Is Digital Development the Answer to Energy Poverty? Evidence from China

Abstract

Energy poverty is one of the major challenges to global sustainable development, while digital development, as a significant trend of the current era, is considered a key pathway to transcend traditional energy governance frameworks. Anchored in provincial panel data spanning 30 regions across China from 2003 to 2023, this study systematically examines the impact and heterogeneity of digital development on energy poverty and further explores the underlying mechanisms and nonlinear characteristics. The findings show that digital development can significantly alleviate energy poverty, and this conclusion remains valid after addressing endogeneity issues and conducting a series of robustness tests. However, the poverty reduction effect of digital development exhibits significant regional heterogeneity: the mitigation effect in central and western regions is significantly stronger than that in eastern regions, the effect in northern regions is higher than that in southern regions, and the effect in energy-disadvantaged regions is better than that in advantageous regions. Additionally, digital development alleviates energy poverty through mediating pathways such as promoting non-agricultural employment, improving human capital levels, and driving technological innovation. Notably, digital development demonstrates threshold effects and quantile heterogeneity in relation to energy poverty, characterized by diminishing marginal returns as digital development progresses; regions with higher levels of energy poverty experience more significant poverty reduction effects from digital development. This research provides a theoretical basis for energy poverty governance under the global energy crisis and offers empirical references for other countries to achieve energy sustainability goals (SDG7) through context-specific digital transformations.

1. Introduction

Energy serves as the central driving force behind economic and social development. It not only acts as a vital pillar for the progress of human civilization but also forms a crucial basis for attaining sustainable development goals [1]. Nevertheless, the global energy landscape is undergoing profound changes. The volatility in energy prices and the instability in supply have worsened the energy crisis. For developing countries in particular, the issue of energy poverty has become increasingly severe [2]. Energy poverty is primarily defined as the lack of access to modern energy services that are affordable, dependable, and environmentally sustainable. Its consequences not only impede economic growth but also have far-reaching effects on social equity, public health, and the ecological environment [3]. According to a report by the International Energy Agency (IEA), although the number of people worldwide without access to electricity has decreased by 925 million since 2000, approximately 750 million people still lack electricity. Furthermore, more than two billion people lack access to clean cooking technologies, with a substantial proportion residing in developing nations across sub-Saharan Africa and Asia [4]. Sustainable Development Goal 7 (SDG7) of the United Nations’ 2030 Agenda emphasizes universal access to modern energy services that are affordable, dependable, and sustainable by the year 2030 [5]. However, the IEA predicts that by 2030, around 650 million people globally will still be without electricity, and about 1.7 billion people will lack clean cooking facilities [4]. China, being the world’s most populous developing country, faces energy poverty issues that have become a focal point of extensive research and debate [6]. Data from the World Health Organization (WHO) shows that there are still approximately 293 million people in developing countries, including China, who rely on solid fuels [7]. Energy poverty not only restricts economic growth in these regions but also exacerbates social inequality, especially in rural and remote areas [8]. The eradication of energy poverty is therefore a matter of significant importance for developing economies, with China being no exception.

Facing the multidimensional challenges of energy poverty, digital transformation is regarded as a critical pathway to transcend traditional energy governance frameworks [9,10,11]. By embedding advanced tools like big data, artificial intelligence, and the Internet of Things, digital development reconfigures the organization of production, delivery, and consumption activities. Its core characteristics include data-driven elements, cross-domain collaborative innovation, and dynamic adaptive optimization [12]. In the energy sector, digitization influences energy consumption, efficiency, structure, and security [13,14]. Research demonstrates that the application of digital technologies contributes to tackling wider energy sustainability issues by improving efficiency in energy use [15]. Beyond advancing the efficiency of energy system oversight and control, digitization supports global cooperation, facilitates international energy trade, refines governance and policy design, and disseminates exemplary approaches to energy efficiency [16,17]. The application of internet-based technologies diminishes reliance on traditional energy in industry and simultaneously advances upgrading processes and improvements in the energy mix [18]. By integrating digital technologies into traditional energy sectors, particularly in production and transmission, innovations such as the energy internet, distributed platforms, automated production lines, and intelligent regulation have emerged, transforming energy supply processes, improving regulatory practices, and decreasing transaction costs [19]. Donner (2008) [20] posits that information and communication technologies (ICT) can bring higher productivity and social equity to impoverished populations. Their application in developing countries positively impacts information access, financial inclusion, and economic growth [21]. Acting as a key catalyst, the digital economy promotes development across economic and industrial sectors while advancing the modernization of energy systems and networks [22]. Under the strategic context of “Digital China” construction, the Chinese government addresses energy poverty by deeply integrating the internet with energy industrial chains and leveraging digital technologies to promote clean and efficient energy use [23]. Numerous factors contribute to energy poverty [24]. Research findings have shown that key attributes of household heads, such as their income, gender, level of education, age, and family composition, play a crucial role in determining household energy poverty [25]. According to Drescher and Janzen (2021) [26], factors such as family structure, labor market participation, education level, low energy efficiency, and housing heating arrangements are closely associated with energy poverty. Paradoxically, energy poverty may intensify with human development progress [27]. Research on Australia demonstrates that greater ethnic diversity within households corresponds to elevated risks of energy poverty [28]. Governments play a crucial role in addressing energy poverty. Belaïd (2022) [29] found that poorly designed policies may worsen energy poverty, whereas effective government interventions can alleviate it by raising personal incomes and reducing social inequality [30]. According to Dimnwobi et al. (2022) [31], energy poverty in Nigeria is exacerbated by public capital expenditure, which hampers both urban electrification and renewable energy generation. Barnes et al. (2011) [32], using 2004 data from rural Bangladesh, found that energy poverty affected approximately 58% of the rural population, and that electrification substantially narrowed this gap. Ren et al. (2022) [33] demonstrate that marketization processes can mitigate rural energy poverty, while natural gas availability and usage exert substantial positive effects [6]. Xia et al. (2022) [34] found that financial decentralization and escalating country risks may intensify energy poverty. From a public health standpoint, the COVID-19 pandemic intensified energy poverty in Pakistan [27]. According to Munyanyi and Churchill (2022) [35], limited energy supply and high energy prices serve as major constraints on alleviating energy poverty in developing countries. For example, despite achieving universal electricity access, a significant portion of rural Chinese residents still face energy poverty due to perceived high electricity costs [36]. In China, energy poverty reflects unequal distribution of resources and imbalances in the development of the energy sector [37].

The primary drivers of energy poverty include restricted availability of modern energy services, low uptake of clean energy, and inefficient energy consumption, all of which can potentially be alleviated through advancements in digital technologies [8]. The role of digital development in mitigating energy poverty has garnered significant academic interest [15], and research exploring how the digital economy and related technological innovations influence energy poverty has emerged as an important area within energy economics [8,38]. Based on household survey microdata from Lhasa, Zang et al. (2023) [39] demonstrate that mobile phone use helps mitigate energy poverty by strengthening residents’ social capital, increasing non-agricultural income, and raising household expenditures. Enhanced financial market development increases households’ capacity to meet energy costs, thereby alleviating energy poverty [40,41]. Likewise, digital finance facilitates energy efficiency by providing financial services that support the adoption of energy-saving technologies, particularly in areas where access to modern clean fuels is expensive [42]. The integration of digital tools encompassing telecommunications, internet, and data analytics is especially effective in mitigating energy poverty in areas characterized by weak governance and public administration [43]. Wang and Lin (2022) [15] develop a digital adoption index covering 65 countries and regions, demonstrating that digital development plays a key role in decreasing dependence on fossil fuels and electricity consumption, thus mitigating energy poverty. Lyu et al. (2023) [8] assess provincial energy poverty in China and demonstrate that the digital economy alleviates energy poverty by fostering technological innovation and enhancing energy affordability, efficiency, and sustainability. Pelz and Urpelainen (2020) [44] examine the relationship between ICT availability and grid connectivity in rural northern India, finding that ICT adoption facilitates more equitable energy access and helps alleviate poverty. This conclusion is validated in China, where rural digitalization significantly reduces energy poverty primarily by improving farmers’ income quality—a more effective strategy than mere income growth [45]. Policies addressing energy poverty must incorporate digital transformation, including improved internet access and integrated digital-renewable energy strategies [46].

However, some argue that digitalization-enabling technologies may act as barriers to energy poverty alleviation [47], suggesting digital development could exacerbate rather than mitigate energy poverty. The process of digitalization may generate issues like the digital divide, leaving those with insufficient technical readiness or limited connectivity marginalized, especially in low-income and low-education areas [48]. According to Luan et al. (2023) [49], the digital divide negatively influences household energy poverty, thereby impeding poverty reduction initiatives. Privacy concerns, personal freedom risks, and cybersecurity issues may reduce willingness to adopt new technologies, undermining the effectiveness of digital solutions for energy poverty [10]. Zhou et al. (2018) [50] and Salahuddin and Alam (2015) [51] reveal that advancements in information and communication technology (ICT) lead to higher power consumption and energy intensity, thereby worsening energy poverty. Studies indicate that rapid ICT development in EU countries may increase unemployment and exacerbate energy poverty [52]. Ackermann et al. (2023) [28] found that although high-speed internet can help mitigate energy poverty by enabling online education, telemedicine, and remote work, it may simultaneously increase time spent indoors and energy consumption, thereby exacerbating poverty. Similar findings exist in China: although digital development promotes economic growth and industrial upgrading, it significantly increases energy consumption [23]. While numerous studies examine energy poverty and the impact of digital technologies, the connection between digital development and energy poverty remains unresolved. Our understanding of energy poverty and effective solutions for its mitigation continues to be limited, warranting further research [47]. As China enters a period of accelerated digital development [53], urgent research is needed to provide international references for achieving SDG7 through digital transformation [8].

Overall, existing research examines how digital development impacts energy poverty across multiple dimensions. Despite providing valuable theoretical support and empirical evidence, considerable gaps persist. In particular, although a substantial number of studies address the role of digitalization or the digital economy in shaping energy poverty, the distinction between digital development and the digital economy is rarely clarified. As a result, their indicator systems remain nearly identical. However, digital development and the digital economy differ significantly in both connotation and denotation. Accurate measurement of digital development could yield novel insights into this relationship. Second, most literature focuses on direct effects from micro-level household (or rural household) perspectives or macro-level cross-national, provincial, and regional (rural areas or specific cities) perspectives. However, they overlook the nonlinear effects of digital development levels and energy poverty severity on poverty reduction outcomes. These nonlinear relationships may alter the marginal effectiveness of digital solutions in different contexts. Third, existing studies on China’s energy poverty concentrate data samples between 2011–2019. While this timeframe highlights the accelerated adoption of digital technologies, it fails to account for the extended development path of the “Digital China” initiative. Since 2003, marked by the “Digital Zhejiang” initiative, local governments have pioneered integration of digitalization with socioeconomic development, gradually evolving into the national “Digital China” strategy. The exclusion of the early stage (2003–2010) risks severing the long-term causal chain between digitalization and energy poverty alleviation, potentially underestimating the true impact of digital development, or misidentifying critical mechanisms.

The primary contributions and innovations of this study are threefold: (1) By clarifying the theoretical distinctions between digital development and the digital economy, the study develops a seven-dimensional indicator system to measure China’s digital development, thereby enhancing the existing theoretical framework. (2) The research systematically examines the relationship between digital development and energy poverty, considering not only direct effects but also heterogeneous characteristics and underlying mechanisms. It further investigates the nonlinear effects of varying digital development levels and energy poverty severity on poverty alleviation outcomes, thus refining the analytical framework and expanding research in this domain. (3) Utilizing sample data spanning 2003–2023, the study provides a comprehensive and detailed empirical assessment of the link between digital development and energy poverty, offering robust evidence for the poverty reduction effects of digital development under the “Digital China” strategy.

The study is organized as follows: Section 2 outlines the theoretical framework and formulates the research hypotheses. Section 3 presents a detailed description of the methodology and data employed. Section 4 reports the empirical findings, while Section 5 examines the mediating effects of non-agricultural employment, human capital, and technological innovation capacity. Section 6 explores the nonlinear relationship between digital development and energy poverty. Finally, Section 7 concludes the study and discusses policy implications within the context of China. The detailed research framework is illustrated in Figure 1.

2. Theoretical Analysis and Hypothesis

2.1. Mechanisms of Direct Effect

The causes of energy poverty can be decomposed into insufficient accessibility to energy services, low cleanliness of consumption, inadequate management completeness, and overloaded energy expenditure burdens combined with low efficiency. First, regarding energy service accessibility, based on the technology diffusion theoretical model [54], digital technologies break geographical constraints and enhance energy consumption convenience by constructing smart energy networks and distributed energy systems. Second, in terms of clean energy consumption, digitalization drives a low-carbon and cleaner energy consumption structure through technological empowerment. Additionally, digital platforms enhance consumer awareness and preference for clean energy through environmental information disclosure and carbon footprint tracking mechanisms, accelerating the clean energy transition—a behavioral change consistent with the attitude–subjective norm–perceived behavioral control framework [55]. Third, improving energy management completeness relies on enhanced managerial and investment capabilities enabled by digitalization. According to transaction cost theory [56], governments integrate energy data through digital governance platforms to establish dynamic monitoring and early-warning systems, improving the scientificity and responsiveness of energy planning. Enterprises optimize energy production processes using digital twin technology to reduce operational costs. Finally, improvements in energy affordability and efficiency manifest as reduced energy costs and enhanced energy utilization efficiency through digital technologies. Mobile payment and energy e-commerce platforms lower transaction costs, making energy services more accessible to low-income groups. Shared energy models and energy service outsourcing enabled by the digital economy generate significant positive externalities [57], creating employment opportunities, raising household incomes, and indirectly strengthening energy affordability. Based on this, the following research hypothesis is proposed:

H1. 

Digital development has a positive effect on alleviating energy poverty.

2.2. Mechanisms of Indirect Effect

This study investigates how digital development may influence energy poverty through three key channels: expansion of non-agricultural employment, enhancement of human capital, and promotion of technological innovation.

First, regarding the non-agricultural employment mechanism, digitalization creates substantial non-agricultural job opportunities through new digital economy formats (e.g., e-commerce, sharing economy, remote services), transforming traditional single-income structures reliant on agriculture. The migration of rural surplus labor into digital economy sectors boosts household disposable income and strengthens the capacity to afford clean and efficient energy through occupational transitions. For example, the digital platform economy lowers employment barriers, enabling low-skilled groups to participate in emerging industries like logistics and data labeling. Their income growth directly translates into consumption capacity for clean energy products (e.g., induction cookers, solar lamps), aligning with the traditional-to-modern sector dual structural transformation logic [58]. Notably, digital technology-enabled flexible employment models (e.g., gig economy) allow poor households to earn supplementary income while managing agricultural production, significantly enhancing budget flexibility for energy expenditure. Based on this, the following hypothesis is proposed:

H2. 

Digital development indirectly alleviates energy poverty by promoting non-agricultural employment.

Second, human capital enhancement constitutes a critical mediating pathway. The downward penetration of digital educational resources (e.g., online courses, vocational skill training platforms) breaks geographical constraints and significantly improves human capital accumulation in rural areas. Digital technology-enabled vocational education can specifically cultivate professional skills such as energy management and smart device operation, enhancing individual employability in the energy sector. For example, emerging job demands like photovoltaic power station operation and maintenance, smart home system management are directly linked to digital skill training. Groups mastering such skills not only earn higher incomes but also optimize household energy efficiency through technological applications, a knowledge-skill-income chain, consistent with the law of increasing returns in human capital investment [59]. Importantly, enhanced digital literacy allows low-income households to effectively employ digital tools, such as energy management applications and efficiency monitoring devices, thereby improving energy efficiency through greater technical proficiency. Based on this, the following hypothesis is proposed:

H3. 

Digital development indirectly alleviates energy poverty by improving human capital levels.

Third, strengthening technological innovation capabilities constitutes the core mechanism through which digitalization influences energy poverty. According to endogenous growth theory [60], deep integration of digital technologies with the energy sector—such as AI algorithms optimizing grid dispatch and blockchain enabling distributed energy transactions—drives energy technological innovation and reduces clean energy development costs. These innovations improve traditional energy efficiency while accelerating clean energy adoption. Notably, digital platforms facilitate industry–university–research collaboration, shortening technology transfer cycles and enabling low-cost clean energy solutions to reach poor regions faster. Additionally, energy internet models spawned by digital technology restructure energy production–consumption relationships, enhancing energy efficiency through user-side energy storage and demand response mechanisms. Quality improvements and cost reductions in energy services brought by technological innovation fundamentally address energy poverty challenges. For more specific analysis, please refer to Appendix A.1 and Appendix A.2. Based on this, the hypothesis is proposed:

H4. 

Digital development indirectly alleviates energy poverty by enhancing technological innovation capabilities.

2.3. The Mechanism of Nonlinear Effect

2.3.1. Theoretical Analysis of the Impact of Digital Development Disparities on the Effect of Digital Development in Alleviating Energy Poverty

The impact of digital development on energy poverty may exhibit heterogeneity across different development levels. According to technology diffusion theory [54], initial-stage technological innovations play a critical role in alleviating energy poverty by improving energy supply-demand coordination and managerial effectiveness. At this point, the marginal benefits of technological applications are higher than the marginal costs [61]. However, as the level of digitalization crosses a certain threshold, the marginal costs of technology penetration accelerate, leading to a gradual weakening of the alleviating effect. When digital development enters the stage of in-depth application, the energy system’s dependence on specific technological paths deepens, forming a technological-institutional symbiotic lock-in [62]. This technological lock-in effect makes it difficult for subsequent technological improvements to break through the existing framework, and the marginal benefits show a decreasing trend. Digital development requires the transformation of energy governance models from administrative-led to data-driven, but institutional changes have significant time lags. At higher stages of digitalization, institutional obstacles such as data ownership, privacy protection, and cross-departmental coordination become major constraints. This institutional stickiness offsets the benefits of technological innovation with institutional friction costs, creating a paradox where technology is advanced but its impact is diminished [63]. Based on this, the following hypothesis is proposed:

H5. 

The impact of digital development on energy poverty may exhibit a threshold effect of digital development.

2.3.2. Theoretical Analysis of the Impact of Energy Poverty Disparities on the Effect of Digital Development in Alleviating Energy Poverty

Variations in regional energy poverty may cause inconsistent effects of digital development on its alleviation. The law of diminishing marginal utility suggests that in regions with elevated energy poverty, digital development initially addresses core survival requirements, including availability and cleanliness of energy services, resulting in markedly higher marginal benefits than in regions with lower energy deprivation. For example, the remote operation and maintenance of photovoltaic equipment in extremely impoverished areas made possible by digital technology can directly overcome geographical limitations of traditional energy supply, and this “timely assistance” effect has higher marginal value compared to “icing on the cake” in areas with well-established energy infrastructure [64]. According to the poverty trap framework [65], households experiencing more severe energy poverty exhibit greater energy system fragility and are more sensitive to interventions involving digital technologies. When the level of energy poverty exceeds a critical threshold, impoverished groups are more inclined to adopt digital solutions that can directly alleviate survival pressures; the marginal benefits of infrastructure increase [66]; Policymakers in areas with severe energy poverty are more likely to break through conventional constraints and promote innovative applications of digital technology. The higher the degree of poverty, the greater the room for adaptive adjustment of the energy system to digital technology, and the marginal benefits of technological innovation show exponential growth [67]. Based on this, the following hypothesis is proposed:

H6. 

The alleviating effect of digital development on energy poverty may exhibit differential heterogeneity in energy poverty.

The visualization of the mediation mechanism is shown in Figure 2.

3. Methods and Data

3.1. Model Settings

3.1.1. Baseline Econometric Model

To empirically examine the impact of digital development on energy poverty, this study sets up the following model (Equation (1)) (to mitigate potential heteroscedasticity and ensure the stationarity of variables, this study applies logarithmic transformation to all relevant variables):

$${lnPoverty}_{it}=\alpha +\beta {lnDigitization}_{it}+\delta {lnX}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

where i represents the region and t represents the year. The dependent variable ${lnPoverty}_{it}$ indicates the energy poverty status of region $i$ in $t$ year, and the independent variable ${lnDigitization}_{it}$ represents the digital development level of region $i$ in year $t. {lnX}_{it}$ represents a series of control variables that may affect regional energy poverty. ${\mu }_i$ denotes regional fixed effects, ${\lambda }_t$ denotes time fixed effects. ${\epsilon }_{it}$ denotes random disturbance terms. $\alpha$ is the constant term.

3.1.2. Mediation Effect Model

To empirically verify whether potential transmission mechanisms exert mediating effects during the process of digital development alleviating energy poverty, the following mediating effect model is introduced for verification:

$${lnPoverty}_{it}=Cons+\beta {lnDigitization}_{it}+\delta {lnX}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

$${lnM}_{it}=Cons+\alpha {lnDigitization}_{it}+\delta {lnX}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(3)

$${lnPoverty}_{it}=Cons+{{\beta }^{\prime }lnDigitization}_{it}+\gamma {lnM}_{it}+\delta {lnX}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(4)

where ${lnM}_{it}$ represents mediating variables encompassing three potential mechanisms: non-agricultural employment, human capital level, and technological innovation capacity. $Cons$ denotes the constant term. Definitions of remaining variables are consistent with those in Equation (1).

3.1.3. Panel Threshold Model

To capture the nonlinear relationship between digital development and energy poverty, this study utilizes a panel threshold model. The model is formulated as follows:

$$\begin{array}{ll}{lnProverty}_{it}& =\alpha +{\beta }_1{lnDigitization}_{it}I\left({lnDigitization}_{it}\le {\gamma }_1\right)\\ & +{\beta }_2{lnDigitization}_{it}I\left({\gamma }_1<{lnDigitization}_{it}\le {\gamma }_2\right)\\ & +\cdots \\ & +{\beta }_{n+1}{lnDigitization}_{it}I\left({lnDigitization}_{it}>{\gamma }_n\right)\\ & +\delta {lnX}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}\end{array}$$

(5)

The variable definitions remain consistent with previous sections. In Equation (5), I (·) represents the indicator function of the panel threshold model, which takes the value of 1 when the condition in the parentheses is true, and 0 otherwise; The parameter $\gamma$ denotes the threshold value to be tested.

3.2. Variables and Data Sources

3.2.1. Explained Variable

Energy poverty (Poverty): While there is no universally agreed-upon standard for measuring energy poverty, some scholars have adopted multidimensional evaluation indicators to assess this issue, which has gradually become an important benchmark for measuring energy poverty in recent years [68]. There is a need to construct a unified assessment system for energy poverty in China that integrates its theoretical concept with the prevailing situation and aligns with current national indicators and standards. Building on the IEA’s energy development index for developing countries and informed by Dong et al. (2021) [6] and Lyu et al. (2023) [8], a multidimensional assessment system was established, incorporating 17 indicators grouped into four dimensions: energy service access, consumption cleanliness, management robustness, and household affordability and efficiency (see

Table 1. Energy poverty indicator system construction.

Goal	Dimension	Element	Specific Indicator	Unit	Property
Energy poverty	Accessibility of energy services	Energy consumption	Per capita electricity consumption	kWh/cap	-
			Per capita natural gas consumption	m3/cap	-
		Energy supply	Per capita electricity generation	kWh/cap	-
			Gas penetration rate	%	-
			Per capita LPG supply volume	t/cap	-
			Per capita hot water supply capacity	W/cap	-
	Cleanliness of energy consumption	Low carbonization of energy consumption structure	Per capita installed hydropower capacity in rural areas	kW/cap	-
		Clarification of energy consumption structure	Biogas production volume per rural household	m3/household	-
			Per capita clean energy consumption	kgce/cap	-
	Comprehensiveness of energy management	Energy management capability	Per capita number of legal entities in relevant industries	units/106 people	-
			Proportion of employees in relevant industries	%	-
		Energy investment capability	Per capita investment in energy industry	yuan/cap	-
			Per capita energy funding investment in rural areas	yuan/cap	-
	Energy affordability and efficiency	Energy affordability	Family income	10k yuan	-
		Energy efficiency	Air conditioner ownership per 100 households	units/100 households	-
			Range hood ownership per 100 households	units/100 households	-
			Refrigerator ownership per 100 households	units/100 households	-

). Considering the instability inherent in the selected indicators, the entropy approach is employed to estimate the energy poverty index. By utilizing information from data variation, it derives indicator weights and supports comprehensive evaluation across multiple dimensions. For more detailed information, please refer to Appendix A.3, Appendix A.4 and Appendix A.5. According to the characteristics of the entropy value method measurement, the study categorized energy poverty into five levels (when 0 < Poverty Index ≤ 0.2, it is classified as Marginal poverty; 0.2 < Poverty Index ≤ 0.4 corresponds to Moderate poverty; 0.4 < Poverty Index ≤ 0.6 indicates Significant poverty; 0.6 < Poverty Index ≤ 0.8 represents Severe poverty; and 0.8 < Poverty Index ≤ 1 is defined as Extreme poverty), with the specific status of energy poverty in China presented in Figure 3 (due to space limitations, only the data for the years 2003, 2010, 2017, and 2023 are presented).

3.2.2. Core Explanatory Variable

This research defines digital development as an integrated state covering multiple fields such as digital technology, digital economy, digital application, and digital governance. It refers to the systematic development level of digitalization at the regional level and reflects the overall evolutionary trend of digitalization. Theoretically, digital development represents a comprehensive process encompassing technological penetration, institutional restructuring, and social transformation. Fundamentally, it reflects the extensive fusion and systematic restructuring of digital tools such as big data, AI, and IoT across both productive sectors and social life. This manifests in multidimensional scenarios, including the digital upgrading of infrastructure (e.g., 5G networks, smart grids), the transformation of organizational and management models (e.g., e-government, telemedicine), and the shift in social behavioral paradigms (e.g., the popularization of digital literacy, the expansion of online education) [69].

In contrast, the digital economy is essentially the concrete manifestation of digital progress in the economic sphere. The official definition of the digital economy is as follows: “The digital economy refers to a series of economic activities that take the use of digital knowledge and information as key production factors, modern information networks as important carriers, and the effective use of information and communication technologies as an important driving force for efficiency improvement and economic structure optimization” (Source: G20 Initiative on Digital Economy Development and Cooperation, https://www.cac.gov.cn/2016-09/29/c_1119648520.htm (accessed on 25 August 2025)). The term refers to a novel economic paradigm characterized by data-driven development, reliance on digital platforms as key intermediaries, and the fusion of digital innovations with traditional economic activities. It is principally structured around two dimensions: the advancement of digital industries (e.g., cloud computing, blockchain) and the digitalization of conventional sectors (e.g., smart manufacturing, e-commerce) [70].

The two form a logical relationship between the “whole” and the “part”: digital development serves as the overarching framework for technological and social change [71,72], providing the technological foundation and institutional environment for the digital economy; while the digital economy is the value realization form of digital development at the economic operation level. Clarifying the distinction between the two holds significant theoretical value: in energy poverty research, focusing solely on the income growth effects brought by the digital economy may underestimate the fundamental improvements in energy accessibility driven by digital development, potentially leading to structural deviations in policy design.

Digital development (Digitization): Compared with existing measurements related to digitization, this study places greater emphasis on the connotations and quality of digitization development, striving for a more comprehensive evaluation. Specifically, by synthesizing numerous existing studies and considering the characteristics of digitization development in China, while referencing the approaches of Li and Ruan (2025) [73] and Tian and Liao (2024) [74], this study establishes a comprehensive evaluation system for digitization development that integrates seven dimensions: digital infrastructure, digitization input, digitization application, digital innovation capacity, digital industrialization, industrial digitization, and digital governance. Detailed indicator descriptions are provided in

Table 2. Digital development indicator system construction.

Goal	Dimension	Specific Indicator	Unit	Property
Digital development	Digital infrastructure	Number of internet broadband access ports	104 units/km2	+
		Capacity of mobile telephone switching centers per square kilometer	104 households/km2	+
		Capacity of local telephone switching centers per square kilometer	104 households/km2	+
		Length of long-haul optical fiber cable lines	104 km	+
	Digital investment	Fixed asset investment in digital-related industries	108 yuan	+
		Proportion of employees in digital-related industries	%	+
		Average wage of employees in digital-related industries	yuan	+
	Digital applications	Mobile phone penetration rate	%	+
		Number of internet users	104 people	+
		Digital TV subscribers	104 people	+
		Computer usage per 100 people	units/100 people	+
	Digital innovation capability	Number of digital invention patent applications	104 items	+
		Number of digital utility model patent applications	104 items	+
		Number of registered software copyrights	items	+
	Digital industrialization	Total volume of telecommunications services	108 yuan	+
		Output value of software and information services industry	108 yuan	+
		Number of legal entities in digital-related industries	units/104 people	+
	Industrial digitalization	Number of websites per 100 enterprises	nits/100 enterprises	+
		Sales revenue from e-commerce	108 yuan	+
		Proportion of enterprises with e-commerce transaction activities	%	+
	Digital governance	Frequency count of digitalization-related terms in government work reports	items	+
		Number of government open data platforms	units	+
		Establishment date of provincial big data management platform	years	+

. Since the measurement of digitization development level involves a comprehensive assessment of numerous factors, this study continues to employ the entropy method, which has the advantages of strong objectivity and high precision, to calculate the weights of each indicator, thereby avoiding deviations caused by subjective human factors. For more detailed information, please refer to Appendix A.6. The digitization development level index in China, calculated based on the entropy method, is shown in Figure 4 (due to space limitations, only the average digitization development level from 2003 to 2023 (a span of 21 years) is presented, along with the digitization development levels for the four specific years of 2003, 2010, 2017, and 2023.).

3.2.3. Other Variables

Control variables. Building upon the research by Feng et al. (2025) [24], He et al. (2023) [75], Tang (2024) [76], and Mahumane and Mulder (2022) [77], this study incorporates a series of control variables to identify the independent effect of digitization development more accurately on energy poverty. Specifically, the control variables include: regional economic development level (PerGDP), relative energy price (Enepri), urban-rural disparity (Urbrur), population density (Popden). Mediating variables. After an in-depth analysis of the impact mechanisms, drawing on the research by He et al. (2023) [75], (Wu and Lu, 2023) [78], and (Hu et al., 2020) [79], this study examines the transmission paths through which digitization development alleviates energy poverty from three possible dimensions: non-agricultural employment, human capital level, and technological innovation capability (For more detailed information, please refer to Appendix A.7).

3.2.4. Data Sources

To analyze the influence of digital development on energy poverty in China, this study employs panel data from 30 provinces, excluding Tibet, Hong Kong, Macau, and Taiwan due to data constraints. The sample period spans 2003–2023, reflecting the evolution of China’s digitalization process. The empirical analysis primarily relies on official statistical sources (see Appendix A.8 and Appendix A.9 for details,

Table A2. Data Sources for Specific Indicators in China’s Energy Poverty Comprehensive Index System.

Specific Indicator	Source of Indicators
Per capita electricity consumption	China Energy Statistical Yearbook, National Bureau of Statistics, Computation
Per capita natural gas consumption	China Energy Statistical Yearbook, National Bureau of Statistics, Computation
Per capita electricity generation	China Energy Statistical Yearbook, National Bureau of Statistics, Computation
Gas penetration rate	National Bureau of Statistics
Per capita LPG supply volume	China Energy Statistical Yearbook, Computation
Per capita hot water supply capacity	China Energy Statistical Yearbook, Meteorological Science Data Sharing Service Network, National Bureau of Statistics, Computation
Per capita installed hydropower capacity in rural areas	China Energy Statistical Yearbook, Computation
Biogas production volume per rural household	China’s Agricultural Statistical Data
Per capita clean energy consumption	China Energy Statistical Yearbook, Meteorological Science Data Sharing Service Network, Computation
Per capita number of legal entities in relevant industries	China Energy Statistical Yearbook, National Bureau of Statistics, Computation
Proportion of employees in relevant industries	Agricultural statistics and calculations in China
Per capita investment in energy industry	China Energy Statistical Yearbook, National Bureau of Statistics, Computation
Per capita energy funding investment in rural areas	China Agricultural Statistical Data, China Population and Employment Statistical Yearbook, Computation
Family income	National Bureau of Statistics, Computation
Air conditioner ownership per 100 households	National Bureau of Statistics
Range hood ownership per 100 households	National Bureau of Statistics
Refrigerator ownership per 100 households	National Bureau of Statistics

has been systematically sorted out). Missing values are supplemented through linear interpolation to ensure data completeness. Descriptive statistics of the variables are provided in

Table 3. Statistical descriptions of variables.

Variables	Obs	Mean	Std. Dev.	Min	Max
lnPoverty	630	−0.533	0.197	−1.025	−0.109
lnDigitization	630	−2.714	0.996	−4.693	−0.404
ln PerGDP	630	10.460	0.772	8.218	12.207
ln Enepri	630	−4.407	1.236	−9.690	−2.112
lnUrbrur	630	0.974	0.172	0.585	1.495
lnPopden	630	5.440	1.273	2.000	8.281
lnNonagr	630	−0.466	0.262	−1.510	−0.013
lnHuman	630	6.076	1.023	2.573	8.319
lnScitec	630	13.814	1.635	9.079	17.364

.

4. Results and Discussion

4.1. Baseline Regression

This study investigates the direct effect of digital development on energy poverty, with the findings reported in

Table 4. Estimated results of baseline regression.

Variables	(1)	(2)	(3)	(4)	(5)
lnDigitization	−0.026 ***	−0.026 ***	−0.027 ***	−0.027 ***	−0.025 ***
	(0.008)	(0.007)	(0.007)	(0.007)	(0.008)
lnPerGDP		−0.514 ***	−0.503 ***	−0.441 ***	−0.439 ***
		(0.076)	(0.075)	(0.073)	(0.073)
lnEnepri			0.011 ***	0.010 ***	0.010 ***
			(0.003)	(0.003)	(0.003)
lnUrbrur				0.251 ***	0.260 ***
				(0.041)	(0.043)
lnPopden					−0.024
					(0.033)
Constant term	−0.413 ***	0.657 ***	0.675 ***	0.269	0.389
	(0.031)	(0.160)	(0.158)	(0.167)	(0.237)
Control variables	No	Yes	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	630	630	630	630	630
Adjusted R2	0.936	0.941	0.942	0.946	0.946

Note: Standard errors are in parentheses (); *** denotes p < 0.01, ** denotes p < 0.05, and * denotes p < 0.1.

. To assess the robustness of the estimates, model (1) excludes control variables, while models (2)–(5) sequentially incorporate them. The results from model (1) indicate that the coefficient of digital development is −0.026, statistically significant at the 1% level, suggesting that digital advancement reduces energy poverty. This provides empirical support for H1, namely that higher levels of digital development significantly mitigate energy poverty. Models (2)–(5) consistently yield significantly negative coefficients, even after including the full set of control variables. In the fully specified model, the coefficient remains −0.025, still significant at the 1% level. These outcomes demonstrate that the results are stable across specifications, further confirming H1.

The estimation results for the control variables reveal several key relationships. First, regional economic development is negatively associated with energy poverty, significant at the 1% level, implying that economic growth contributes to reducing energy poverty. Specifically, a 1% rise in economic development corresponds to a 0.439% decline in energy poverty. This effect can be explained by the fact that higher economic growth enhances residents’ financial capacity to adopt modern energy, thereby reducing deprivation. Second, the relative price of energy is positively linked to energy poverty, also significant at the 1% level. A 1% increase in energy prices raises energy poverty by 0.01%. Since energy is a necessity with low demand elasticity, particularly for low-income households, rising prices constrain their ability to reduce consumption. Consequently, higher prices lower real household income (income effect) and may force substitution toward cheaper energy sources. However, when substitutes are limited or of inferior quality, the severity of energy poverty intensifies. Third, the urban-rural income gap exhibits a positive and significant correlation with energy poverty at the 1% level. A 1% widening of this gap results in a 0.26% increase in energy poverty. This outcome reflects the disparity in infrastructure: urban areas typically enjoy advanced energy facilities (e.g., grid access, gas pipelines) and reliable supply, whereas rural regions depend more heavily on biomass or decentralized energy sources, which are costly and unstable. Moreover, with lower average incomes, rural households face higher relative energy expenditure burdens, heightening their vulnerability to energy poverty. Finally, the coefficient of population density does not demonstrate statistical significance in the benchmark regression, indicating no measurable effect on energy poverty in this context.

4.2. Endogeneity Treatment

4.2.1. Instrumental Variable Approach

Endogeneity may produce biased and inconsistent estimates. To address this issue, this study employs the number of post offices and fixed telephones in each province (autonomous region or municipality) in 1984 as instrumental variables for digital development. Local telecommunications infrastructure directly influences the subsequent development and adoption of digital technologies; thus, these historical measures effectively capture regional digital infrastructure, satisfying the relevance condition for instrumental variables. At the same time, these historical data do not directly affect contemporary energy poverty, meeting the exogeneity requirement.

Given that both instruments are cross-sectional, they cannot be directly applied in a panel data context. Following the approach of Nunn and Qian (2014) [80], we construct interaction terms by multiplying the previous year’s number of broadband internet users (which varies over time) by the 1984 counts of post offices and fixed telephones in each province. These interactions serve as instrumental variables for provincial digital development levels, denoted IV1 and IV2, respectively. The study then applies a two-stage least squares (2SLS) panel instrumental variable approach to re-estimate the model, with results presented in

Table 5. Regression results obtained using instrumental variables.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	IV1 2SLS		IV2 2SLS		IV1 LIML	IV2 LIML
	lnDigitization	lnPoverty	lnDigitization	lnPoverty	lnPoverty	lnPoverty
IV1	0.244 *** (0.013)
IV2			0.241 *** (0.016)
lnDigitization		−0.047 *** (0.001)		−0.031 *** (0.012)	−0.047 *** (0.001)	−0.031 *** (0.012)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Anderson-LM	217.268 [0.000]		168.289 [0.000]
Cragg-Donald Wald F	328.482 <16.38>		227.441 <16.38>
Sargan	0.074 [0.786]
Observations	630	630	630	630	630	630

Note: The values in parentheses () are standard errors; *** denotes p < 0.01, ** denotes p < 0.05, and * denotes p < 0.1. The values in brackets [] are the p-values of the Anderson-LM statistic and Sargan statistic, respectively. The values in angle brackets <> are the critical values of the Stock-Yogo weak instrumental variable test at the 10% level.

.

Column (1) of Table 5 reports the first-stage 2SLS regression using IV1, showing a significant positive relationship between IV1 and digital development. The Anderson-LM statistic has a p-value of 0.000, confirming this correlation, while the Cragg-Donald Wald F statistic is 328.482, well above the 10% Stock-Yogo critical value of 16.38, indicating that IV1 is a strong instrument. Similarly, column (3) presents the first-stage results for IV2, which also exhibits a significant positive correlation with digital development. The Anderson-LM statistic p-value is 0.000, and the Cragg-Donald Wald F statistic is 227.441, exceeding the critical threshold and confirming the strength of IV2. The Sargan test yields a p-value of 0.786 for both instruments, supporting their exogeneity.

Columns (2) and (4) display the second-stage 2SLS results for IV1 and IV2, respectively, with coefficients of −0.047 and −0.031. These estimates are close to the benchmark coefficient of −0.025, all significant at the 1% level, and consistent with the baseline regression, indicating that digital development continues to reduce energy poverty after accounting for endogeneity. Additionally, the study employs the limited information maximum likelihood (LIML) method as an alternative instrumental variable approach, with results shown in columns (5) and (6). A comparison of the LIML and 2SLS estimates reveals strong consistency, further confirming the robustness of the findings.

4.2.2. Dynamic Panel Model

The potential endogeneity issues in this study may arise from omitted variables and reverse causality. The dynamic panel model effectively overcomes the problems of omitted variables and reverse causality [81] (Arellano and Bond, 1991). Considering that the current energy poverty in a region is closely related to the previous period’s energy poverty status, meaning that energy poverty has certain inertia and lag, this study introduces lagged terms of the dependent variable into Equation (1) to construct a dynamic panel model. This study re-examines the impact of digitization development on energy poverty in China through the dynamic panel model. The dynamic panel model Equation (6) is as follows:

$${\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{t}\mathrm{y}}_{\mathrm{i}\mathrm{t}}=\mathrm{\alpha }+\mathrm{\phi }{\mathrm{l}\mathrm{n}\mathrm{P}\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{t}\mathrm{y}}_{\mathrm{i},\mathrm{t}-1}+\mathrm{\beta }{\mathrm{l}\mathrm{n}\mathrm{D}\mathrm{i}\mathrm{g}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{z}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}\mathrm{t}}+\mathrm{\delta }{\mathrm{l}\mathrm{n}\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\lambda }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(6)

In this equation, the meanings of the variables are the same as those in Equation (1), with the only difference being the introduction of the lagged term of lnPoverty in the dynamic panel model. Since the system GMM can utilize more information than the difference GMM, resulting in smaller estimation biases and more efficient estimation results, this study employs the system GMM to estimate the newly constructed dynamic panel model. The results are presented in

Table 6. Regression results of dynamic panel model for the impact of digital development on energy poverty.

Variables	System GMM		Difference GMM
	(1)	(2)	(3)	(4)
L.lnPoverty	0.945 *** (0.006)	0.877 *** (0.026)	0.897 *** (0.005)	0.852 *** (0.012)
lnDigitization	−0.017 *** (0.001)	−0.027 *** (0.005)	−0.026 *** (0.001)	−0.029 *** (0.002)
Constant term	−0.099 *** (0.006)	−0.261 *** (0.079)	−0.149 *** (0.005)	−0.152 (0.125)
Control variables	No	Yes	No	Yes
Observations	600	600	600	600
AR (1)	−3.4097 *** [0.001]	−3.395 *** [0.001]	−3.419 *** [0.001]	−3.374 *** [0.001]
AR (2)	0.487 [0.626]	0.688 [0.492]	0.556 [0.578]	0.667 [0.505]
Sargan test-p value	1.000	1.000	1.000	1.000

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1; values in brackets [] are p-values for AR (1) and AR (2).

.

The premise for both the system GMM and the difference GMM to hold is that there is no autocorrelation in the disturbance term. The test results, as reported in Table 6 under AR (2), show that the p-values are all greater than 0.05, which are not significant, indicating the absence of second-order autocorrelation. Since the GMM estimation utilizes internal instrumental variables to construct moment conditions and the number of instrumental variables exceeds the endogenous explanatory variables (digitization development), an over-identification test is required. The test results, as reported in Table 6 under Sargan test-p values, show that the p-values are all greater than 0.05, indicating that the instrumental variables are valid. This confirms that the dynamic panel model meets the prerequisite standards for using GMM estimators. Column (1) in Table 6 presents the regression results of the system GMM without control variables, and column (2) presents the regression results of the system GMM with control variables. Regardless of whether control variables are included, the estimated coefficients of the lagged dependent variable are significantly positive at the 1% statistical level, with coefficient values reaching 0.945 and 0.877, respectively. This reflects that the previous period’s energy poverty level has a positive effect on the current period’s energy poverty level. This also indicates that energy poverty is a dynamic and continuous process with significant lag effects. After including the lagged energy poverty, the negative impact of digitization development on the current energy poverty status remains unchanged and is still significant at the 1% statistical level. After including control variables, the coefficient is −0.027, which is basically consistent with the coefficient of −0.025 in the benchmark regression. This once again proves that, after considering endogeneity issues, the mitigating effect of digitization development on energy poverty remains significant, meaning that endogeneity issues have not significantly impacted the research conclusions. Furthermore, this study also estimates the dynamic panel model using the difference GMM. The results are shown in columns (3) and (4) of Table 6, which are consistent with the system GMM estimation results, indicating the robustness of the research findings.

4.3. Robustness Tests

In order to ensure the robustness of the research findings, robustness tests are conducted, considering the replacement of the explained variable, replacing the core explanatory variable, core explanatory variables lagged by one period, model specification change, exclusion of municipalities, bilateral winsorization of data, and incorporation of new control variables such as family size [75], energy infrastructure level [82] and forest resource abundance [76] (see Appendix A.10 for details). The results presented in

Table 7. Robustness test results.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
lnDigitization	−0.018 ***	−0.029 ***	−0.020 **	−0.106 ***	−0.018 **	−0.023 ***	−0.026 ***
	(0.006)	(0.006)	(0.008)	(0.005)	(0.007)	(0.007)	(0.007)
lnPerGDP	−0.338 ***	−0.400 ***	−0.389 ***	−0.497 ***	−0.372 ***	−0.503 ***	−0.402 ***
	(0.062)	(0.073)	(0.079)	(0.056)	(0.069)	(0.076)	(0.073)
lnEnepri	0.011 ***	0.009 ***	0.008 ***	0.017 ***	0.005 *	0.010 ***	0.009 ***
	(0.002)	(0.003)	(0.003)	(0.003)	(0.002)	(0.003)	(0.003)
lnUrbrur	0.205 ***	0.245 ***	0.276 ***	0.376 ***	0.207 ***	0.264 ***	0.221 ***
	(0.037)	(0.043)	(0.047)	(0.039)	(0.043)	(0.044)	(0.043)
lnPopden	−0.001	−0.106 ***	−0.065 *	0.039 ***	−0.300 ***	−0.024	−0.189 ***
	(0.028)	(0.033)	(0.036)	(0.012)	(0.036)	(0.033)	(0.058)
lnFamsca							0.174 ***
							(0.052)
lnEneinf							−0.013 ***
							(0.004)
lnForest							0.096 ***
							(0.030)
Constant term	0.220	0.741 ***	0.502 *	−0.241	1.746 ***	0.528 **	−0.818 **
	(0.201)	(0.220)	(0.265)	(0.158)	(0.229)	(0.240)	(0.388)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	630	630	600	630	546	630	600
Adjusted R2	0.931	0.946	0.944	0.931	0.962	0.946	0.949

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

indicate that the coefficients of digital development remain significantly negative at the 5% level, confirming that digital advancement effectively mitigates regional energy poverty and is consistent with the findings of the benchmark regression.

4.4. Heterogeneity Analysis

Given China’s vast geographic and socioeconomic diversity, the impact of digital development on energy poverty cannot be assumed to be uniform. A nuanced analysis considering various sources of heterogeneity is therefore essential to identify differential effects. Accordingly, the following section examines the relationship between digital development and energy poverty across three dimensions of heterogeneity.

4.4.1. Alleviation Effects of Digital Development on Energy Poverty in Different Economic Regions

Following the classification standards of the National Bureau of Statistics (see Appendix A.11), the study divides China’s 30 provinces into three economic regions: eastern, central, and western. The primary differences in digital development and energy poverty are observed between the eastern region and the combined central and western regions. Accordingly, the analysis focuses on this contrast, merging the central and western regions for comparative purposes. The regression results are presented in

Table 8. Regression results of economic regional heterogeneity.

Variables	Eastern Regions		Central and Western Regions
	(1)	(2)	(3)	(4)
lnDigitization	−0.016	−0.022	−0.030 ***	−0.029 ***
	(0.016)	(0.017)	(0.008)	(0.008)
Constant term	−0.491 ***	−0.997 **	−0.362 ***	0.360 **
	(0.054)	(0.502)	(0.033)	(0.181)
Control variables	No	Yes	No	Yes
Regional fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Observations	231	231	399	399
Adjusted R2	0.907	0.911	0.959	0.962

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Columns (1) and (2) display the effects of digital development on energy poverty in the eastern region without and with control variables, respectively, while columns (3) and (4) report the corresponding results for the central and western regions.

The regression results indicate that digital development alleviates energy poverty across regions and regardless of whether control variables are included, with effects in the central and western regions being significant at the 1% level. However, the impact demonstrates notable regional heterogeneity. In the eastern region, the effect is not statistically significant, likely due to its advanced economy and well-developed energy infrastructure, which results in relatively milder energy poverty and smaller marginal gains from digital development. Conversely, the central and western regions, characterized by less developed economies and incomplete energy infrastructure, face more severe energy poverty. In these areas, digital development can improve income for low-income households, enhance energy affordability, and thereby more effectively reduce energy poverty.

The eastern region, characterized by a developed economy and well-established energy infrastructure, experiences relatively mild energy poverty. Consequently, the marginal impact of digital development in this area is limited, making significant alleviation of energy poverty difficult. Initiatives such as the “West-to-East Power Transmission” and “West-to-East Gas Pipeline” have largely secured energy service availability in the region. Additionally, the energy consumption structure in the east is dominated by electricity and clean energy, with energy poverty more often manifesting as high energy costs or insufficient service quality rather than lack of accessibility, which may constrain the effectiveness of digital development. In contrast, the central and western regions feature less developed economies and incomplete energy infrastructure, resulting in more severe energy poverty. Data from the National Energy Administration indicate that both energy accessibility and service quality in these regions lag behind the eastern region. The application of digital technologies in the central and western regions is therefore more likely to enhance energy access and utilization efficiency. As a result, the marginal effects of digital development are greater in these areas, enabling a more substantial reduction in energy poverty.

4.4.2. Alleviation Effect of Digital Development on Energy Poverty in Different Geographical Regions

China’s vast territory encompasses considerable variations in climate, historical culture, energy structure, and energy demand between the northern and southern regions. Consequently, the effect of digital development on energy poverty may differ geographically. In this study, the 30 provinces are divided into northern and southern regions, using the Qinling Mountains-Huaihe River as the boundary (see Appendix A.12 for details). The regression results are presented in

Table 9. Regression results of geographical location heterogeneity.

Variables	Northern Regions		Southern Regions
	(1)	(2)	(3)	(4)
lnDigitization	−0.046 ***	−0.031 ***	−0.014	−0.003
	(0.011)	(0.012)	(0.011)	(0.010)
Constant term	−0.496 ***	1.122 ***	−0.259 ***	−1.150 ***
	(0.046)	(0.368)	(0.041)	(0.354)
Control variables	No	Yes	No	Yes
Regional fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Observations	315	315	315	315
Adjusted R2	0.915	0.932	0.957	0.962

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Columns (1) and (2) report the effects of digital development on energy poverty in the northern regions without and with control variables, respectively, while columns (3) and (4) provide the corresponding results for the southern regions.

The regression results reveal pronounced geographical heterogeneity in the effect of digital development on energy poverty. In the northern regions, the coefficients are significantly negative at the 1% level regardless of the inclusion of control variables, indicating that digital development effectively reduces energy poverty. In contrast, while the coefficients in the southern regions are consistently negative, they are not statistically significant, suggesting a limited impact of digital development in these areas. This difference can be partly explained by regional energy structures. The northern regions rely more heavily on traditional energy sources such as coal, where digital technologies can more effectively optimize production, distribution, and consumption, thus significantly alleviating energy poverty. In the southern regions, energy systems are more diversified, incorporating both traditional and cleaner sources such as hydropower and nuclear energy. The relative stability and diversity of these systems may reduce the marginal impact of digitalization on energy poverty. Additionally, the cold climate in the north necessitates substantial winter heating, providing greater scope for digital technologies to improve heating efficiency and further reduce energy deprivation.

4.4.3. Alleviation Effect of Digital Development on Energy Poverty in Regions with Different Energy Endowments

Differences in energy endowment not only influence energy accessibility and pricing but may also affect the efficacy of digital technology applications within the energy sector. To address this issue, the study conducts a heterogeneity analysis based on regional energy endowment (see Appendix A.13 for details). The regression results are presented in

Table 10. Regression results of energy endowment heterogeneity.

Variables	Energy-Disadvantaged Regions		Energy-Advantaged Regions
	(1)	(2)	(3)	(4)
lnDigitization	−0.054 ***	−0.067 ***	−0.013 *	−0.013 *
	(0.010)	(0.010)	(0.007)	(0.007)
Constant term	−0.562 ***	−0.090	−0.217 ***	−0.903 ***
	(0.037)	(0.322)	(0.029)	(0.260)
Control variables	No	Yes	No	Yes
Regional fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Observations	315	315	315	315
Adjusted R2	0.931	0.944	0.979	0.979

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Columns (1) and (2) report the effects of digital development on energy poverty in energy-disadvantaged regions without and with control variables, respectively, while columns (3) and (4) display the corresponding results for energy-advantaged regions.

As presented in Table 10, the coefficients of digital development are negative in both energy-advantaged and energy-disadvantaged regions, indicating a mitigating effect on energy poverty. However, the impact is substantially stronger in energy-disadvantaged regions, where the coefficients are −0.054 and −0.067, significant at the 1% level. In contrast, in energy-advantaged regions, the coefficient is −0.013 and significant only at the 10% level, reflecting a comparatively weaker effect. This suggests that the impact of digital development on energy poverty has significant energy endowment heterogeneity. Energy-disadvantaged regions face more difficulties and challenges in energy access and utilization due to their relatively scarce energy resources and have a more urgent need to improve energy efficiency and optimize energy structure. Digital technology can improve this situation through various means, such as utilizing digital energy monitoring systems to control energy consumption and optimize energy distribution networks, thereby significantly alleviating energy poverty more accurately. In contrast, energy-advantaged regions have relatively abundant energy resources, the potential to further improve energy utilization efficiency through digital technology is relatively limited, and the marginal benefits are less pronounced.

5. Impact Mechanism Test

5.1. Non-Agricultural Employment Mechanism Test

This subsection employs a stepwise regression approach to examine the mediating role of non-agricultural employment, with results presented in

Table 11. Test results of non-agricultural employment mechanism.

Variables	(1) lnPoverty	(2) lnNonagr	(3) lnPoverty
lnDigitization	−0.025 ***	0.121 ***	−0.012
	(0.008)	(0.015)	(0.008)
lnNonagr			−0.114 ***
			(0.021)
Constant term	0.389	−0.879 *	0.289
	(0.237)	(0.463)	(0.232)
Control variables	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes
Observations	630	630	630
Adjusted R2	0.946	0.736	0.948

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Column (1) shows that the coefficient of digital development is −0.025 and statistically significant at the 1% level, indicating that digital development substantially alleviates regional energy poverty. Column (2) reports a coefficient of 0.121 for digital development, also significant at the 1% level, demonstrating its positive effect on promoting non-agricultural employment. In column (3), the coefficient of non-agricultural employment is −0.114, significant at the 1% level, suggesting that increased non-agricultural employment reduces energy poverty. Moreover, the coefficient of digital development in column (3) declines to −0.012 and is no longer significant, indicating that non-agricultural employment partially mediates the relationship between digital development and energy poverty, thereby supporting H2. For robustness, the mediating effect is further validated using the Bootstrap method, with results presented in

Table 12. Test results of non-parametric bootstrap method for non-agricultural employment.

Effect	Coefficient	Standard Error	Z-Value	p-Value	95% Confidence Interval
Mediation effect	−0.014	0.003	−4.540	0.000	[−0.020, −0.008]
Direct effect	−0.012	0.008	−1.37	0.172	[−0.028, 0.005]

.

5.2. Human Capital Level Mechanism Test

This subsection applies stepwise regression to examine the mediating role of human capital, with results presented in

Table 13. Test results of human capital level mechanism.

Variables	(1) lnPoverty	(2) lnHuman	(3) lnPoverty
lnDigitization	−0.025 ***	0.089 ***	−0.016 **
	(0.008)	(0.026)	(0.007)
lnHuman			−0.102 ***
			(0.011)
Constant term	0.389	−0.102	0.379 *
	(0.237)	(0.822)	(0.222)
Control variables	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes
Observations	630	630	630
Adjusted R2	0.946	0.972	0.952

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Column (1) indicates that the coefficient of digital development is −0.025, significant at the 1% level, suggesting that digital development effectively alleviates regional energy poverty. In column (2), the coefficient of digital development is 0.089 and significant at the 1% level, demonstrating its positive influence on enhancing human capital. Column (3) shows that human capital has a coefficient of −0.102, significant at the 1% level, indicating that higher human capital levels contribute to reducing energy poverty. Additionally, in column (3), the coefficient of digital development decreases to −0.016 and remains significant at the 5% level. Although the direct effect persists, the reduction in magnitude supports the mediating role of human capital as a pathway through which digital development impacts energy poverty, thereby confirming H3. The robustness of this mediating effect is further validated using the Bootstrap method, with results reported in

Table 14. Test results of non-parametric bootstrap method for human capital level.

Effect	Coefficient	Standard Error	Z-Value	p-Value	95% Confidence Interval
Mediation effect	−0.009	0.003	−2.830	0.005	[−0.015, −0.003]
Direct effect	−0.016	0.008	−2.080	0.038	[−0.032, −0.001]

.

5.3. Technological Innovation Mechanism Test

This subsection employs a stepwise regression to examine the mediating role of technological innovation, with results presented in

Table 15. Test results of technological innovation mechanism.

Variables	(1) lnPoverty	(2) lnScitec	(3) lnPoverty
lnDigitization	−0.025 ***	0.245 ***	−0.014 *
	(0.008)	(0.049)	(0.007)
lnScitec			−0.046 ***
			(0.006)
Constant term	0.389	5.829 ***	0.655 ***
	(0.237)	(1.543)	(0.229)
Control variables	Yes	Yes	Yes
Regional fixed effects	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes
Observations	630	630	630
Adjusted R2	0.946	0.935	0.950

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

. Column (1) shows that the coefficient of digital development is −0.025, significant at the 1% level, indicating that digital development effectively mitigates regional energy poverty. In column (2), the coefficient of digital development is 0.245, significant at the 1% level, demonstrating its positive impact on promoting technological innovation. Column (3) reveals that technological innovation has a coefficient of −0.046, significant at the 1% level, suggesting that improvements in technological innovation contribute to reducing energy poverty. Additionally, in column (3), the coefficient of digital development decreases to −0.014 and is significant at the 10% level. Although the direct effect remains modestly significant, the reduction in magnitude supports the mediating role of technological innovation as a mechanism through which digital development influences energy poverty, thereby validating H4. The robustness of this mediating effect is further confirmed using the Bootstrap method, with results reported in

Table 16. Test results of non-parametric bootstrap method for technological innovation.

Effect	Coefficient	Standard Error	Z-Value	p-Value	95% Confidence Interval
Mediation effect	−0.011	0.003	−3.350	0.001	[−0.018, −0.005]
Direct effect	−0.014	0.009	−1.650	0.100	[−0.031, 0.003]

.

6. Empirical Test of Nonlinear Effects of Digital Development on Energy Poverty

6.1. Examining the Impact of Digital Development Disparities on the Alleviation Effect of Digital Development on Energy Poverty

6.1.1. Threshold Effect Test

Prior to the formal regression analysis, a threshold effect test is conducted to ascertain the presence and number of thresholds, thereby specifying the appropriate form of the threshold panel model for examining the impact of digital development on energy poverty. The results of this test are presented in

Table 17. Threshold effect test results.

	F	P	BS	Critical Value			Threshold Estimate	95% Confidence Interval
				1%	5%	10%
Single	63.640	0.036	1000	85.827	59.233	48.563	−4.016	[−4.037, −4.000]
Double	24.730	0.276	1000	77.273	50.498	38.658

.

Subsequently, a verification test is conducted to assess whether the estimated threshold aligns with the true value. As illustrated in Figure 5, the threshold estimate is consistent with the actual value.

6.1.2. Empirical Results Analysis

Table 18. Digital development threshold regression results.

Variables	(1)	(2)	(3)	(4)
lnDigitization	−0.026 ***	−0.025 ***
	(0.008)	(0.008)
lnDigitization (1)			−0.033 ***	−0.029 ***
(Digitization ≤ 0.018)			(0.007)	(0.007)
lnDigitization (2)			−0.019 ***	−0.019 ***
(Digitization > 0.018)			(0.007)	(0.007)
lnPerGDP		−0.439 ***		−0.393 ***
		(0.073)		(0.072)
lnEnepri		0.010 ***		0.007 ***
		(0.003)		(0.003)
lnUrbrur		0.260 ***		0.229 ***
		(0.043)		(0.042)
lnPopden		−0.024		−0.033
		(0.033)		(0.032)
Constant term	−0.413 ***	0.389	−0.413 ***	0.367
	(0.031)	(0.237)	(0.029)	(0.231)
Control variables	No	Yes	No	Yes
Regional fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Observations	630	630	630	630
Adjusted R2	0.936	0.946	0.942	0.948

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

reports the findings from the single-threshold panel regression. Columns (3) and (4) show that when digital development is below the threshold value of 0.018, the coefficients of digital development on energy poverty are −0.033 without control variables and −0.029 with control variables, both significant at the 1% level. A comparison across thresholds indicates that the impact of digital development above the threshold is smaller than that below it, suggesting that although digital development effectively mitigates energy poverty, its marginal alleviation effect diminishes at higher levels. These results confirm H5.

6.2. Impact Test of Energy Poverty Disparities on the Effect of Digital Development in Alleviating Energy Poverty

Given the considerable regional disparities in energy poverty, the alleviating effects of digital development may vary across different levels of energy deprivation. To capture this potential heterogeneity, this study employs a quantile regression model to examine the nonlinear impact of digital development on energy poverty.

Table 19. Panel quantile regression results of the impact of digital development on energy poverty.

Variables	(1)	(2)	(3)	(4)	(5)
	QR_10	QR_25	QR_50	QR_75	QR_90
lnDigitization	−0.011	−0.013 ***	−0.017 **	−0.023 ***	−0.025 ***
	(0.011)	(0.004)	(0.007)	(0.008)	(0.007)
lnPerGDP	−0.269 **	−0.154 ***	−0.275 ***	−0.195 **	−0.212 ***
	(0.108)	(0.040)	(0.071)	(0.079)	(0.063)
lnEnepri	0.004	0.003 *	0.005 *	0.004	0.007 ***
	(0.004)	(0.001)	(0.003)	(0.003)	(0.002)
lnUrbrur	0.171 ***	0.158 ***	0.151 ***	0.125 ***	0.046
	(0.064)	(0.023)	(0.042)	(0.047)	(0.037)
lnPopden	−0.066	0.022	0.072 **	−0.131 ***	−0.153 ***
	(0.049)	(0.018)	(0.032)	(0.036)	(0.029)
Constant term	0.391	−0.597 ***	−0.620 **	0.727 **	1.048 ***
	(0.425)	(0.156)	(0.277)	(0.311)	(0.249)
Regional fixed effects	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	630	630	630	630	630

Note: Standard errors are in parentheses (); *** indicates p < 0.01, ** indicates p < 0.05, * indicates p < 0.1.

presents the results of the panel quantile regression. The findings indicate that the magnitude of the regression coefficient for digital development differs across quantiles. Specifically, as the quantile level increases, the absolute value of the coefficient rises, demonstrating heterogeneity in the poverty alleviation effects of digital development. In regions experiencing higher levels of energy poverty, digital development exerts a stronger mitigating impact, whereas its effect is comparatively weaker in regions with lower energy poverty. These results suggest that the alleviating influence of digital development intensifies with the severity of energy poverty, thereby confirming H6.

7. Conclusions and Policy Implications

Using panel data from 30 Chinese provinces spanning 2003 to 2023, this study systematically investigates the effects and heterogeneity of digital development on energy poverty, while also exploring the underlying mechanisms and nonlinear characteristics. The key findings are as follows: digital development significantly mitigates energy poverty, a conclusion robustly supported by endogenous analyses and multiple robustness checks. The poverty reduction effects of digital development vary markedly across economic regions, geographic locations, and energy endowment conditions. In particular, the alleviating impact is substantially stronger in the central and western regions compared to the eastern region, and more pronounced in the northern regions relative to the southern regions, highlighting a higher marginal effect in areas with weaker energy infrastructure and greater reliance on traditional energy sources. Moreover, the poverty alleviation effect is more pronounced in regions with disadvantaged energy endowments, suggesting that resource-poor areas can achieve greater marginal benefits by overcoming constraints on energy accessibility through digital development. Digital development indirectly enhances households’ energy affordability and system efficiency by fostering non-agricultural employment, improving human capital, and stimulating technological innovation, thereby contributing to reductions in energy poverty. Additionally, the influence of digital development exhibits threshold effects and quantile heterogeneity: the alleviating effect is stronger when digital development is below the threshold, but marginal benefits decline once the threshold is surpassed, with the impact being more significant in regions experiencing higher levels of energy poverty.

Drawing on the foregoing findings, this study advances the following policy recommendations:

We should focus on promoting balanced and coordinated digital development in different regions and address the technological gap between the east and the west. In light of the significant regional disparities in digital development, especially the prominent gap between the east and the west and the insufficient technology spillover, policies should target structural imbalances. The top priority is to implement precise regional digital infrastructure investment strategies. In the western regions, especially those with concentrated energy poverty, priority should be given to deploying high-speed broadband networks, cloud computing centers, and IoT nodes, among other new digital infrastructures, and to providing special fiscal transfer payments and tax reduction policies to significantly lower their access and usage costs. A long-term mechanism for targeted technical support and talent exchange from the east to the west should be established. We should encourage eastern universities, research institutions, and high-tech enterprises to set up branches or joint laboratories in the west to promote the deep integration of advanced digital technologies and management models with the actual needs of the west. At the same time, the “bridge” function of the central regions should be strengthened. We should support central provinces in building high-level digital industry platforms, optimizing the business environment, and attracting the gradient transfer of digital industries to form a digital development corridor that links the east and the west, effectively curbing the further expansion of regional disparities and ultimately achieving a convergent development of the national digital level.

We should build a precisely adapted regional differentiated digital empowerment system to maximize the marginal benefits of energy poverty alleviation. Given the significant regional heterogeneity and energy endowment heterogeneity in the alleviation effect of digital development on energy poverty, as well as the law of diminishing marginal returns, policy design must abandon the “one-size-fits-all” thinking approach. In the developed eastern regions, the policy focus should shift to deepening the integration of digital technologies with existing mature energy systems, with a focus on supporting high-end applications such as smart energy management platforms, demand-side response systems, and distributed energy smart microgrids to promote the refined improvement of energy efficiency. In the northern regions, digital means should be used to upgrade traditional energy transmission and distribution networks, develop intelligent grid condition monitoring and optimal dispatching systems, and enhance the resilience and reliability of energy supply. In the southern regions, especially rural areas, digital technology-based household clean energy solutions should be vigorously promoted, such as solar IoT monitoring and biomass energy intelligent management platforms, to improve the accessibility and cleanliness of energy services. For regions with energy endowment advantages, policies should guide them to go beyond simple resource dependence and use big data, artificial intelligence, blockchain, and other technologies to reconstruct the energy value chain and develop new models such as energy internet, virtual power plants, and digital carbon asset management to explore new growth points. It is particularly crucial to establish a dynamic monitoring and evaluation system for energy poverty and digital development levels and prioritize the allocation of limited digital resources to regions with deep energy poverty and low digital development levels to ensure the precision of policy intervention and maximize marginal benefits.