The economic benefits of empowering women have been extensively discussed by researchers since the 1990s. Improving women’s education and health, as well as enhancing their participation in the labor market could increase women’s productivity and income, which in turn raised household income and alleviating household poverty (Morrison et al., 2007). An increase in mothers’ earnings and their decision-making power within the household further facilitates improvements in child welfare, ensuring better education and nutrition outcomes. Consequently, this helped prevent the intergenerational transmission of poverty and mitigates the risk of future household poverty (ibid.). Although definition and measurement variables for women’s empowerment varied, most review studies reported a consistent positive association between women’s empowerment and food and nutrition security (Aziz et al., 2022), which are the fundamental elements in the elimination of poverty. Women’s empowerment, particularly in the sense of bargaining power and decision-making on consumption and human capital investment, has proved to be a strong pathway to poverty reduction through ensuring women’s land rights (Meinzen-Dick et al., 2019). Further, research evidence demonstrated that women empowered by microfinance institutions developed greater entrepreneurship and greater decision-making authority and control over their economic stability, thus reduced their poverty level and also boosted regional economic development (Chen et al., 2025); “left-behind” women in rural China, with enhanced literacy on internet use and e-commerce through government-led training programs, tackled poverty by marketing local agricultural, cultural products (Zhu et al., 2022).

Apart from the direct economic benefits of the empowering women, the social benefits also gained interest of researchers so that the relations between women’s empowerment and social capital were explored. Effective women’s empowerment requires the breaking of gendered social norms, such as the recognition of women’s access to resources, opportunities, and decision making (Galiè et al., 2022), the shifting of perceptions on reproductive work as petty and low-skilled (Singh et al., 2018). Interventions that promoting the family and economic benefits of women’s empowerment could lead to behavior change and alternative social expectations for both men and women thus bringing about some level of norm changes (Hillenbrand and Miruka, 2019). It was argued in a qualitative study that well-designed micro banks that genuinely empowered women—such as those granting women freedom and direct control over the use of loans without requiring spousal approval—could foster women’s social bonding and networks, thus encouraging their collective efforts through its extensive networks, such as village organizations, and community development associations (Khatun and Hasan, 2015). In addition, the implementation of community-based, education-related development programs improved cooperation and trust within project villages and this effect was particularly pronounced among impoverished groups and was directly correlated to the increased agency and empowerment of women (Janssens, 2010). Women’s cooperatives in Indonesia provided women with opportunities for savings and access to credit; the trust built in women’s cooperatives forms a social capital, which helps women obtain loans to start and expand their businesses (Widiyanti et al., 2018). Further, varying relationships were found between women’s empowerment and different types of social capital, for example, bonding social capital was significantly correlated with women’s participation in agricultural decisions; bridging social capital was applied by women to diversify their training and information sources (Po and Hickey, 2020).

Based on the above, we frame the following two hypotheses:

Social capital could also benefit poverty reduction. A study in rural Malaysia found that an increase of one unit in social capital reduced the probability of household poverty by 18.3% (Abdul-Hakim et al., 2010). Chinese scholars have identified the resource attributes of social networks— business relationships, political connections, and accessible social organizations—which have made significant contributions to poverty reduction, with residents’ trust in institutions playing a moderating role (Zhang et al., 2017). In Bangladesh, households with lower levels of social networks, norms of reciprocity, and civic engagement were more likely to be poor compared to those with higher levels (Islam and Alam, 2018). The impact of social capital on poverty reduction may not always be direct. Compared to physical and human capital, social capital plays a greater role in improving household livelihoods through the mediating effect of intrinsic motivation, including life satisfaction, confidence in improving future living standards, and the ability to face life’s challenges with resilience (Liu et al., 2020). However, the relationship between social capital and poverty reduction were not always consistent: according to the theories of “externalities” of social capital, richer individuals tend to collaborate to those who possess similarly extensive networks, as such ties yield greater access to information, resources, and influence (Collier, 2002). This mechanism of shared cognition and mutual selection inherently disadvantages the poor, who often lack access to high-quality networks.

The impact of social capital on poverty reduction was also explored from the gender perspectives. Women with higher levels of social capital were found less likely to be consumption poor as social capital ensures access to information and other capital resources that enables households to invest more in either farm or off-farm income opportunities (Sharaunga and Mudhara, 2021). Ul-Hameed et al. (2018) found that the positive effect of social capital could be enhanced where vulnerability factors co-existed, i.e., gender discrimination, shortage of financial resources and less focus on development activities by government, etc., indicating that social capital has more significance for women in poor regions to fight against poverty. Women’s agricultural production organization generated closer bonding between family members and friends, helping each other with collective labor and aiding old mothers by donating agricultural products, thus generating more income for the family particularly for daughters and obtaining food security for the poor (Naughton et al., 2017). It was argued that social capital could stimulate rural women entrepreneurs to achieve business success through the enhancement of entrepreneurship and social innovation abilities, enabling them to increase their incomes and improve the welfare of their families, thus fully participating in solving social problems such as poverty and inequality (Osei and Zhuang, 2020). Social capital was found as a mediating factor for the effectiveness of women’s self-help groups. It was found that more mature groups with established trust and economic security enabled women’s active participation in development interventions compared with the poorer and newer groups (Nichols, 2021). Interestingly, the same study also pointed out the “dark side” of social capital: the confinement of women’s traditional roles as caregivers in the domestic sphere and their relatively low educational level leads to women’s exclusion from high-value social networks. This to some extent illustrated the trade-off role of women’s empowerment in mitigating the dark side of social capital.

The above literature suggested that empowered women—who have higher agency, decision-making power, and resource access—are more enabled to leverage social capital effectively for economic gains. Therefore, we frame the following two hypothesis:

Figure 1 illustrates the Hypotheses Framework of this paper.

Data used in this study came from a Rural China and Food Security Household Longitudinal Survey in 7 poverty-stricken counties of 4 western provinces of China conducted in 2021. The Survey was started in 2010 and revisited every 3 years. In the 2021 Survey, the research team refined the “women’s empowerment” module with more questions related with social capital, allowing authors to construct the social capital variable for the purpose of this study. In addition, the 2021 dataset provides a timely perspective for analyzing these questions within the broader context of China’s policy transition from elimination of extreme poverty to the prevention of large scale return to poverty.

The 7 surveyed counties were: Zhen’an and Luonan counties in Shaanxi Province, Wuding and Huize counties in Yunnan Province, Panzhou and Zheng’an counties in Guizhou Province and Qingshui county in Gansu Province. The above four provinces held almost one-third of the 592 previously poverty-stricken counties on the 1994 government list for poverty reduction interventions in China’s national program. County selection prioritized the following criteria: alignment with 271 counties identified by Xiao and Nie (2010) as the least food-secure counties, based on a cluster analysis of food security status across 592 poverty-stricken counties; ensuring coverage of diverse agroecological zones in Western China; feasibility of implementation, including the willingness to cooperate of local authorities, although it brought limitations to sampling procedure. 16 or 19 villages in each county were selected through Probability Proportional to Size (PPS) sampling. 12 households were selected randomly in each village.

Data was obtained through interviews with a multi-module questionnaire, including household demographic characteristics, household expenditure, women’s participation in household decision-making as well as women’s group participation and community engagement. We also conducted interviews with village committee leaders with a structured questionnaire on village characteristics, including its transport infrastructure and market access. These data allow us to explore the relationship between women’s empowerment, social capital and household poverty, while controlling for individual, household and village characters.

For this study, we restricted samples to dual-adult households, who had decision-making activities on all selected indicators. This was because household decision-making indicators in the Survey was designed to involve the level of decision input between couples. By restricting the sample to dual-adult households, the study ensures that the analysis captures intra-household bargaining dynamics, which would not be observable in single-adult households. As a result, we got 1,239 samples out of the 1,556 surveyed households surveyed. Caution is warranted when generalizing conclusions to all household types in the study areas.

Table 2 is the statistical description of variables for this study.

We first estimate a probit model to examine the relationship between women’s empowerment and household poverty:

Y i ∗ = α 1 + β 1 WE i + ξ 1 X i + ε i 1 , ε i 1 ∼ N ( 0 , 1 )

Next, we employ a causal mediation analysis (Imai et al., 2010) with social capital as the mediator:

Y i ∗ = α 3 + β 3 WE i + γM i + ξ 3 X i + ε i 3 , ε i 3 ∼ N ( 0 , 1 )

where Y i =1 indicates that household i is poor; WE i represents women’s empowerment; X i includes individual-, household- and village-level controls; M i denotes social capital. α 1 , α 2 , α 3 , β 1 , β 2 , β 3 , ξ 1 , ξ 2 , ξ 3 and γ are parameters to be estimated, and ε i 1 , ε i 2 and ε i 3 are error terms.

We performed Confirmatory Factor Analysis (CFA) for the women’s empowerment variable (WE). We got a good fit model by setting the error term of two indicators related, namely decision on farm activities and decision on non-farm activities. The goodness-of-fit indices included the chi-squared test, the root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the standardized root mean square residual (SRMR). According to Nasser-Abu Alhija (2019), we considered a model to be fit if CFI > 0.95, RMSEA<0.08, SRMR<0.06. Although a non-significant χ² and a small χ²/df ratio are also expected for a well-fitting model, it can be exempted if the sample size is big enough. We manually calculated the Composite Reliability (CR) and got 0.778. CR is considered as good it is more than 0.70. Our result suggested satisfactory internal consistency of a construct, meaning the items are closely related and capture the same underlying concept.

Table 3 presents the standardized coefficients (factor loadings) with their standard errors and the value of good fit indices.

A new variable (Latent WE) was then created, containing observation-by-observation values of estimated factor scores (meaning predicted values of the latent variable WE). These factor scores were calculated with an analog of regression scoring method following CFA, which combined information from all observed indicators (StataCorp, 2017). The new variable is a continuous one with a mean of approximately zero and a standard deviation of 0.90 (See Table 2).

To test H1-that women’s empowerment positively contributes to poverty reduction, we got estimates under six model specifications, as presented in Table 4. Columns (1)–(4) adopted a probit approach, with latent WE as the key explanatory variable, and control variables added progressively at women’s individual, household and community levels. Across all these models, the coefficient of latent WE is consistently negative and statistically significant, indicating that higher level of women’s empowerment reduces the probability of a household being poor. The magnitude of the marginal effect has decreased slightly as more covariates are introduced; however, the negative relationship remains robust.

Column (5) replaced latent WE with composite WE in the probit model. The results again reveal a negative and significant relationship between women’s empowerment and household poverty, though with a smaller marginal effect. This finding further confirms the robustness of the relationship.

To address the potential endogeneity concerns, such as omitted variables that may affect both empowerment and poverty, or the reverse causality between the two, we adopted an instrumental variable approach. Here, “decision-making power of a woman’s natal family prior to marriage” was selected as the instrumental variable. Theoretically, this variable is relevant to women’s empowerment: women exposed to female decision-making authority in their natal households are more likely to develop autonomy and bargaining capacity, which in turn shapes empowerment within their marital households. Meanwhile, decision-making power of a woman’s natal family prior to marriage cannot directly affect the current household’s poverty status, satisfying the exogeneity condition.

The first-stage results of the IV-Probit confirm the relevance of the instrument: the coefficient of natal-family decision-making power is 0.194 (p < 0.001), which is statistically significant. Additionally, the Wald test of exogeneity (chi2(1) = 3.10, p = 0.078) indicates potential endogeneity in the probit model, thereby supporting the preference for the IV specification.

Column (6) presents the IV-probit estimates: the coefficient on women’s empowerment remains significantly negative (p < 0.01), confirming that empowerment exerts a causal effect in lowering the likelihood of household poverty. The corresponding marginal effect is substantially larger than the simple probit estimate, underscoring the downward bias when potential endogeneity is ignored.

For control variables, square of women’s age became statistically insignificant once the household- and community-level variables were included in the model. In contrast, women’s education was consistently associated with lower poverty probability. Additionally, a larger number of children in the household increased the likelihood of poverty. At the community level, village access to markets significantly reduced household poverty probability while access to bus services showed no significant association.

Two specifications were employed for the causal mediation analysis, corresponding to two measures of social capital: a continuous social capital score (SOC) and a binary social capital variable (BSOC). The results presented in Table 5 reveal two key findings that support the relevant hypotheses. First, women’s empowerment significantly increases women’s social capital (SOC: = 0.211, p < 0.01; BSOC: = 0.286, p < 0.01), which provides evidence to accept H2. Second, women’s social capital significantly contributes to household poverty reduction (SOC: = − 0.128, p < 0.01; BSOC: = − 0.189, p < 0.1), and these results further support the acceptance of H3.

Table 6 summarizes the mediation effects of social capital. Specifically, SOC mediates 23.4% of the total effect of women’s empowerment on poverty, whereas BSOC mediates only 7.1% of this total effect. These results confirm H4: women’s social capital mediates the relationship between women’s empowerment and household poverty reduction.

Notably, the substantial gap between the mediation effects of SOC and BSOC underlines a key insight: it is the breadth and density of women’s social networks—not merely their affiliation with a single group—that effectively translates gains in women’s empowerment into sustained household poverty reduction.