables (δ = 1); δ and β are calculated using $R_{\max }$=1.3$\tilde{\mathrm{R}}.$ A negative value of delta is generated when the correlations between observables and the treatment, and between unobservables and treatment, have opposing signs. The variables were constructed by aggregating corresponding items. See Oster (2019) for details.