跳到正文
原文
Nature 健康护理主题·· 9 天前AI 评分31

COVID-19 对中国家庭能源贫困和清洁能源转型的影响

COVID-19 impact on households’ energy poverty and clean energy transition in China

AI 导读

研究发现,COVID-19 疫情改变了中国家庭的能源转型轨迹。自 2014 年以来,现代能源使用比例逐步增加,但疫情导致这一趋势中断。研究通过空间和时间维度评估了能源贫困的发生率、深度和强度,并估计了多维能源贫困指数,揭示了不同收入群体和城乡家庭在能源选择和不平等上的变化。

正文

Introduction

Energy poverty is defined as the lack of access to affordable, reliable, and modern energy services1. This includes (a) the inability to afford clean cooking solutions, leading to reliance on harmful fuels like coal and wood, (b) the lack of access to electricity or reliable heating, which can affect education2, health3,4, and wellbeing5,6, and (c) high energy expenditures that consume a disproportionate share of a household’s income. Energy poverty remains a pressing global challenge, with nearly 1.2 billion people lacking access to electricity in 20207, a situation further worsened by the COVID-19 pandemic. Lockdown measures sharply reduced commercial and industrial energy demand while increasing residential consumption, straining household budgets—especially for low-income families whose energy bills rose as incomes fell. Energy suppliers also faced reduced supply capacity and liquidity problems from delayed or unpaid bills, further threatening service continuity. While wealthier households could absorb higher energy costs, vulnerable groups—such as the unemployed, elderly, less educated, low-income, rural residents, and minorities—often lost access to basic energy services due to rising prices8,9. This disruption deepened inequalities, as energy inaccessibility compounded other socio-economic hardships, creating a vicious cycle where reduced income, higher household energy needs, and supply instability reinforced one another. Empirical evidence further confirms the growing disparity in energy poverty both between advantaged and vulnerable groups10 and across EU member states11,12. Anti-pandemic measures adopted by the Global North and South countries are markedly different, yet all exhibited, to varying degrees, a lack of equity — for instance, support was provided only to certain groups of consumers or limited to specific fuels13. Additionally, different measures lead to varying effects. For example, in Spain, the impact of social measures (e.g., the electrical social bonus) was far less effective than that of unemployment aids14. In Ghana, although the electricity-transfer programme provided rapid relief, it was regressive, with low-consuming households, tenants paying through landlords, and households sharing meters receiving less support15.

As one of the countries where energy poverty is most pronounced, China has implemented numerous policies and regulatory actions to address energy poverty and transition towards sustainable and clean energy (Fig. 1). Since the 1990s, China has implemented a series of targeted policies and programs that have significantly alleviated household energy poverty, particularly in rural and underserved areas. Initiatives such as the “Household Biogas Program”16 and the “Clean Heating Plan in Northern Region”17 have expanded access to modern energy services and improved the affordability and reliability of household energy. These efforts, alongside rapid urbanization and income growth, have contributed to a notable decline in energy poverty levels nationwide. However, various forms of shocks—including climate-related events (e.g., extreme heatwaves, cold spells, and floods) and economic disruptions (e.g., job loss and rising energy prices)—continued to disproportionately affect energy-poor households. Studies have documented that such shocks often exacerbate energy insecurity by increasing household energy uses or limiting access to critical energy services, thereby undermining the resilience of vulnerable communities18,19. These dynamics highlight the importance of designing adaptive and inclusive energy policies that account for multi-dimensional vulnerabilities and the compounding effects of external shocks.

Fig. 1: Timeline of China’s response to energy poverty, where a chronological overview of key milestones is presented from 1990s to 2030.

Milestones cover rural electrification, renewable energy development, and inclusive energy services. The grey arrow indicates chronological progression, and grey markers denote years or planning periods. Red headings identify policies, with accompanying text summarizing their relevance to energy poverty.

However, in-depth analyses of the impacts of the COVID-19 pandemic on energy poverty in China remained scant, particularly the changes before and after the outbreak and the underlying reasons received little attention. To fill this gap, we first present the energy transition during 2014–2022 and compare the changes before and after COVID-19. Secondly, we examine how urban and rural households’ energy consumption has further changed in the post-pandemic era. Thirdly, we assess energy poverty at the spatial-temporal levels for the three indices of energy poverty incidence, depth, and intensity. Lastly, we estimate the multidimensional energy poverty index, evaluate the local government’s anti-pandemic policy effect on Chinese households’ energy poverty, and identify its driving forces (Fig. 2).

Fig. 2: The research framework for examining the impact of COVID-19 on Chinese household energy poverty and clean energy transition.

The framework links COVID-19 to fuel choices and energy inequalities across income groups and urban–rural settings. Household financial, housing, and energy-use indicators inform a multidimensional energy poverty index, assessed through incidence, depth, and intensity from a spatial–temporal perspective. Arrows indicate analytical links between components.

Results

COVID-19 altered China’s trajectory of household energy transition

Since 2014, overall household energy consumption trends demonstrated steady improvement with the adoption of modern energy sources. Available panel data showed a gradual increase in electricity and clean fuel usage, with corresponding declines in reliance on traditional biomass (Fig. 3). However, the pandemic in 2019 disrupted this trend. Our Sankey diagram indicated that although the traditional fuels of coal and wood/crop residues use continued to decline, the pace slightly slowed after the pandemic in 2020, by comparing their declining rate of 15.56% during 2016–2018 and 13.39% during 2020–2022. The share of households using electricity for cooking declined since 2020 and returned to 2014 levels by 2022. Notably, before and after the pandemic (in 2018 and 2020, respectively), the proportions of households that shifted from using electricity for cooking to using coal and wood/crop residues were the same, accounting for 0.5% and 1.9% of the total sample, respectively; yet the proportion that turned to using liquefied petroleum gas (LPG) and natural gas (NG) decreased from 7.1% to 6.1%. Considering that the proportion of ‘electricity-to-gas’ transitions was increasing year by year before the pandemic, the clean energy transition was affected. In addition, although the share of households using LPG/NG continued to increase, its growth slowed after 2020: the share increased by 7.0 percentage points from 47.1% in 2018 to 54.1% in 2020, but by only 4.1 percentage points to 58.2% between 2020 and 2022. The corresponding residential energy transitions for urban and rural households are presented in Figs. S1 and S2 (see Supplementary information).

Fig. 3: Residential energy transitions, 2014–2022.

The width of each flow in the Sankey diagram represents the proportion of households using a specific energy type for cooking, providing a clear visualization of energy transitions in China before (2014, 2016, and 2018) and after COVID-19 (2020 and 2022). The figure in the brackets denotes the percentage of household flow from electricity use to electricity/other types of energy use. (1) We used data from 6,114 households that participated in all five waves of the survey. The numbers of rural and urban households in each wave were as follows: 3343 and 2771 in 2014; 3242 and 2872 in 2016; 3146 and 2968 in 2018; 3138 and 2976 in 2020; and 3135 and 2979 in 2022, respectively. (2) Traditional fuels refer to the combined use of coal, wood, and crop residues. Their combined share decreased from 34.7% in 2016 to 29.3% in 2018, representing a decline of 5.4 percentage points, or 15.56% relative to the 2016 level. It subsequently decreased from 23.9% in 2020 to 20.7% in 2022, equivalent to 3.2 percentage points, or a relative decline of 13.39%. (3) Urban–rural status was determined separately in each survey wave using the contemporaneous CFPS classification. Households whose recorded status changed from rural to urban were classified as urban households in subsequent waves, contributing to the gradual decline in the rural subsample.

Household energy-burden inequality declined before the pandemic but reversed after the outbreak in 2020 and widened further by 2022, with a marked rural–urban divergence (Fig. 4). Nationally, the weighted household Gini coefficient fell from 0.592 in 2014 to 0.558 in 2018, before rising to 0.596 after the outbreak in 2020 and 0.622 in 2022. This reversal was more pronounced among rural households, where the Gini coefficient decreased from 0.617 in 2014 to 0.578 in 2018 and then increased to 0.605 in 2020 and 0.642 in 2022, whereas urban inequality remained comparatively stable, changing from 0.565 in 2020 to 0.554 in 2022. The Lorenz-type curves show the same pattern, with the 2022 distribution shifting further from the line of equality nationally and especially in rural areas, while the urban distribution changed little. Matched-household bootstrap comparisons further showed that the 2020–2022 increase was statistically detectable nationally and among rural households, but not among urban households. These results indicate that the post-pandemic widening of household energy-burden inequality was concentrated primarily in rural China, highlighting greater distributional vulnerability among rural households during the pandemic period20.

Fig. 4: Temporal changes in household energy-burden inequality in China, 2014–2022.

Survey-weighted conventional Lorenz-type curves of household energy burden for national (a), rural (b) and urban (c) households in 2014, 2016, 2018, 2020 and 2022, respectively, by referring to Wang et al. 53 and Jiang et al. 41. Household energy burden is defined as annual household energy expenditure divided by annual household income41,42,54. Households are ranked from lowest to highest energy burden; the dashed diagonal indicates perfect equality37,43,55. d–f, Survey-weighted Gini coefficients of household energy burden for national (d), rural (e) and urban (f) households over the same period. Points show weighted Gini estimates, and error bars indicate 95% bootstrap confidence intervals. CFPS cross-sectional household survey weights are applied throughout. To limit the influence of extreme ratios associated with very low reported incomes, energy burden is winsorized at the 99th percentile within each survey wave. Paired-household bootstrap comparisons between 2020 and 2022 show statistically detectable increases in energy-burden inequality nationally (ΔG = 0.026, 95% CI 0.008–0.045) and among rural households (ΔG = 0.033, 95% CI 0.011–0.055), whereas the change among urban households is not statistically distinguishable from zero (ΔG = −0.011, 95% CI−0.043–0.021).

Energy expenditures grew steadily across all income groups in the pre-pandemic period, with the national average increasing from RMB2,464 in 2014 to RMB2,886 in 2018, reflecting an increase rate of nearly 17% (Fig. 5). Meanwhile, energy expenditures for both urban and rural households continued to rise, with urban households consistently spending more, highlighting the persistent urban–rural divide in energy consumption and access to modern energy services. This is consistent with the findings that Chinese urban households generally have higher energy needs due to more energy-intensive lifestyles21. In contrast, the post–2020 period marked a significant shift in energy cost growth. The national average energy expenditure in 2022 had surged to RMB3,729, a nearly 30% increase compared to 2018, where urban and rural households increased 22% and 38%, respectively, during the same period. In particular, the energy burden for rural households has increased (steeper growth) at an alarmingly rapid pace.

Fig. 5: Household energy expenditure by urban–rural divide for multiple income groups.

The vertical bar plots represent energy costs (in RMB per year) for each income group across the years, with individual data points marked by colored dots. A thick grey trendline, with open-circle markers, represents the annual average. Data used in this graph are the same as Fig. 3.

Assessing China’s energy poverty

We quantified three common indices to measure energy poverty, namely energy poverty incidence, depth, and intensity. All three indices declined between 2014 and 2022, but their trajectories differed (Fig. 6a). The reduction in energy poverty incidence had already begun to moderate before the pandemic, whereas the reductions in depth and intensity remained at approximately 28–30% through 2020 before slowing sharply to 13.96% and 14.03%, respectively, during 2020–2022. The sustained pre-2020 improvements were consistent with China’s poverty alleviation efforts, which achieved “a significant reduction in energy poverty.”22 The subsequent slowdown was therefore concentrated in energy poverty depth and intensity rather than shared equally across all three indices.

Fig. 6: a Temporal variations and b spatial heterogeneity of Chinese household energy poverty during 2014–2022.

a Absolute values and growth rates are illustrated as the bars and broken lines, respectively. Year 2020 is the pandemic period. Growth rates reported for each survey wave represent changes relative to the preceding wave; for example, the 2022 growth rates refer to changes between 2020 and 2022. b Each province’s vertical three-bar indicator highlights a different facet of energy poverty. The relatively low estimates in Xinjiang and Ningxia should be interpreted cautiously because of their small provincial samples, while energy-resource endowment, supply infrastructure, and higher heating-related expenditure may also partly explain the pattern; however, these mechanisms cannot be separately identified using the current data. Tibet and Qinghai do not have enough data for estimation. The map was created by the authors using Python with administrative boundary data from Natural Earth (1:10 m Admin 1 – States, Provinces; public domain). Data and calculation methods can be found in the Materials and Methods section.

Regionally, by comparison, a clear spatial pattern emerged: western and certain central provinces consistently exhibited higher incidence of energy poverty alongside greater depth of deprivation, resulting in markedly higher intensity values, whereas the more developed eastern coastal provinces showed lower incidence and shallower energy poverty depth (Fig. 6b). This finding is in line with Jia and Wu23, who have argued that China’s spatial distribution of energy poverty could be generally characterized as “high in the western and central regions and low in the eastern regions”. For example, southwestern provinces such as Guizhou and Yunnan (together with northwestern Gansu and a central province like Henan) rank among the most energy-poor regions, indicating both widespread and acute energy deficits, while affluent eastern provinces like Zhejiang and Shanghai have among the lowest energy poverty levels with only minimal intensity.

Identifying the driving forces of Chinese households’ energy poverty

We estimated the pandemic effect on our multidimensional energy poverty index using a Difference-in-Difference (DiD) design (Model 1; N = 10,857)) and two machine learning-based Difference-in-Difference (ML-DiD) approaches (Models 2 and 3; N = 12,027). The DiD method, a quasi-experimental approach, estimates causal effects by comparing changes in energy poverty over time between a treatment group (i.e., more exposed to the pandemic) and a ‘control’ group (i.e., less exposed). Because the pandemic affected all regions, the treatment vs. control is divided based on the median mortality rate, and the control is not a blank control per se (see Materials and Methods in detail). The ML-DiD method enhances this by using machine learning algorithms to flexibly estimate counterfactual outcomes and control for high-dimensional covariates. Our results generated significant and positive signs, illustrating that Chinese households that have been more severely impacted by the pandemic faced a higher incidence of energy poverty (Fig. 7). Among the control variables, income, family size, and electricity expenditure are negatively associated with energy poverty, while reliance on firewood or coal exhibited a large positive coefficient. These findings suggested that advantaged households—those with higher incomes, larger family sizes, and greater electricity consumption—were less likely to fall into energy poverty even under the shock of the pandemic. In contrast, marginalized households that rely on traditional fuels were more vulnerable and prone to deeper energy deprivation during the crisis. These findings are consistent with the results mentioned above. In addition, the PSM-DiD estimates and the placebo test (see Supplementary Information) further justified the robustness of these findings.

Fig. 7: COVID-19 impact on household energy poverty.

(1) Household and year effects are fixed, ***p < 0.01, **p < 0.05, *p < 0.1; (2) We use the conventional DiD (Model 1; N = 10,857) and two machine learning techniques (Lasso Regression in Model 2 and Elastic Net in Model 3; N = 12,027) to examine the COVID-19 impact on household energy poverty; (3) The regularization of control variables by machine learning algorithms may shrink some coefficients to zero, so the regression results of control variables are typically not reported; (4) The estimation result was obtained by taking the median of the adjusted results from 101 repeated samplings; (5) The PSM-DiD estimates in the Supplementary Information further support the findings that the pandemic shock had significant effects on household energy poverty; (6) The placebo test presented in the Supplementary Information eliminated the potential “saturation effect” of pre-pandemic poverty alleviation policies and illustrates the robustness of our regressions.

Discussion

The COVID-19 pandemic exposed and deepened existing patterns of energy inequality in China, defined as the unequal distribution of energy access across socio-economic and geographic lines24,25. Our results showed that energy-deprived families and rural low-income households, already disadvantaged, were disproportionately affected. As incomes contracted and energy costs rose, rural households may have become increasingly trapped in a cycle where energy access fell below socially acceptable thresholds26 In this cycle, limited energy use undermines health, education, and economic resilience, further constraining future incomes and perpetuating deprivation. The pandemic thus did not merely disrupt energy security but entrenched structural inequalities, highlighting the need for integrated policies that address both income instability and energy affordability in vulnerable communities.

The widening urban-rural divide in energy expenditure confirmed that, although urban households consistently spent more on energy in absolute terms, rural households experienced a significantly faster growth rate after the pandemic. Two underlying factors may have contributed to this trend. First, the surge in rural energy demand partly reflected increased time spent at home during pandemic lockdowns, as work-from-home and remote education became more widespread. Second, rural areas faced a more pronounced rise in residential energy use due to stricter travel restrictions and the large-scale return of ‘urban’ workers to their rural hometowns. This shift led to higher energy consumption for essential needs such as cooking, heating, and lighting, thereby accelerating the growth of rural household energy expenditures relative to their urban counterparts.

The deceleration in the growth rates of energy poverty incidence, depth, and intensity following the COVID-19 outbreak suggested a partial reversal in China’s previous progress toward alleviating household energy poverty since 2014. This trend was especially pronounced in the western and certain central regions, which exhibited higher vulnerability compared to the more developed eastern provinces. Multiple lines of evidence supported the view that the pandemic undermined China’s trajectory of poverty reduction. COVID-19 precipitated the country’s slowest economic growth in decades and sharply suppressed household incomes. The World Bank estimated that, absent additional interventions, 8 to 20 million fewer Chinese individuals would escape poverty in 2020 relative to pre-pandemic projections, highlighting how the crisis disrupted poverty alleviation efforts27. These widespread income shocks likely translated into diminished ability to afford adequate energy services, particularly for already vulnerable groups. Indeed, studies have noted that the pandemic “represents an unprecedented challenge for low-income populations, especially those that are already energy insecure,”28 as household earnings declined while residential energy needs increased due to extended time spent at home during lockdowns.

Globally, 2022 marked the first increase in the number of people lacking access to electricity in over a decade, reflecting the wider disruption wrought by COVID-1929. Within China, the pandemic exposed and intensified structural regional disparities in energy poverty. Western and certain central provinces—characterized by lower income levels, weaker energy infrastructure, and higher dependence on traditional fuels—experienced a sharper deterioration compared to eastern regions, where more robust economic foundations and better-developed energy systems offered greater resilience. These enduring regional imbalances underscored the need for differentiated policy responses. High levels of energy poverty in western and central regions should be prioritized through accelerated investments in clean and decentralized energy infrastructure, targeted subsidies and financial support mechanisms to enhance household energy affordability, and programs that promote the adoption of modern and efficient fuels22,30,31,32. Tailoring interventions to local economic and infrastructural conditions is critical not only for addressing acute energy poverty deficits but also for closing the persistent urban-rural and regional gaps that have been exacerbated by the pandemic shock.

The consistently positive and statistically significant treatment effects across all models suggest that the pandemic intensified multidimensional energy poverty for vulnerable populations. Households with structural advantages, such as higher income, larger family size, and higher electricity consumption, appeared to have maintained their resilience during the crisis. These groups likely benefited from diversified energy access and better economic buffers, shielding them from downward mobility in energy conditions. In contrast, households who are dependent on traditional fuels like firewood or coal faced disproportionate energy poverty. In particular, the strong negative associations of income and electricity expenditure with energy poverty pointed to the dual importance of both financial capacity and modern energy access in mitigating economic deprivation. The use of firewood or coal, by contrast, is not merely a symptom of poverty but a driver of continued energy exclusion. These patterns highlighted the multidimensional nature of energy poverty—where fuel choice, household composition, and consumption behavior interacted with exogenous shocks to shape outcomes.

Policy responses should focus on preventing socioeconomic shocks from reversing previous progress in energy poverty alleviation. The slower decline in energy poverty after 2018, together with the deceleration of LPG/NG adoption during 2020–2022, indicates that households close to the deprivation threshold remain particularly vulnerable, consistent with evidence that external energy shocks disproportionately increase the burden on low-income households33. Energy poverty monitoring should therefore be incorporated into existing poverty-alleviation systems to identify households at risk of falling back into deprivation and maintain their access to modern household energy during disruptions.

The provincial disparities call for two complementary policy packages. A household-oriented package should combine vulnerability monitoring, targeted financial support, energy-efficiency retrofits, and clean-energy appliances, particularly in high-burden provinces such as Henan. A place-based package should improve last-mile infrastructure and align household energy transitions with local resources: Guizhou could link its hydropower endowment with grid improvements and affordable electric cooking34, while Gansu could combine wind and solar resources with distributed generation, storage, and clean heating35,36. These packages should be combined at the county level according to household affordability, infrastructure reliability, and resource availability37.

China’s experience provides a broader lesson for the Global South: improvements in clean energy use may remain fragile because sustained LPG consumption depends on household income, fuel availability, and reliable supply systems38. Achieving SDG7 therefore requires attention not only to expanding modern energy access but also to the resilience and equitable distribution of these gains.

Methods

Data

The China Family Panel Studies (CFPS)39 is the most comprehensive large-scale multidisciplinary social survey launched by Peking University, China, and followed up every two years since 2010. By tracking individuals, families, and communities over time, the CFPS provides high-quality longitudinal data for academic investigation and policy research. Data employed in this study are mainly extracted from the family economics surveys section and include household income, expenditure, and assets. The CFPS dataset has found widespread application in several key studies37,40.

Notably, considering the questionnaires have considerable changes between 2010 and 2014, this study selected data after 2014 to ensure consistency, namely the five survey years of 2014, 2016, 2018, 2020, and 2022. This enabled us to compare the energy poverty changes before and after the COVID-19 pandemic. 12,027 valid household values were finally selected, including up to 7046 urban households’ values and 4981 rural households’ values; 1846 values in 2014, 2527 values in 2016, 2875 values in 2018, 2228 values in 2020, and 2551 values in 2022. Table S1 presented the variables employed, their descriptions (including sourced questions), and their characteristics. Potential confounding from energy-price fluctuations was limited because residential energy prices in China are strictly regulated, while common temporal shocks from international and domestic energy markets were absorbed by the year fixed effects.

Calculating household energy burden, Lorenz-type curves, and Gini coefficients

Following established household-energy affordability measures, this research defined household energy burden as the share of annual household income devoted to residential energy expenditure:

$${{EB}}_{{it}}=\left({E}_{{it}},|,{Y}_{{it}}\right)\times 100 \%$$

(1)

where \({E}_{it}\) denotes annual household expenditure on electricity, household fuels, and central heating, and \({Y}_{it}\) denotes annual household income. This income-normalized definition separates affordability pressure from differences in absolute energy spending and follows recent household energy-burden and energy-inequity studies37,41,42. Annual energy expenditure was constructed as 12 times monthly expenditure on electricity and household fuels plus annual central-heating expenditure, consistent with the CFPS-based household energy-cost construction used by Wang et al. 37.

To quantify distributional inequality, this research used the Lorenz–Gini framework. Jiang et al. 41 explicitly distinguish the Lorenz curve and Gini coefficient, which measure the overall dispersion of an indicator after households are ranked by that indicator, from income-ranked concentration measures. Accordingly, households were ranked from lowest to highest energy burden within each survey wave and separately for the national, rural, and urban samples. This approach is consistent with established applications of Lorenz curves and Gini coefficients to household energy and energy-related distributions41,43. Let \({w}_{i}\) denote the CFPS cross-sectional survey weight and EBᵢ household energy burden. The survey-weighted Lorenz-type curve plots the cumulative household share \({P}_{k}={\sum }_{i\le k}{w}_{i}/{\sum }_{i}{w}_{i}\) against the corresponding cumulative energy-burden share \({L}_{k}={\sum }_{i\le k}{w}_{i}{{EB}}_{i}/{\sum }_{i}{w}_{i}{{EB}}_{i}\). The weighted Gini coefficient was calculated from the area under the Lorenz curve as \(G=1-2{\int }_{0}^{1}L\left(p\right){dp}\), where larger values indicate greater inequality and the 45° diagonal represents perfect equality.

CFPS wave-specific cross-sectional household survey weights were applied throughout the Lorenz-type curve and Gini calculations. Statistical uncertainty was quantified using 1200 household-level nonparametric bootstrap resamples while retaining the survey weights within each resample, and percentile-based 95% confidence intervals were reported for the weighted Gini coefficients. For the 2020–2022 matched-household comparison, households observed in both waves were linked using the CFPS panel identifier. The pair-level weight was defined as the arithmetic mean of the household’s 2020 and 2022 cross-sectional weights, and 2000 paired bootstrap resamples were used to obtain the 95% confidence interval for \(\triangle G={G}_{2022}-{G}_{2020}\).

Estimating the thresholds of household energy poverty

Kez et al. 44 classified energy poverty measurement methods as consensus-based, expenditure-based, and direct measurement. We chose to use the expenditure-based method to estimate China’s household energy poverty thresholds because of its objectivity and quantifiable nature, and the energy-related variables employed are mostly expenditure data. This study defined the basic energy demand as the annual total household expenditures on electricity, fuel, and heat, respectively:

$${{ED}}_{i}={Bil}{l}_{e}+{Bil}{l}_{f}+{Bil}{l}_{h}$$

(2)

Referring to Wang and Lin45, we divided the overall sample into ten income quantiles and regressed the sub-samples according to the following Eq. (3). The energy poverty threshold is measured by using the income group where significant results emerged for the first time from low to high, by considering all the household samples across the research years. By doing so, the threshold is defined as the average annual energy expenditure of households in that income group. Households’ expenditure lower than the threshold is defined as energy poverty. The formula that estimated the threshold is shown below:

$${{ED}}_{i}=\alpha +{\sum}_{k}{\beta }_{k}{Z}_{{ik}}+{\sum}_{j=2}^{10}{\gamma }_{j}{{Income}}_{{{decile}}_{{ij}}}+{\varepsilon }_{i}$$

(3)

where \({{Income}}_{{{decile}}_{{ij}}}\) is the 0/1 dummy variable of income deciles; \({Z}_{{ik}}\) is a vector of control variables comprising k types of household characteristics, include family size (FS), dwelling scales (DS), expense for the durable goods (Exp_dg), and whether to use solid fuel for cooking (Cook_sf). \({\varepsilon }_{i}\) is the unobserved random error. Results and explanations are presented in Table S2. We estimated the electricity poverty threshold to be RMB1,336.068 per year per household.

Assessing Chinese households’ energy poverty incidence, depth, and intensity

We quantified three common indices to measure energy poverty. Energy poverty incidence reflected the proportion of households lacking access to adequate modern energy. Energy poverty depth denoted the average degree of energy deprivation among those energy-poor households. Energy poverty intensity captured overall severity by incorporating both how widespread and how deep the deprivation was. These three indicators are interrelated: incidence shows the breadth of the problem, depth reveals its intensity among the affected, and intensity synthesizes both dimensions into a single measure. Studying all three is essential, as focusing on only one could mask important aspects of the problem—for example, a low incidence rate could coincide with severe deprivation among the affected, or a high incidence rate might involve only mild deprivation. Together, they provide a more comprehensive and nuanced understanding of energy poverty, enabling more targeted and effective policy interventions.

This study modified the Foster-Greer-Thorbecke metrics proposed by Foster et al. 46 and proposed the formula below:

$${P}_{a}\left(x,z\right)=\frac{1}{n}{\sum }_{i=1}^{q}{\left(\frac{z-{x}_{i}}{z}\right)}^{a}$$

(4)

where \(n\) denotes total households, \(q\) denotes energy-poor households, \(z\) denotes energy poverty threshold, and \({x}_{i}\) denotes annual energy expenditure for household \(i\). The parameter \(a\) assumes the values 0, 1, and 2. When a = 0, \({P}_{0}\) denotes the incidence rate of energy poverty; when a = 1, \({P}_{1}\) represents the depth of energy poverty, signifying the gap of the energy expenditure of energy-poor households compared to the threshold; when a = 2, \({P}_{2}\) represents the intensity of energy poverty, indicating the degree of imbalance in energy consumption among energy-poor households.

Identifying the driving factors of households’ energy poverty

We estimated the multidimensional energy poverty index (MEPI) by referring to Wang and Lin45: Firstly, we chose five dimensions, including clean resource, energy service, availability, dwelling condition, basic demand, and burden of energy consumption, then assigned weights to each dimension respectively. Secondly, we set the cutoff criterion for deprivation of the indicators. Lastly, we estimated the MEPI for each household, and a higher level of MEPI meant a more deprived status (see Table S3). Below is the calculation formula of MEPI:

$${{MEPI}}_{i,t}={\sum}_{j=1}^{d}{w}_{j,t}{g}_{i,j,t}$$

(5)

where \({{MEPI}}_{i,t}\) denotes multidimensional energy poverty index in household i at year t; \({w}_{j,t}\) denotes the weight of indicator \(j\) at year t; and \({g}_{i,j,t}\) values 1 if the household i belongs to the energy poor group in indicator \(j\) at year t. Therefore, a higher level of \({{MEPI}}_{i,t}\) means a more deprived status.

In our regressions, MEPI was treated as the dependent variable. We employed both the conventional Difference-in-Difference (DiD) and the Machine Learning-based Difference-in-Difference (ML-DiD) method47 by using Stata 18 and Python 3.13.3 to explore the COVID-19 impact on Chinese household energy poverty. The conventional DiD model compares changes in outcomes over time between treated and control groups. The basic specification in this study is:

$${{MEPI}}_{i,t}=\alpha +\delta {{treat}}_{i}\times {{post}}_{t}+\varphi {X}_{i,t}+{\mu }_{i}+{\gamma }_{t}+{\varepsilon }_{i,t}$$

(6)

where \({{MEPI}}_{i,t}\) is multidimensional energy poverty index for household i at year t; \(\delta\) captures the average treatment effect on the treated; \({{treat}}_{i}\times {{post}}_{t}\) is the interaction term indicating treatment exposure after the COVID-19 impact; \({X}_{i,t}\) denotes a vector of control variables; \({\mu }_{i}\) and \({\gamma }_{t}\) denote household and year fixed effects; and \({\varepsilon }_{i,t}\) denotes the error term.

The core of the DiD model consists of the division of the experimental group and control group, as well as the determination of the exogenous shock. Since our topic is to explore the COVID-19 impact on household energy poverty, we selected COVID-19 as the exogenous shock. The \({{post}}_{t}\) variable therefore takes 0 in 2014, 2016, and 2018, and 1 in 2020 and 2022. Considering that the pandemic-induced deaths48 can truly reflect the degree of impact suffered by the given region (provinces with a high mortality rate indicate a greater pandemic impact, and vice versa)49, we adopted the median mortality rate as the dividing criterion since no region is unaffected by the pandemic: provinces with mortality rates above the median were defined as the higher-severity group, while those below the median constituted the lower-severity comparison group. Similar median-based classifications have been used to examine heterogeneous responses to pandemic-related shocks. Nguyen et al. 50 divided US states into above- and below-median baseline COVID-19 mortality groups to estimate heterogeneous responses to reopening policies, while Brzezinski et al. 51 divided US counties by median science skepticism in event-study and staggered DiD analyses of responses to shelter-in-place policies. This choice is primarily based on the fact that mortality is one of the most direct and objective outcome measures of pandemic severity. During the early stage of the outbreak, it effectively reflected the level of pandemic pressure experienced by each province, which could, in turn, affect household electricity consumption and other energy uses. Moreover, the primary aim of this study is to estimate the causal effects of differences in pandemic shock severity on residential energy use, rather than to evaluate the independent effects of specific containment policies. Therefore, distinguishing between high- and low-impact provinces based on mortality rates is appropriate within the analytical framework of this study.

In the ML-DiD framework, we relaxed parametric assumptions and allowed more flexibility in modeling the outcome variable by replacing or augmenting the outcome model with machine learning predictions. The goal is to improve control for observed covariates and to reduce model misspecification. The ML-DiD approach is advantageous to traditional DiD in the following two aspects: on the one side, it avoids the prerequisite assumption of linear relationships among variables and permits the existence of nonlinear relationships; on the other side, this approach can handle a greater number of control variables, thereby enabling a more accurate estimation of causal relationships (40). Chernozhukov et al. 52 have demonstrated that the coefficients estimated by ML-DiD are unbiased. The specification is as follows:

$${{MEPI}}_{i,t}=\alpha +\delta {{treat}}_{i}\times {{post}}_{t}+g({X}_{i,t})+{\mu }_{i}+{\gamma }_{t}+{\varepsilon }_{i,t}$$

(7)

$$E\left({\varepsilon }_{i,t}|{{treat}}_{i}\times {{post}}_{t},{X}_{i,t}\right)=0$$

(8)

where \(g({X}_{i,t})\) denotes the operations of two selected machine learning techniques–namely Lasso Regression and Elastic Net. Additionally, to enhance the stability and reliability of the model estimation, we employed a 5-fold cross-validation fitting method for the regression samples, and the results are reported in Fig. 7.

Data availability

The primary household-level data analyzed in this study were obtained from the China Family Panel Studies (CFPS), administered by the Institute of Social Science Survey, Peking University, and are available through the official CFPS data platform subject to its data access and use requirements. Other data used in the analyses were obtained from publicly available sources cited in the References. The derived data and calculation files supporting the findings of this study are available in Zenodo at https://zenodo.org/records/15428485.

Code availability

 Codes used in this study are available in Zenodo at https://zenodo.org/records/15428485.

References

  1. IEA. Energy Poverty: How to Make Modern Energy Access Universal? OECD/IEA (2010).

  2. Apergis, N., Polemis, M. & Soursou, S. E. Energy poverty and education: Fresh evidence from a panel of developing countries. Energy Econ. 106, 105430 (2022).

    Article  Google Scholar 

  3. Pan, L., Biru, A. & Lettu, S. Energy poverty and public health: Global evidence. Energy Econ. 101, 105423 (2021).

    Article  Google Scholar 

  4. Lee, C. C. & Yuan, Z. Impact of energy poverty on public health: A non-linear study from an international perspective. World Dev. 174, 106444 (2024).

    Article  Google Scholar 

  5. Samarakoon, S. A justice and wellbeing-centered framework for analysing energy poverty in the Global South. Ecol. Econ. 165, 106385 (2019).

    Article  Google Scholar 

  6. Li, Y. et al. Would energy poverty affect the wellbeing of senior citizens? Evidence from China. Ecol. Econ. 200, 107515 (2022).

    Article  Google Scholar 

  7. Min, B., O’Keeffe, Z. P., Abidoye, B. & Gaba, K. M. More than a billion people live in ‘energy poverty’. Nature. 630, 11 (2024).

  8. Shao, Q. Pathway through which COVID-19 exacerbates energy poverty and proposed relief measures. Energy Sustain. Dev. 74, 1–5 (2023).

    Article  Google Scholar 

  9. Memmott, T., Carley, S., Graff, M. & Konisky, D. M. Sociodemographic disparities in energy insecurity among low-income households before and during the COVID-19 pandemic. Nat. Energy 6, 186–193 (2021).

    Article  Google Scholar 

  10. Burguillo, M., del Río, P. & Juez-Martel, P. How have the COVID pandemic and the war in Ukraine affected energy poverty?. Appl. Energy 377, 124690 (2025).

    Article  Google Scholar 

  11. Carfora, A., Scandurra, G. & Thomas, A. Forecasting the COVID-19 effects on energy poverty across EU member states. Energy Policy 161, 112597 (2022).

    Article  Google Scholar 

  12. Śmiech, S. & Karpińska, L. Energy poverty response to COVID-19 and energy crisis in European NUTS 1 Regions. Soc. Indic. Res. https://doi.org/10.1007/s11205-025-03660-4 (2025).

  13. Hesselman, M., Varo, A., Guyet, R. & Thomson, H. Energy poverty in the COVID-19 era: Mapping global responses in light of momentum for the right to energy. Energy Res. Soc. Sci. 81, 102246 (2021).

    Article  Google Scholar 

  14. Bienvenido-Huertas, D. Do unemployment benefits and economic aids to pay electricity bills remove the energy poverty risk of Spanish family units during lockdown? A study of COVID-19-induced lockdown. Energy Policy 150, 112117 (2021).

    Article  Google Scholar 

  15. Berkouwer, S. B., Biscaye, P. E., Puller, S. & Wolfram, C. D. Disbursing emergency relief through utilities: Evidence from Ghana. J. Dev. Econ. 156, 1–9 (2022).

  16. Li, Y., Qing, C., Zhou, W. & Xu, D. Exploring the role of environmental regulations in biogas development in China, from the perspective of peer effect. Energy 315, 134393 (2025).

    Article  Google Scholar 

  17. Zhang, H. et al. Trade-offs between cold protection and air pollution-induced mortality of China’s heating policy. PNAS Nexus 2, 1–7 (2023).

    Article  Google Scholar 

  18. Li, X., Smyth, R., Xin, G. & Yao, Y. Warmer temperatures and energy poverty: Evidence from Chinese households. Energy Econ. 120, 106575 (2023).

    Article  Google Scholar 

  19. Zhang, S. et al. The 2023 China report of the Lancet Countdown on health and climate change: taking stock for a thriving future. Lancet Public Health 8, e978–e995 (2023).

    Article  Google Scholar 

  20. Brosemer, K. et al. The energy crises revealed by COVID: Intersections of Indigeneity, inequity, and health. Energy Res. Soc. Sci. 68, 101661 (2020).

    Article  Google Scholar 

  21. Han, X., Wei, C. & Cao, G. Y. Aging, generational shifts, and energy consumption in urban China. Proc. Natl. Acad. Sci. USA 119, 1–9 (2022).

    Article  Google Scholar 

  22. Chen, X., Xie, Q., Cao, X. & Li, Q. Examining the effectiveness of China’s energy poverty alleviation policies: A text analysis on inter-provincial panel data. Energy Policy 186, 113978 (2024).

    Article  Google Scholar 

  23. Jia, W. & Wu, S. Spatial Differences and Influencing Factors of Energy Poverty: Evidence From Provinces in China. Front. Environ. Sci. 10, 1–16 (2022).

    Article  Google Scholar 

  24. Sovacool, B. K., Lipson, M. M. & Chard, R. Temporality, vulnerability, and energy justice in household low carbon innovations. Energy Policy 128, 495–504 (2019).

    Article  Google Scholar 

  25. Apergi, M. et al. An energy justice index for the energy transition in the global South. Renew. Sustain. Energy Rev. 192, 1–14 (2024).

  26. Bouzarovski, S. & Petrova, S. A global perspective on domestic energy deprivation: Overcoming the energy poverty-fuel poverty binary. Energy Res. Soc. Sci. 10, 31–40 (2015).

    Article  Google Scholar 

  27. Zhao, L., Eckardt, S., Vashakmadze, E., Lugo, M. A. & Nair, D. China’s Recovery Strategy Should Focus on Inclusive and Sustainable Growth: World Bank Report. (2020).

  28. Graff, M. & Carley, S. COVID-19 assistance needs to target energy insecurity. Nat. Energy 5, 352–354 (2020).

    Article  Google Scholar 

  29. International Energy Agency. Access and affordability. https://www.iea.org/topics/access-and-affordability (accessed 4 October 2026).

  30. Wang, F., Geng, H., Zha, D. & Zhang, C. Multidimensional Energy Poverty in China: Measurement and Spatio-Temporal Disparities Characteristics. Social Indicators Research vol. 168 (Springer Netherlands, 2023).

  31. Li, S. & Shao, Q. Exploring the determinants of renewable energy innovation considering the institutional factors: A negative binomial analysis. Technol. Soc. 67, 101680 (2021).

    Article  Google Scholar 

  32. Li, S. & Shao, Q. Greening the finance for climate mitigation: An ARDL–ECM approach. Renew. Energy 199, 1469–1481 (2022).

    Article  Google Scholar 

  33. Guan, Y. et al. Burden of the global energy price crisis on households. Nat. Energy 8, 304–316 (2023).

    Article  Google Scholar 

  34. Li, Q. et al. Generation expansion planning for Guizhou province based on the complementary characteristics of wind and solar. Energy Rep. 8, 574–584 (2022).

    Article  Google Scholar 

  35. Wang, Y. et al. Accelerating the energy transition towards photovoltaic and wind in China. Nature 619, 761–767 (2023).

    Article  Google Scholar 

  36. Wang, Y., Chao, Q., Zhao, L. & Chang, R. Assessment of wind and photovoltaic power potential in China. Carbon Neutrality 1, 1–11 (2022).

  37. Wang, Q. et al. Examining energy inequality under the rapid residential energy transition in China through household surveys. Nat. Energy 8, 251–263 (2023).

    Article  Google Scholar 

  38. Shupler, M. et al. Modelling of supply and demand-side determinants of liquefied petroleum gas consumption in peri-urban Cameroon, Ghana and Kenya. Nat. Energy 6, 1198–1210 (2021).

    Article  Google Scholar 

  39. CFPS. China Family Panel Studies. Peking University http://www.isss.pku.edu.cn/cfps/ (2025).

  40. Loyalka, P. et al. Skill levels and gains in university STEM education in China, India, Russia and the United States. Nat. Hum. Behav. 5, 892–904 (2021).

    Article  Google Scholar 

  41. Jiang, K. et al. Unclean but affordable solid fuels effectively sustained household energy equity. Nat. Commun. 15, 1–13 (2024).

  42. Forrester, S. P., Montañés, C. C., O’Shaughnessy, E. & Barbose, G. Modeling the potential effects of rooftop solar on household energy burden in the United States. Nat. Commun. 15, 4676 (2024).

    Article  Google Scholar 

  43. Oswald, Y., Owen, A. & Steinberger, J. K. Large inequality in international and intranational energy footprints between income groups and across consumption categories. Nat. Energy 5, 231–239 (2020).

    Article  Google Scholar 

  44. Al Kez, D., Foley, A., Lowans, C. & Del Rio, D. F. Energy poverty assessment: Indicators and implications for developing and developed countries. Energy Convers. Manag. 307, 118324 (2024).

    Article  Google Scholar 

  45. Wang, Y. & Lin, B. Can energy poverty be alleviated by targeting the low income? Constructing a multidimensional energy poverty index in China. Appl. Energy 321, 119374 (2022).

    Article  Google Scholar 

  46. Foster, J., Greer, J. & Thorbecke, E. A Class of Decomposable Poverty Measures. Econometrica 52, 761–766 (1984).

    Article  Google Scholar 

  47. Chang, N. C. Double/debiased machine learning for difference-in-differences models. Econ. J. 23, 177–191 (2020).

    MathSciNet  Google Scholar 

  48. JHU. Pandemic Data Initiative. Johns Hopkins Coronavirus Resource Center 1 https://github.com/CSSEGISandData/COVID-19/tree/master/csse_covid_19_data/csse_covid_19_time_series (2024).

  49. Jiang, Y., Xie, Y. & Shao, Q. How did Internet usage affect life satisfaction before and after COVID-19? Mediating effects and heterogeneity analysis. Socioecon. Plan. Sci. 95, 102007 (2024).

    Article  Google Scholar 

  50. Nguyen, T. D. et al. Impacts of state COVID-19 reopening policy on human mobility and mixing behavior. South. Econ. J. 88, 458–486 (2021).

    Article  Google Scholar 

  51. Brzezinski, A., Kecht, V., Van Dijcke, D. & Wright, A. L. Science skepticism reduced compliance with COVID-19 shelter-in-place policies in the United States. Nat. Hum. Behav. 5, 1519–1527 (2021).

    Article  Google Scholar 

  52. Chernozhukov, V. et al. Double/debiased machine learning for treatment and structural parameters. Econ. J. 21, C1–C68 (2018).

    MathSciNet  Google Scholar 

  53. Wang, Q., Kwan, M. P., Fan, J. & Lin, J. Racial disparities in energy poverty in the United States. Renew. Sustain. Energy Rev. 137, 1–12 (2021).

  54. U.S. Department of Energy. Low-Income Energy Affordability Data (LEAD) Tool. 1–2 https://www.energy.gov/cmei/scep/low-income-energy-affordability-data-lead-tool (2025).

  55. Wang, Y. et al. Global energy inequality across energy types, expenditure groups, and consumption items. Nexus 2, e100086 (2025).

Download references

Acknowledgements

We acknowledge the data mainly provided by the China Family Panel Studies (CFPS), which is funded by Peking University and the NSFC, and maintained by the Institute of Social Science Survey of Peking University.

Funding

Q.S. thanks for the supported from the Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM902), the Guangdong S&T Programme (2026A1010020010), the National Natural Science Foundation of China (Grant No. 71903131), the Alexander von Humboldt Foundation (CHN 1194898 HFST-P), and the Erasmus+ Jean Monnet Modules (ERASMUS-JMO-2025-MODULE, 101234644 — DoughEU).

Author information

Authors and Affiliations

  1. School of Ecology, Sun Yat-sen University – Shenzhen Campus, Shenzhen, China

    Qinglong Shao, He Li, Xiaotong Wang & Tien Ming Lee

  2. School of Business Administration, Southwestern University of Finance and Economics, Chengdu, China

    Yong Xie

  3. School of Economics, Xihua University, Chengdu, China

    Guikun Yin

Authors

  1. Qinglong Shao
  2. Yong Xie
  3. Guikun Yin
  4. He Li
  5. Xiaotong Wang
  6. Tien Ming Lee

Contributions

Qinglong Shao: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft, Writing - Review & Editing, Supervision, Project administration, Funding acquisition. Yong Xie: Methodology, Software, Formal analysis, Data Curation, Writing - Original Draft, Visualization. Guikun Yin: Software, Formal analysis, Data Curation, Visualization. He Li: Methodology, Plot-making, Writing - Review & Editing.Xiaotong Wang: Methodology, Writing - Review & Editing.Tien Ming Lee: Validation, Investigation, Resources, Writing - Original Draft, Writing - Review & Editing, Supervision, Funding acquisition.

Corresponding authors

Correspondence to Qinglong Shao or Tien Ming Lee.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Shao, Q., Xie, Y., Yin, G. et al. COVID-19 impact on households’ energy poverty and clean energy transition in China. npj Clean Energy 2, 36 (2026). https://doi.org/10.1038/s44406-026-00052-6

Download citation

  • Received: 24 April 2026

  • Accepted: 25 September 2026

  • Published: 11 November 2026

  • Version of record: 11 November 2026

  • DOI: https://doi.org/10.1038/s44406-026-00052-6

来源:Nature 健康护理主题 · nature.com