To stimulate farmers’ HSEIBs, it is necessary to clearly know the attribute of RHSE. Firstly, RHSE is a communal and collective public good, and its public good attribute determines that transaction costs are higher and property rights are difficult to determine (Li et al., 2021). Secondly, the funding for promoting farmers’ HSEIBs is raised through a joint payment system between the public and private sectors, demonstrating the club goods characteristics of a “two-part tariff” (Feldstein, 1972; Zhang F. et al., 2024). Specifically, the “two-part tariff” theory advocates separating the annual membership fee from the cost of purchasing club goods, thereby reducing free-rider behavior and improving the efficiency of public good provision. In the case of household toilet hygiene renovation and sewage treatment, the public sector is responsible for constructing wastewater disposal infrastructure, providing farmers with the prerequisite conditions for scientifically treating wastewater and transforming their toilets for hygiene. And farmers must bear the cost of connecting household pipelines to the public wastewater system and acquiring necessary ancillary equipment. In the “two-part tariff” model, whether increasing village-level public investment in the “first part” or improving household income in the “second part” can significantly promote households to properly manage domestic wastewater and use sanitary toilets (Shao et al., 2024). In addition to economic factors, according to behavioral mimicry theory, farmers’ HSEIBs are also influenced by social factors, including their own perception of policies and the influence of surrounding individuals (Lede et al., 2019; Sun et al., 2020).

Based on previous Chinese data, it suggests that there may be a positive correlation between DVD and farmers’ HSEIBs. So, how does DVD influence farmers’ HSEIBs? By incorporating the attributes of farmers’ HSEIBs, we discussed the three theoretical mechanisms, including effects of enriching villages, enriching villagers and social interaction. The theoretical framework of this study is shown in Figure 1.

The research data used in this article is sourced from the CLES and the CDRI jointly released by CIRDNC and Alibaba Research Institute. We ultimately selected the 2020 and 2021 CLES data, as well as the corresponding CDRI data for the study.

The CLES is conducted by Nanjing Agricultural University in Jiangsu Province, covering aspects like agricultural production, household welfare and participation in public affairs. The survey uses the PPS sampling method, first selecting two counties from each city and then selecting households from two administrative villages in each county. Given the significant economic disparities in the northern, central and southern regions of Jiangsu Province, the CLES sample is nationally representative (Chen and Ye, 2024). The CDRI database presents the development of digital rural areas in Chinese counties, including the comprehensive index of DVD, four secondary indicators and multiple tertiary indicators. For research purposes, we matched these two databases based on the county address information and removed samples with missing data for key variables, resulting in 1054 observations of farm households.

The dependent variable is farmers’ HSEIBs. The treatment of domestic sewage and the use of sanitary toilets are key priorities of RHSE. Based on existing research (Shao et al., 2024; Zhang F. et al., 2024), we measure farmers’ HSEIBs from the two aspects of scientifically treating domestic sewage and using sanitary toilets. Specifically, methods of domestic wastewater treatment primarily include discharging into sewers, using dedicated containers for collection, and direct discharge. We define the method of directly discharging domestic wastewater as non-scientific treatment, while other methods are considered scientific treatment. Regarding the use of sanitary toilets, if a farming household has a toilet with harmless treatment capabilities, it is considered that they are using a sanitary toilet; otherwise, it is regarded as not using one.

The core explanatory variable is DVD. We adopt the “comprehensive Index for Digital Village Development” to represent it. In addition, in the exploration of heterogeneity, we also selected four secondary indicators for analysis, including the “Rural Infrastructure Digitalization Index,” “Rural Economic Digitalization Index,” “Rural Governance Digitalization Index” and “Rural Life Digitalization Index.”

For control variables, we refer to existing studies (Sun et al., 2020; Li et al., 2021; Li Z. et al., 2024) and research objectives and select the following: household characteristics, including the gender, age, education level and political status of the household head; family characteristics including family size and the proportion of family labor force; county-level characteristics, including per capita GDP and population density. Notably, in China, the political status of whether one is a Party member will affect their leading role in HSEIBs (Zhang Y. et al., 2024).

Mechanism variables include three aspects. The first is the development of village collective economy, represented by per capita operating income and per capita environmental governance expenditure in the village. The second is household income, represented by the logarithm of per capita household income. The third is social interaction variables, including farmers’ policy perception, village cadres’ leadership and trust among villagers, measured by farmers’ understanding of RHSE policies, farmers’ trust in village cadres and farmers’ trust in neighbors, respectively.

The empirical work mainly estimates the impact of county DVD on farmers’ HSEIBs. HSEIBs in this study are binary dummy variables, and the Probit model is often used. However, the efficient operation of sanitation toilets typically relies on the availability and functionality of supporting public sewage infrastructure. Therefore, the toilet renovation in rural areas is seamlessly integrated with domestic wastewater treatment. Estimating the two behaviors of domestic wastewater treatment and sanitary toilet use separately using a Probit model fails to accurately capture their interaction and does not allow for effective comparison (Zhang F. et al., 2024). Considering the interrelationship between two types of HSEIBs, we adopt Multivariate probit (Mvprobit) models that can simultaneously handle multiple binary choices. The models are specified as follows:

For the dependent variable, the equation can be set as:

In Equations 1–3, Y m i =1 and Y m i =0 indicate, respectively, that farmers i adopt or do not adopt the m t h behavioral change in RHSE. Y 1 i ∗ and Y 2 i ∗ denote, respectively, whether farmers i use a sanitary toilet and whether they scientifically treat domestic wastewater. D V D i is the core explanatory variable for the digital village development level of the county. X i represents a series of control variables. δ j denotes fixed effects for cities. θ j denotes fixed effects for years. β 0 is the constant term. ω i is an error term that follows a multivariate normal distribution with zero means and unit variances. By maximizing the likelihood function in the above equation, we can obtain the estimated parameter β ′ .

Then, to validate effects of enriching villages, enriching villagers and social interaction, we construct the linear Ordinary Least Squares (OLS) model, as follows:

In Equation 4, M v i represents the per capita operating expenditure and per capita environmental governance expenditure of the village where household i is located, as well as the per capita household income of household i . Additionally, it encapsulates the policy perception, trust in village cadres and trust in neighbors of farmer i . α 0 is a constant term. ϵ i is the random error term, α 1 and α 2 are the parameters to be estimated. The meanings of the other symbols are roughly the same as in Equation 1.

In the application of OLS models, two points need to be clarified. First, given that the income and expenditure variables are continuous, we decided to use the OLS model. Prior to formally applying the OLS model, we conducted preliminary regression analyses on the income and expenditure variables separately and confirmed that these variables meet the prerequisites of the OLS model, which specifically include linearity, independence of observations, normal distribution of residuals, homoscedasticity, and the absence of multicollinearity. Second, regarding the variables related to social interaction mechanisms, they fall into the category of ordinal categorical variables. Ordinal logit models are typically an appropriate choice for such variables. However, following the study by Güneri et al. (2022), it is necessary to verify the proportional odds assumption before using ordinal logit models. Unfortunately, our data did not pass this test, and even when attempting to use a generalized ordered logit model with relaxed assumptions, the regression results failed to converge. Therefore, we opted to use the OLS model for validation. The reason is that, when the dependent variable is ordinal, if the model is correctly specified, the econometric regression results from the OLS model, including the coefficients and their significance levels, are substantively not significantly different from those obtained through an ordinal logit model (Ferrer-i-Carbonell and Frijters, 2004). Similarly, we also conducted regression analyses on the variables related to policy perception and trust, and confirmed that they satisfy the prerequisites of the OLS model.

Descriptive statistics of the main variables in the empirical analysis are shown in Table 1. Meanwhile, to visually observe the positive correlation between DVD and farmers’ HSEIBs, we conducted comparisons of differences between groups. The results are presented in Table 2. Overall, the results are as expected. However, these results have not considered the interference of other factors and thus require further empirical analysis. It is worth noting that the test between groups for using sanitary toilet did not pass, which indirectly reflects the necessity of establishing the Mvprobit model. That is, the use of sanitary toilet relies on the establishment of a sewage system.

Table 3 reports the impact of DVD on farmers’ HSEIBs. Columns (1) and (2) report the estimation results of Probit models, while columns (3) and (4) report the estimation results of the Mvprobit regression. Mvprobit models revealed that the correlation coefficient between farmers’ use of sanitation toilets and domestic wastewater treatment was 0.370, significant at the 1% level. This suggests a strong association between farmers’ HSEIBs, indicating that the adoption of sanitary toilets is often accompanied by the scientific treatment of domestic wastewater. Regression results in columns (3) and (4) show that DVD significantly increases the probability of scientifically treating domestic wastewater and using sanitary toilets. Based on the above analysis, H1 is confirmed.

According to the results of controlling variables, households with a higher level of education are more likely to use sanitary toilets, which is consistent with previous research findings (Sun et al., 2020). The reason lies in two main aspects. On the one hand, individuals with higher education levels possess more health knowledge and have a deeper understanding of the health risks associated with not using sanitary toilets. On the other hand, they generally have a stronger awareness of environmental protection and sustainable development, and are more clearly aware of the negative impacts that unhygienic waste disposal can have on the environment. Therefore, they are more likely to choose environmentally friendly sanitation facilities. In addition, the estimated coefficient of per capita GDP at the county level is significantly negative. This may be due to some regions prioritizing GDP growth while neglecting environmental protection (Duan et al., 2024), lacking public investment in environmental governance, which is not conducive to HSEIBs.

Firstly, replacing the dependent variable. We use “farmers’ satisfaction with the ecological livability of their village” as a substitute variable for farmers’ HSEIBs. This is because if DVD improves RHSE, it will ultimately enhance farmers’ satisfaction with the ecological livability of their village. Given that the categorical variable did not satisfy the parallel regression assumption of the ordinal probit model, but did meet the prerequisites of the OLS model, we chose to use the OLS model for regression analysis. Secondly, eliminating potential sources of interference. Farmers’ HSEIBs may be influenced by the level of village development. For instance, villages that have been awarded the title of “Civilized Village” generally possess a higher level of ecological civilization construction, which in turn encourages farmers’ actions to be more environmentally friendly. Thus, we exclude the sample of civilized villages. The empirical estimation results are shown in Table 4. The estimated coefficients of DVD are all positive and statistically significant. In conclusion, the results of the benchmark regression can be considered robust.

Although a range of control variables were selected in the baseline regression, there may still be potential omitted variable bias. Thus, we conducted an omitted variable bias test. Specifically, based on the sensitivity analysis used in existing studies (Aidt and Franck, 2015; Li X. H. et al., 2024), the Ratio coefficient is constructed to identify the strength of bias caused by unobservable variables through observable variables. The Ratio coefficient is constructed as follows: Ratio = | β ̂ 2 / β ̂ 2 − β ̂ 1 | . β ̂ 2 and β ̂ 1 represent the estimated coefficients of the core explanatory variables under all possible control sets and finite variable sets, respectively. When Ratio > 1 , it can be considered that the selected observable variables have strong explanatory power and potential omitted variable bias can be ignored.

Due to the complexity of the probit model, the OLS model can be used here to estimate the ratio parameter. As shown in Table 5, the minimum Ratio value calculated for different sets is 1.833, indicating that if the interference of omitted variables is considered, the explanatory power is at least 1.83 times that of the selected control variables. Therefore, it can be considered that the impact of omitted variables on the regression results is relatively small.

As evidenced by theoretical analysis, DVD can promote farmers’ HSEIBs through effects of enriching villages, enriching villagers and social interaction. Based on Equation 4, we further conduct empirical analysis and the regression results are shown in Tables 6, 7. It is worth noting that some mechanism variables are missing, so the regression sample and the benchmark regression are not completely consistent.

In Table 6, the impact coefficients of DVD in columns (1)–(3) are all positive and have passed the significance test. With the advancement of DVD, the financial strength of villages has been enhanced, which has led to an increase in public expenditure on environmental governance in villages, meaning the “first part” investment in RHSE. At the same time, DVD also creates favorable conditions for increasing farmers’ income, providing a foundation for the “second part” investment. Therefore, H2 and H3 are verified.

Table 7 columns (1)–(3) report the estimated results of DVD on policy perception, trust in village cadres and trust in neighbors, with coefficients of 0.045, 0.060 and 0.042, respectively, all passing the significance test at the 1% level. Thus, H4 is confirmed.

Existing studies mostly focus on social interaction mechanisms. For example, Liu et al. (2022) and Yi et al. (2024) explore how to promote farmers’ HSEIBs, from the perspectives of village cadres’ public service motivation and leadership. Sun et al. (2020) investigated the willingness and behavior of farmers to improve RHSE from the dual perspectives of social norms and social supervision. Unlike previous studies, our analysis is grounded in the “two-part tariff” model for RHSE, which reflects both club goods’ characteristics and the collective action dynamics. It not only embraces the social interaction mechanisms highlighted in prior research, but innovatively examines economic mechanisms, thereby offering a more comprehensive perspective.