Digital capability, household income, and intra-rural income inequality: micro-level evidence from China

1 Introduction

Common prosperity is a defining feature of Chinese modernization, and income provides one of its most direct measures. Yet the continued rise in rural residents’ per capita disposable income has not eliminated disparities within rural China (Luo et al., 2020). Data from the China Statistical Yearbook show that the Gini coefficient for rural per capita income reached 0.443 in 2022, exceeding the urban coefficient of 0.421. The income ratio between the highest and lowest quintiles of rural households also increased from 6.47 in 2000 to 9.52 in 2023. These figures point to persistent income inequality among rural residents. Narrowing disparities within rural communities is therefore central to advancing common prosperity. As digital village initiatives expand, digital technologies are becoming embedded in rural life, agricultural production, and business activities, creating new opportunities for agricultural development and rural prosperity. Their benefits, however, are not automatic. Effective adoption depends partly on whether farmers possess the capabilities needed to access, understand, and apply digital tools. Smallholders remain the backbone of China’s agricultural production and management. Strengthening their digital capability, alongside the digital transformation of agriculture, may therefore provide a new source of rural income growth. Against this background, the relationship between digital capability and farm household income has attracted growing academic attention. Early studies largely viewed individual digital capability as the ability to access information and communication technologies (Martin, 2006). As digital technologies have spread, the concept has broadened. Recent research generally defines digital capability as the capacity to acquire and use digital technologies to generate economic benefits and regards it as a new form of livelihood capital in the digital environment (Ragnedda et al., 2020).

However, academic consensus remains elusive regarding the association between digital capability and income among farmers. Digital dividend theory suggests that digital empowerment can raise farm household income, narrow income disparities, and reduce multidimensional poverty (Leng et al., 2020). Digital divide theory offers a different account. Because of unequal access, use, and skills, digital benefits may be captured disproportionately by better-endowed groups, leaving marginalized households unable to share in these gains (Helsper, 2012). Digital capability may therefore be associated with either a more equal income distribution or wider income differentiation, depending on who can acquire digital skills and convert them into economic returns. Qualitative reasoning alone cannot resolve this tension. Household-level evidence is needed to clarify how digital capability is related to income inequality and whether these relationships differ across rural groups. This issue matters not only for common prosperity and rural revitalization in China. Given the scale of China’s rural economy, sustained rural income growth and narrower disparities also contribute to global poverty reduction and sustainable development.

Existing empirical research exhibits the following limitations. Much of the existing evidence relies on aggregate data at the provincial or county level, leaving household-level differences insufficiently examined. Studies using household survey data often measure digital capability with a small number of simple indicators, which may not capture its multidimensional nature and can introduce measurement error. Research on income inequality presents another limitation. Common approaches focus on aggregate indicators, such as the urban-rural income ratio, or use quantile regression to examine distributional differences. Although informative, these methods reveal less about relative deprivation among individual farm households. Drawing on data from the 2020 China Rural Revitalization Survey, this study examines how digital capability is associated with two dimensions of common prosperity among farm households: income growth and income inequality. It also investigates the potential pathways underlying these relationships. The marginal contributions of this research are mainly reflected in three aspects. (1) The Kakwani index is introduced to measure the relative deprivation of farm household income. Compared with macro-level indicators, this index accurately captures micro-level inequality within the farm household group, providing more detailed empirical evidence for targeted policy formulation and the promotion of common prosperity among farm households. (2) Drawing on nationally representative household survey data, this study examines two potential pathways linking digital capabilities to common prosperity among farm households: stronger social networks and lower credit constraints. The findings provide micro-level evidence on the processes associated with farm household income growth. (3) A multi-dimensional measurement index system for farm households’ digital capabilities is constructed and incorporated into the analytical framework of common prosperity, and the intrinsic mechanisms through which digital capabilities influence the inclusive growth of farm household income are explored in depth, thus expanding and deepening the conclusions of relevant existing research.

2 Theoretical analysis and hypothesis

2.1 Theoretical analysis

Digital capability exerts multifaceted positive effects on farm household income growth. (1) Creating non-agricultural employment opportunities. With the vigorous development of the digital economy, numerous new digital business models continue to emerge. Enhancing farmers’ digital capabilities facilitates their integration into the digital economy, helps overcome traditional labor market constraints, and is associated with greater participation in non-farm employment and entrepreneurial activities, thereby contributing to income diversification (Wang et al., 2025). (2) Streamlining agricultural production and distribution chains. On the production side, digital capabilities assist farmers in more readily mastering advanced production techniques, thereby enhancing agricultural productivity. On the distribution side, digital capabilities encourage farmers to actively engage in e-commerce, broadening agricultural product sales channels and consequently increasing agricultural sales revenue (Sargani et al., 2025). (3) Enhancing financial market participation. Digital capabilities facilitate farmers’ access to financial knowledge, strengthening their capacity for financial transactions and investments. By optimizing household financial asset allocation efficiency and expanding diversified investment channels, this promotes growth in property income (Wang et al., 2023).

Because of limited initial endowments and insufficient capacity to use digital tools effectively, some low-income farm households remain unable to share fully in the benefits of the digital economy. This divide extends beyond access to digital infrastructure. It also involves systematic disparities in digital skills, patterns of use, and subsequent economic returns. Better-endowed households can more readily convert digital capability into tangible economic gains, increasing the risk of elite capture and wider income inequality. This distributional divergence, however, is not necessarily permanent. As the Digital Village initiative advances and digital technologies spread throughout rural areas, wider use of digital tools can reduce market frictions, improve information sharing, and support more efficient resource allocation across household groups (Van Deursen and Helsper, 2018). The inclusive and spillover effects of these technologies may also weaken the dominance of high-income households over key resources, narrowing disparities in resource access and creating additional income opportunities for low-income households. For low-income farm households, which generally face more severe capability gaps and resource constraints, stronger digital capability is particularly important for overcoming development bottlenecks and diversifying income sources. Their marginal income gains may therefore be more pronounced (Ma et al., 2020). High-income households, by contrast, already possess stronger resource endowments and more diversified income streams, leaving less scope for additional gains from digital empowerment. Differences in these marginal returns provide one potential channel through which stronger digital capability may narrow income disparities among farm households.

Based on the aforementioned theoretical analysis, research hypothesis H1 is proposed.

H1: Digital capability is positively associated with per capita gross income and negatively associated with income inequality among farm households.

2.2 Mechanism analysis

2.2.1 Social network expansion

Farmers with stronger digital capability are generally more likely to adopt digital technologies. Once adopted, these technologies allow real-time communication within social networks, reducing the costs of maintaining traditional social ties and helping farmers reinforce existing relationships. Internet-based tools also make it easier to overcome geographical and social barriers, thereby lowering the barriers to forming new connections (Neves, 2013). Farmers with greater digital capability can therefore use online platforms to establish relationships beyond their immediate communities and expand the reach of their social networks. Digital capability may thus contribute to both network expansion and stronger social ties.

Social networks provide important channels through which farmers obtain external resources. By supplementing or partially substituting for productive assets, these networks may improve labor productivity and support farm household income growth (Lu et al., 2013). The expansion of social networks associated with digital capability may be particularly valuable for disadvantaged farmers. Broader networks can help them address development constraints, access digital opportunities, and share more fully in the digital dividend, thereby contributing to common prosperity. The benefits of social networks, however, are unlikely to be evenly distributed. High-income farmers typically possess stronger initial resource endowments and more diversified income sources, making them less dependent on social networks. For low-income farmers, social ties can substantially reduce the costs of information search and factor acquisition, so their income growth may rely more heavily on network access. Knowledge and skills embedded in these networks also generate positive spillovers (Banerjee et al., 2024). Such spillovers may help low-income farmers overcome barriers associated with social stratification, gain access to higher-quality networks, and obtain greater economic returns (Fafchamps and Minten, 2002). Because these benefits may be larger for low-income farmers, the social network pathway can contribute to narrowing income disparities across household groups. Based on this reasoning, we propose the following research hypothesis:

H2: Social network expansion serves as a potential mediating pathway linking digital capability to higher per capita gross income and lower income inequality among farm households.

2.2.2 Credit constraint alleviation

Credit constraints are a major bottleneck to income growth and a source of widening income disparities among farm households. Traditional rural credit markets are characterized by severe information asymmetry, inefficient credit assessment, and a shortage of eligible collateral. Together, these problems create a pronounced “threshold effect” in the allocation of credit resources (Stiglitz and Weiss, 1981). Stronger digital capability may help ease this constraint through three pathways. First, digital capability can facilitate farmers’ deeper integration into the digital inclusive financial system. Digital verification mechanisms reduce the operating costs of financial institutions and the transaction costs faced by farmers, thereby improving credit accessibility from both the supply and demand sides (Li et al., 2020). A second pathway operates through information access. Stronger digital capability improves the efficiency with which farmers obtain financial information and helps break down the information silos common in traditional rural credit markets. This not only alleviates credit constraints arising from information asymmetry but also reduces passive credit limitations stemming from farmers’ tendency to apply blindly or their reluctance to pursue such opportunities(Munyegera and Matsumoto, 2016). Third, stronger digital capability encourages farmers to participate more deeply in digital production and consumption. The data generated through these activities form a “financial digital footprint,” which can partially substitute for the credit enhancement function traditionally provided by collateral. This lowers the threshold for credit access and increases farmers’ likelihood of obtaining loans (Berg et al., 2020).

The alleviation of credit constraints may have a stronger inclusive effect on low-income farmers. Adequate credit support can help them expand production, lower market entry barriers, and broaden sales channels, creating additional income-generating opportunities. At the same time, stronger digital capability improves farmers’ ability to use digital technologies effectively. The wider adoption of these technologies may, to some extent, curb elite capture in agricultural credit markets and correct the uneven allocation of credit resources. This has created fairer income-generating opportunities for low-income farmers and contributed to narrowing the income gap between different groups (Chen et al., 2024) (Figure 1).

Based on this, we propose the following research hypothesis:

H3: Credit constraint alleviation serves as a potential mediating pathway linking digital capability to higher per capita gross income and lower income inequality among farm households.

3 Data, variable and model

3.1 Data

The data in this paper is derived from the China Rural Revitalization Survey (CRRS) organized and implemented by the Institute of Rural Development at the Chinese Academy of Social Sciences in 2020. The survey encompasses multiple dimensions including agricultural production and management, farmers’ livelihoods, and social welfare. It employs a stratified multistage equal-probability random sampling method: firstly, 10 sample provinces were selected across eastern, central, western, and northeastern China (Guangdong, Zhejiang, Shandong, Anhui, Henan, Heilongjiang, Guizhou, Sichuan, Shaanxi, and Ningxia); Secondly, counties (cities, districts) were selected from each sample province by grouping them according to per capita regional GDP using equal-interval sampling. Thirdly, three townships were randomly selected from each sample county. Administrative villages within these townships were categorized into better-developed and less-developed groups based on development levels, with one administrative village randomly selected from each group. Finally, 12 to 14 farm households were selected from each sample village through systematic sampling and included in the field survey. In total, the survey covered 50 counties, 156 townships, and 300 administrative villages, collecting information from 3,833 households. The resulting sample is broadly representative and provides reliable information on the production and living conditions of farm households. After data completeness and usability were assessed, observations with missing values or outliers were excluded, leaving a final analytical sample of 2,703 households (Figure 2).

3.2 Variable

3.2.1 Dependent variable

This study uses two dependent variables: per capita gross income and income inequality. Per capita gross income is log transformed in the regression analysis to reduce right skewness in the income distribution and potential heteroskedasticity. Its untransformed value is retained when constructing the income inequality measure. Income inequality at the household level is measured using the Kakwani Relative Deprivation Index, which is grounded in relative deprivation theory (Kakwani, 1984). Consistent with previous research (Yang et al., 2019), the farm household serves as the unit of analysis. For each focal household, the reference group comprises all sampled farm households within the same province, including the focal household itself. Within this reference group, only households with higher per capita gross income than the focal household contribute to its relative deprivation score. The index is dimensionless and scale invariant, with values ranging from 0 to 1. A higher value indicates greater relative deprivation and a more disadvantaged position within the provincial income distribution. The resulting index thus captures each household’s relative income position and serves as a microlevel measure of income inequality. The index is calculated as follows:

RD ( y , y ip ) = 1 n p μ p { ∑ j = i + 1 n p ( y jp − y ip ) } = γ y ip + ( μ y ip + − y ip ) μ p (1)

In Equation 1, p denotes the province in which household i is located, and n p denotes the total number of sampled farming households in province p. The term y denotes the provincial distribution of per capita gross income. Households are arranged in ascending order of per capita gross income, such that y 1p ≤ y 2p ≤ ⋯ ≤ y n p p. The term y ip denotes the per capita gross income of household i; while j indexes households with higher per capita gross income than household i. The term μ p denotes the provincial mean per capita gross income. The terms μ y ip + and γ y ip + denote, respectively, the mean income and proportion of households whose per capita gross income exceeds y ip. Finally, RD ( y , y ip ) denotes the Kakwani relative deprivation index for household i, calculated using all sampled farming households in province p as the reference group.

3.2.2 Core explanatory variable

Digital capability. Digital capability refers to the ability to utilize digital technologies for information acquisition, evaluation, and processing, to engage in effective communication, and to solve real-world problems (Hargittai, 2001). Drawing upon existing relevant research (Lythreatis et al., 2022) and considering the actual economic characteristics of China’s rural areas, this paper constructs a digital capability indicator system for farmers comprising ten metrics across three dimensions: digital access, information acquisition, and application transformation. Details are presented in Table 1.

Table 1. Digital capability indicator system.

3.2.3 Control variables

To account for potential confounding factors, the model includes controls at the household head, household, and village levels (Wang, 2024). These variables capture household head demographics and human capital, household demographic and production characteristics, and village economic, geographic, and institutional conditions. To mitigate omitted-variable bias arising from unobserved province-level factors, province dummies are additionally included in the model. Detailed definitions are provided in the variable definition table.

3.2.4 Mechanism variables

(1) Social Networks. Existing research predominantly measures social network size by the number of close friends and relatives with whom individuals maintain frequent contact (Granovetter, 1973). Drawing upon this classical approach and considering the availability of survey data, this paper selects ‘the number of relatives and friends from whom one can borrow over CNY 5,000 (whether interest is charged or not)’ as a proxy variable for social network size. A higher value for this indicator indicates that the farmer possesses more abundant social network resources and a larger social network. (2) Credit Constraint Alleviation. Prior research generally identifies credit constraints when households require external financing but either do not apply for credit or have their applications rejected (Ren, 2021). Based on this definition, the study constructs a binary proxy for credit constraint alleviation using households’ access to formal bank loans. The variable equals 1 if a household had credit demand, applied for a bank loan, and obtained approval, and 0 otherwise. A value of 1 indicates successful access to formal credit and, correspondingly, a lower observed degree of credit constraint.

Definitions and descriptive statistics for the above variables are detailed in Table 2.

Table 2. Descriptive statistics results.

All variables are based on the same sample of 2,703 valid observations.

3.3 Model

To examine how digital capabilities are related to farmers’ income and income inequality, this paper constructs the following econometric model (see Equation 2):

Y i = α 1 + β 1 Digita l i + γ 1 X + ε i (2)

Where: Y i represents either Per capita gross income or the income inequality index per capita for households, characterized using the Kakwani index, corresponding to two separate regressions; Digita l i denotes the core explanatory variable representing household i’s digital capability level; X denotes the control variable.ε i denotes the random disturbance term;β, α and γ are the parameters to be estimated; β represents the adjusted association between digital capability and the dependent variable and is the primary coefficient of interest in this study. If β 1 > 0, it indicates that enhanced digital capability increases household income; if β 1 < 0, it suggests that improved digital capability mitigates income inequality among households.

To further analyze the mechanism through which digital capabilities influence farmers’ income and income inequality, drawing upon existing research (Jiang, 2022), Equation 3 was constructed to examine whether the core explanatory variables act through the mediating variables. Building upon this, to avoid potential issues where theoretical arguments alone may prove insufficient in addressing the causal effect of mediating variables on the dependent variable, reference was made to relevant studies (Dong et al., 2024), we further examine the effect of the mediating variables on the dependent variable by constructing Equation 4. As specified in Equations 3, 4 below, this study draws on the causal mediation analysis framework (Imai et al., 2010). The total effect of digital capability on the dependent variable is decomposed into the natural indirect effect, the natural direct effect, to test whether the mediating effect is statistically significant. To account for potential heteroskedasticity and within-village correlation in the error terms, all regressions report village-level cluster-robust standard errors.

M i = α 2 + β 2 Digita l i + γ 2 X + ε i (3)

Y i = α 3 + β 3 M i + c ′ Digita l i + γ 3 X + ε i (4)

M i denotes the mediating variable, representing social networks and credit constraint alleviation. As social networks are continuous variables, the OLS model is employed; credit constraint alleviation is a binary variable (0, 1), hence the Probit model is used; the meanings of other variables remain consistent with (1).

4 Results

4.1 Baseline regression results

The results in columns (1) to (3) of Table 3 indicate that digital capability is positively associated with household income at the 1% significance level, regardless of whether control variables and province dummy variables are included. The stability of the coefficient across specifications supports the robustness of this relationship. In economic terms, after retransformation from the log scale, column (3) shows that a one-standard-deviation increase in digital capability, equivalent to 0.297 units, is associated with a 17.48% increase in per capita gross income. The corresponding 95% confidence interval ranges from 13.16% to 21.80%. A shift in digital capability from the 25th percentile (0.471) to the 75th percentile (0.676) is likewise associated with an 11.74% increase in predicted per capita gross income. Several factors may account for this pattern. Stronger digital capability can help farmers integrate resources and reach markets more efficiently, widen their access to employment and entrepreneurial opportunities, and facilitate their participation in financial markets.

Table 3. Baseline regression results.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively. Brackets in columns (3) and (6) contain 95% confidence intervals.

An equally consistent pattern emerges in columns (4) to (6). Whether or not control variables and province dummy variables are included, digital capability is negatively associated with income inequality at the 1% significance level. Column (6) indicates that a one-standard-deviation increase in digital capability is associated with a 6.88% reduction in the income inequality index relative to its sample mean of 0.519, with a 95% confidence interval of 5.37% to 8.39%. Similarly, an increase in digital capability from the 25th to the 75th percentile is associated with a decline in the predicted income inequality index from 0.539 to 0.514. This change amounts to 0.025 points, with a 95% confidence interval of 0.019 to 0.030 points. Taken together, these estimates show that stronger digital capability is associated with lower income inequality among farm households. One possible explanation is that digital capability functions as a production factor subject to diminishing marginal returns. Digital empowerment may allow low-income farmers to overcome development constraints and achieve leapfrog development, generating larger marginal gains. High-income farmers, by contrast, already possess stronger foundations and may therefore experience more limited marginal improvements. Differences in these marginal returns may contribute to narrowing the overall income gap.

To distinguish the roles of the three digital capability dimensions, technological access, information acquisition, and application transformation are entered simultaneously into the regression model. This specification identifies the adjusted association of each dimension while controlling for the other two. Multicollinearity does not appear to be a concern: all variance inflation factors are well below the critical threshold of 5, and the pairwise correlation coefficients remain within acceptable ranges. As reported in Table 4, all three dimensions are positively associated with per capita gross income and negatively associated with income inequality at the 1% significance level, although the magnitudes differ. For household income, technological access capability shows the strongest positive association, consistent with its foundational role in allowing rural households to participate in the digital economy. The association for information acquisition capability is slightly smaller and may reflect reduced information constraints and wider access to off-farm employment opportunities. Application transformation capability displays the weakest association. One possible explanation is that the relatively advanced age and lower educational attainment of sampled farmers, together with limited digital application scenarios, restrict the extent to which such capability can be converted into income. A different ranking emerges for income inequality. Information acquisition capability shows the strongest negative association, suggesting that low-income households may benefit more from improved access to information. Technological access capability ranks second and provides the basic conditions for rural households to participate in the digital economy. As rural digital infrastructure matures and access gaps continue to narrow, however, the marginal contribution of hardware diffusion alone may decline. Application transformation capability has the weakest association with lower income inequality. Its potential to support digital catch-up may remain constrained because many low-income households have yet to cross the capability threshold required to convert digital applications into economic returns. Rural digital development must therefore extend beyond technological access alone. Strengthening information acquisition and application transformation capabilities may support a more equitable distribution of digital dividends and more inclusive income growth among rural households.

Table 4. Regression results for the dimensions of digital capability.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.2 Endogeneity test

4.2.1 Instrumental variable method

Although the baseline regressions include a comprehensive set of control variables and province dummy variables, potential endogeneity may remain because unobserved household characteristics cannot be fully controlled. Reverse causality between digital capability and household income may also bias the estimates. Drawing on relevant research (Ma et al., 2020), this study therefore adopts a village-level instrumental variable: the mean digital capability of all other sampled households in the same village, excluding the focal household. The choice of this instrument rests on both shared local conditions and peer interactions. Households within the same village generally face similar digital infrastructure and technological environments. Dense social ties and peer effects in rural communities also encourage communication, learning, and imitation among villagers. The digital capability of an individual household is therefore likely to be correlated with that of other households in the same village, supporting the relevance condition. The exclusion restriction assumes that, after accounting for the control variables and province dummy variables, the digital capabilities of other village households are not directly related to the focal household’s income outcomes except through its own digital capability.

The Kleibergen–Paap rk LM test rejects the null hypothesis of underidentification at the 5% level (p = 0.015), supporting the relevance of the instrument. The Kleibergen–Paap rk Wald F statistic is 10.134, exceeding the conventional threshold of 10 and suggesting that the instrument is not weak under this commonly used criterion. As a sensitivity check, the models were re-estimated with standard errors clustered at the township level. The resulting coefficients and significance levels remain consistent with the baseline IV estimates. Weak-instrument-robust Anderson–Rubin tests also reject the null hypothesis that the coefficient on digital capability is zero in both the per capita gross income and income inequality models (p = 0.008) and (p = 0.003), respectively. As shown in the second-stage results in Table 5, digital capability remains positively associated with per capita gross income and negatively associated with income inequality, with both coefficients statistically significant. The instrumental variable estimates retain the signs and significance of the baseline estimates, supporting the robustness of the main findings under the IV specification.

Table 5. Results of the instrumental variable method.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.2.2 Propensity score matching

To reduce potential selection bias arising from observable characteristics, this study applies propensity score matching (PSM). Three matching algorithms are used: 1:1 caliper matching with a caliper of 0.05 and without replacement, kernel matching with a bandwidth of 0.06 and with replacement, and radius matching with a caliper of 0.03 and with replacement. The sample mean of the digital capability score serves as the cutoff (MacCallum et al., 2002). Farm households with scores above the mean are assigned to the treatment group, while those with scores at or below the mean form the control group.

• (1) Common support and balance tests. Propensity scores span [0.058, 0.888] for the treatment group and [0.038, 0.874] for the control group, yielding a common support region of [0.058, 0.874]. This substantial overlap indicates that the matching requirement is well satisfied, as visually confirmed by Figure 3. As reported in Table 6, all three matching methods substantially reduce the pseudo R², while both the mean and median standardized biases fall well below the 5% threshold. After matching, the overall tests of joint covariate significance are not statistically significant (p > 0.1), indicating that the observable covariates no longer jointly predict treatment assignment. These results suggest that no statistically detectable systematic differences in observable characteristics remain between the treatment and control groups.

Table 6. Balance test results.

• (2) Treatment effect estimation. Table 7 reports the average treatment effects on the treated (ATT) obtained from the three matching approaches. Across all three matching methods, the ATT estimates are positive for per capita gross income and negative for income inequality, with both sets of estimates statistically significant at the 1% level. After accounting for selection based on observable characteristics, treated rural households therefore exhibit higher per capita gross income and lower income inequality than their matched counterparts. The stability of the signs and significance levels across the three matching approaches lends further support to the robustness of these findings.

Table 7. Average treatment effect on the treated (ATT) estimates.

*** p < 0.01, ** p < 0.05, * p < 0.1. Standard errors are obtained via bootstrap with 200 replications, clustered at the village level.

• (3) Post-matching regression analysis. The post-matching regression results in Table 8 are consistent with the baseline estimates. Across all three matched samples, digital capability remains significantly positively associated with per capita gross income and significantly negatively associated with income inequality at the 1% level. These findings indicate that the core results remain robust after mitigating potential endogeneity arising from selection based on observable characteristics.

Table 8. Results of the propensity score matching method.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.3 Robustness tests

Table 9 reports robustness tests for income inequality and per capita gross income. Income inequality. Panel A presents five robustness tests. (1) Alternative Construction of Digital Capability (PCA). Digital capability is reconstructed using PCA instead of the entropy weighting method, following a widely used approach to constructing household composite indices (Filmer and Pritchett, 2001). The KMO statistic of 0.817 and Bartlett’s test result (p < 0.001) support the suitability of PCA. (2) Fractional Response Model. Because the Kakwani index is bounded between 0 and 1, the baseline model is re-estimated using a fractional response model (Papke and Wooldridge, 1996). (3) Podder Index. The Podder index replaces the Kakwani index as the measure of income inequality (Podder, 1996). (4) Outlier Treatment. Per capita gross income is winsorized at the 1st and 99th percentiles, after which the Kakwani index is recalculated using the province as the reference group. (5) Alternative Reference Group (County). The Kakwani index is recalculated using households within the same county, as the reference group (Stark and Taylor,1991).

Table 9. Robustness tests.

All models include the full set of control variables and province dummy variables (N = 2,703). Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

Per capita gross income. Panel B presents three robustness tests. (1) Alternative Construction of Digital Capability (PCA). Digital capability is reconstructed using PCA. (2) Net of Transfers. Drawing on the literature on income source decomposition, per capita gross income is recalculated after excluding transfer income (Yitzhaki, 1979). (3) Outlier Treatment. Per capita gross income is winsorized at the 1st and 99th percentiles. Across all specifications, digital capability remains negatively associated with income inequality and positively associated with per capita gross income at the 1% significance level. These results confirm the robustness of the baseline findings.

4.4 Mechanism tests

To explore the potential mechanisms linking digital capability to per capita gross income and income inequality among rural households, this study examines social networks and credit constraint alleviation as potential mediating channels. The empirical results align with the theoretical expectations of H2 and H3.

Column (1) of Table 10 shows a positive association between digital capability and social networks at the 1% significance level, suggesting that farmers with stronger digital capability tend to maintain broader social ties. Social networks, in turn, are positively associated with per capita gross income in column (2) and negatively associated with income inequality in column (3). The corresponding indirect effects are statistically significant, providing evidence consistent with partial mediation. Stronger digital capability may lower the communication and transaction costs involved in establishing and maintaining social ties, allowing farmers to develop connections beyond their immediate communities. Broader networks can then alleviate information asymmetries and facilitate the diffusion of practical knowledge and skills. These relationships may link digital engagement to higher household income and narrower income disparities.

Table 10. Mechanism test.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively. The natural direct effect (NDE) and natural indirect effect (NIE) are estimated via causal mediation analysis based on 999 bootstrap replications. Column (4) reports coefficients from the Probit model.

A similar pattern emerges for credit constraint alleviation. Column (4) reports a positive association between digital capability and the easing of credit constraints, while columns (5) and (6) show that credit constraint alleviation is positively associated with per capita gross income and negatively associated with income inequality, respectively. The statistically significant indirect effects again provide evidence consistent with partial mediation. By improving access to financial information and digital financial services, stronger digital capability may increase the likelihood of obtaining formal credit. The resulting relief from credit constraints can provide liquidity for production and business activities and may be associated with expanded operations and greater efficiency. This pathway may be particularly relevant to lower-income households, which typically face tighter financing constraints and depend more heavily on external credit. Across all specifications, the direct effects remain statistically significant. Social networks and credit constraint alleviation therefore account for part, but not all, of the observed relationships.

4.5 Heterogeneity analysis

4.5.1 Heterogeneity in relative poverty status

Informed by relevant research on the classification of relatively poor and nonpoor households (Sun and Xia, 2019), this study defines the relative poverty line as 45% of the median per capita income among sampled households. Households below this threshold are classified as relatively poor, while the remaining households are classified as nonpoor.

The regression estimates indicate that digital capability is significantly associated with lower relative income deprivation and higher per capita income among nonpoor households. The 95% confidence intervals for both estimates exclude zero. In contrast, neither association is statistically significant among relatively poor households. Formal tests further indicate significant differences between the two groups. These findings suggest that the favorable income related associations of digital capability are concentrated among nonpoor households and are not evident among relatively poor households.

To identify possible sources of this heterogeneity, the analysis compares the human capital characteristics of the two groups. Relatively poor households contain a higher proportion of older members, and their household heads are older on average. Educational attainment is also lower: fewer household members have completed senior high school or above, and average years of schooling are shorter. Taken together, these patterns point to cumulative human capital disadvantages in digital adoption. Older individuals may have less capacity or willingness to acquire new technological skills. Limited education, meanwhile, can restrict their ability to understand and apply digital information. Digital skills may therefore be harder to convert into tangible returns from production and business activities among relatively poor households. Nonpoor households, by comparison, possess more favorable human capital endowments and better conditions for practical application, placing them in a stronger position to translate digital technology into income-generating opportunities.

The heterogeneity analysis also helps clarify the distributional pattern underlying the baseline results. Nonpoor households make up the majority of the sample, so the aggregate association between digital capability and lower relative income deprivation may largely reflect the relationship observed within this group. Among relatively poor households, by contrast, the associations are not statistically significant—pointing to potential barriers arising from limited human capital and digital skills. These findings suggest that digital inclusion policies should provide targeted skills training and accessible technical support for older and less-educated rural residents. Such measures may strengthen the ability of disadvantaged households to turn digital access and skills into tangible economic opportunities (Table 11).

Table 11. Results of relative poverty heterogeneity.

Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

4.5.2 Heterogeneity in income sources

To assess whether the relationship between digital capability and household income differs across income sources, total income is divided into five components: agricultural operating income, wage income, nonfarm business income, property income, and transfer income (Benjamin et al., 2005). Panel A of Table 12 presents the coefficient estimates together with their 95% confidence intervals. A clear pattern emerges. Digital capability is positively and significantly associated with wage income and nonfarm business income, and both confidence intervals exclude zero. The estimates for agricultural operating income, property income, and transfer income are not statistically significant, with their confidence intervals containing zero. The positive relationship between digital capability and household income is therefore most evident in wage employment and nonfarm business activities.

Table 12. Results of income sources heterogeneity.

In Panel A, income components are expressed in thousands of yuan and transformed using the inverse hyperbolic sine (IHS) function to accommodate zero or negative values. In Panel B, the corresponding Kakwani indices are calculated from the original, untransformed income values. Village-clustered standard errors are reported in parentheses. *, , and * indicate significance at the 10%, 5%, and 1% levels, respectively.

Panel B turns to relative deprivation within each income component. The reference group for each measure consists of households in the same province with valid observations for the corresponding income source. After controlling for relevant covariates and province dummy variables, digital capability is significantly associated with lower relative deprivation in wage income and nonfarm business income; again, both confidence intervals exclude zero. No statistically significant relationships are found for relative deprivation in agricultural operating income, property income, or transfer income, and the corresponding confidence intervals include zero. Taken together, the distributional association of digital capability is concentrated primarily in wage income and nonfarm business income.

5 Discussion

Farm household income has long been a central concern in rural development research. Less well understood is how digital capability relates to household income formation and distribution in the digital economy.The study develops a multidimensional measure of farmers' digital capability using household-level data. This approach provides a micro-level framework for examining rural household income during the digital transition. The findings are broadly consistent with previous studies linking digitalization to higher income and improved income distribution (Ma et al., 2020). The full-sample estimates indicate that greater digital capability is associated with higher Per capita gross income and lower relative income deprivation. These patterns may partly reflect improvements in rural digital infrastructure and declining barriers to digital access in China. Unlike studies focusing solely on basic digital access or aggregate digitalization, the measurement approach here captures three distinct dimensions: technological access, information acquisition, and application transformation. The household-level Kakwani index further measures relative income deprivation within provincial reference groups. Together, these measures provide more detailed micro-level evidence on the relationships among digital capability, household income, and income distribution.

The favorable pattern is not universal, however. Digital capability is significantly associated with higher income and lower relative deprivation only among nonpoor households. Neither association is statistically significant among relatively poor households. Because nonpoor households account for nearly 80% of the sample, the full-sample estimates may largely reflect the pattern observed within this group rather than an even distribution of digital benefits. Differences in human capital offer one possible explanation. Relatively poor households also have a larger share of older members, older household heads, and lower educational attainment. These disadvantages may constrain the acquisition of digital skills and their conversion into tangible economic benefits (Lythreatis et al., 2022). The resulting pattern is consistent with the outcome divide identified in the digital divide literature, whereby the benefits of digital technology vary with household resources and capabilities (Van Deursen and Van Dijk, 2015).

The analysis by income source provides further insight into this heterogeneity. Digital capability is positively associated with wage income and nonfarm business income and negatively associated with relative deprivation in these two income categories. No statistically significant associations are found for either the level or relative deprivation of agricultural operating income, property income, or transfer income. Thus, the favorable income associations of digital capability are most apparent in wage employment and nonfarm business activities. One possible explanation is that digital capability may improve access to employment and market information, reduce search costs, and expand market access. Other income sources may depend more heavily on landholdings, capital endowments, initial wealth, and institutional arrangements. Digital capability alone may therefore play a more limited role in alleviating these constraints. Taken together, these findings show that favorable aggregate associations can coexist with limited inclusiveness and provide new household-level evidence on how the relationship between digital capability and income distribution varies across socioeconomic groups and income sources.

This study also has several limitations. First, while farmers’ incomes are influenced by multiple complex factors, this paper focuses on the role of digital capability, with relatively limited consideration of other potential variables. Second, income fluctuations exhibit dynamic and long-term cumulative characteristics; however, constrained by data availability, this study was unable to employ panel data for longitudinal tracking analysis. Additionally, because the CRRS 2020 data were collected during the early phase of normalized pandemic prevention and control, confounding macroeconomic disruptions cannot be entirely eliminated. Future research could focus on collecting multi-period panel data on farmers to enhance the robustness of related findings.

6 Conclusion

Drawing on data from the China Rural Revitalization Survey (CRRS), this study develops a multidimensional measure of farm households’ digital capability and examines its associations with per capita gross income and relative income deprivation. Greater digital capability is significantly associated with higher income and lower relative deprivation. Mediation analysis points to two potential channels underlying these relationships: broader social networks and the alleviation of credit constraints. The pattern, however, is not uniform across households. Favorable associations are statistically significant among nonpoor households but not among relatively poor households. Results also vary across income sources. Digital capability is positively associated with wage income and nonfarm business income and negatively associated with relative deprivation in both components. By contrast, no statistically significant associations are found between digital capability and agricultural operating income, property income, or transfer income.

Several policy implications follow from these findings. First, develop a tiered framework for strengthening rural households’ digital capabilities. Policy design should account for differences in households’ initial resources and direct support to relatively poor households and other disadvantaged groups. Greater investment in rural basic education can lower barriers to acquiring digital skills. Accessible training tailored to these groups can further narrow the digital divide and broaden their participation in the digital economy. Second, integrate digital technologies more deeply into agricultural production and farm management. Priority should be given to smart devices and digital agriculture platforms suited to smallholder farmers. Practical training can help households expand their social networks through digital tools and improve access to rural digital financial services. These measures could create better conditions for digital capabilities to contribute to a more equitable distribution of income. Third, foster an inclusive environment for nonfarm employment and entrepreneurship. Stronger oversight of rural e-commerce platforms can help prevent excessive concentration of resources. Targeted digital support for low-income households and other disadvantaged groups can reduce the costs and barriers to participating in the digital economy. Such measures could help limit the risk of widening income gaps in nonfarm activities.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI was used for language editing, improving clarity and readability, and assisting in the refinement of the manuscript. The authors reviewed and revised all AI-assisted content and take full responsibility for the final version of the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mengying Feng, University of Roehampton London, United Kingdom

Reviewed by: Armand Fréjuis Akpa, Institut Supérieur Agronomique et Aétérinaire Valéry Giscard d'Estaing, Guinea

Surahman Surahman, Politeknik Negeri Samarinda, Indonesia