Introduction

In the context of increasing global economic uncertainty and accelerating digital transformation, poverty vulnerability has emerged as a significant impediment to sustainable human development. In 2020, China successfully eliminated absolute poverty based on existing standards, making a substantial contribution to global poverty reduction. However, significant economic inequality persists, and relative poverty has not been effectively addressed. Furthermore, many rural households that have escaped poverty remain at high risk of falling back into deprivation. In recent years, digital technologies have profoundly reshaped production methods and daily life in rural China. By the end of 2021, China’s digital village development index had reached 39.1%Footnote 1, and by 2023, the number of rural broadband users had grown to 192 millionFootnote 2. Digital embeddedness has transformed information flows, expanded access to financial services, reshaped market behavior, and created new opportunities to reduce vulnerability to poverty. In this context, a comprehensive investigation into how digital embeddedness affects rural households’ vulnerability to poverty—and the mechanisms through which it operates—is essential for consolidating China’s poverty alleviation achievements and reducing the risk of poverty recurrence.

The concept of poverty vulnerability originates from the notion of “vulnerability” in disaster science and was first introduced by the World Bank to measure the probability that a household will fall into poverty in the future due to exposure to risk shocks. Risk shocks refer to uncertain events—such as illness, natural disasters, and market fluctuations—that may threaten individual or household welfare. Risk responses, by contrast, are the actions taken to mitigate the decline in welfare following such events. These typically include ex-post coping mechanisms and strategies for recovery and development. Within this framework, risk shocks act as the triggers of vulnerability, while the resulting decline in welfare constitutes the outcome. Households with high vulnerability are often characterized by internal structural imbalances and limited capacity to manage or recover from these shocks (Hernández and Zuluaga, 2022). Compared with traditional poverty indices that statically assess household welfare at a specific point in time, the concept of poverty vulnerability provides a dynamic perspective by linking current living conditions with the risk of future poverty. It serves both as a theoretical foundation for understanding risk-induced poverty and as a practical tool for preventing poverty recurrence through improved risk management.

Recent research on poverty vulnerability and the poverty-reducing effects of digital embeddedness has produced valuable insights, yet several limitations remain. First, most existing studies focus narrowly on digital finance or equate digital development at the micro level with the mere adoption of internet technologies. In reality, digital technology and digital content are deeply embedded in individuals’ lifestyles and social interactions in multidimensional and holistic ways, continuously shaping their identities, cultural affiliations, social images, and behavioral patterns. As such, a more comprehensive and nuanced conceptualization of individual digital embeddedness is necessary to better understand its influence on rural households’ susceptibility to poverty. Second, digitalization at the micro level is often measured using a binary indicator of Internet usage, which inadequately reflects the richness and depth of digital engagement. This simplistic measure fails to account for the extent and intensity of online participation and overlooks the broader impact of digital embeddedness on rural residents’ production practices and daily life. Finally, in the context of poverty vulnerability, risk shocks serve as the initiating factors, inadequate coping capacity represents the core issue, and reduced welfare constitutes the ultimate consequence. While most current studies emphasize the impact of digital development on poverty vulnerability, they frequently overlook the pivotal role of risk itself. Insufficient attention has been paid to how digital embeddedness interacts with the dynamics of risk exposure and response mechanisms—an aspect that is critical for achieving a more holistic understanding and effective mitigation of poverty vulnerability.

This study aims to systematically reveal the specific directional impact and magnitude of digital embeddedness on rural households’ vulnerability to poverty, based on a scientific quantification of their digital embeddedness levels, and to clarify its critical role in the formation mechanisms of poverty vulnerability. By doing so, it seeks to provide more explanatory empirical evidence and theoretical frameworks for understanding the relationship between digital embeddedness and poverty vulnerability. To achieve this, the research utilizes micro-level statistical data from China to construct a multidimensional digital embeddedness index, thoroughly examining the pathways through which digital embeddedness enhances rural households’ capacity to avoid risk shocks and strengthen their risk-coping abilities. The findings indicate that digital embeddedness significantly reduces poverty vulnerability, with digital productive practice embeddedness playing a pivotal role. Regarding the impact mechanism, within the risk shock dimension, digital embeddedness facilitates income growth in agricultural production and promotes the transition of family members into non-agricultural employment. These effects contribute to economic risk diversification. In the risk response dimension, digital embeddedness enhances rural households’ ability to accumulate social capital and access financial capital, further improving their capacity to withstand external economic disruptions.

The potential contributions of this study are outlined as follows: 1. With regard to the research subjects, this study primarily focuses on the integration of digital production and digital life in rural households. Drawing on the concept of “embeddedness,” it clearly defines digital embeddedness at the micro-individual level and constructs a digital embeddedness index using factor analysis. This provides a new perspective for examining the influencing factors of poverty vulnerability. 2. In terms of the analytical framework for poverty vulnerability, this study focuses on the key risk factors contributing to poverty vulnerability and examines how digital embeddedness impacts poverty vulnerability in rural households through two key dimensions: risk shocks and responses. This approach provides an analytical framework for future research on the impact of digital development on poverty vulnerability. 3. The study conducts an in-depth analysis of the heterogeneous effects of digital embeddedness on household poverty vulnerability across groups with different educational backgrounds. Furthermore, digital embeddedness is decomposed into three dimensions: digital value perception embeddedness, digital lifestyle embeddedness, and digital productive practice embeddedness. The study explores the differentiated impacts of these dimensions on poverty vulnerability, deepening the understanding of the role of digital embeddedness and providing a theoretical reference for more targeted policy formulation.

The structure of this paper is organized as follows. Section “Literature review” presents a literature review that synthesizes current research on poverty vulnerability and the impact of digital embeddedness on poverty alleviation. Section “Theoretical analysis and research hypotheses” provides a theoretical framework, clearly defining the concept of digital embeddedness and assessing, from a theoretical perspective, how it directly affects the poverty vulnerability of rural households. This section also explores the underlying mechanisms through the lens of the risk shock–response framework. Section “Research design” describes the research design, including the data sources and the construction of the empirical model. Section “Empirical findings and analysis” reports the empirical results, covering the baseline regression, the examination of mediating mechanisms, and robustness checks. Section “Further analysis” examines the heterogeneous effects of digital embeddedness on poverty vulnerability, specifically investigating how these effects vary across households with different educational attainments and exploring the impacts from multiple dimensions of digital embeddedness. Finally, the section “Conclusions and implications” concludes the paper by summarizing the key findings and discussing the policy implications.

Literature review

Measurement and determinants of poverty vulnerability

Current academic research related to poverty vulnerability mainly centers on the measurement of poverty vulnerability and the factors affecting it. Regarding the measurement of poverty vulnerability, the academic community has developed three primary approaches based on different measurement criteria: vulnerability as exposure to risk, vulnerability as expected utility, and vulnerability as expected poverty (VEP) (Gallardo, 2018). Among these, the VEP method is widely adopted because of its ability to provide ex ante predictions with relatively low data requirements (Jha et al., 2018).

In examining the factors that influence household poverty vulnerability, limited resource endowments emerge as a key determinant. These include poor health status (Jacob et al., 2012), inadequate physical assets (Lewis and Lewis, 2014), limited financial capital (Gang et al., 2018), insufficient human capital (Banks et al., 2021; Ngepah et al., 2022), and weak social capital (Chantarat and Barrett, 2012). In addition, external risk shocks—such as extreme climate events (Eriksen and O’Brien, 2007; Kiptum, 2024), floods (Mahanta and Das, 2017), and other natural disasters—further exacerbate vulnerability. To mitigate this, policy efforts have focused on enhancing economic participation and strengthening households’ capacity to respond to shocks. Effective interventions include on-the-job training programs (Elise and Timothy, 2022), access to health insurance (Atake, 2018), promotion of non-agricultural employment opportunities in rural areas (Lohmann and Liefner, 2009), the development of social safety nets (Msuha and Kissoly, 2024), and investments in public infrastructure (Sumargo et al., 2024). Collectively, these measures contribute to reducing household poverty vulnerability.

However, despite significant progress in identifying key factors and proposing coping strategies, the existing literature predominantly adopts a static perspective centered on resource endowments, focusing on their direct associations with vulnerability. It tends to overlook the dynamic processes through which households respond to risk shocks, which limits a deeper understanding of the fundamental characteristics and underlying logic of poverty vulnerability.

Digital embeddedness and poverty vulnerability

Growing academic attention has been devoted to the impact of digital embeddedness on poverty vulnerability, with particular emphasis on the advancement of digital finance and the adoption of digital technologies among rural households. In the realm of digital finance, research indicates that the digital transformation of the financial sector has significantly reduced credit access barriers, eased credit constraints, and thereby lowered the risk of poverty for rural households. This improvement is largely attributed to the widespread use of online payment platforms, including mobile payments, which have reduced the cost of financial services while enhancing both accessibility and convenience (Daud et al., 2024). Furthermore, digital embeddedness has strengthened the connection between rural residents and financial markets, supporting the development and dissemination of inclusive digital financial products (Chen and Liu, 2024). As a result, rural households have experienced improvements in asset allocation strategies, risk management capabilities, and income-generating potential (Liu and Guo, 2023; Mushtaq and Bruneau, 2019). In addition, innovations in digital inclusive finance have diversified funding channels for rural entrepreneurship and stimulated the expansion of rural e-commerce, thereby contributing to increased household income (Wang et al., 2024). The mitigating effect of digital finance on household poverty vulnerability has been empirically validated in several developing countries and regions, including China (Xu and Zhang, 2025), India (Gautam et al., 2022), and Sub-Saharan Africa (Traore and Moussa, 2025; Onyejiaku et al., 2024).

With regard to the adoption of digital technologies, existing research has primarily focused on the Internet usage behavior of rural residents (Djibo and Malam, 2024; Nguyen et al., 2022). Prior studies have shown that the widespread availability of the Internet has significantly improved access to information for rural populations (Verma et al., 2023), yielding various benefits. On the one hand, Internet usage has been found to enhance financial literacy in rural areas (Li et al., 2022). On the other hand, it facilitates the expansion of social networks and the accumulation of social capital (Antoci et al., 2015), which in turn contribute to increased household labor productivity (Phan, 2023; Rajkhowa and Qaim, 2021) and ultimately reduce poverty vulnerability (Kuhn and Mansour, 2014). However, concerns have emerged regarding the risks associated with digital inequality. Some scholars argue that the widening digital divide may exacerbate disparities in resource access among rural households (Lissitsa and Chachashvili-Bolotin, 2022). Moreover, the shift toward a digital lifestyle has been linked to potential adverse effects on both the physical and mental health of rural residents, increasing their vulnerability to health-related shocks (Wang, 2025).

In summary, existing studies have made significant progress in examining the impact of digitalization on poverty vulnerability, particularly emphasizing its role in promoting income growth and improving financial accessibility. However, existing studies tend to overlook the dynamic effects of digitalization on household resource allocation and response strategies in the context of risk shocks. Moreover, the measurement of digitalization is often restricted to basic indicators, such as internet access, which fail to capture the heterogeneity and depth of rural residents’ actual digital engagement. To address these limitations, this study constructs a “digital embeddedness index” to quantify rural residents’ digital engagement and adopts a “risk shock–response” framework to systematically analyze its impact on household poverty vulnerability. This approach helps overcome shortcomings in both the measurement of digitalization and the understanding of the mechanisms through which digital embeddedness influences poverty vulnerability.

Theoretical analysis and research hypotheses

Analysis of how digital embeddedness directly influences the poverty vulnerability of rural households

An actor simultaneously engages in multiple relationships within a network, each of which collectively influences their behavior. With the continuous advancement and widespread adoption of digital information technologies, the rules and forms of social embeddedness have undergone profound transformations (Zhao et al., 2023). In the digital era, digitized information and resources have penetrated every aspect of personal and social life, including but not limited to work, social interaction, culture, consumption, and education. These digital components are deeply integrated into individuals’ daily lives, continuously shaping and influencing their modes of production and ways of living.

Building upon this, digital embeddedness is defined in this paper as the process of individuals deeply restructuring their social relationships, economic behaviors, lifestyles, and cognitive models in the process of production and life, relying on digital technology, data resources, and information networks. This process not only includes the reconstruction of the way individuals produce, process, transmit, and apply information based on digital technology, but also emphasizes the impact of digitalization on individuals’ identity, behavior, economic decision-making, and other content.

Poverty vulnerability refers to a household’s ability to maintain its economic status following a risk shock and reflects its resilience in the face of economic uncertainty. Digital embeddedness enhances individuals’ capacity to produce, store, process, and integrate data, facilitating the transformation of complex, unstructured data into actionable information. This process influences behavioral judgments and decision-making, thereby exerting a significant impact on rural households’ capital accumulation and total factor productivity.

First, in terms of information capital, digital embeddedness has transformed the ways in which information is transmitted and utilized, alleviating the information asymmetry that rural households often encounter in modern society. This transformation significantly improves their access to information capital. Reduced information access costs and enhanced transmission efficiency enable rural households to access educational resources and participate in online training programs, thereby enhancing the human capital of household members. In addition, digital technologies promote remote communication and online interaction, contributing to the development of rural households’ social capital (Stern and Adams, 2010). Second, regarding total factor productivity, digital technologies enhance the flexibility of economic production and facilitate collaboration across geographical boundaries and among different actors. The integration of digital technologies into agricultural practices has increased capital intensity, reducing dependence on manual labor. Digital embeddedness allows rural households to better adapt and optimize their production processes, leading to more efficient resource allocation and higher productivity (Varzaru, 2025).

Building on this, the study proposes Research Hypothesis 1: Digital embeddedness has a significant effect on reducing the vulnerability of rural households to poverty.

Analysis of the mechanisms by which digital embeddedness affects the vulnerability of rural households to poverty

When the intensity of a risk shock exceeds a household’s capacity to absorb it, household income declines. This process can be conceptualized within the “shock–resistance–outcome” framework. Drawing on the poverty vulnerability framework proposed by Yao and Xie (2022), this study explores how digital embeddedness affects rural households’ vulnerability to poverty through two interrelated dimensions: risk shocks and risk responses.

First, digital embeddedness helps reduce and mitigate the risk shocks that rural households may face in their production and livelihood activities. On the one hand, digitalization plays a crucial role in alleviating information asymmetries in agricultural production. Through digital platforms, farmers can access timely information on weather conditions, pest outbreaks, agricultural technologies, government policies, and market prices. This enables them to optimize production decisions and reduce transaction costs in agricultural product marketing, thereby diminishing the adverse effects of natural and market risks and ultimately improving their operational profitability (Peng et al., 2021). On the other hand, digital platforms and tools expand access to off-farm employment opportunities, allowing rural households to reinvest in human capital through the acquisition of new knowledge and professional skills (Wang and Zhang, 2022). Increased off-farm work opportunities and enhanced employability enable rural households to transition into more productive sectors such as modern industry and services, thereby achieving ex ante risk diversification (Duong et al., 2020).

As illustrated in Fig. 1a, two households, M1 and M2, start with the same initial level of welfare. However, household M2, characterized by a higher level of digital embeddedness, gains better access to information, allowing for more informed decision-making. Consequently, a larger proportion of M2’s household production is focused on off-farm activities. When a risk event occurs at time t0, household M1 suffers a significant shock and falls into poverty. In contrast, household M2, benefiting from ex ante risk mitigation and reduced exposure to risk through a diversified production structure, only experiences a mild shock and successfully avoids poverty.

Fig. 1: Schematic representation of the mechanisms by which digital embeddedness affects poverty vulnerability.
figure 1

a The impact of digital embeddedness on poverty vulnerability in the risk shock phase. b The impact of digital embeddedness on poverty vulnerability in the risk response phase.

Full size image

Thus, this paper proposes Research Hypothesis 2: In the risk shock dimension, digital embeddedness reduces the vulnerability of rural households to poverty through two primary mechanisms: facilitating the transfer of off-farm labor and increasing income from agricultural production.

Digital embeddedness strengthens rural households’ risk response capabilities. First, digital technologies have eliminated the constraints of time and space in social interactions, thereby narrowing the “information gap” among individuals. Through the Internet and various digital platforms, rural residents can not only reinforce traditional social networks based on kinship and locality but also establish new social ties driven by economic activities (Noble et al., 2024). The expansion of social capital significantly improves rural households’ access to information and knowledge, enhancing their ability to manage risks and increase income. On the one hand, farmers can leverage broader social networks to gain advantages in labor markets and business ventures. On the other hand, social capital also serves as a source of informal financial support, such as mutual guarantees and informal insurance (Mesay, 2015; Wuepper et al., 2018).

Moreover, digital embeddedness has substantially improved financial literacy and broadened access to financial products among rural households. These advancements have enhanced both the availability and efficiency of financial services in rural areas. Improved financial access enables rural households to better secure resources, thereby strengthening their capacity to absorb and recover from external shocks (Weng et al., 2023). For instance, in Fig. 1b, both households N1 and N2 start with the same level of welfare and face a risk shock at t0. Despite their efforts to cope (e.g., using savings, selling assets) during the period t0–t1, both households descend into poverty. However, due to its higher level of digital embeddedness, greater social capital, and increased financial resources, household N2 exhibits greater resilience. Following recovery and development from t1 to t2, N2’s welfare level rises above the poverty line, allowing it to escape poverty.

Thus, this paper proposes Research Hypothesis 3: In the risk response dimension, digital embeddedness decreases rural households’ vulnerability to poverty by strengthening social capital and alleviating credit limitations.

Research design

Data sources

This study utilizes data from the China Family Panel Studies (CFPS), conducted by the Institute of Social Science Survey at Peking University. The selection of this dataset is based on the following considerations. First, the CFPS adopts an implicitly stratified, multi-stage, multi-level, and probability-proportional-to-size sampling method, covering 25 provinces, municipalities, and autonomous regions—excluding Hong Kong, Macau, Taiwan, Xinjiang, Tibet, Qinghai, Inner Mongolia, Ningxia, and Hainan—which together represent approximately 95% of China’s population. The survey is therefore highly representative and reliable. Second, the CFPS provides detailed information on individuals’ internet usage across various domains of daily life, including work, social interaction, learning, business, and entertainment. This enables the construction of a robust measure of individual-level digital embeddedness. In addition, the dataset offers comprehensive data on household economic conditions—including income, employment, assets, and liabilities—as well as key demographic characteristics such as educational attainment, marital status, and age, all of which are closely aligned with the analytical objectives of this study.

This study uses data from the 2020 and 2022 waves of the CFPS. These two waves were selected primarily because the CFPS questionnaire is revised in each survey round, yet the internet-related modules in 2020 and 2022 are relatively comprehensive and exhibit strong consistency in indicator design and question wording. This continuity facilitates the stable measurement and comparison of individuals’ digital embeddedness across time. During data processing, the household economic and individual questionnaires from both survey years were first cleaned, and the sample was filtered. Specifically, respondents who completed the financial section of the household economic questionnaire were identified as household heads. Only those with rural household registration (hukou) were retained. Their individual-level demographic information was matched with corresponding household-level economic variables. Observations with missing values for key variables or those not meeting the study criteria were excluded. As a result, separate rural household sample databases were constructed for each wave. Next, data from the two periods were merged to form an unbalanced two-period panel dataset. In the final dataset, each household is represented by exactly one rural household head. The dataset contains 11,341 valid observations in total, including 5863 rural households from 2020 and 5478 from 2022, with 2731 rural households included in both survey waves.

Selection of variables

Explained variables

Among the three primary approaches to measuring poverty vulnerability, the VEP method is distinguished by its relatively low data requirements and broad applicability. It estimates the likelihood that a household will fall into poverty in the future due to exposure to various risks. Unlike conventional poverty indicators that reflect only current poverty status, the VEP method, as a typical ex ante predictive approach, focuses on assessing the risk of future poverty rather than explaining existing conditions. This forward-looking perspective aligns closely with the theoretical essence of “vulnerability,” offering a predictive assessment with enhanced early-warning capacity and greater policy relevance. Therefore, this study adopts the VEP measure to evaluate the likelihood that a household’s future income will fall below the relative poverty line following a risk shock, thereby capturing its susceptibility to poverty. The formula for this measure is:

$${{Vul}}_{h,t}=Pr \left({Y}_{h,t+1}\le Z\right)={\int }_{-{\infty }}^{z}{f}_{t}\left({Y}_{h,t+1}\right)d\left({Y}_{h,t+1}\right)$$
(1)

In Eq. (1), the poverty vulnerability of household h at time t, denoted as \({{Vul}}_{h,t}\), is defined as the probability that the per capita income of household h in t + 1 (\({Y}_{h,t+1}\)) will decline below the relative poverty threshold (Z). The function \({f}_{t}(\cdot )\) represents the probability density function describing the distribution of household per capita income.

Currently, there is no consensus in academia on how to define relative poverty. Researchers commonly define the relative poverty threshold using different percentiles of per capita household income or the median income. Following the method of Li and Liao (2023), we define the relative poverty line based on 0.4, 0.5, and 0.6 times the per capita disposable income of rural households in that year. We set the vulnerability threshold at 50%, where predicted poverty vulnerability equals 0 if it is below 50%, and 1 if it is above 50%. Accordingly, we construct three poverty vulnerability indicators—Vul1, Vul2, and Vul3—based on these criteria.

In the measurement of poverty vulnerability, the temporal dimension plays a central role. The VEP approach employed in this study focuses on whether a household’s expected income in period t + 1 falls below the poverty line, rather than merely assessing its current poverty status. Here, t represents the point in time when the data are collected, while t + 1 denotes a future period. To calculate these estimates, we follow Chaudhuri et al.’s (2002) approach for deriving unbiased estimates of future income and variance. The estimation of household poverty vulnerability is conducted using the feasible generalized least squares (FGLS) technique, following a systematic procedure. Initially, an ordinary least squares (OLS) regression is employed to estimate the logarithm of household per capita income. Subsequently, income volatility is assessed by applying OLS to the squared residuals. The corresponding estimation equations are presented as follows:

$${{LnY}}_{h}={X}_{h}\beta +{e}_{h}$$
(2)
$${\hat{e}}_{h}^{2}={X}_{h}\rho +{\varepsilon }_{h}$$
(3)

Where \({{LnY}}_{h}\) is the logarithm of per capita household income, and \({X}_{h}\) represents a series of control variables affecting income at the individual, household, and regional levels. Second, using the fitted values from these regressions, we construct the FGLS estimation, which allows us to calculate the expectation and variance of future log income as follows:

$$\hat{E}\left({Ln}{Y}_{h}|{X}_{h}\right)={X}_{h}{\hat{\beta }}_{{FGLS}}$$
(4)
$$\hat{V}\left({Ln}{Y}_{h}|{X}_{h}\right)={\hat{e}}_{h}^{2}\,={X}_{h}{\hat{\rho }}_{{FGLS}}$$
(5)

Where \(\hat{E}\left({Ln}{Y}_{h}|{X}_{h}\right)\) represents the expected value of household h’s future log income, and \(\hat{V}\left({Ln}{Y}_{h}|{X}_{h}\right)\) represents the variance of that income. Assuming household income follows a lognormal distribution, we substitute the expected value and variance from Eqs. (4) and (5) into Eq. (6) to derive the household’s poverty vulnerability:

$${{Vul}}_{h,t}=Pr \left({Y}_{h,t+1}\le Z\right)=\varphi \left(\frac{{LnZ}-{X}_{h}{\hat{\beta }}_{{FGLS}}}{\sqrt{{X}_{h}{\hat{\rho }}_{{FGLS}}}}\right)$$
(6)

After the estimation was completed, the predicted household poverty vulnerability data from the 2020 dataset was compared with the actual poverty status of rural households in 2022 to evaluate the accuracy of the model’s predictions. The results show that, under relative poverty vulnerability thresholds set at 0.4, 0.5, and 0.6 times the rural per capita income, the overall prediction accuracies were 88.602%, 82.778%, and 75.875%, respectively—all surpassing the 70% benchmark. These findings suggest that the model employed in this study is generally effective in accurately predicting the poverty vulnerability of rural households.

Core explanatory variables

The core explanatory variable in this study is the digital embeddedness index. Digital embeddedness refers to the transformation and diversification of rural households’ production and daily life driven by digital technologies. The internet plays a central role in this process. On one hand, the internet, with its inherent connectivity, provides the necessary infrastructure for the development and deployment of digital technologies. On the other hand, digital technologies depend on the internet as both a platform for iterative innovation and an interface for responding to market dynamics. Digital technology facilitates the rapid and efficient transmission of diverse forms of data and information via the internet, enhancing the accessibility and reach of information exchange and contributing to the formation of a highly connected communication network for users.

The CFPS survey includes multiple items on respondents’ internet access, usage scenarios, and frequency. However, some variation exists in the specific questions across different years. To ensure consistency, this study utilizes 19 variables common to both the 2020 and 2022 waves and employs exploratory factor analysis (EFA) to construct a digital embeddedness index for rural households and uncover its underlying structure.

Prior to analysis, the suitability of the data for factor analysis was assessed. The Kaiser–Meyer–Olkin measure yielded a value of 0.896, indicating strong correlations among variables and confirming the appropriateness of EFA. Furthermore, Bartlett’s test of sphericity produced a p-value of 0.000, providing additional support for the use of factor analysis. During the factor extraction stage, principal component analysis was employed to estimate the initial factor loading matrix. This method is a widely used factor extraction technique that applies linear transformations to the original variables, generating new variables that capture as much of the original variance as possible. Factors were selected based on the “eigenvalue greater than 1” criterion and further validated using a scree plot, resulting in the retention of three common factors. To enhance the interpretability of the factor structure, orthogonal rotation was applied, and factor scores were subsequently calculated. These scores were then aggregated using a weighted average method, with weights determined by the proportion of variance explained by each factor, to generate a comprehensive score representing the digital embeddedness index.

Since negative values appear in the final digital embeddedness index, the Min-Max standardization method is applied to rescale the composite factor scores to a range between 0 and 1 for easier comparison and interpretation. The common factors are named according to the magnitude of their factor loadings and the meanings of the associated variables. Through EFA, three common factors with clear substantive interpretation were extracted. These factors together account for 80.18% of the total variance. This indicates that the constructed factor model demonstrates strong explanatory power and robustness.

Table 1 lists the observed variables representing different dimensions of digital embeddedness. Based on factor loadings and variable interpretations, three latent dimensions are identified—cognitive, lifestyle, and productive layers—corresponding to digital value cognition embeddedness, digital lifestyle embeddedness, and digital productive practice embeddedness, respectively. Specifically, digital value cognition embeddedness is primarily characterized by social tool usage (e.g., WeChat) and perceived value of internet technologies in various functional domains, including work, daily life, social interactions, and learning. This dimension reflects individuals’ dependence on social networking tools and their recognition of the broader value of digital life, serving as the motivational foundation for digital engagement. Digital lifestyle embeddedness captures high-frequency entertainment and consumption behaviors, including gaming, online shopping, and short-form video consumption. It is also positively associated with fragmented learning activities, indicating the extent to which digital lifestyles permeate individuals’ daily routines. Digital productive practice embeddedness is strongly correlated with PC-based usage patterns and structured learning behaviors, underscoring the role of digital technologies as instrumental tools for productivity. This dimension is directly linked to skill acquisition and professionalization, highlighting the integration of digital tools into productive and specialized economic activities.

Table 1 Extracted common factors and their representative variables with high factor loadings.
Full size table

The normalized digital embeddedness index, which ranges from 0 to 1, represents the relative position of a household within the digital embeddedness hierarchy. Higher values indicate a greater degree of digital technology usage and Internet access, while lower values reflect a lower level of digital embeddedness. A change in the value of a household’s digital embeddedness index from 0 to 1 signifies a complete shift in the household’s level of digital embeddedness from the lowest to the highest position.

Mediating variables

The theoretical analysis above suggests that digital embeddedness reduces the poverty vulnerability of rural households by facilitating off-farm employment. Given that the household head is typically the primary labor force in rural households, this study uses whether the household head is engaged in off-farm work as a proxy variable for labor transfer. This variable is derived from the personal survey questionnaire: “What is the nature of your current or most recent job?” Respondents are classified as employed in agriculture if their primary activity involves agriculture, forestry, animal husbandry, sideline production, or fisheries. Otherwise, they are categorized as participating in off-farm employment. A value of 1 is assigned to off-farm employment, and 0 is assigned to agricultural employment or non-employment. In addition, digital embeddedness can help rural households mitigate risks associated with agricultural production and reduce economic losses. This effect is captured through agricultural income. Accordingly, the mediator variable for agricultural income is constructed using responses to the family economic questionnaire question: “What is the total value of your family’s agricultural by-products?”

The essence of social capital lies in social networks, reciprocal norms, and trusting relationships. In rural societies, interpersonal relationships are a defining feature of social culture. Kinship and neighborhood networks are sustained through the exchange of gifts, forming social structures rooted in blood relations, familial ties, and geographic proximity (Su and Duan, 2025). The frequency and value of gift-giving often reflect the scope and depth of an individual’s social network, serving as a key proxy for social capital (Zhao and Yao, 2017). Existing studies have shown that household-level social expenditures can reduce transaction costs in production and market exchanges, improve access to informal credit, and mitigate income volatility (Zhang et al., 2020). Compared with subjective indicators such as trust and community participation, interpersonal expenditures offer the advantage of being expressed in monetary terms, which enhances the feasibility and accuracy of data collection and measurement. Accordingly, this study uses household interpersonal expenditures as a proxy variable for social capital. In addition, bank loans and private borrowing are the primary channels through which rural households obtain financial resources. The alleviation of credit constraints through digital embeddedness is typically manifested in these two forms. Therefore, credit constraints are measured based on whether the household has access to either bank loans or private borrowing, with a value of 1 indicating the existence of at least one of these channels, and 0 otherwise.

Control variables

Drawing on relevant literature, this study incorporates control variables at the individual, household, and regional levels that may affect the poverty vulnerability of rural households. At the individual level, the household head typically serves as the primary financial decision-maker and plays a pivotal role in shaping the household’s economic status. Key characteristics of the household head—such as gender, age, educational attainment, and health status—can influence social participation, cognitive capacity, and labor productivity, all of which are closely associated with household income and vulnerability to poverty. Furthermore, marital status can impact access to social support networks. Therefore, this study includes the following individual-level control variables: gender, age, age squared, education level, marital status, and health condition of the household head (Zhang et al., 2024).

At the household level, demographic and asset-related factors constitute critical resources for income generation and risk mitigation. With respect to demographic characteristics, this study considers both household size and structure. Specifically, the number of household members and the intra-household dependency ratio—defined as the ratio of non-working-age individuals to working-age individuals—are used to reflect labor supply capacity and intergenerational support burdens. In terms of household assets, four categories are included as control variables: land assets, cash holdings, financial assets, and fixed productive assets.

At the regional level, disparities in infrastructure and industrial development across provinces significantly affect rural households’ vulnerability to poverty. To capture these regional differences, provincial per capita GDP is employed as a proxy for regional economic development. Additionally, rural households are classified into four major regions—eastern, central, western, and northeastern—to control for regional heterogeneity. Table 2 presents the descriptive statistics for all variables.

Table 2 Descriptive statistics of variables.
Full size table

Model design

Benchmark regression model design

Based on the theoretical hypotheses outlined in the previous section, the main objective is to assess whether digital embeddedness helps alleviate rural households’ vulnerability to poverty. In this analysis, rural household poverty vulnerability serves as the dependent variable, expressed as a binary outcome (0 or 1). The Logit model, as a variant of multiple regression analysis, is employed to investigate the association between diverse influencing factors and a dependent variable, which is either binary or multinomial in nature. This approach is appropriate for analyzing rural household vulnerability to poverty. Referring to the research of Xu and Zhang (2025), the empirical model is constructed as follows:

$${{Vul}}_{{it}}={\beta }_{0}+{\beta }_{1}{{DE}}_{{it}}+{\beta }_{2}{X}_{{it}}+{{\rm{\mu }}}_{i}+{\tau }_{t}+{\varepsilon }_{{it}}$$
(7)

where \({{Vul}}_{{it}}\) represents the dependent variable, measuring the poverty vulnerability of household i in period t. The key explanatory variable, \({{DE}}_{{it}}\), captures the level of digital embeddedness of the household head in the same period. \({X}_{{it}}\) denotes a set of control variables, including characteristics of the household head, household attributes, and regional factors. The parameters \({\beta }_{0}\), \({\beta }_{1}\), and \({\beta }_{2}\) are to be estimated. Additionally, \({{\rm{\mu }}}_{i}\) accounts for region-fixed effects, \({\tau }_{t}\) captures time fixed effects, and \({\varepsilon }_{{it}}\) represents the error term.

Mediated effects modeling

To further investigate the mechanism by which digital embeddedness mitigates rural households’ vulnerability to poverty, the mediation effect model proposed by Wen and Ye (2014) is utilized in this study. This model tests the mediation effects by sequentially analyzing the regression coefficients. The mediation model is specified as follows:

$${M}_{{it}}={\delta }_{0}+{\delta }_{1}{{DE}}_{{it}}+\varphi {X}_{{it}}+{{\rm{\mu }}}_{i}+{\tau }_{t}+{\varepsilon }_{{it}}$$
(8)
$${{Vul}}_{{it}}={\alpha }_{0}+{\alpha }_{1}{{DE}}_{{it}}+{\alpha }_{2}{M}_{{it}}+\lambda {X}_{{it}}+{{\rm{\mu }}}_{i}+{\tau }_{t}+{\varepsilon }_{{it}}$$
(9)

Among these, \({M}_{{it}}\) represents the mediating variable. The first step in testing the mediating effect involves conducting the baseline regression as specified in Eq. (7), where the estimated coefficient \({\beta }_{1}\) represents the total effect. In Eq. (8), \({\delta }_{1}\) captures the effect of digital embeddedness on the mediating variable, while in Eq. (9), \({\alpha }_{1}\) and \({\alpha }_{2}\) represent the direct effects of digital embeddedness and the mediating variable on household poverty vulnerability, respectively. The mediating effect is calculated as the product of \({\delta }_{1}\) and \({\alpha }_{2}\). To mitigate the impact of correlations between household-level factors and the regression results, robust standard errors are clustered at the household level.

Empirical findings and analysis

Analysis of baseline regression results

Stata 18.0 software is employed in this study for empirical estimation. The regression results examining the impact of digital embeddedness on the poverty vulnerability of rural households are presented in Table 3. After controlling for relevant variables, the marginal effects of digital embeddedness on vul1, vul2, and vul3 are −0.04, −0.06, and −0.08, respectively. All coefficients are statistically significant at the 1% level. These findings indicate that digital embeddedness significantly reduces rural households’ vulnerability to poverty. Moreover, as the relative poverty line standard increases, the mitigating effect of digital embeddedness becomes increasingly pronounced. Therefore, research hypothesis 1 is empirically supported.

Table 3 Regression results on the impact of digital embeddedness on the poverty vulnerability of rural households.
Full size table

Among the personal characteristic variables, the gender and education level of the household head have a significantly negative effect on the poverty vulnerability of rural households. Notably, this negative effect becomes more pronounced as the relative poverty line becomes more relaxed. These findings suggest that rural households headed by women or individuals with lower levels of education are at a relative disadvantage in achieving sustainable family development. The age of the household head shows a significant positive effect only on Vul3, indicating that older age increases household vulnerability under more lenient poverty thresholds. Health status exerts a significant negative effect on Vul1 but shows no significant impact on Vul2 and Vul3. This implies that health is a more critical factor under stricter poverty standards, while its influence diminishes when poverty lines are more relaxed—possibly due to the availability of more resources or support mechanisms. Marital status consistently and significantly reduces household poverty vulnerability across different poverty standards, with its protective effect intensifying as the thresholds are relaxed. This suggests that marriage plays a stable and strengthening role in safeguarding households against poverty. This effect may be attributed to the multifaceted advantages of marriage, including more efficient household resource allocation, better economic coordination, labor division, and stronger social network support, all of which contribute to greater household resilience in the face of shocks.

Among the family characteristic variables, both the number of household members and the dependency ratio have a significant positive effect on poverty vulnerability. This effect becomes more pronounced as the poverty line standard increases. Larger household size and a higher dependency ratio typically indicate greater financial burdens and heightened exposure to economic risks. These pressures are particularly magnified under more relaxed poverty standards, leading to an increase in vulnerability. In contrast, increases in household financial assets and productive fixed assets significantly reduce poverty vulnerability, with their mitigating effect growing stronger as the poverty threshold rises. This suggests that asset accumulation plays an increasingly important role in shielding households from poverty when standards become more lenient. Interestingly, a significant positive effect on the vulnerability of rural households to poverty is associated with an increase in land assets within the household. This may be because households with more land assets are more likely to engage in traditional agricultural production, and the inherent disadvantages of such agricultural practices may hinder efforts to reduce poverty vulnerability. Finally, the amount of cash assets is found to have a significant positive effect only on Vul1. This indicates that, among the extremely poor, cash holdings may primarily serve as emergency reserves rather than stable sources of income, thus correlating positively with poverty vulnerability under stricter poverty standards.

Among the regional characteristic variables, an increase in the per capita GDP of the province where a rural household is located significantly reduces its poverty vulnerability, with the effect becoming more pronounced as the relative poverty threshold becomes more relaxed. Economically developed regions provide rural households with greater access to employment opportunities, entrepreneurial ecosystems, and resource support, thereby enhancing household income levels. Moreover, higher per capita GDP is often linked to more comprehensive social security systems, which help reduce the cost of essential services such as healthcare, education, and pensions. These systems strengthen a household’s capacity to cope with economic shocks. In particular, they help mitigate the risk of poverty resulting from serious illness or educational expenses, thereby contributing effectively to the reduction of poverty vulnerability.

Analysis of the results of the mediation effect test

The theoretical analysis suggests that digital embeddedness reduces poverty vulnerability by increasing agricultural income, facilitating off-farm labor transfer, alleviating credit constraints, and enhancing social capital in the context of both risk shocks and risk responses. However, does empirical evidence support these theoretical conclusions? This paper uses the widely applied poverty vulnerability indicator, defined as 0.5 times per capita household income, to verify the mechanisms through which digital embeddedness operates.

The results of the test on the impact mechanism of the risk shock dimension are presented in Table 4. It is found that digital embeddedness has a significant positive effect on both the off-farm employment status and the agricultural income level of rural households. Additionally, digital embeddedness exerts a significant negative effect on the poverty vulnerability of households. Furthermore, increases in off-farm employment and agricultural income significantly reduce household poverty vulnerability. These findings provide evidence of a mediating effect, whereby digital embeddedness mitigates poverty vulnerability through promoting off-farm labor transfer and increasing agricultural income.

Table 4 Results of testing the impact mechanism of risk shock dimension.
Full size table

The robustness of the mediating effect is assessed using the Bootstrap method. Bootstrap is a non-parametric resampling technique that effectively mitigates the limitations of normal distribution assumptions by repeatedly calculating the standard error of the mediating effect through random sampling. The mediating effect is considered significant when the confidence interval of the test result excludes 0. In this study, the confidence intervals were calculated using the bias-corrected percentile method with 500 bootstrap replications. The results indicate that the confidence intervals for the mediating effects of off-farm employment and increased agricultural income exclude zero, thereby confirming the presence of these mediating effects. Consequently, it can be concluded that under the risk shock dimension, digital embeddedness reduces rural household poverty vulnerability by facilitating off-farm labor transfer and increasing agricultural income. This finding supports Hypothesis 2.

The results of the test on the impact mechanism of the risk response dimension are presented in Table 5. It is found that digital embeddedness significantly enhances rural households’ social capital and alleviates household credit constraints. Additionally, digital embeddedness exerts a significant negative effect on household poverty vulnerability. Furthermore, improvements in social capital and the alleviation of credit constraints significantly reduce household poverty vulnerability. These findings provide evidence of a mediating effect, whereby digital embeddedness reduces poverty vulnerability through enhancing social capital and easing credit constraints.

Table 5 Results of testing the impact mechanism of the risk response dimension.
Full size table

To ensure the robustness of the mediating effect, the Bootstrap method was utilized. Confidence intervals were calculated using the bias-corrected percentile method, based on 500 resampled iterations. The confidence intervals for the mediating effects of social capital improvement and credit constraint alleviation did not include zero, confirming the significance of these effects. Therefore, it can be concluded that within the risk response dimension, social capital improvement and credit constraint alleviation serve as key mediating channels through which digital embeddedness mitigates poverty vulnerability. Research hypothesis 3 is thus supported.

Robustness tests

Endogenous treatment

While panel data can help mitigate certain endogeneity issues, it still encounters challenges such as omitted variable bias and bidirectional causality, which hinder its ability to fully resolve endogeneity concerns. Therefore, this study further adopts the instrumental variable (IV) approach to address these remaining endogeneity problems. Currently, no widely accepted method exists in the academic literature for handling instrumental variables in nonlinear panel data models. To address endogeneity, an extended regression model (ERM), based on a multivariate normal distribution and maximum likelihood estimation, is employed. The ERM approach offers the advantage of accommodating both continuous and binary endogenous variables by specifying appropriate model options. Specifically, this study adopts the extended probit model (Eprobit) within the ERM framework. Following the approaches of Yang (2021) and Xin et al. (2024), this study employs the proportion of village-level Internet access and household expenditures on postal and telecommunication services as instrumental variables for digital embeddedness. Instrumental variable regression models are then estimated under the extended ERM framework, respectively.

The proportion of village-level Internet access is defined as the proportion of surveyed households within a village that have Internet access, excluding the respondent’s own household, relative to the total number of surveyed households in that village. Internet access is defined as the use of a computer or mobile device to connect to the Internet.

The proportion of village-level Internet access is chosen as an instrumental variable for two main reasons. First, it is strongly correlated with individual-level digital embeddedness. In rural communities, digital infrastructure is typically shared, and households within the same village generally depend on a common network environment. Given the characteristics of acquaintance-based societies, individual Internet usage behavior is likely influenced by surrounding peers. Thus, the proportion of Internet users at the village level significantly influences the extent of individual digital embeddedness, thereby satisfying the relevance condition for instrument validity (Agarwal et al., 2009; Nakada et al., 2024).

Second, the proportion of village-level Internet access primarily reflects the broader digital environment and Internet accessibility of the village, rather than individual household characteristics. Changes in this variable are largely driven by exogenous factors such as government policies, market dynamics, and infrastructure development, which makes it unlikely to directly affect household poverty vulnerability. Therefore, the exogeneity condition is likely to be satisfied (Zhang et al., 2024). Furthermore, compared to individual Internet usage, the proportion of village-level Internet access is less susceptible to confounding factors such as household income and education level. This reduces the risk of omitted variable bias and further reinforces its validity as an instrument for addressing the endogeneity of digital embeddedness. In conclusion, the proportion of village-level Internet access serves as an effective instrument for addressing the endogeneity of digital embeddedness.

Table 6 presents the regression results on poverty vulnerability under different criteria using the proportion of village-level Internet access as an instrumental variable. In the first-stage regression of IV-Eprobit, a significant positive relationship is observed between the proportion of village-level Internet access and the digital embeddedness index of the sampled households. This finding demonstrates a strong correlation between the instrumental variable and digital embeddedness. In the second-stage regression of IV-Eprobit, the correlation coefficient of the residual term is found to be statistically significant at the 5% level, confirming the reliability of the instrumental variable approach. The marginal effect of the instrumental variable on rural poverty vulnerability remains significantly negative, indicating that, even after addressing endogeneity, digital embeddedness continues to have a significant mitigating effect on the poverty vulnerability of rural households. These findings reinforce the robustness and credibility of the baseline regression results.

Table 6 Regression results of the ERM using the proportion of village-level internet access as a tool variable.
Full size table

The selection of monthly household expenditure on postal and telecommunication services as an instrumental variable for digital embeddedness is based on the following reasons: First, such expenditure represents a primary cost associated with mobile network usage among rural households. Higher spending indicates a deeper level of integration into the mobile internet, thus satisfying the relevance condition required of instrumental variables. Second, this expenditure category includes costs for mail delivery and telephone services. For rural residents, mail-related expenses account for only a small portion, while telephone costs constitute the majority. Given that telephone service payments in China often follow prepaid models—such as annual contracts or monthly data packages—monthly postal and telecommunication expenses are largely predetermined and therefore relatively exogenous (Zhang et al., 2025). In the empirical analysis, this paper uses the logarithm of “ monthly household expenditure on postal and telecommunication services ” as the instrumental variable for digital embeddedness.

Table 7 presents the regression results on poverty vulnerability under different criteria, using household postal and telecommunications expenditure as an instrumental variable. In the first-stage regression of IV-Eprobit, a significant positive relationship is observed between household expenditure on postal and telecommunication services and the digital embeddedness index of the sampled households. This demonstrates a strong correlation between the instrumental variable and digital embeddedness. In the second-stage regression of IV-Eprobit, the correlation coefficient of the residual term is found to remain statistically significant at the 1% level, supporting the reliability of the instrumental variable approach. The marginal effect of the instrumental variable on rural poverty vulnerability is significantly negative, indicating that, even after addressing endogeneity, digital embeddedness continues to exert a significant mitigating influence on the poverty vulnerability of rural households. These findings reinforce the robustness and credibility of the baseline regression results.

Table 7 Regression results of the ERM using household postal and telecommunications expenditure as an instrumental variable.
Full size table

Robustness test using the Heckman two-stage method

In the preliminary data processing stage, this study obtained a total of 13,602 rural household samples. However, due to missing values in key variables for some samples, 2261 samples were subsequently excluded, including 1384 samples removed because the household poverty vulnerability index could not be calculated. Considering that such sample exclusion based on missing dependent variable data may lead to sample selection bias and thus affect the validity of the regression results, this paper employs the Heckman two-stage estimation method for robustness testing. Specifically, a binary indicator variable denoting whether the household poverty vulnerability variable is missing (coded 0 for missing and 1 for non-missing) is constructed as the dependent variable in the first stage. To satisfy the identification requirements of the model, the degree of respondents’ eagerness to complete the survey is introduced as an exclusion restriction variable. This variable is highly correlated with the completeness of the survey data but is not directly related to poverty vulnerability itself, thereby ensuring its exogeneity.

First, the equation in the first stage is estimated using a Probit model to calculate the Inverse Mills Ratio (IMR). The calculated IMR is then included in the second-stage baseline regression model for estimation to test for the presence of sample selection bias. The second-stage regression results of the Heckman two-stage method, presented in Table 8, show that after controlling for the IMR, the effect of digital embeddedness on the three poverty vulnerability measures remains significantly negative, while the IMR term is insignificant across all models. This indicates that no systematic sample selection bias is detected, demonstrating the robustness of the baseline results.

Table 8 Second-stage regression results of the Heckman two-stage model.
Full size table

In addition, during the study, the main characteristic variables (such as household assets, years of education, household size, etc.) of the excluded samples were compared with those of the retained samples. The results showed no significant differences between the two groups in most variables, further supporting the conclusion that the risk of sample selection bias is low.

Robustness test with alternative regression methods

To further ensure the robustness of the research findings, an alternative method is employed to calculate poverty vulnerability. Instead of setting a poverty vulnerability threshold, this approach generates poverty vulnerability as a continuous variable. The impact of digital embeddedness on rural household poverty vulnerability is re-assessed using OLS regression. As demonstrated in Table 9, even with a change in the regression method, a significant negative influence of digital embeddedness on rural household poverty vulnerability persists. This finding further corroborates the robustness and credibility of the baseline regression results.

Table 9 Regression results after replacing with OLS estimation.
Full size table

Regression estimation with an alternative dataset

To ensure that the empirical results are not driven by a specific sample dataset, an alternative dataset, the 2018 wave of the CFPS, is used to re-estimate the impact of digital embeddedness on rural household poverty vulnerability. Table 10 presents the regression results based on this alternative dataset. The findings indicate that across various relative poverty thresholds, digital embeddedness consistently exhibits a significantly negative impact on rural household poverty vulnerability. These results further support the robustness of the baseline findings.

Table 10 Regression results after replacing the dataset.
Full size table

Further analysis

Further analysis 1: heterogeneous impact of digital embeddedness on rural household poverty vulnerability

There exists a widespread inequality among different entities in bearing disaster risks. This inequality primarily stems from systematic differences in risk sensitivity and coping capacity, resulting in significant heterogeneity in both the degree of risk exposure and the outcomes of risk response. Previous studies have shown that the educational attainment of household heads, as a core component of human capital, not only enhances individuals’ cognitive abilities and technological adaptability (Harker Roa et al., 2023) but also significantly influences their social network structures and access to financial resources, thereby shaping differentiated capacities for risk response (Gesthuizen et al., 2008).

Building on this foundation, the present study further finds that digital embeddedness expands rural households’ access to information and enhances their informational capital. This, in turn, facilitates stronger social connections and improved access to financial resources, ultimately contributing to a significant reduction in poverty vulnerability. This finding raises a critical and yet underexplored question: Does the educational level of household heads moderate the poverty-alleviation effect of digital embeddedness? In other words, does education amplify or weaken the capacity enhancement and resource acquisition effects brought about by digital embeddedness? It is necessary to conduct an in-depth analysis and empirical examination of this question.

Given the generally low educational attainment among rural residents in China, this study categorizes households into two groups based on the educational attainment of the household head: those with education up to the primary school level and below, and those with education at the junior high school level and above. Using the poverty vulnerability index based on the 0.5 times per capita income threshold, an interaction term between digital embeddedness and education level is introduced to explore the differential effect of digital embeddedness on the poverty vulnerability of households with different educational backgrounds. To enhance the robustness of the heterogeneity analysis and account for potential variation in control variable effects across groups, a cross-term between the control variable and the education level group is also generated during the calculation.

Table 11 presents the results of the heterogeneity analysis based on the educational attainment of household heads. Column (1) reports the baseline regression results without the interaction term, while Column (2) incorporates the interaction between digital embeddedness and education level. The estimates in Column (2) show that the marginal effect of digital embeddedness on poverty vulnerability for the low-education group is −0.0480, indicating that a one-unit increase in digital embeddedness corresponds to an average 4.80% reduction in poverty vulnerability for this group. Furthermore, the coefficient of the interaction term is −0.0479, suggesting that the poverty-alleviating effect of digital embeddedness is significantly stronger among households with more highly educated heads. Specifically, for the high-education group, a one-unit increase in digital embeddedness is associated with an approximate 9.59% reduction in poverty vulnerability.

Table 11 Results of heterogeneity analysis by the education level of household head.
Full size table

The heterogeneity analysis reveals that the higher levels of household head education significantly amplify the poverty-reducing effect of digital embeddedness. This moderating role likely arises from differences in cognitive adaptability linked to education levels. Higher education enhances an individual’s capacity to understand and effectively utilize digital technologies. On one hand, it enables more efficient use of digital information for early household risk warning and adaptive production strategies, thereby mitigating potential risk shocks. On the other hand, educational stratification affects the quality and access to social capital, as individuals with higher education are more capable of leveraging digital platforms to obtain valuable resources, which in turn strengthens their households’ risk response capabilities.

Importantly, this variation in the poverty-alleviating impact of digital embeddedness across educational levels may contribute to emerging patterns of social stratification. Since higher-educated individuals are better equipped to harness digital technologies, the gap in poverty vulnerability between them and less-educated groups may widen. This selective technological empowerment potentially exacerbates economic inequality across social strata. These findings provide critical insights for policymakers striving to balance the benefits of digital embeddedness with the pursuit of social equity.

Further analysis 2: the impact of different dimensions of digital embeddedness on rural households’ poverty vulnerability

The preceding analysis has confirmed that digital embeddedness reduces rural households’ vulnerability to poverty and has elucidated the underlying mechanisms through mediation effects. However, digital embeddedness is a multifaceted concept encompassing dimensions such as digital value cognition embeddedness, digital lifestyle embeddedness, and digital productive practice embeddedness. Each dimension reflects distinct resource endowments that affect rural households’ risk resilience in unique ways. Therefore, a comprehensive examination of the heterogeneous effects of these different dimensions on rural households’ poverty vulnerability is crucial. Such analysis not only clarifies the specific pathways through which digital embeddedness mitigates poverty vulnerability but also provides a solid theoretical basis for targeted policy interventions.

Table 12 reports the regression results examining the effects of different dimensions of digital embeddedness on rural households’ vulnerability to poverty. The results reveal that all three dimensions significantly reduce poverty vulnerability, albeit with varying magnitudes. Notably, digital productive practice embeddedness exhibits the strongest effect, with its marginal impact increasing alongside the poverty threshold—from −0.0309 at Vul1 to −0.0526 at Vul3—highlighting its vital role in alleviating structural constraints on household poverty.

Table 12 Regression results on the impact of different dimensions of digital embeddedness on rural household poverty vulnerability.
Full size table

Digital lifestyle embeddedness shows a steadily rising effect, with marginal impacts growing from −0.0135 to −0.0283. This pattern likely reflects the enhanced consumption-smoothing capacity afforded by digital platforms, which helps buffer expenditure shocks among near-poor households. Although digital value cognition embeddedness consistently demonstrates a statistically significant negative coefficient across all specifications, its relatively modest marginal effect suggests that the informational and social capital it provides primarily serve as supplementary resilience mechanisms.

These findings emphasize the importance of differentiated policy responses. Specifically, skill-oriented digital engagement should be prioritized to tackle extreme poverty, while lifestyle-related digital integration calls for improved content governance to bolster its protective role for populations vulnerable near the poverty threshold.

Conclusions and implications

With the continuous advancement and widespread application of digital technology, the embedding of digital technology in the production and daily lives of rural residents has provided new opportunities to alleviate the poverty vulnerability of rural households. Drawing on theoretical analysis, this study employs data from the 2020 and 2022 China CFPS to conduct an in-depth investigation into the impact of digital embeddedness on rural households’ poverty vulnerability.

The key findings of this study are as follows. Firstly, digital embeddedness significantly reduces rural households’ vulnerability to poverty, and this effect becomes more pronounced as the relative poverty threshold increases. Secondly, mechanism analysis reveals that digital embeddedness influences poverty vulnerability through both risk shock and risk response dimensions. In the risk shock dimension, digital embeddedness increases rural households’ agricultural income and facilitates the transition of household members to off-farm employment, thereby mitigating risk shocks and reducing poverty vulnerability. In the risk response dimension, digital embeddedness enhances rural households’ ability to cope with risks by fostering the accumulation of social capital and improving access to financial resources, thereby effectively lowering their poverty vulnerability. Thirdly, Further analysis indicates that the impact of digital embeddedness on poverty vulnerability varies by education level. Rural households whose heads have attained at least a junior high school education experience a stronger inhibitory effect of digital embeddedness on poverty vulnerability. Additionally, the impact of digital embeddedness varies across its dimensions: It is strongest in digital productive practice embeddedness, followed by digital lifestyle embeddedness, with digital value cognition embeddedness having the weakest effect.

Building on the conceptualization of digital embeddedness and the development of a composite index, this study investigates its impact on the vulnerability of rural households to poverty from two dimensions: risk shocks and risk responses. On the one hand, it provides a novel analytical tool for evaluating the level of digital development at the micro-individual level. On the other hand, it also provides a new perspective for investigating the determinants of poverty vulnerability, thereby deepening the theoretical understanding of how digital technology development influences the poverty vulnerability of rural households. In order to better harness the poverty-alleviating potential of digital embeddedness, and based on the empirical findings of this study, the following policy recommendations are proposed:

First, efforts should be directed toward strengthening rural digital infrastructure and enhancing farmers’ digital competencies by establishing a multi-tiered support system for digital transformation. At the county level, “digital agricultural service stations” should be established to integrate resources such as drone-based crop protection, smart farming technologies, and livestream-based agricultural product marketing. Furthermore, targeted digital skills certification programs should be implemented for farmers with at least a junior high school education, with training modules covering topics such as the operation of smart devices and e-commerce platform management. To improve the precision and effectiveness of technological dissemination, participation in certification programs should be linked to eligibility for agricultural subsidies.

Second, a digital risk-buffering mechanism should be established to enhance the integration and coordination of social and financial capital. Local governments, in collaboration with financial institutions, can develop a “digital credit assessment system” that quantifies farmers’ digital engagement—such as e-commerce transaction records and IoT device usage data—into assessable credit assets. In addition, a mutual aid production network can be created through algorithm-based matching, thereby expanding rural households’ access to low-interest credit and community-based social support.

Third, targeted support should be strengthened for digitally disadvantaged groups to systematically narrow the digital capability gap among rural households. The education sector should collaborate with the agriculture and rural affairs departments to develop a localized digital literacy training curriculum, featuring step-by-step learning plans tailored to farmers with varying educational backgrounds. To facilitate digital comprehension among low-education groups, short videos, virtual simulations, and other digital tools should be leveraged, thereby gradually bridging the digital capability gap across households.

This study has several limitations. First, due to data constraints, the digital embeddedness index mainly reflects individuals’ internet usage. Future research could include broader indicators like digital infrastructure and public services for a more comprehensive measure. Second, this study uses an ERM model with village-level internet penetration and household telecom expenditures as instruments to address endogeneity. However, the validity of these instruments cannot be fully assured due to exogeneity testing limitations. Future research should refine the model and seek better instruments to improve robustness.