Addressing Health Inequities in Energy-Poor Households: Evidence from China’s Photovoltaic Poverty Alleviation Program

Abstract

Energy poverty, defined as a households’ limited ability to secure affordable energy, has become a key issue in the energy transition debate due to its impact on well-being, health risks, and social inequality. We investigate its health impacts using two-way fixed effects and high-dimensional fixed effects models, two-stage least squares, and quantify additional medical costs during extreme cold events with grouped fitting regression. We assess the effect of China’s Photovoltaic Poverty Alleviation Program using the Difference-in-Differences-in-Differences technique. Our results show that energy poverty significantly worsens household health deprivation, extreme cold events further increase medical costs in southern regions, while northern energy-poor families reduce healthcare spending to cover heating costs, and the Photovoltaic Poverty Alleviation Program significantly reduces both the medical burden and catastrophic medical expenditure among energy-poor households. These findings underscore the need for energy transition policies that combine targeted subsidies, health support during cold seasons, and wider deployment of modern heating technologies to protect vulnerable families and ensure a fair, resilient energy transition.

1. Introduction

Global climate change is one of the most urgent challenges facing the world today. As the largest developing country and a major emitter of carbon dioxide, China has experienced rapid economic growth accompanied by accelerated industrialization and rising energy consumption. In 2023, non-fossil energy made up just 17.9% of China’s total energy mix, while coal, oil and gas remained dominant [1]. From 2013 to 2023, national CO₂ emissions increased by 565 million tons, reaching 1.26 billion tons in 2023; emissions from fuel combustion rose by 5.2% over the same period [2]. It is important to note that the energy transition has addressed many challenges beyond the shift to renewables and the associated reductions in CO₂ emissions [3]. Progress in the energy transition is also closely linked to reducing energy poverty [4,5], and this connection is especially clear in China [6].

Energy poverty has attracted growing attention in the energy transition debate because it concerns fair household participation and the protection of family welfare during this shift [7]. It refers to a household’s limited ability to secure energy that is both accessible and affordable [8]. Energy poverty in China remains inadequately addressed, with rural communities experiencing higher incidence than their urban counterparts, and the severity of natural disasters influencing regional differences [9]. Globally, energy poverty persists: the European Union has made substantial progress in its governance, yet research on the most affected countries—particularly Bulgaria and the broader Balkan region—remains limited [10,11]. In the United Kingdom, extensive studies employ diverse measurement approaches that reveal spatial heterogeneity and have led to increasingly multidimensional and precise energy poverty indices. Evidence from other Asian nations shows that external crises can exacerbate energy poverty; for example, fluctuations in Japan’s energy (or fuel) poverty closely correlate with seismic events [12]. In India, energy poverty has declined in recent years, with socio-economic status identified as the principal determinant [13]. Households in energy poverty often report lower subjective health scores, showing that inadequate energy services lead to measurable declines in well-being when temperature extremes occur [14]. Moreover, health declines triggered by extreme climate events worsen this burden and distribute its effects unevenly across social groups [15]. Differences in energy consumption and the quality of appliances across income brackets mean that energy-poor households face not only income inequality but also systemic injustice [16]. Adopting clean energy and reducing solid-fuel use enable households to upgrade their energy systems and improve their health outcomes [17]. However, extreme climate events can cause major disruptions in energy infrastructure and efficiency [18], increasing the risk of energy poverty [19]. As a result, energy-poor families face higher energy costs and worsening health conditions due to inadequate heating, which in turn drives up medical expenses [20]. This combination of energy deprivation and rising healthcare needs reduces their ability to afford treatment, widens welfare gaps, and entrenches social inequality [21].

Energy-efficient buildings and reduced energy consumption constitute essential strategies for mitigating energy poverty and inequality, and a building’s energy performance under varying climatic conditions further shapes levels of energy vulnerability [22]. Another crucial and effective measure involves the development and implementation of targeted policy interventions. Policy interventions can either ease or worsen these vulnerabilities. A notable example is the Photovoltaic Poverty Alleviation Program, which builds solar farms in pilot regions to meet the energy needs of low-income households and boost their financial resilience [23,24]. By linking solar photovoltaics with poverty alleviation, the program promotes economic development and the energy transition in impoverished areas through subsidies, investment in photovoltaic power plant infrastructure, and expanded employment opportunities [25]. We examine China’s Photovoltaic Poverty Alleviation Program, launched under the Guiding Opinions on the Implementation of Photovoltaic Power Generation for Poverty Alleviation issued jointly by the National Development and Reform Commission of the People’s Republic of China [26]. Operated within a three-party framework involving government, enterprises, and beneficiary households, the program uses two main delivery mechanisms. At the village level, centralized photovoltaic stations channel electricity revenues into local management funds that support public-interest projects, wage subsidies for registered poor households, and improvements to village infrastructure. Meanwhile, participating families install rooftop photovoltaic systems—typically rated at 3–5 kW—manage their own arrays, and sell surplus power to the grid [27]. In areas with limited arable land, agrivoltaic systems combine solar panels with agricultural production, allowing farmers to lease previously unused slopes or greenhouses for electricity generation [28]. The program’s geographic rollout aligns with China’s solar-resource distribution, which is most favorable in the western and northern interior regions where economic development lags and poverty rates remain high [29]; the program has delivered tangible benefits in these areas [30]. By harnessing abundant rural solar resources to generate both electricity and household income, the program addresses the link between energy poverty and health vulnerability, providing a sustainable way out of long-term hardship [31].

Although computable general equilibrium models (CGE) [32,33], integrated assessment models (IAM) [34], data envelopment analysis (DEA) [35,36], game-theoretic approaches [37,38], the Greenhouse Gas and Air Pollution Interactions and Synergies model (GAINS) [39], the Community Multiscale Air Quality Modeling System (CMAQ) [40], and the Long Range Energy Alternatives Planning system (LEAP) [41] are widely used to assess climate and energy policies, few studies have looked at their indirect effects on the most vulnerable households.

To address this gap, we ask three research questions: (1) How does household energy poverty affect health outcomes? (2) Do extreme cold events produce regionally varied impacts on the medical expenses of energy-poor households? (3) What effect does the Photovoltaic Poverty Alleviation Program have on the health-related economic burden of energy-poor households? After clarifying how energy poverty leads to health deprivation, we trace how the financial pressures of healthcare shift under extreme cold conditions and test whether the Photovoltaic Poverty Alleviation Program can reduce this vulnerability. We use a two-way fixed effects (TWFE) model and a high-dimensional fixed effects (HDFE) model, supplemented by two-stage least squares (2SLS), to measure the causal impact of energy poverty on health. Then, we run grouped fitting regression to estimate the health-related economic burden on energy-poor households during extreme cold events. Finally, we apply the Difference-in-Differences-in-Differences (DDD) method [42] to assess the program’s mitigating effect.

This study makes three main contributions. First, we quantify the extra health-related financial burden that extreme cold events impose on energy-poor households. Second, we add a regional inequality perspective by showing how the medical payment capacity of energy-poor families varies across China’s climate zones, highlighting spatial disparities that earlier studies missed [43]. Third, we move the evaluation of the Photovoltaic Poverty Alleviation Program from the aggregate level to the household level. Together, these contributions provide new evidence on the combined vulnerabilities of energy-poor households and suggest equity-focused policies that align energy transition goals with public health protection.

2. Data and Methods

2.1. Data and Indicators

2.1.1. Survey Design, Sampling and Execution

To conduct a detailed micro-level household analysis, we used the China Family Panel Studies (CFPS) administered by the China Social Science Survey Center at Peking University (the CFPS is conducted by the Institute of Social Science Survey of Peking University (www.isss.pku.edu.cn/cfps, accessed on 5 January 2025)). We processed the raw CFPS data to address missing and anomalous values, thereby enhancing overall data quality. Next, we constructed core indicators of energy poverty and adverse health outcomes by merging and aggregating the relevant survey variables. Building on this foundation, we integrated regional climate metrics into the dataset using provincial identifiers. Following these data cleaning and integration steps, we characterized households’ energy poverty status and socio-economic attributes for subsequent analysis. The CFPS survey was launched in 2010; however, because several key variables were missing from that initial wave, we selected the 2012–2020 CFPS waves as our sample. Because our analysis involves extensive derivation of income, energy consumption, and self-reported health indicators, respondents under 16 or over 80 are unlikely to provide these measures accurately. To ensure analytical precision, we merged the household and individual datasets and excluded any households whose head was younger than 16 or older than 80. The CFPS is a comprehensive nationwide survey based on a complex sampling design and a large sample size. After cleaning, our sample comprised 12,313 households in 2012, 13,396 in 2014, 13,416 in 2016, 13,247 in 2018, and 10,198 in 2020.

To ensure high data quality, the CFPS team implemented three main procedures [44]. Firstly, all interviewers were required, upon completing each interview, to verify every variable and datum in the questionnaire, following explicit standards to identify any potential issues or deviations in interview conduct. Concurrently, the CFPS team developed a data access management system that enables interviewers to transmit data to headquarters in real time; this system allows headquarters to swiftly detect any data anomalies and liaise directly with the interviewer to implement corrections. To ensure interviewer competence, CFPS established a rigorous recruitment, training, and assessment process. Candidates underwent résumé screening, telephone and on-site interviews, followed by a six-day training program that included lectures, group exercises, simulation tests, and field practice. Only those who passed the final assessment conducted the survey. Second, CFPS applied a robust data verification protocol. Completed questionnaires were cross-checked via audio recordings, telephone follow-ups, and on-site reviews, with verification rates of 15% for audio, 25% for telephone, 15% for on-site, and 5% for random checks. Every seven days, all variables underwent statistical consistency tests. Data from each interviewer and all related documentation were reviewed to ensure that outcomes—such as refusals, vacant addresses, or communication barriers—were accurately recorded. Finally, post-survey cleaning removed invalid or missing entries at multiple stages for each wave.

2.1.2. Extreme Cold Events: Data and Measurement

The Global Surface Summary of the Day (GSOD) is a station based, gridded weather product that provides daily temperature and precipitation observations worldwide (the GSOD is provided by the NOAA National Centers for Environmental Information (https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C00516, accessed on 22 January 2025)). We used GSOD data to assess extreme cold events, defined as short term cold snaps and long duration cold waves. We obtained raw station records from GSOD, interpolated them onto a 0.1° by 0.1° grid, and rasterized the results. We then averaged the gridded values by province to obtain annual, monthly, and daily temperature and precipitation series. Using these processed weather data, we calculated the incidence of extreme cold events by applying a temperature and precipitation bin approach, a method widely used in climate economics [45,46]. Specifically, we defined extreme cold events according to two complementary criteria: (1) Using 1980–2010 as the baseline period, we computed the 5th percentile of historical monthly average temperatures for January through March and November through December. Any month whose average temperature fell below this threshold was classified as experiencing extreme cold conditions. (2) We recorded each day with a minimum temperature of 5 °C or below as an extreme cold day.

For each province, we then derived the total annual count of extreme cold days by combining long-term trend comparisons (average monthly conditions against historical records) with short-term event identification (individual days exhibiting significant cold). This dual approach allowed us to capture both gradual shifts and immediate shocks in extreme cold exposure across regions. Due to CFPS data constraints, we merged the extreme cold events and CFPS databases using the directly available province identification codes (national standard province codes).

2.1.3. Measurement of Energy Poverty Indicators

We introduced a set of energy poverty indicators to analyze the impact of the carbon emissions trading policy on households and to ensure robustness across multiple dimensions. To distinguish energy poverty from low income alone, we focused on the share of energy expenditure in household income, using this ratio as a comprehensive measure rather than considering energy consumption by itself. First, the primary indicator was the energy burden threshold (energy_poverty). A household was classified as energy poor if its ratio of annual energy expenditure to income exceeded 10% [47]. Second, we defined a binary variable (solid_fuel) to flag households that remained reliant on traditional fuels such as coal or firewood. Third, we calculated the energy bill to income ratio (energy_burden), which directly measured the level of energy-cost burden. Fourth, we applied the Low-Income High-Cost method (LIHC) [48]. A household was deemed energy poor under LIHC if its energy expenditure exceeded the provincial median and its residual income after paying energy costs fell below 50% of the province median income. Fifth, we identified low-income households at risk of high energy expenditure (low_inc_enbill > 10%) [49].

Specifically, we classified as energy poor those households with income below the 30th percentile in their province and energy expenditure exceeding 10% of income. We did not adopt the criterion of energy expenditure exceeding twice the provincial median income, because that measure was negatively correlated with health vulnerability and thus unsuited to capture genuine energy poverty.

Because several indicators rely on the energy burden threshold, we calculate energy_burden using Equation (1).

$$energy\_burden={\displaystyle \frac{12\times \left(electricity\_cost+fuel\_cost\right)+central\_heating\_cost}{household\_income}},$$

(1)

where $electricity\_cost$, $fuel\_cost$ and $central\_heating\_cost$ denote the monthly electricity bill (yuan/month), monthly fuel cost (yuan/month) and annual central heating cost (yuan/year) as recorded in CFPS. The term “$12\times \left(electricity\_cost+fuel\_cost\right)+central\_heating\_cost$” represents each household’s total annual energy bill. $energy\_burden$ thus measures the ratio of that bill to net household income ($household\_income$).

2.1.4. Indicator Definition for Household Health Damage

We measure household health damage (health_deprivation) by calculating the proportion of household members classified as unhealthy. Unhealthiness is defined according to three criteria: self-rated health, body mass index (BMI), and recent hospitalization. This approach combined subjective and objective standards and accounted for socioeconomic and medical criteria [50]. Household health deprivation was defined as the share of household members who met any of the following criteria [51]: (1) Self rated health was assessed through an individual’s perception of their own health on a five-point scale ranging from very good to poor. Compared with conventional measures such as mortality and morbidity rates, self-rated health provided a broader understanding of overall well-being [52]. Previous studies demonstrated that self-rated health predicted objective outcomes such as mortality, disease incidence, and functional limitations [53,54], and captured issues like chronic pain, fatigue, or psychological distress that standard medical evaluations might miss. Individuals reporting fair or poor health were classified as unhealthy. (2) Body mass index (BMI) was calculated from CFPS data on current height (cm) and weight (jin). BMI is used as the second criterion for identifying unhealthy household members. Individuals with BMI values below 18.5 or above 24 were considered unhealthy. BMI is a widely accepted indicator for assessing whether weight is appropriate for height [55]. Deviations from the normal range (18.5 to 24) may indicate malnutrition or, conversely, an elevated risk of chronic conditions such as cardiovascular disease, hypertension, and diabetes. (3) Recent hospitalization was determined by asking respondents if they had been hospitalized for illness in the past 12 months. A positive response indicated an acute health event, reflecting both the severity of health conditions and the household’s access to medical care.

We computed the health deprivation indicator at the household level as the proportion of members meeting any of the criteria listed above. To further capture the economic dimension of health deprivation, we used two indicators: (1) Catastrophic medical expenditure status (catastrophic_medical_expenditure_status), also referred to as extreme health poverty. This binary variable equals 1 if a household’s out-of-pocket medical spending exceeded 40% of its net income, consistent with the World Health Organization definition. A value of 1 indicates that the household experienced severe financial pressure from health risks and medical expenses. (2) Medical expenditure ratio (medical_expenditure_ratio), defined as the proportion of annual household income spent on medical expenses. This ratio highlights the combined effect of health shocks and financial burdens. Both indicators were calculated at the household level using CFPS data on medical expenditures and household income.

All explanatory and control variables were log-transformed where appropriate, based on their distributions. Models were controlled for the following household characteristics: expenditure on equipment and daily goods (expenditure_equipment_daily_goods), net household assets (net_household_assets), and number of properties owned (num_properties_owned), as well as household-head characteristics such as age (age), gender (gender), residence type (residence_type), marital status (marital_status), and social standing (social_standing).

We also introduced three instrumental variables to address potential endogeneity issues. The first instrumental variable was access to clean water (no_clean_water), defined as a dummy variable based on responses to the CFPS question: “Which water source does your household mainly use for cooking?” Households relying on sources other than tap or bottled/filtered water were classified as lacking clean water access (no_clean_water = 1). The second instrumental variable was Engel’s coefficient (engel_ratio), defined as the share of household expenditure spent on food, serving as a proxy for overall economic pressure. The third instrumental variable was non-agricultural employment (non_agric_emp), measured as the proportion of household members engaged in non-agricultural labor, reflecting household economic structure and income stability.

Table 1. Description of the database and indicators.

Type	Variables	Definition
Explanatory variables	energy_poverty	Equals 1 if a household’s annual energy expenditure exceeds 10% of its income; 0 otherwise.
	solid_fuel	Dummy variable based on CFPS question “Which fuel does your household mainly use for cooking?” (1 firewood; 2 coal; 3 bottled coal gas/liquefied gas; 4 natural gas/piped gas; 5 solar/biogas; 6 electricity; 77 other). Equals 1 if response is 1 or 2 (traditional fuel), 0 otherwise.
	energy_burden	Share of household net income spent on energy
	LIHC	Low-Income High-Cost method: equals 1 if household energy expenditure exceeds the provincial median and residual income after paying that bill falls below 50% of the provincial median income; 0 otherwise.
	low_inc_enbill >10%	Equals 1 for households whose income lies below the 30th provincial percentile and whose energy expenditure still exceeds 10% of income; 0 otherwise.
	cold_wave	Defined as extreme cold events.
	pv_poverty_alleviation	China’s Photovoltaic Poverty Alleviation Program.
Dependent variables	health_deprivation	The proportion of household members classified as unhealthy.
	catastrophic_medical_expenditure_status	Dummy variable: equals 1 if out-of-pocket medical spending exceeds 40% of net household income (extreme health poverty), 0 otherwise.
	medical_expenditure_ratio	Share of annual household net income devoted to medical expenses.
Instrumental variables	no_clean_water	Dummy variable based on CFPS question “Which water source does your household mainly use for cooking?” (1 river/lake water; 2 well water; 3 tap water; 4 bottled/filtered water; 5 rainwater; 6 cistern water; 7 pond/spring water). Equals 1 if response is neither 3 nor 4, 0 otherwise.
	engel_ratio	Engel’s coefficient: share of household expenditure spent on food.
	non_agric_emp	Proportion of household members engaged in non-agricultural labor.
Household characteristic controls	expenditure_equipment_daily_goods	Household expenditure on equipment and daily necessities.
	net_household_assets	Total net assets held by the household.
	num_properties_owned	Number of real-estate properties owned by the household.
Household head controls	age	Age of the household head.
	gender	Gender of the household head.
	residence_type	Urban or rural residence (household head).
	marital_status	Marital status of the household head.
	social_standing	Self-assessed social status of the household head.

provides detailed descriptions of all indicators included in our database.

2.2. Methodology

2.2.1. Model for Assessing the Impact of Energy Poverty on Health Deprivation

Household energy poverty deepens health deprivation by increasing vulnerability to both energy and health shocks and by imposing additional financial strain. To examine this relationship, we proposed three linear specifications that estimate how energy poverty affects household health deprivation. In all specifications, health_deprivation was the dependent variable and energy_poverty (the 10% energy-expenditure-to-income threshold) served as the primary independent variable. Additional energy poverty indicators (solid_fuel, energy_burden, LIHC, low_inc_enbill > 10%) were included in sensitivity analyses.

Our baseline specification was the TWFE model, which included household and time fixed effects to account for unobserved heterogeneity across households and over time. The second specification was the HDFE model, which further added province fixed effects. The TWFE and HDFE specifications are given in Equations (2) and (3), respectively.

$${health\_deprivation}_{it}={\alpha }_0+{\alpha }_1\left({energy\_poverty}_{it}\right)+{\alpha }_2{X}_{it}+{\mu }_i+{\lambda }_t{+\epsilon }_{it},$$

(2)

$${health\_deprivation}_{it}={\alpha }_0+{\alpha }_1\left({energy\_poverty}_{it}\right)+{\alpha }_2{X}_{it}+{\mu }_i+{\lambda }_t+{\sigma }_j{+\epsilon }_{it},$$

(3)

The dependent variable ${health deprivation}_{it}$ represents the health deprivation of household i in year t, where ${\alpha }_0$ is the intercept and ${\alpha }_1$ is the coefficient for the explanatory variable ${energy poverty}_{it}$. $X_{it}$ denotes control variables. ${\mu }_i$ represents household fixed effects, ${\lambda }_t$ captures time fixed effects, ${\sigma }_j$ accounts for provincial fixed effects, and ${\epsilon }_{it}$ is the random error term.

The third model we employ is the 2SLS model. Given the potential endogeneity issues arising from reverse causality or omitted variable bias, we identify three instrumental variables to address this concern within the 2SLS framework. The instrumental variables are whether the household uses non-clean water for cooking (no_clean_water), the household’s Engel coefficient (engel_ratio), and the proportion of household members engaged in non-agricultural labor (non_agric_emp). All three instrumental variables exhibit strong relevance and pass the overidentification test, confirming their exogeneity and establishing them as valid and reliable instruments. The first-stage estimation of the 2SLS model is described in Equation (4).

$${energy\_poverty}_{it}={\gamma }_0+{\gamma }_1{no\_clean\_water}_{it}+{\gamma }_2{engel\_ratio}_{it}+{\gamma }_3{non\_agric\_emp}_{it}+{\gamma }_4{X}_{it}{+\epsilon \prime }_{it},$$

(4)

where ${no\_clean\_water}_{it}$, ${engel\_ratio}_{it}$, and ${non\_agric\_emp}_{it}$ serve as instrumental variables. ${\gamma }_0$ is the intercept, and ${\epsilon \prime }_{it}$ represents the random error term.

The second-stage estimation is Equation (5).

$${health\_deprivation}_{it}={\delta }_0+{\delta }_1\widehat{\left({energy\_poverty}_{it}\right)}+{\delta }_2{X}_{it}{+\epsilon ″}_{it},$$

(5)

where $\widehat{{energy\_poverty}_{it}}$ is the predicted value of ${energy\_poverty}_{it}$ obtained from the first stage. ${\delta }_0$ is the intercept, and ${\epsilon ″}_{it}$ is the random error term.

2.2.2. Identification Strategy and Impact Evaluation Model for the Photovoltaic Poverty Alleviation Program

The Photovoltaic Poverty Alleviation Program was introduced within the broader framework of clean energy transition and low-carbon development to support impoverished and low-income groups, aiming to achieve both energy transition and social equity. We employ the DDD design, treating the Photovoltaic Poverty Alleviation Program as a quasi-natural experiment. The policy mandated pilot implementation by 2020; accordingly, we define the implementation period as 2016–2020 (${post}_t$ = 1). We identified fourteen provinces as the primary pilot regions (${pilot}_i$ = 1): Shanxi, Qinghai, Ningxia, Jilin, Hainan, Heilongjiang, Shaanxi, Gansu, Anhui, Inner Mongolia, Sichuan, Hebei, Xinjiang, and Yunnan. Henan was excluded because its pilot was limited to a single county, and Tibet was omitted due to insufficient CFPS coverage. We defined the beneficiary group as energy-poor households (${energy\_poor}_j$ = 1), on the premise that these households face the greatest constraints in energy consumption, access, and affordability. The triple interaction ${pilot}_i\times {{post}_t\times energy\_poor}_j$ thus captures the effect of the Photovoltaic Poverty Alleviation Program (${pv\_poverty\_alleviation}_{it}$) on energy-poor households. We estimated Equations (6) and (7) to evaluate the policy’s impact on household’s catastrophic medical expenditure status (catastrophic_medical_expenditure_status) and medical expenditure ratio (medical_expenditure_ratio).

$$\begin{array}{c}{catastrophic\_medical\_expenditure\_status}_{it}\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}={\beta }_0+{\beta }_1\left({{pilot}_i\times post}_t\right)+{\beta }_2\left({{pilot}_i\times energy\_poor}_j\right)+{\beta }_3\left({{post}_t\times energy\_poor}_j\right)\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_4\left({pv\_poverty\_alleviation}_{it}\right)+{\rho X}_{it}{+\gamma }_i+{\lambda }_t+{\epsilon }_{it},\hfill \end{array}$$

(6)

$$\begin{array}{c}{medical\_expenditure\_ratio}_{it}\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}={\beta }_0+{\beta }_1\left({{pilot}_i\times post}_t\right)+{\beta }_2\left({{pilot}_i\times energy\_poor}_j\right)+{\beta }_3\left({{post}_t\times energy\_poor}_j\right)\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_4\left({pv\_poverty\_alleviation}_{it}\right)+{\rho X}_{it}{+\gamma }_i+{\lambda }_t+{\epsilon }_{it},\hfill \end{array}$$

(7)

In this specification, the term ${pv\_poverty\_alleviation}_{it}$ is equivalent to the triple interaction ${pilot}_i\times {{post}_t\times energy\_poor}_j$, and its coefficient ${\beta }_4$ is the primary parameter of interest. A statistically significant negative ${\beta }_4$ indicates that the photovoltaic poverty alleviation program reduced health vulnerability among energy-poor households, for example, by lowering the share of medical spending or the incidence of catastrophic medical expenditure. If ${\beta }_4$ is insignificant or positive, this suggests no beneficial effect or even a short-term worsening of health burdens, perhaps owing to higher energy bills under cold conditions or inadequate subsidy levels. ${\beta }_0$ is intercept. $X_{it}$ is a vector of household and household-head control variables, ${\gamma }_i$ and ${\lambda }_t$ denote household and time fixed effects, respectively, and ${\epsilon }_{it}$ is the idiosyncratic error term.

3. Results

3.1. The Consequences of Energy Poverty on Health Damage

Energy poverty and health deprivation vary significantly across provinces and geographic regions in China (Figure 1). The proportion of energy-poor households is particularly high in Shanxi, Hebei, and Heilongjiang, with Shanxi alone accounting for nearly one-third of all such households. This prevalence reflects the energy profiles of these provinces: they are rich in fossil fuels, located in northern China, and heavily dependent on single-resource industries, especially coal. Paradoxically, despite their natural resource wealth [56], these provinces exhibit disproportionately high rates of energy poverty. Analysis of provincial health deprivation shows that Qinghai, Heilongjiang, and Jilin report the highest average rates. Located in China’s northwest and northeast regions, these provinces are characterized by lower levels of economic development and under-resourced healthcare infrastructure. Heilongjiang stands out with both a high share of energy-poor households and elevated health deprivation. Its harsh climate, remote location, and aging energy infrastructure further exacerbate household vulnerability. Across all provinces, a greater prevalence of energy poverty corresponds with more severe health deprivation, suggesting that energy poverty may contribute substantially to worsening health outcomes at the household level.

Table 2. Multi-model estimation of the impact of energy poverty on health deprivation.

Variables	TWFE Model	HDFE Model	2SLS Model
energy_poverty	0.00848 **	0.00851 **	0.318 ***
	(0.00352)	(0.00412)	(6.87)
Household characteristic controls	√	√	√
Household head controls	√	√	√
Family fixed effects	√	√
Province fixed effects		√
Time fixed effects	√	√
F			407.56 ***
Kleibergen-Paap rk LM			297.634 ***
Cragg-Donald Wald F			105.56
Kleibergen-Paap rk Wald F			100.911
Hansen J			4.443
			(0.108)
Constant	0.548 ***	0.620 ***	0.189 ***
	(0.0214)	(0.0317)	(4.53)
Observations	53,256	49,870	50,832

Note: the dependent variable is health_deprivation. In the TWFE and HDFE models, values in parentheses are t-statistics; in the 2SLS model, they are z-statistics. Significance levels are indicated by ** for p < 0.05 and *** for p < 0.01. A check mark (√) indicates that the variable or effect is included as a control. In the 2SLS model, the F-statistic tests the joint significance of the explanatory variables. The Kleibergen–Paap rk LM statistic assesses under-identification. The Cragg–Donald Wald F-statistic and the Kleibergen–Paap rk Wald F-statistic evaluate instrument strength under homoscedasticity and heteroscedasticity, respectively. The Hansen J statistic tests over-identifying restrictions.

presents detailed results from three regression models, controlling for both household and household-head characteristics. Columns (1) and (2) report estimates from the TWFE and HDFE models, respectively. Both specifications include household-by-year fixed effects, while the latter also incorporates province fixed effects. Column (3) shows 2SLS estimates, again controlling for the same covariates. In the instrumental variables regression, we use non-clean cooking water, the Engel coefficient, and the share of non-agricultural employment as instruments. The Kleibergen–Paap rk LM statistic of 297.634 and the first-stage Wald F statistic of 100.911 confirm that the instruments are highly relevant. The Cragg–Donald Wald F statistic of 105.56 further underscores their strength. A Hansen J statistic of 4.443 indicates that the instruments pass the over-identification test. Finally, the overall F statistic of 407.56 confirms the joint significance of the model. These diagnostics suggest that the two-stage least squares specification effectively addresses endogeneity and yields results consistent with those from the fixed-effects models.

Across all specifications, the coefficient on energy_poverty is positive and statistically significant at the 5% level. This finding indicates that energy poverty critically impacts household health by exacerbating health_deprivation among families already facing energy shortages. The financial and logistical stress associated with inadequate energy access and high energy costs hinders these households from addressing health issues effectively. In our analysis, health_deprivation is measured using three indicators that capture both subjective and objective dimensions of physical and mental well-being. Households experiencing severe energy poverty exhibit poorer self-rated health, body mass index values outside the normal range, and higher rates of hospitalization.

Our results clearly show that energy-poor households face heightened health risks. Energy poverty not only aggravates health problems but also creates a compounded vulnerability, intertwining inadequate energy access with health deprivation. This dual burden significantly increases financial pressures on these families and can trap them in a long-term poverty cycle. At its core, this issue stems from their limited economic means and lack of alternative energy options to secure efficient energy and improve usage. Consequently, insufficient energy supply undermines their ability to maintain basic living conditions and good health, further worsening adverse health outcomes.

However, for energy-poor households, the worsening of health outcomes does not mean they have the financial capacity to address illness or prevent future health risks. When extreme cold events strike, their medical burden shifts and displays pronounced regional inequality (Figure 2). Our analysis reveals marked regional heterogeneity in the effects of extreme cold events. Figure 2 shows that the annual distribution of extreme cold days is tightly clustered in northern provinces but more dispersed in the south. Although both regions exhibit similar clustering in medical expenditure ratios, southern households experience a significant increase in medical spending during extreme cold events, reflecting their vulnerability stemming from higher winter temperatures and limited access to heating. By contrast, energy-poor households in the north reduce their medical expenditures to cover essential heating costs, thereby deepening their financial strain. This coping strategy is not observed among non-energy-poor households in northern provinces, which maintain stable medical expenditures throughout extreme cold events, indicating sufficient disposable income to absorb additional health risks.

3.2. Impact of the Photovoltaic Poverty Alleviation Program on Health Vulnerability of Energy-Poor Households

The Photovoltaic Poverty Alleviation Program constructs grid-connected solar power plants that harness solar energy to provide stable economic returns to impoverished regions. Its implementation modes include utility-scale solar farms, rooftop distributed installations, village-level power stations, and agrivoltaics systems that integrate crop production with electricity generation. Supported by dedicated poverty alleviation funds, budgetary investments, and government interest subsidies, these initiatives have driven the rapid expansion of photovoltaic installations [26]. Although photovoltaic output declines during winter and rainy seasons, most photovoltaic projects are fully grid-connected, allowing surplus generation during solar-rich periods to be exported to the grid and enabling households to draw power from the grid when sunlight is limited, thereby smoothing seasonal fluctuations. Figure 3 shows that pilot regions are concentrated in western and northern China, areas characterized by lower levels of economic development and abundant solar resources. These provinces have traditionally relied on coal and other fossil fuels, adjusting the essential energy mix. Solar power provides a clean, renewable alternative. Through targeted subsidies and technical assistance, the photovoltaic poverty alleviation program has alleviated both energy access and cost burdens for rural households.

For example, in Sichuan, Gansu, and Inner Mongolia, remote areas with relatively underdeveloped local economies, households have benefited from reduced energy expenditures. The construction and operation of photovoltaic installations have generated local employment opportunities and boosted household incomes [57]. Under the agrivoltaics model, the co-location of solar arrays and crops has also enhanced agricultural productivity and diversified income sources.

We found that household energy poverty exacerbated health damage and that rising energy burdens led to adverse health outcomes. In particular, energy-poor families were forced to cut medical spending in order to pay energy bills, leaving them ill-equipped to cope with the health impacts of climate change [58]. When facing health problems induced by extreme cold events, these households reduced medical expenditure to cover the high energy demand and heating costs.

In the pilot regions of the Photovoltaic Poverty Alleviation Program, local authorities built solar power plants and connected solar generation to the grid, thereby providing impoverished and vulnerable families with stable income and subsidy benefits. We further tested the program’s effect on this micro group, and the results are presented in

Table 3. The impact of the Photovoltaic Poverty Alleviation Program on household health vulnerability.

Variables	Medical_Expenditure_Ratio	Catastrophic_Medical_Expenditure_Status
pilot × post	0.155 ***	0.0360 ***
	(0.0323)	(0.00535)
post × energy_poor	0.711 ***	0.0807 ***
	(0.0527)	(0.0114)
pilot × energy_poor	1.204 ***	0.230 ***
	(0.0398)	(0.00900)
pv_poverty_alleviation	−0.960 ***	−0.173 ***
	(0.0777)	(0.0177)
Family fixed effects	√	√
Time fixed effects	√	√
Household characteristic controls	√	√
Household head controls	√	√
Constant	−3.522 ***	0.0678 *
	(0.199)	(0.0366)
Observations	48,942	56,790

Note: significance levels are denoted by the following symbols: * p < 0.1; *** p < 0.01. A check mark (√) indicates that the variable or effect was included as a control in the model.

. In all specifications, the coefficient on pv_poverty_alleviation was negative and statistically significant at the 1% level, indicating that the Photovoltaic Poverty Alleviation Program succeeded in protecting energy-poor households in the pilot areas. Specifically, it reduced both the medical expenditure ratio (medical_expenditure_ratio) and the incidence of catastrophic medical expenditure (catastrophic_medical_expenditure_status).

The Photovoltaic Poverty Alleviation Program and its associated infrastructure were constructed in pilot regions, driving an energy transition and structural adjustment that curtailed reliance on fossil fuels. This led to reductions in both outdoor and indoor air pollutant emissions, thereby mitigating health impacts. Pilot areas, characterized by low economic development and household incomes, benefited from distributed photovoltaic systems and microgrids, which generated local employment and increased residents’ capacity to address health needs. At the same time, lower energy prices ensured access to basic public health services. Throughout the program, direct subsidies and other measures reduced energy costs (Figure 4). Our analysis showed that severe energy poverty had heightened health risks; in the pilot regions, households that received financial support through the Photovoltaic Poverty Alleviation Program experienced a corresponding reduction in the economic burden of health-related expenses.

3.3. Sensitivity Analyses

We include the remaining four indicators measuring energy poverty in the TWFE and HDFE models to test the robustness of our results. In the sensitivity analysis, the coefficients for energy_burden and low_low_inc_enbill > 10% models are significant. While the coefficients for the other two variables are not significant, their direction is consistent with the results from the main regressions in both models (

Table 4. Sensitive test results.

	TWFE Model				HDFE Model
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Variables	solid_fuel	energy_burden	LIHC	low_inc_enbill > 10%	solid_fuel	energy_burden	LIHC	low_inc_enbill > 10%
energy_poverty	0.00411	0.00425 ***	0.00459	0.0129 ***	0.00385	0.00426 **	0.00469	0.0129 **
	(0.00440)	(0.00136)	(0.00475)	(0.00416)	(0.00704)	(0.00183)	(0.00600)	(0.00503)
Family characteristics	√	√	√	√	√	√	√	√
Household head characteristics	√	√	√	√	√	√	√	√
Family fixed effects	√	√	√	√	√	√	√	√
Province fixed effects					√	√	√	√
Time fixed effects	√	√	√	√	√	√	√	√
Constant	0.551 ***	0.561 ***	0.552 ***	0.546 ***	0.623 ***	0.634 ***	0.620 ***	0.618 ***
	(0.0215)	(0.0219)	(0.0214)	(0.0214)	(0.0318)	(0.0332)	(0.0317)	(0.0321)
Observations	53,256	51,435	53,256	53,256	49,870	48,011	49,870	49,870

Note: The dependent variable is health_deprivation. Significance levels are denoted by ** for p < 0.05 and *** for p < 0.01. A check mark (√) indicates that the variable or effect was included as a control.

). The switch between the two models and the use of multiple explanatory variable measurement methods both demonstrate the robustness of the findings in this study.

4. Discussion

Household energy poverty and health deprivation create compounded vulnerabilities that significantly amplify families’ economic burdens, potentially trapping them in a persistent cycle of poverty [59]. Our study showed that energy poverty exacerbated household health issues because affected families lacked sufficient economic capacity and access to reliable energy or efficiency improvements. This shortfall reduced their resources and capabilities to maintain basic living standards and health, thereby increasing the risk of illness [60]. When climate extremes occurred, the energy crisis facing these already vulnerable households intensified, leading to further health problems and higher medical burdens. We also uncovered marked regional disparities in the impact of extreme cold events. In southern provinces, where winter temperatures are milder and households faced no formal heating costs, extreme cold events drove up the medical expenditure ratio. By contrast, energy-poor households in northern provinces had to bear substantial heating costs during the cold season and, as a result, reduced their medical spending in order to cover those energy expenses. This reduction in medical expenditure further deepened their financial strain. Non-energy-poor households in the north did not cut back on healthcare spending during extreme cold events, since they possessed sufficient income to absorb both energy and health-related costs. Previous research has highlighted strong links between illness-induced poverty and catastrophic medical expenditure in China, with vulnerable families exhibiting regional disparities in both health and financial outcomes [61]. Economically disadvantaged and health-vulnerable households therefore display greater inequality across provinces. Identifying the compound effects of climate extremes and energy poverty on overall household vulnerability is essential for designing effective climate adaptation and energy policies [62,63].

Because extreme cold events forced northern energy-poor households to divert limited resources from healthcare to heating, their health needs were often neglected. We found that the Photovoltaic Poverty Alleviation Program offered a practical remedy. By constructing grid-connected solar installations and providing targeted subsidies, the program relieved the economic burden of both energy and health expenditures for energy-poor families. Pilot regions saw significant reductions in the medical expenditure ratio and in the incidence of catastrophic medical expenditure, alongside improvements in overall health status. In winter—especially during extreme cold events—rising heating costs posed a major challenge for energy-poor households. The rollout of clean energy equipment and associated subsidy schemes under the program led to a marked decline in the use of traditional fuels such as coal [64]. Improved air quality and the adoption of clean heating technologies not only transformed household energy portfolios but also contributed to better health outcomes [65].

Moreover, by focusing financial support on vulnerable households in designated regions, the Photovoltaic Poverty Alleviation Program advanced social equity. It reduced the dual economic pressures of energy consumption and healthcare spending, improved living standards for low-income groups, and narrowed health disparities across income cohorts. Although the transition to sustainable energy can raise household energy costs, in this case, those higher costs strengthened the program’s health vulnerability mitigation effect by triggering additional subsidies and efficiency measures. Under the program’s incentives, local governments and agencies installed more efficient appliances, promoted clean energy technologies, and advanced household energy transitions. These shifts improved both energy equity and health resilience [66]. This evidence suggests that policymakers must balance cost burdens and equity considerations when crafting energy transition strategies to achieve both fair and effective pathways to climate mitigation.

5. Conclusions

We found that although an accelerated energy transition benefits the adjustment of the energy mix and supports climate change mitigation, extreme cold events exacerbate household energy poverty and add a new dimension to health deprivation. The shift from direct health damage to rising health-related costs underscores the need to account for energy poverty in energy transition policies. We quantified the specific effects of extreme cold events on populations suffering both energy and health deprivation and assessed the impact of the Photovoltaic Poverty Alleviation Program on the health-related economic burden of energy-poor households. These findings offer policymakers and the public an initial model and empirical evidence that household vulnerability for the energy-poor and health-deprived is worsening. For countries and economies pursuing energy transition while aiming to meet climate mitigation targets, it is vital to safeguard the interests of vulnerable groups and optimize existing policies. Promoting clean, efficient, and affordable energy projects, rather than focusing solely on emission reduction intensity, can meet energy transition and climate mitigation goals while also benefiting disadvantaged households.

Although energy transition policies have delivered measurable progress in emission reduction and performance improvement, significant regional disparities persist. The policy impact is weakest in economically underdeveloped areas, cold climate regions, and among vulnerable groups, all of which still face major transition challenges. Strengthening policy coordination, promoting interregional exchanges of clean energy technology, and improving local governments’ climate governance capacity are therefore essential to meeting long-term mitigation targets. At the same time, existing energy transition strategies should be optimized to achieve climate justice and a just transition. Our findings show that, during the energy transition and climate mitigation process, energy-poor households shoulder a higher energy burden. Greater energy subsidies and financial support are needed for low-income groups, so that vulnerable families can absorb the cost increases associated with the clean energy transition. A comprehensive health protection scheme is also required, particularly during the winter heating season and periods of frequent extreme cold events, to ensure that health risks and damage in vulnerable households are effectively alleviated.

While promoting structural energy change, governments should deploy air-source and ground-source heat pumps and other modern heating technologies in energy-poor areas, expand technical training on clean energy use, and increase subsidies for clean energy projects. Improving the availability and accessibility of efficient, energy-saving equipment will further reduce household energy burdens and strengthen the capacity and resilience of residents in climate-sensitive regions to adapt to climate change. These steps will enhance fairness in micro-level participation in climate governance and help optimize policy design, achieving both sustainable high-quality development and the dual goals of energy justice and a just transition.