Does digital literacy reduce the risk of returning to poverty? Evidence from China

China's use of digital technology for poverty alleviation has yielded substantial results. In establishing and improving long-term mechanisms for poverty eradication governance, changing capabilities is the key and ultimate goal in alleviating the problem of poverty. In the context of the widespread adoption of digital technology and rapid development of the digital economy, the lack of digital literacy is a signi�cant obstacle that hinders rural residents from reaping the bene�ts of the digital economy. However, few studies explore the speci�c impact of improving digital literacy on poverty alleviation governance. Based on data from the China Family Panel Studies from 2018, this paper explores the impact of digital literacy on the risk of impoverished households returning to poverty, and examines the underlying mechanism. The study �nds that digital literacy can reduce the risk of impoverished households returning to poverty in China by promoting family entrepreneurship, improving entrepreneurial performance, and expanding entrepreneurial scale. Further analysis shows that this effect varies among households with different regional and capital endowments, and that digital business literacy has the most signi�cant effect in terms of alleviating the risk of returning to poverty. This study has signi�cant implications for understanding and improving the governance mechanism of sustainable poverty alleviation through digital literacy.


Introduction
Poverty is a serious social problem facing China and other countries (Zhou et al., 2018;Griggs D, 2013).The rst goal of the United Nations Sustainable Development Goals is to eliminate all forms of poverty around the world.By the end of 2020, China announced that it had comprehensively established a moderately prosperous society; at the same time, more than 70% of the world's poverty had been eliminated.China has made signi cant contributions to the global poverty reduction cause, and provided important inspiration and bene cial reference for many developing countries.However, China's poverty eradication is not a one-time event, and China No. 1 central document of 2023 states that "we will resolutely guard the bottom line of not returning to poverty on a large scale".It can be seen that the focus of China's poverty alleviation work has shifted from "poverty alleviation" to "poverty prevention", and China will continue to face the major challenges of consolidating the results of poverty eradication, preventing large-scale poverty return, and ensuring the sustainability of poverty eradication (Pan et al., 2020).
At the same time, based on the continuous development of information and communication technology (ICT) such as mobile phones, computers and the Internet, digital transformation has become a trend.According to the "2022 Digital Rural Development Work Points" released by the China Internet Network Information Center, as of the end of December 2022, the Internet penetration rate in rural areas of China exceeded 60%, and the gap between accessibility and availability of digital technology had narrowed.China's implementation of poverty alleviation through digital technology has achieved substantial results.With the continuous introduction of digital technology into people's daily lives, the difference re ected by the digital divide has evolved from the accessibility and availability of digital technology to the inability to obtain equal bene ts from digital technology due to differences in personal abilities and literacy.Therefore, an increasing number of scholars has begun to pay attention to the importance of digital literacy (Scheerder et al., 2019;Philip, 2017a;World bank ,2016).Digital literacy is not uniquely de ned; its core characteristic is an ability and attitude that helps people to collect, understand, and use relevant digital resources.According to the core idea of Sen's (1982) poverty theory, changing capabilities is the key and ultimate aim of alleviating poverty in establishing and improving long-term mechanisms for poverty eradication governance.In particular, in the context of widespread digital technology and the rapid development of the digital economy, the lack of digital literacy is a major obstacle that restricts rural residents from reaping the dividends of the digital economy.Speci cally, the application of digital technology provides equal opportunities for rural residents, but it does not necessarily mean that each rural resident has an equal chance to bene t from the digital development outcomes in the digital economy.Studies have shown that most of the groups that can obtain the "digital dividend" are those who possess the necessary skills and quali cations, while the majority of rural poor are excluded due to their lack of digital literacy (Bhavnani et al., 2008a;Scheerder et al., 2017a).
Digital literacy-the basic abilities of individual farmers in regard to digital applications such as smartphone,computer and so on-can effectively bridge the digital divide and enhance the internal development dynamics of digitalization.Promoting the comprehensive coverage of Internet infrastructure is the external guarantee for preventing poverty under the background of digitalization, while improving the digital literacy of farmers is the internal impetus for preventing poverty recurrence.With the continuous increase in the rate of inclusiveness in digital information technology in China, improving the digital literacy of farmers has long-term theoretical and practical signi cance for poverty alleviation, prevention of poverty recurrence, and establishment of a long-term effective poverty alleviation mechanism.It also offers reference value for poverty governance in various developing countries around the world.However, most extant studies only focus on comparing super cial effects, and rarely analyze the role path and in uence mechanism of digital literacy on poverty recurrence.Against this backdrop, exploring the relationship between digital literacy and poverty recurrence risk undoubtedly has signi cant theoretical value and practical signi cance for achieving stable poverty alleviation in vulnerable households.Due to differences in digital literacy, digital dividends do not reach most groups, especially the rural poor, resulting in half-hearted efforts to alleviate poverty and an increased risk of returning to poverty.Will the improvement of digital literacy have a restraining effect on the risk of returning to poverty?If so, what are the mechanisms underlying this effect?Speci cally, which aspects of digital literacy have the greatest impact?Does this impact vary signi cantly across different regions?For different poverty alleviation groups, does this impact difference result in "bene ting the poor" or "making them more vulnerable"?The answers to these questions will not only help China consolidate its poverty alleviation achievements from the perspective of development and endogenous dynamics, and establish anti-poverty governance and prevention mechanisms in the post-poverty alleviation period, but also provide important reference and inspiration for other developing countries to develop more effective policies and measures against poverty To this end, starting from the perspective of poverty recurrence risk, this paper uses household micro data from the China Family Panel Studies (CFPS) from 2010 to 2018 to explore the effect of digital literacy on the risk of returning to poverty from the impoverished households, and further examines the pathways and structural effects of digital literacy on the risk of returning to poverty.At the same time, the heterogeneous role of digital literacy on the risk of returning to poverty is explored for different regions and capital endowments.
The main contributions of this paper are as follows.First, in terms of research objects, this study attempts to explore the impact of household digital literacy levels on the risk of returning to poverty by directly starting from the micro level of households.In extant studies, scholars have started from the macro level and studied the impact of digitalization on poverty risks (Aker and Mbiti, 2010; Atasoy, 2013a) and relative poverty (Kwilinski et al., 2020), but have not reached a consistent conclusion.This may be because these studies all started with digitalization as the object rather than digital literacy.Therefore, this paper attempts to directly investigate the role of impoverished households' digital literacy in the risk of returning to poverty.Because entrepreneurship is an important means of connecting poor households to the market and is manifested as a new form of digital marketing, this paper attempts to empirically test the transmission mechanism of "digital literacy-household entrepreneurship-risk of returning to poverty" and analyses the mediating transmission mechanism by dividing household entrepreneurship into three components: entrepreneurial entry, entrepreneurial performance, and entrepreneurial scale.
Second, in terms of research methods, most studies to date have explored the impact of digital literacy on poverty reduction by incorporating it into absorptive capacity or endogenous motivation with other factors (Pangrazio, 2016;Neumeyer et al., 2020), which weakens the speci city of digital literacy research.Meanwhile, this paper focuses on exploring the impact of digital literacy on the risk of returning to poverty as an independent core explanatory variable, and further investigates its regional differences and "bene ting-the-poor" characteristics under different regions and capital endowments, thus enriching the literature on the evaluation of the role of digital literacy.
Third, in terms of research content, this paper not only examines the causal relationship between digital literacy levels of impoverished households and the risk of returning to poverty, but also further dissects digital literacy into different dimensions to explore the impact of different sub-dimensions of digital literacy on the risk of returning to poverty.The few literatures that have been written only theoretically and logically suggest that digital skills literacy, which enhances work ability and builds extended social networks, can be capital-enhancing, whereas digital skills for recreation and leisure have no signi cant positive effect (Bonfadelli, 2002;DiMaggio and Bonikowski, 2008).Simply using digital technology cannot automatically enhance people's ability to use it, and more attention should be paid to creative activities such as digital nance and social participation to maximize the advantages of digital resources (James, 2021a).Therefore, this paper further divides digital literacy into ve front-end variables: digital learning literacy, digital work literacy, digital social literacy, digital entertainment literacy, and digital business literacy; analyzes the impact of different sub-dimensions of digital literacy on the risk of returning to poverty; and deepens the understanding of the impact of digital literacy on poverty recurrence risk.
The remaining structure of this paper is arranged as follows: the second section provides a literature review and theoretical analysis; the third section introduces the research design; the fourth section presents the empirical test results and analysis; and the fth section concludes with recommendations.
2. Literature review and theoretical analysis

Research on the risk of returning to poverty
The subject of returning to poverty has always been a focal issue for scholars.The risk of returning to poverty is similar to poverty vulnerability, which is described in the World Development Report (2000/2001) as the likelihood that a household or individual will fall from non-poverty to poverty or continue to fall deeper into poverty in the future.The risk of returning to poverty can therefore be taken to mean not only that poor households are at risk of remaining poor or even falling deeper into poverty, but also the possibility that households that are not poor or that have escaped poverty will fall below the poverty line.Reviewing existing research, the reasons for poverty recurrence can be broadly categorized as follows: (1) Overreliance-induced returning to poverty is a type of poverty recurrence.Overreliance on poverty alleviation policies not only increases pressure on government nances, but also reduces the e ciency of poverty alleviation efforts (Fang et al., 2020a).Its essence lies in the fact that those lifted out of poverty lack endogenous creative abilities and have a natural preference for "blood-transfusion style" poverty alleviation efforts.Once such efforts are discontinued, they will quickly return to poverty.
(2) Risky-accidents-induced returning to poverty is another type of poverty recurrence.According to Barbier (2010), rural poverty is concentrated in regions with poor environments or high vulnerability, where people's livelihoods may depend closely on natural resource utilization and ecosystems.Therefore, extreme weather and natural disasters can have a devastating impact on the production activities of impoverished populations, increasing their risk of falling back into poverty (Xu et al., 2017;Xiang et al., 2023).
(3) Human-capital-induced returning to poverty is a type of poverty recurrence.Nobel laureate Amartya Sen (1999) emphasized in his "Theory of Feasible Ability" that low income is essentially attributable to a lack of capability.Among these, health and education are important components of human capital, and poverty recurrence due to illness, aging, and dropping out of school has become a prominent issue in the context of rural poverty (Yang and Guo, 2020;Fang et al., 2020b;Zhou et al., 2021 ).
(4) Intergenerational-transfer-induced returning to poverty is a type of poverty recurrence based on social genetic codes.
The elements of social genetic codes are personal endowments and the comprehensive effects of the growth environment, especially at the psychological level.Family poverty can have direct or indirect impacts on children's growth, and parents' beliefs and ideas can also make it di cult for their children to accept new knowledge and ways of thinking, leading them to refuse external opportunities for development and falling back into poverty (Lewis, 1959;Yoshikawa, 2012).
To sum up, the return to poverty caused by natural disasters and accidents is based on force majeure factors.However, the other three reasons for returning to poverty can be summarized as poverty in spirit and ability, which essentially arise from the lack of endogenous motivation.

Research hypothesis
Referring to the de nition in the World Development Report (2000/2001), this paper de nes the risk of return to poverty for households as the likelihood that the future standard of living of a household that has been lifted out of poverty, i.e. a household that has emerged from poverty, will again fall below the poverty line when subjected to a negative shock.Based on the de nition of the risk of returning to poverty, digital literacy mitigates the risk of returning to poverty in three main ways: First, digital literacy can enhance the endogenous motivation of those living below the poverty line.Endogenous motivation can enable people to fully exert their initiative and enthusiasm for self-development, allowing them to independently improve their sustainable "self-generating" capacity and avoid relapse into poverty due to interrupted external support (Prior et al., 2016).Theoretically, good digital literacy helps family members understand and use digital information and IT tools to gain expertise and learn the skills needed to do their jobs, which can effectively reduce the cost of acquiring knowledge (Atasoy, 2013b), increase social activities, in uence political participation (Campante, 2018), thus increasing their own development capacity and endogenous motivation.
Digital technology as a source of information channels to promote farmers' skills and as a medium to facilitate the transformation of their labour, abilities and skills into socio-economic values (Lechman and Popowska, 2022b).Improving digital literacy levels can help impoverished households reduce the cost of searching for market demand information, improve the e ciency of market information matching, and meet the needs of impoverished households for employment, entrepreneurship, and business opportunities-for example, through nding job positions via online recruitment platforms, opening online shops or participating in e-commerce-related work through e-commerce platforms, using digital technology to empower management systems and production systems in agriculture and individual private enterprises, impoverished households can reap the digital dividends of the digital technology revolution, improving their production e ciency and enhancing their economic capacity.
Third, improving digital literacy can help impoverished households enhance their market risk identi cation and avoidance capabilities, thereby enhancing resource allocation e ciency in the context of poverty alleviation.Market is a key factor in improving resource allocation.However, under market economy conditions, the distribution of economic results does not inherently favor the poor, which requires impoverished populations to participate in various development projects to realize their own interests.However, adequate digital literacy can help households emerging from poverty to improve their ability to identify and avoid market risks, thereby optimising the distribution of market economy outcomes.For example, increased digital literacy can promote digital nancial inclusion, thereby facilitating household nancial market participation, optimising household asset allocation and improving risk resilience (Xu et al., 2022).Speci cally, by using online social and knowledge exchange platforms and learning about nancial products such as loans, insurance and wealth management offered by digital nancial platforms, households can develop nancial literacy, improve their risk perceptions and gain more access to the fruits of the market economy, thereby reducing the risk of returning to poverty.Based on the above theoretical mechanism analysis, we propose hypothesis 1 H1: Digital literacy can reduce impoverished households' risk of returning to poverty.
The consensus that entrepreneurship reduces poverty risks has been established, and related content on entrepreneurship and poverty reduction is constantly enriched with the development of digital technology.Current cutting-edge theoretical research focuses on the New BOP Model, Resource Replenishment Theory, Platform Empowerment Theory, Institutional Change Theory and Learning Change Theory, which generally refers to the marketplace for the poor and is a grassroots entrepreneurial model or theory.The main difference between the new BOP (Base of the Pyramid Update) model and the BOP model (Base of the Pyramid) is the addition of digital technology and the use of digital platforms, which is closely related to digital literacy.The core idea of the Remediation Perspective is that entrepreneurship can reduce poverty when resources are directly supplied to impoverished households (Sutter et al., 2019a).Thus, when poor people with entrepreneurial needs are provided with material or other social resources to help them start their own businesses, they will not only increase their own and household income, but will also have a positive impact on the non-economic performance of the household, including the health and education and even the social status of the children of the impoverished households.The Platform Empowerment Perspective (PEP) is a hot topic of theoretical exploration in recent years in the context of digital technology.Empowerment refers to the process by which individuals or organisations have greater control over their environment and conditions to replace a sense of powerlessness (Perkins and Zimmerman, 1995).
Platform empowerment can help poor people to build diversi ed entrepreneurial channels on the platform, such as through the sale of their own products to obtain transactions to increase income and achieve a model of poverty reduction (Si et al., 2015).Institutional Reform Perspective (IRP) suggests that the ultimate goal of reducing poverty through entrepreneurship should be to increase social equality (Sutter et al., 2019b), which ts with digital literacy as a strategy to eliminate social inequality (Bach et al., 2018a).Entrepreneurship for poverty reduction involves differences in resources, institutions, etc., but how to improve the personal literacy of the poor to break down individual developmental constraints involves the Learning and Changing Perspective (LCP).This theory assumes that the endogenous motivation of impoverished households is insu cient, but that an increase in digital literacy will stimulate their endogenous motivation and learning potential.Through learning, impoverished households can change their attitudes and actions towards entrepreneurship and trial and error, and can gain useful knowledge and increased awareness through learning (Ashford and Tsui, 1991; Goel and Karri, 2020), thus increasing entrepreneurial income.
Based on the above understanding, it is clear that as digitisation continues to advance, the entrepreneurial poverty reduction model is also closely linked to digital empowerment, with entrepreneurship emerging as an important channel for households or individuals to connect to the market and unlock digital dividends.We hypothesized that the improvement of digital literacy can digitally empower impoverished households, optimize the entrepreneurial poverty reduction mode, bring better human capital, considerable material capital and relatively abundant social capital, and thus alleviate the risk of returning to poverty.Speci cally, digital literacy should be able to mitigate the risk of falling back into poverty for impoverished households through voluntary entry into entrepreneurship, increased performance, and scaling up, for the following reasons: First, digital literacy will give families out of poverty the courage and con dence to start their own businesses.Digital literacy will change the backward marketing knowledge and traditional consumption habits of impoverished households, help families actively understand the new digital service mode and new entrepreneurial form, and further perceive the potential business opportunities in the market, so as to effectively stimulate the entrepreneurial courage and con dence of impoverished households.Secondly, good digital literacy can foster the management ability of impoverished households.
Digital literacy can cultivate the management awareness and ability of impoverished households and improve the e ciency of management production.Thus, with lower production costs, it can quickly gain competitive advantages, improve pro t margins and signi cantly increase entrepreneurial performance.Finally, adequate digital literacy can further scale up entrepreneurship.Adequate digital literacy can enable impoverished households to break through the geographical restrictions of traditional social networks, fully collect and reasonably use digital social network resources, and further expand business channels.And attract more consumers through the Internet platform, digital logistics system and diversi ed sales channels, so as to expand the scale of entrepreneurship and alleviate the risk of returning to poverty.Thus, we put forward hypothesis 2.
H2: Digital literacy can promote household entrepreneurship and reduce the risk of returning to poverty for households that have escaped poverty by promoting entry into self-employment, increasing entrepreneurial performance and entrepreneurial scale.
Based on the above theoretical analysis,the mechanism pathway is drawn in Fig. 1: 3. Research design

Data
We empirically analyzed the impact of digital literacy of impoverished households on the risk of returning to poverty using survey data from the CFPS from 2010 to 2018.The CFPS database re ects the economic and non-economic welfare of Chinese society through data at the individual, family, and community levels.It is a national and comprehensive social tracking survey project that includes information on various aspects such as the economic situation, culture, and education level of sample households.The CFPS database has a wide collection scope, covering 25 provinces (municipalities, autonomous regions) in China.Through implicit strati cation and multi-stage equal probability sampling, it represents household data from these 25 provinces well, which comprehensively re ects China's poverty status and digital literacy situation.
We selected the 2018 survey data from the CFPS, where information on the risk of returning to poverty and control variables primarily come from the Household and Individual databases, while information on digital literacy comes from the Individual database.The data processing procedure was as follows: Data cleaning: removing samples with missing indicators and data anomalies; Horizontally merging the Household and Adult databases to obtain corresponding variables; Identi cation of impoverished households: The research object of this paper is impoverished households.
Households were included in the sample if they had experienced both poverty and then poverty alleviation.We identi ed impoverished households by comparing household consumption levels with the consumption poverty line.Based on the World Bank's 2018 poverty standard of $1.90 per capita consumption per day, a total of 7,757 households whose consumption levels were below the poverty line between 2010 and 2016 were identi ed, among which 4,180 households had been lifted out of poverty by 2018.

Explained variable
The dependent variable is the risk of returning to poverty for households.The risk of returning to poverty refers to the probability that a household that has already lifted itself out of poverty will fall below the poverty line in the future due to risk shocks.This de nition is similar to that of poverty vulnerability, with the only difference being the research object.Hence, this paper used the measurement method of poverty vulnerability to replace the measurement of the risk of returning to poverty for impoverished households.For the selection of feature variables and estimation methods, we referred to a series of variables that affect households' per capita consumption levels.Speci cally, these variables include household characteristics such as household size, whether they are in possession of self-owned housing, logarithm of household net assets, logarithm of per capita household net income, total value of durable goods, natural logarithm of total household debt, whether any family members receive retirement or pension bene ts, whether they receive social donations or government subsidies, whether any family members work outside of the home, land rental expenses, and province virtual variables at the regional level.
Speci c estimation methods are as follows: First, the consumption equation was estimated, and the natural logarithm of the residual square obtained after regression was taken as the OLS estimation of consumption uctuation.The estimated equation is shown as follows: Where, is the per capita household consumption expenditure, is the variables related to the per capita household consumption such as the size of the family population selected above, and is the household consumption uctuation estimated by Formula (1).
In the second step, the tting value obtained in the rst step is used to construct the weight for FGLS estimation, and the expected value and consumption uctuation of logarithmic consumption are obtained as follows: The third step is to select the poverty line and calculate the poverty vulnerability of the h family.Assuming that the consumption level follows the lognormal distribution, the following can be obtained: 5 Where lnZ represents the natural logarithm of the poverty line, and the per capita daily consumption of $1.90 published by the World Bank in 2018 is used as the basis for dividing the poverty line.According to the above steps, the risk of returning to poverty of impoverished households-in other words, the probability value of impoverished households returning to poverty-is nally obtained.

The core explanatory variable
The core explanatory variable of this paper is digital literacy (digital), which re ects an individual's attitude and ability to correctly and reasonably use digital tools and equipment, utilize digital resources, learn new knowledge, and communicate with others socially.
We divided the identi cation of digital literacy into two steps: First, according to the two questions in the questionnaire, "Does the family have Internet access" and "Does the family have mobile access", a preliminary judgment is made on whether a family uses digital tools.If both answers are no, the value of digital literacy of the family is assigned as 1.
Second, digital literacy includes multiple dimensions of digital skills, and the level of digital literacy is measured according to ve questions in the questionnaire: "frequency of using the Internet for learning", "frequency of using the Internet for work", "frequency of using the Internet for social networking", "frequency of using the Internet for entertainment" and "frequency of using the Internet for business".The frequency of use ranges from "never" to "almost every day", with a score of 1 to 7.
Because the digital literacy data were obtained from the personal database-that is, each family member has a digital literacy score for each dimension-we took the average of the digital literacy score of each family member to obtain the digital literacy score of the family under each dimension.In addition, considering the differences in digital infrastructure among provinces and cities, the degree of contribution of each sub-dimension of digital literacy to the overall level of digital literacy is different.It is not appropriate to take the mean value or directly sum the scores of each dimension of digital literacy, which would probably aggravate the gap in digital literacy ability among families.In this paper, the different structural literacy of each province and city in the sample is added and the proportion of literacy of each dimension is calculated, so as to assign equal weighting values within the family, and, nally, the digital literacy score of the family is calculated.

Control variables
To control for other factors affecting the risk of household return to poverty as much as possible, and to avoid bias in the estimation results of the regression model caused by differences in individual, family, and region factors, a series of characteristic variables was selected as control variables for model estimation in this paper.
At the individual level, the head of household age (age), household status (houkou), employment (employ), and marital status of the household head (marriage); the family level includes the size of the household (family size), the proportion of healthy population (health), elderly population (old) and minor population (teen), whether the household owned a car (car), wage income (wage), whether the household owned a house with independent property rights (house), the expenditure on human gifts (expense) and nancial products ( nance).And, whether the household was engaged in self-employment (operation) or agriculture (agriculture).At the area level, community nature (community) and province dummy variables (region) were included.
The meaning and descriptive statistics of speci c variables are shown in Table 1.It can be seen that the higher the poverty standard, the higher the risk of returning to poverty of impoverished households, which is consistent with empirical observation.Table 2 shows the regression results for the in uence of digital literacy on the risk of returning to poverty for impoverished households.The rst column includes only digital literacy, and the estimated coe cient of digital literacy is -0.0167, which is signi cant at the 1% level.The second column includes a series of control variables, and the third column includes control variables and province dummy variables whose estimated coe cient of digital literacy is -0.018, which is signi cant at the 1% level, indicating that the effect of digital literacy on the risk of returning to poverty of impoverished households is signi cantly negative; the results also show that when the level of digital literacy of impoverished households increases by 10 point, the risk of returning to poverty of impoverished households can be reduced by 18%, and the level of digital literacy has a suppressive effect on the risk of returning to poverty of impoverished households. (expense) V ul

Estimation of results: Using Instrumental variable method
To overcome potential endogeneity problem the above benchmark regression model, we used instrumental variables to perform two-stage least squares estimation (2SLS) in order to solve the potential estimation result bias problem.The selection principle of instrumental variables needs to meet the criteria of correlation and exogenesis.Therefore, we selected the "Average Internet usage within the same village except for the interviewees" (internet)as the instrumental variable of digital literacy.In terms of correlation, village-level internet usage re ects the extent to which communication infrastructure is built and improved in the village as a whole, and is highly correlated with the level of digital literacy of households in the village.In addition, according to the principle of exogeneity, the average Internet usage rate in the same village, except for the outdoor area, can in uence households' risk of returning to poverty through their digital literacy level, but the villagelevel Internet usage rate is additive and does not change signi cantly due to the Internet usage behaviors of individual households, nor can it have a direct impact on the risk of returning to poverty of the households interviewed.In summary, the instrumental variable is more desirable and satis es the correlation and exogeneity conditions for the next step of analysis.
First, the instrumental variables were tested for correlation.Column (1) of Table 3 shows the results of the rst stage regression of 2SLS.The F-statistic of 52.473, which is greater than the critical value of 16.38 at the 10% bias level, rejects the original hypothesis that endogenous variables are not correlated with instrumental variables, and also indicates that there is no weak instrumental variable problem.Column (2) of Table 3 shows the regression results of the second stage of the 2SLS, where the coe cient of the effect of digital literacy on the risk of returning to poverty for households that have escaped poverty is -0.0362 after controlling for endogeneity using instrumental variables.Second, the instrumental variables are tested for exogeneity, and the exogeneity of the instrumental variables cannot be tested directly at this point because the instrumental variables are aptly identi ed.Referring to Ashraf (2013), the endogenous variables are replaced with instrumental variables into the baseline model for regression, and the results are shown in column (3) of Table 3: the average Internet usage in villages except for the outdoor area surveyed (internet) has a signi cantly negative effect on the risk of returning to poverty.Further, after controlling for the level of digital literacy and the village Internet usage rate of impoverished households and regressing them, we found that the effect of digital literacy on the risk of returning to poverty of impoverished households was still signi cantly negative, but the effect of village Internet usage rate on the risk of returning to poverty of the impoverished households was not signi cant.The regression results indicate that the village Internet usage rate affects the risk of returning to poverty through the digital literacy level of the impoverished households, and does not directly affect the risk of returning to poverty of the impoverished households; therefore, the hypothesis of exogeneity of the instrumental variable holds.

Substitution the poverty line standard
In the benchmark regression mentioned above, the explained variable is calculated according to the poverty standard of $1.9 per day, which may not be fully applicable to the current national context of China.We therefore used different poverty standards in the robustness test to recalculate the risk of returning to poverty of impoverished households.The regression results after replacing the explained variables are shown in columns ( 1) and ( 2) of Table 4.It can be seen that when different poverty lines are adopted, the digital literacy level has a inhibitory effect on the risk of returning to poverty of impoverished households, and the regression result is robust.

of the criteria for measuring digital literacy
Digital literacy is both an ability and an attitude, and the degree of recognition and attention paid to digital and internet applications re ects the level of a household's digital literacy.In the regression section of this paper, digital literacy is characterized based on the frequency of internet technology usage.However, the measurement method involves assessing the attitude of impoverished households towards the use of digital technology through their evaluation of the importance of various internet functions.The questionnaire includes questions on the "the importance of internet learning", "the importance of internet work", "the importance of internet socializing", "the importance of internet entertainment", and "the importance of internet business activities".These questions are used to gauge the attitude of impoverished households towards using digital skills for learning, work, entertainment, socializing, and business activities, thereby re ecting their level of digital literacy.The scales range from "not important" to "very important" and are assigned scores from 1 to 7. As before, we calculated the proportion of different structural literacies in each province or city in the sample, weighted them equally within households, and nally calculated the digital literacy score for each household.
The regression results with the replaced explanatory variable are presented in column (2) of Table 4.The results show that digital attitude has a signi cant negative impact on the risk of returning to poverty of impoverished households.
Furthermore, using digital attitude to represent the level of digital literacy still yields the same conclusion as the benchmark regression, indicating the results of the benchmark regression are robust.

Exclusion of areas with well-developed Internet infrastructure
Considering that some regions well-developed infrastructure, high level of digital technology and Internet development, high urbanization rate, and high degree of digitalization, the frequency of digital technology usage of households in these regions may be in uenced by external environmental factors rather than solely driven by individual subjective abilities and intentions.Therefore, using the speci c behavioral frequency to represent the level of digital literacy for impoverished households in these areas could lead to biased estimation results.According to the "Internet and Related Services Operation" data jointly released by the Ministry of Industry and Information Technology, the top ve provinces in China with high levels of digital development are Fujian, Shanghai, Beijing, Zhejiang, and Guangdong.Therefore, we attempted to eliminate regions with high levels of digital economic development and then conduct regression; the results are shown in column (3) of Table 4.It can be seen that even after eliminating the ve provinces with more developed internet infrastructure, the regression results are still signi cant.The level of digital literacy among sub-samples has a negative impact on the risk of returning to poverty for impoverished households, which is signi cant at the 1% level and again veri es the robustness of the benchmark regression.

Exclusion of the impact of government subsidies
According the logical framework of the theory of sustainable poverty alleviation, the ability to resist risk shocks may come from both the household itself and external sources such as the government safety net policy or subsidies for basic medical insurance, serious illness insurance, and medical assistance.In this case, even if the household lacks assets to resist risk shocks, the function of buffering income shocks for farmers can also protect them from returning to poverty after experiencing risk shocks.However, this may lead to "welfare dependence", causing residents to ignore the awareness and cultivation of digital literacy, which in turn may reduce the anti-poverty effect of digital literacy.Therefore, the sample households that received government subsidies were excluded and the remaining samples of families that did not receive government subsidies were regressed.The results showed that the coe cient of digital literacy (-0.021) under the subsamples is signi cant at the 1% level, indicating that the improvement of digital literacy has a signi cant effect on the reduction of the risk of returning to poverty for impoverished households.Therefore, we once again veri ed the robustness of the baseline regression.

The functional mechanisms of digital literacy
Although has already been demonstrated that digital literacy can signi cantly reduce the risk of returning to poverty, the speci c pathways that affect the risk of returning to poverty remain unknown.From the theoretical analysis above, it is clear that the entrepreneurial poverty reduction model has become an important channel for households or individuals to connect to the market and unlock the digital dividend in the context of the digital process.Therefore, we attempted to test the transmission mechanism of "digital literacy-household entrepreneurship-risk of returning to poverty" in hypothesis 2, with home entrepreneurship as a mediating mechanism.According to the previous theoretical analysis, entrepreneurial entry is only a way to connect to the market, and whether and how much it can bene t from digital development and digital literacy is mainly re ected in entrepreneurial performance and entrepreneurial scale.Therefore, this paper divides the mediating variable of household entrepreneurship into three sub variables, namely entrepreneurial entry (entrance), entrepreneurial performance (performance) and entrepreneurial scale (scale).The models are constructed in turn to test and further explore the mechanism of the role of digital literacy.
Columns (1), (3), and (5) of Table 5 indicate that increased digital literacy can signi cantly increase the likelihood, performance, and business size of household entrepreneurship.Digital literacy, as an important component of human capital, has a signi cant impact on the integration of social resources, screening of information and entrepreneurial opportunities, and acquisition of knowledge and skills of entrepreneurs, and is an important driver of entrepreneurial entry and further expansion of entrepreneurship to increase pro ts.The results in columns (2), (4), and (6) show that digital literacy, entrepreneurial entry, entrepreneurial performance, and entrepreneurial scale all have a signi cant negative effect on the risk of returning to poverty among households that have escaped poverty, indicating the mediating effects of entrepreneurial entry, performance, and scale.In summary, digital literacy promotes entrepreneurship, improves entrepreneurial performance, increases the scale of entrepreneurship, and thus mitigates the risk of returning to poverty.H1 is veri ed.The improvement of digital literacy can help impoverished households to start their own businesses, participate more in market activities, share digital dividends, and further increase the pro ts and scale of entrepreneurship, which can reduce the risk of returning to poverty.

Heterogeneity analysis
Although the uence of digital literacy on the risk of a family's return to poverty has already been demonstrated, it is not known whether there are differences in this effect.Therefore, we perform the heterogeneity analysis below to examine both macro and micro perspectives.At the macro level, we examine the urban-rural differences in the impact of digital literacy on the risk of returning to poverty; at the micro level, we examine which groups bene t more from digital literacy in terms of physical, human, and social capital.

Universal access does not mean universal bene ts: Heterogeneous analysis of urban-rural differences
Although the diffusion of digital technologies such as the Internet in China's rural areas has yielded substantial results, if urban residents bene t more than rural residents from the increase in digital literacy, the urban-rural gap will further widen, running counter to China's goal of minimizing poverty and extending the fruits of development to all people.Therefore, we need to further explore whether there is heterogeneity in the effect of digital literacy on the risk of returning to poverty from the perspective of urban-rural differences.Accordingly, we performed grouped regressions of model ( 1) based on samples from both urban and rural areas.
The regression results are shown in columns ( 1) and ( 2) of Table 6.The results indicate that increased digital literacy reduces the risk of returning to poverty in both urban and rural areas.Digital literacy has a stronger effect on alleviating the risk of returning to poverty for impoverished households in rural regions.In summary, there is a clear "bene cial poverty effect" of digital literacy on the risk of returning to poverty for Chinese households: digital literacy has a stronger mitigating effect on economically backward rural areas and a weaker mitigating effect on economically more developed urban areas.
The reasons for this can be considered from the respective regional standpoint.First, rural inhabitants do not bene t equally from the fruits of digital development in the context of economic activity.This is because these disadvantaged groups must rst acquire a certain level of digital literacy in order to have better access to economic opportunities and social rights (Akerman et al., 2015;Scheerder et al, 2017b); second, as towns and cities in general have a more developed digital economy, the lower limit of digital literacy requirements is higher (Reynolds and Stryszowsk, 2014b; Bejaković and Mrnjavac, 2020b).Furthermore, when relying solely on digital literacy as human capital, access to income or earnings can also be limited to a large extent (Morduch and Sicular, 2000).

Who bene ts more from digital literacy?
Analysis of the literature has shown that physical, human, and social capital all have signi cant effects on entrepreneurship (Hurst and Lusardi, 2004;Lazear, 2005;Knight and Yueh, 2008).Although the development of digital literacy has a positive impact on the probability of successful entrepreneurship among households emerging from poverty, if it is more strongly focused on helping groups with advantages in the "three capitals", digital literacy will widen the gap within households emerging from poverty, which is contrary to China's intention to accelerate the development of a digital society and promote universal prosperity.Therefore, we further explored which groups bene t most from digital literacy by grouping impoverished households according to their capital endowment, namely physical, human, and social capital.
(1) "High "low income" : Heterogeneous analysis of physical capital The essence of income poverty is poverty of ability.literacy can narrow the income gap by creating a back eld advantage for low-income groups (Philip, 2017b).However, if digital literacy is improved at the same time, the degree of bene t to high-income groups is greater than that to low-income groups, indicating that the "bene cial poverty effect"" of digital literacy at the present stage is insu cient; in fact, it will increase the risk of returning to poverty.Therefore, it is necessary to explore whether low-income groups bene t more.Based on this, the family annual income was taken as the proxy variable of material capital; then, impoverished households were divided into a high-income group (high-income) and a low-income group (low-income) according to the median annual income, and the regression is carried out.
The results in columns (3) and (4) of Table 6 indicate that digital literacy signi cantly reduces the risk of returning to poverty for both high-income and low-income groups.In absolute terms, the marginal effect of digital literacy on lowincome households out of poverty is higher than that of high-income households out of poverty, i.e., digital literacy can indeed be a form of "timely assistance" for low-income groups.
This may be due to the fact that the use of digital technology has a more pronounced effect of bene ting poverty for lowincome groups.The use of the Internet can affect poverty vulnerability, especially among relatively poor rural populations, by in uencing off-farm employment.On the one hand it enhances human capital (Pabilonia and Zoghi, 2005) and facilitates the transfer of surplus rural labour.On the other hand, off-farm employment earns wage income and is an important way to increase farmers' income (Deng et al., 2019).By using the Internet, farmers break down information barriers and are able to collect timely and accurate information on off-farm employment, facilitating the effective allocation of surplus rural labour resources, thereby increasing current income and reducing the risk of falling into poverty in the future.

Further analysis
The effect of literacy improvement on the risk of returning to poverty may not be comprehensive if analyzed in isolation, so we disentangled digital literacy and explored the effect of different sub-dimensions of digital literacy on the risk of returning to poverty for impoverished households.Digital literacy no longer refers only to the learning and use of digital skills, but also includes general and creative literacy (Alexander et al., 2016).James (2021b) argued that households should actively participate in creative activities such as digital nance and social engagement to exploit the advantageous effects of digital resources; accordingly, this paper uses ve front-end variables of digital literacy levels, namely digital learning literacy (learning), digital work literacy (work), digital social literacy (social), digital entertainment literacy (entertainment), and digital business literacy (business), and constructs regression models to analyze the effects of different sub-dimensions of digital literacy on the risk of returning to poverty among impoverished households.
The results are shown in Table 7.It can be seen that all ve dimensions of digital literacy have a signi cant inhibitory effect on the risk of returning to poverty among households who have escaped from poverty.Among them, digital business literacy (business) has the greatest impact on the risk of returning to poverty for households out of poverty.Moreover, for every 10-point increase in digital business literacy , the risk of returning to poverty is reduced by 19.7%.This may be because digital business literacy (business) can help impoverished households nd a foothold and opportunities for growth more quickly in more digitally advanced production and business scenarios.Speci cally, as marketization continues to develop, the frequency of digital technology-enabled transactions made by households emerging from poverty increases, and the more likely they are to develop economic opportunities that are conducive to nding hidden deeper in the market, thereby alleviating poverty and reducing the risk of returning to poverty.
Digital entertainment literacy (entertainment) also has a signi cant dampening effect on the risk of returning to poverty among households that have escaped poverty, second only to digital business literacy (business), which is a departure from the commonly held "entertainment futility theory".One reason for this may be that digital entertainment literacy (entertainment) expands an individual's social network.Families can accumulate their own network of contacts and expand their social network through online diversi ed entertainment, which can lead to the accumulation of rich social capital, sharing the risk of entrepreneurship and increasing the likelihood of formal or private nancing, and thus curbing (business) the risk of returning to poverty (Kinnan and 2012).Another reason may be related to the means of knowledge and information access.Along with the digitization of entertainment, more and more knowledge and information dissemination methods are skewed towards entertainment.Compared with traditional paper media and o ine training and education, information dissemination methods based on short entertainment videos are more up-to-date, easier to understand and accept, and less costly to acquire, thus effectively helping family members out of poverty to acquire relevant knowledge and skills and mitigate the risk of returning to poverty.

and recommendations
This paper utilizes the CFPS 2018 dataset to empirically test the effect, and mediating pathway, of digital literacy on the risk of impoverished households returning to poverty.Furthermore, this article analyzes the heterogeneity and effect size of different sub-dimensions of digital literacy on the risk of impoverished households returning to poverty.Through empirical analysis, the following conclusions are drawn: First, digital literacy has a signi cant reducing effect on the risk of returning to poverty overall.For every 10-point increase in digital literacy, the risk of returning to poverty can be alleviated by 18%.Second, this study found that when impoverished households improve their digital literacy, they can mitigate the risk of returning to poverty through the pathway of family entrepreneurship.Speci cally, impoverished households can signi cantly reduce the risk of returning to poverty by independently undertaking entrepreneurship, improving entrepreneurial performance, and expanding entrepreneurial scale.
Third, heterogeneity analysis revealed that improving digital literacy would bring a "bene cial poverty effect" rather than the "Matthew effect".Rural poverty-stricken households would enjoy more bene ts from the improvement of digital literacy, as would poverty-stricken groups with lower capital endowments.Therefore, improving digital literacy can indeed further eliminate social inequality, which is consistent with Bach et al. (2018b).Finally, through analyzing the impact of different dimensions of digital literacy on the risk of poverty-stricken households returning to poverty, it was found that digital business literacy is the most important factor in alleviating the risk of impoverished households returning to poverty, followed by digital entertainment literacy and digital work literacy.
Based on the analysis above, we propose the following.First, in addition to continuing to increase investments in digital Third, supporting policies for digital literacy need to be targeted to key groups.Rural areas with relatively low economic development and poverty-stricken groups with low capital endowment are the groups that need to be paid attention to when preventing large-scale poverty return.Because the level of digital literacy plays a more signi cant role in alleviating the risk of returning to poverty for this group, it has a greater potential space for policy implementation.
Finally, to further leverage the role of digital literacy in alleviating the risk of returning to poverty, it is necessary to guide poverty-stricken populations to cultivate their own digital business and entertainment literacy-for example, through encouraging poverty-stricken households to use digital platforms for entrepreneurship, participate in creative activities such as digital nance and social communication, and actively maximize the effect of digital resources.
Based on the above analysis, there are still some questions to be explored.This paper nds that digital literacy can promote farmers' entrepreneurship and improve the e ciency and scale of entrepreneurship.However, the mechanism underlying the impact of farmers' digital literacy on farmers' entrepreneurial decisions is worth further exploration.

*Ethical Approval
All analyses were based on previous published studies, and the data in our study is collected from public resources through legal channels, which will not violate any ethical rules.Thus no ethical approval and patient consent are required.
The functional mechanisms of digital literacy infrastructure and general digital application platforms, poverty alleviation work should also pay attention to improving the digital literacy of impoverished households.Although China has made substantial progress in the popularization of digital technology, popularity does not equal inclusiveness.Currently, China's digital literacy education system is lagging.Impoverished households ' digital skills can be developed by establishing digital skill training systems, digital technology innovation and entrepreneurship support systems, and other means, fully leveraging the roles of schools, training institutions, social organizations, and other institutions in improving digital literacy levels.Second, entrepreneurship is an important pathway for impoverished households to reduce the risk of returning to poverty by improving their digital literacy.The government should increase support and risk protection for entrepreneurship among poverty-stricken households.Most impoverished households have experienced poverty and are relatively conservative when facing market risks, lacking entrepreneurship information and basic skills.Therefore, the government can take relevant measures to further improve impoverished households' access to entrepreneurship opportunities and basic literacy.For example, conducting training to aid impoverished households in understanding entrepreneurship steps and requirements, helping them use digital technology to understand the business market, identifying business opportunities, and speci cally improving their specialized digital business literacy.

Table 1
Meanings and descriptive statistics of variables Note: The value of the expenditure on human gifts is taken as a natural logarithm.

Table 2
Note: Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.Note: Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.

Table 3
Note: Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.

Table 4
Results of the robustness tests Note: Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.

Table 6
Results of the heterogeneity analysis Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, and *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.
Values in brackets are t-stat.The values in parentheses are standard deviations; *, **, and *** indicate the level of signi cance of 10%, 5%, and 1%, respectively.Data are analyzed by authors using Stata16.