The dependent variable in this paper is farmers’ E-Commerce participation. Referring to the study of Qiu et al. (2024a, 2024b), the questionnaire “whether your family operates and has products traded through the network” is selected for measurement, and the answer of “yes” is assigned as 1, which means that farmers have E-Commerce participation behaviour; the answer of “no” is assigned as 0, which means that farmers do not have E-Commerce participation behaviour. The answer of “no” is assigned as 0, which means that the farmers do not have E-Commerce participation behaviour.

The core independent variables in this paper are credit access, formal credit access and informal credit access. Credit access refers to the behaviour and process of economic entities to obtain funds from capital suppliers to cope with shocks or carry out production and investment. Formal credit acquisition specifically refers to the process where economic entities successfully obtain loans from formal financial institutions. The core characteristics of this type of credit acquisition lie in its institutionalization, regulatory oversight, and contractual nature. In contrast, informal credit acquisition primarily relies on social networks, geographical connections, kinship ties, or personal trust. Its defining features include non-regulatory characteristics, relationship-dependent mechanisms, and flexible adaptability. Referring to the study of Yang et al. (2021), credit access is measured by the indicator of farmers’ credit access, which is regarded as access credit and assigned the value of 1, otherwise it is assigned the value of 0. At the same time, credit access is classified according to the source of borrowing funds. Borrowing from formal financial institutions—including but not limited to commercial banks (e.g., state-owned and joint-equity banks), rural credit unions, and other regulated entities such as licensed microfinance institutions (MFIs) or online lending platforms—is regarded as formal credit access and assigned a value of 1. Borrowing from all other sources (e.g., informal lenders, friends, or relatives) is assigned a value of 0. Borrowing from informal financial institutions, such as relatives, friends and private financing platforms, is regarded as informal credit access and assigned the value of 1, otherwise it is assigned the value of 0.

Based on the previous theoretical analysis, it is known that credit access promotes farmers’ E-Commerce participation by expanding the scale of land operation. Under the land equalization and confirmation system, credit support mainly promotes E-Commerce participations by alleviating the financial constraints of land scale expansion. Specifically, credit is primarily used to pay land rents, level land and build water conservancy facilities, thus directly driving the expansion of land management scale, which is the most direct and core transmission mechanism of credit affecting E-Commerce participations. Referring to the study of Rogers et al. (2021), the indicator of land transfer area, which represents the newly transferred-in land area and is a continuous variable, is used to measure the land management scale.

With reference to the studies of Liu et al. (2020), Li et al. (2021a, 2021b), and Zhang and Zhang (2024), this paper introduces three levels of variables, namely, individual characteristics of the head of the household, household characteristics, and regional characteristics, as control variables. Variables such as gender, age, marital status, education level, political profile, and health status are selected at the individual farm household level; variables such as the number of family labour force, family economic status, and family social capital are selected at the household level; and variables such as village topography, village traffic conditions, village economic level, and regional economic level are selected at the regional level. The descriptive statistical characteristics of all variables are shown in Table 1. We outlined the measurement criteria for each variable in the questionnaire, along with their corresponding mean values and standard deviations.

This paper compares the E-Commerce participation of farmers with and without credit access as shown in Table 2. There is a significant difference in the probability of E-Commerce participation between credit-accessed and non-credit-accessed farmers. Farmers with credit access exhibit a 0.079 higher probability of E-Commerce participation. Those with formal credit access show a 0.071 higher probability, while those with informal credit access show a 0.096 higher probability. From this, it can be preliminarily concluded that credit access plays an important role in promoting E-Commerce participation of farmers. Of course, a more rigorous econometric analysis is still needed to confirm the impact effect of credit access on farmers’ E-Commerce participation. The following section will empirically analyse the impact of credit access on farmers’ E-Commerce participation.

The dependent variable in this study is the E-Commerce participation, represented as a binary categorical variable. Therefore, a binary probit model is constructed as shown in Equation 1:

In Equation 1, where Y denotes farmers’ E-Commerce participation; C denotes credit access, formal credit access and informal credit access, X denotes a series of control variables; α 0 is a constant term; α 1 , α 1 are regression coefficients; σ is the error term.

After estimating the baseline probit models, we pre-registered a multi-pronged robustness protocol in our internal analysis plan. Specifically, we (i) re-estimate all equations with a logit specification to check functional-form sensitivity; (ii) replace key variables with alternative proxies (variable-replacement tests); (iii) exclude potentially contaminated sub-samples (exclusion-sample checks); (iv) construct a balanced sample using Propensity Score Matching (PSM) to mitigate selection bias; and (v) implement the Extended Regression Model (ERM) framework with an instrumental variable to accommodate endogeneity in a non-linear setting. Detailed results of each validation step are reported in Section 4.2 and 4.3.

In order to explore the mechanism of credit access on farmers’ E-Commerce participation, the following mediation effect model is constructed with reference to Baron and Kenny’s (1986) study:

In Equations 2–4, Y denote farmers’ E-Commerce participation; C denote credit access, formal credit access and informal credit access; M denotes the mediator variable, and X denote the control variable; a 0 , b 0 , c 0 , are the constant term, and a 1 , b 1 , c, c 1 , a 2 , b 2 , c 2 , are the coefficient to be estimated; and ε 1 , ε 2 , ε 3 are the error term.

In order to test the robustness of the above results, this paper adopts the binary Logit model for re-estimation, and the results are shown in Table 4. The results show that credit access, formal credit access and informal credit access still have significant positive effects on farmers’ E-Commerce participation, and informal credit access is more able to promote farmers’ E-Commerce participation, which supports the robustness of the previous results.

In addition to replacing the model, this paper further conducts robustness tests by replacing the variables. The amount of credit can reflect the degree of credit access of farmers. Therefore, this paper adopts total credit amount, formal credit amount, and informal credit amount to measure credit access, formal credit access, and informal credit access, respectively, and re-runs the regression analysis, and the results are shown in Table 5. The results show that after replacing the variables, credit access, formal credit access and informal credit access still have a significant positive effect on farmers’ E-Commerce participation, and informal credit access is more likely to promote farmers’ E-Commerce participation, which supports the robustness of the previous results.

Considering the special characteristics of farmers under 16 and over 64 years old in terms of physical ability and health status, according to the judgment standard of full labour force, this paper only retains the sample of farmers aged 16–64 with labour capacity and re-estimates it (Qiu et al., 2024a, 2024b), and the results are shown in Table 6. The results show that after deleting part of the sample, credit access, formal credit access and informal credit access still have a significant positive impact on farmers’ E-Commerce participation, and informal credit access is more able to promote farmers’ E-Commerce participation, further verifying the robustness of the previous results.

Considering that credit access belongs to the self-selected behaviour of farmers, the previous model may have the endogeneity problem caused by self-selection bias, which can be effectively solved by the Propensity Score Matching (PSM) method. Therefore, this paper adopts Propensity Score Matching (PSM) to re-estimate and get the average treatment effect (ATT) of the impact of credit access on farmers’ E-Commerce participation, and the results are shown in Table 7. The results show that credit access, formal credit access, and informal credit access have significant positive effects on farmers’ E-Commerce participation, and informal credit access has a greater effect on farmers’ E-Commerce participation, regardless of whether it is estimated using propensity score matching methods such as nearest-neighbour matching method, caliper matching method, kernel matching method, local linear regression matching, spline matching method, or Mahalanobis matching method.

The PSM model specification needs to satisfy the two prerequisites of overlap assumption and equilibrium properties (Caliendo and Kopeinig, 2008). For the overlap assumption the public support domain needs to be tested and Figure 3 shows the distribution of propensity scores for testing the public support domain. As can be seen from the figure, after matching, the credit access sample and the Non-credit access samples almost overlap in the propensity scores and there are large common support intervals, which suggests that the matching is more reasonable.

In order to examine whether the above propensity score matching method balances the data better, a balance test is needed. Taking the nearest-neighbour matching method and credit access as an example, the test results are shown in Table 8. The results show that compared with the pre-matching period, the standardised deviation rate of the control variables has basically been reduced after matching, and the standardised deviation rate is basically less than 10%. At the same time, the t-test results of most control variables do not reject the original hypothesis that there is no systematic difference between the experimental group and the control group, effectively balancing the differences in the distribution of covariates between the experimental group and the control group. Therefore, the propensity score matching results passed the balance test.

In addition to self-selection bias, there may be a mutually causal endogeneity between credit access and farmers’ E-Commerce participation. On the one hand, farmers use credit access to alleviate financial constraints and then engage in E-Commerce participation activities; on the other hand, E-Commerce participation requires sustained financial investment, which will further stimulate farmers to actively seek financial support. In addition, omitted-variable concerns centre on unobserved entrepreneurial ability, risk appetite and household wealth that simultaneously determine borrowing and the decision to open an on-line store; if these traits are positively correlated with credit take-up, benchmark regression model will overstate the true impact. Measurement error is equally relevant: E-Commerce participation is self-reported and may be under-declared by respondents who fear tax scrutiny, while the credit-use variables capture only loans that households are willing to reveal, creating classical attenuation bias. Both issues motivate the instrumental-variable strategy and robustness checks reported below.

For this reason, this paper adopts Extended Regression Models (ERM) for estimation. Referring to the study of Qiu et al. (2024a, 2024b), the proportion of other farmers’ credit access in the village is selected as an instrumental variable in this paper. This variable satisfies the two core conditions for a valid instrument. Rural credit markets in China are information-intensive and highly referral-based. Loan officers rely on local social networks to screen clients; a higher Peer-share mechanically increases the pool of successful borrowers who can vouch for, co-guarantee, or simply inform their neighbours about upcoming bank visits. Consequently, a one-percentage-point rise in the peer-share raises the probability that any given household is offered a loan contract. However, the peer-share should affect a household’s decision to engage in E-Commerce only through its own credit status. Neighbours’ loans do not directly supply the target household with internet skills, online storefronts, or logistics services; they merely loosen its liquidity constraint.

The results of the endogeneity test of credit access on farmers’ E-Commerce participation are shown in Table 9. Both the F-stat lie well above conventional weak-instrument thresholds, confirming that the peer-share is a strong predictor of individual credit access. The results show that the corr value is significant at 1% statistical level, indicating that the original hypothesis that the variables are exogenous is rejected, i.e., there is an endogeneity problem. After considering the endogeneity problem of mutual causality, credit access, formal credit access, and informal credit access still have a significant positive effect on farmers’ E-Commerce participation, and the promotion effect of informal credit access on farmers’ E-Commerce participation is stronger, which further supports the reliability of the results of the previous study.

However, with only one instrument we cannot perform over-identification tests; the exclusion assumption rests on theoretical reasoning and the rich set of village-level controls. Future work that exploits multiple policy-based instruments (e.g., sequential county-level credit quotas) could further buttress identification.

The three dimensions of topography, agricultural functional zoning and geographical location constitute the basic structural constraints affecting the development of rural E-Commerce. Therefore, this paper conducts group regressions on the samples at the topography, agricultural functional districts, and geographical position, respectively, and the results are shown in Table 11. The results show that the impact of credit access on farmers’ E-Commerce participation is heterogeneous at the level of terrain, functional agricultural subdivisions, and geographic location, and the research hypothesis H3 is confirmed. In particular, credit access promotes E-Commerce participation of farmers in plain areas, food-producing areas and central-eastern areas, and it is worth noting that informal credit promotes E-Commerce participation of farmers in western areas. The reasons may stem from the relatively well-developed infrastructure, efficient logistics networks, and mature agricultural industrialization in plain regions, which enable credit funds to be effectively converted into actual E-Commerce production capacity. In major grain-producing areas, the established large-scale production systems and stable agricultural supply allow credit injections to more easily extend industrial chains and enhance product commercialization rates. Geographically, central and eastern regions demonstrate higher marginal effects of credit utilization due to their greater market potential and more complete digital infrastructure. Notably, in western regions, informal credit has shown unique value. Its flexible and accessible nature better accommodates the relative lack of formal financial services, serving as a crucial supplementary financing channel for farmers to explore E-Commerce participation.

Considering the wide application of digital technology in rural finance, this paper further explores the differences in the effects of different credit access methods on farmers’ E-Commerce participation. It not only overcomes the theoretical limitations of traditional research that homogenizes credit, but also provides empirical evidence for designing inclusive financial policies tailored to farmers’ actual needs, thereby effectively improving the allocation efficiency of financial resources in rural E-Commerce. Referring to the study of Su et al. (2021), based on the digital differences in borrowing methods, the credit access methods of farmers are divided into digital credit and traditional credit, and the effects of these two credit access methods on the E-Commerce participation of farmers are examined separately. Traditional credit access primarily relies on physical branches and collateral guarantees, with standardized offline approval processes. Digital credit access, however, leverages digital platforms and big data-driven risk models to enable full-process online operations. While the former depends on collateral and credit history, creating higher service barriers, the latter conducts online credit assessments, offering greater inclusivity and efficiency. The results are shown in Table 12. The results indicate that digital credit significantly promotes farmers’ E-Commerce participation, while traditional credit also exerts a significant positive impact on farmers’ E-Commerce participation, yet its effect is considerably smaller. The possible reason is that traditional credit access is complicated, requires collateral and other assets, and the threshold for farmers to obtain credit is higher; while digital credit can use information advantages to efficiently match borrowing and lending needs, and the borrowing and lending process is relatively simple, which makes it more convenient for farmers to obtain funds in a timely and stable manner to engage in E-Commerce participation activities.

Referring to the study of Cai et al. (2018), this paper takes the mean value of E-Commerce sales as the cut-off point, divides E-Commerce participation into scale E-Commerce and non-scale E-Commerce, and explores the impact of credit access on different E-Commerce participation scales, and the results are shown in Table 13. The results show that compared with non-scale E-Commerce, credit access can more significantly promote scale E-Commerce participation, which indicates that the initial small-scale E-Commerce participation has a lower demand for funds, but as the scale of E-Commerce participation expands, the importance of credit access grows progressively more salient. It is worth noting that formal credit access has a significant positive effect on large-scale E-Commerce participation, but the effect of informal credit access on large-scale E-Commerce participation is not significant. This suggests that informal credit plays an important role in the initial stage of E-Commerce participation, but as E-Commerce participation scale up, they become more dependent on formal credit support. The possible reason for this is that informal credit carries high implicit costs due to favors, imposing moral constraints on profits and leading entrepreneurs to repay loans quickly rather than reinvesting in operation expansion. Moreover, the scale of capital supply from informal credit, based on trust within a limited circle, is constrained and has a limited impact on expanding E-Commerce participation.

Currently, there are two main E-Commerce sales modes for agricultural products in rural areas, one is the platform E-Commerce mode in which farmers open online shops on E-Commerce platforms such as Jingdong, Taobao and Pindoduo to sell their agricultural products, and the other is the social E-Commerce mode in which farmers sell their agricultural products with the help of social tools such as WeChat, Xiaohongshu and Tik-Tok (Liu et al., 2021). This paper further explores the impact of credit access on different E-Commerce participation models, and the results are shown in Table 14. The findings reveal that credit access, formal credit access, and informal credit access significantly facilitate social e-commerce business. The possible reason for this is the social E-Commerce model offers a strong social aspect, low costs, and asset-light operations, with less financial constraint than platform E-Commerce. It also features lower market entry barriers and simpler operational standards, enabling farmers to engage in E-Commerce with microcredit support.