Measuring and Analyzing the Welfare Effects of Energy Poverty in Rural China Based on a Multi-Dimensional Energy Poverty Index

Abstract

This paper aims to measure and analyze the extent and determinants of energy poverty in China and to examine the effects of electricity accessibility on rural development and welfare. It constructs a multi-dimensional energy poverty index based on five dimensions: household cooking fuel, lighting, household electrical appliance services, entertainment/education, and communication. Using the instrumental variable of the 2SLS method and the hierarchical logit model, this paper also explores the impact of electricity accessibility in rural areas on the multi-dimensional energy poverty index, rural expenditure and income, and individual health and education outcomes. The results indicate that the multi-dimensional energy poverty index has experienced a downward trend over the years, yet it remains higher in rural areas compared with urban areas and in central and western regions compared with eastern regions. Electricity accessibility has a notable effect on reducing multi-dimensional energy poverty and promoting rural development, but it varies by region and rural expenditure quantile. Unclean cooking energy in rural households markedly reduces individual educational levels and increases the probability of ill health. In conclusion, this paper suggests that improving electricity accessibility and promoting clean cooking energy are important policy measures for alleviating energy poverty and improving rural welfare in China.

1. Introduction

Energy poverty signifies the absence of accessible, dependable, sustainable, and contemporary energy services, detrimentally impacting the well-being and life quality of billions worldwide. The United Nations Sustainable Development Goals (SDGs) highlight the imperative of ensuring affordable, reliable, sustainable, and modern energy services for all, a vital prerequisite for global sustainable development. Despite notable enhancements in global energy supply and expenditure, roughly 850 million individuals lack electricity, and almost 3 billion people still employ unclean energy for cooking. These conditions result in heightened environmental pollution, increased greenhouse-gas emissions, and climate change. Moreover, they constrict opportunities for economic growth, social progress, and human development. Energy poverty predominantly affects individuals residing in rural areas, particularly in developing regions like Africa and Asia. According to data from the China Household Follow-Up Survey (CFPS) in 2014, 41% of rural households in China still relied on firewood for cooking. The issue of household energy poverty significantly hampers the robust development of rural areas.

The measurement methods and indicators of energy poverty constitute a complex and multifaceted research domain. Different approaches and metrics capture diverse energy demands, services, and welfare levels. Traditional methods for measuring energy poverty often rely on single or binary indicators, such as electricity access rate or per capita energy expenditure. However, these metrics overlook aspects like the quality, reliability, sustainability, and diversity of energy services. Recently, scholars have proposed composite index methods that consider multiple dimensions for measuring energy poverty. The field of research on measurement methods and indicators for energy poverty is complex and diverse. Various methods and indicators capture different aspects of energy needs, services, and welfare. While traditional approaches to measuring energy poverty rely on single or binary indicators, such as electricity access rates or per capita energy expenditure, they fail to account for factors like energy service quality, reliability, sustainability, and diversity. In response, some scholars have introduced multi-dimensional methods, such as the energy poverty line based on minimum energy requirements and the multi-level energy service framework. Building upon these concepts, this study further develops and refines them to provide a more accurate depiction of energy poverty in rural areas of China.

This article aims to address the following research questions: How can we, based on Sen’s feasibility theory [1], select appropriate dimensions and indicators to construct a multi-dimensional energy poverty index (MEPI) for rural areas in China? What is the energy poverty situation in rural areas of China? What are the differences and characteristics of energy poverty among different regions and groups? How does energy poverty impact the well-being of rural residents in terms of expenditure, income, education, and health? Do these impacts exhibit heterogeneity and endogeneity? What effective measures can be taken to alleviate the issue of energy poverty in rural areas of China? What types of policy interventions and energy poverty alleviation strategies are feasible and sustainable?

To tackle these inquiries, this article’s content is structured as follows: Section 2 provides an overview of the research on measuring energy poverty and the factors that affect it; Section 3 introduces the methodology and data sources for constructing the multi-dimensional energy poverty index; Section 4 empirically estimates the impact of energy accessibility on rural income, expenditure, education and health; Section 5 summarizes the main findings of this article and provides policy recommendations for alleviating rural energy poverty in China.

2. Literature Review

2.1. Measurement of Energy Poverty

Energy poverty pertains to households lacking adequate access to essential energy services, affecting their quality of life and well-being. The definition and measurement approaches for energy poverty differ among countries and regions. In developed nations, energy poverty mainly involves the affordability of fuels, with a notable emphasis on the portion of energy expenses needed to sustain a comfortable indoor temperature [2]. Energy poverty in developing countries like China encompasses energy availability and affordability issues [3,4,5]. As a result, assessing energy poverty requires examining household energy deprivation from multiple dimensions [6]. Historically, early measurement methods assessed household energy poverty using single-dimensional indicators like income or expenditure [7]. While straightforward, this approach overlooks the intricate diversity and complexity of energy poverty [8]. In developing countries, China included, energy poverty primarily centers on energy availability and affordability [3,4,5]. Over time, the measurement of energy poverty has evolved from relying solely on single-dimensional indicators to incorporating multi-dimensional metrics [6]. For instance, the International Energy Agency (IEA) has introduced the Energy Development Index (EDI) to gauge energy poverty. This index evaluates a country’s or region’s energy poverty using four sub-indicators: per capita electricity expenditure, per capita commercial energy expenditure, the percentage of the population lacking electricity access, and the portion of the population reliant on traditional fuels [9]. Another approach comes from Nussbaumer [10], who developed the multi-dimensional energy poverty index (MEPI). This index measures household energy deprivation across five dimensions: cooking, lighting, household appliances, entertainment/education, and communication [11]. The choice of methodology for these indices depends on specific objectives and data scenarios due to their inherent advantages and limitations [12]. In the context of China, the limited availability of micro-level energy expenditure data has led to a scarcity of research on energy poverty [13]. Existing studies in this realm predominantly rely on single indicators or simplifications of multi-dimensional ones to gauge the extent and attributes of energy poverty in China [14,15,16,17]. Therefore, it is imperative to formulate a comprehensive and applicable multi-dimensional index that can effectively evaluate the multifaceted dimensions of energy poverty in China.

2.2. Factors Affecting Energy Poverty

The occurrence and evolution of energy poverty are influenced by various factors, with household demographic and economic characteristics standing out as the most significant ones [18,19]. Household income plays a pivotal role in driving energy poverty [20,21,22], as low-income households struggle to cover energy expenses, leading them into energy poverty. The education level is a key determinant affecting both income and energy expenditure. Higher educational attainment empowers households to access and utilize modern energy resources, thereby alleviating energy poverty [22,23,24] and mitigating the risk of chronic energy poverty [25]. Age, gender, and marital status of the household head also impact the household’s resources and social standing, consequently influencing energy poverty [26,27,28,29,30,31]. The interaction of factors such as household and housing size with variables like income, energy efficiency, and prices has intricate effects on energy poverty, which has been highlighted in studies by Sharma [32] and Moniruzzaman [33]. Additionally, labor conditions further compound the situation by shaping income sources and stability, thereby impacting energy poverty [34]. However, the existing research on energy poverty dynamics lacks sufficient depth. This calls for more comprehensive theoretical and empirical studies to uncover the underlying mechanisms driving these dynamic shifts.

2.3. Economic Effects of Energy Poverty

The energy poverty status in various countries is influenced by living conditions and data availability. As a result, the multi-dimensional energy poverty index (MEPI) often reflects national characteristics and lacks universal applicability. Chinese researchers have progressively enhanced their understanding of multi-dimensional energy poverty by drawing insights from global research. For instance, Wang employed data at the provincial level spanning 2000 to 2011 to create an MEPI for comprehensively assessing urban energy poverty in China [35]. However, the existing literature has established that studies conducted at higher administrative levels tend to obscure inequalities among lower-level entities [36]. Consequently, using provincial- or city-level data could potentially blur disparities among households. Leveraging the China Household Panel Survey (CFPS), Zhang formulated a two-dimensional MEPI centered on cooking energy type and energy expenditure [15]. These dimensions represent accessibility and affordability, respectively, omitting mobility poverty due to the study’s focus on domestic energy poverty. Given the intricate nature of evaluating energy poverty and the lack of a universally accepted definition, China commonly relies on income poverty indicators—such as the officially certified minimum living security level and the poverty line—when establishing objectives for residential energy support policies. In 2012, a tiered electricity pricing system was introduced to address energy deprivation and alleviate electricity expenditure burdens. A preferential policy accompanied this change, offering free monthly electricity supply of 10 to 15 kWh to low-income households. However, the validity of using income poverty as a sole measure for energy deprivation within households comes into question. Scholars have extensively investigated this issue from diverse perspectives. For instance, Marchand and Brennan scrutinized the relationship between various poverty indicators and fuel poverty metrics established by the UK government [37]. Their findings indicate that income poverty inadequately represents fuel poverty for setting national or regional policy goals. Igawa and Managi highlighted the heterogeneous impacts of low income and emphasized the need for distinct criteria to define energy poverty across countries [38]. Similarly, Robinson explored the link between geographical attributes and energy poverty across English regions, yielding comparable outcomes [39]. These findings were reaffirmed in a French context [40]. Echoing these insights, Jiang et al., a group of Chinese scholars, conducted a survey in Qinghai Province, arriving at a parallel conclusion [14].

Regarding the analysis of how electricity accessibility affects welfare, the existing literature primarily employs methods like OLS, IV, or DID to gauge the impact on indicators like expenditure, income, health, and education. For instance, Okushima employed the instrumental variable (IV) approach, using the nearest grid as an instrumental variable for electricity accessibility [41]. His study revealed that in Indonesia, electricity accessibility has a significantly positive impact on rural expenditure and income. However, these methods often overlook potential nonlinear or stratified characteristics of the outcome variable and fail to fully consider heterogeneity effects across regions and expenditure quantiles.

This paper utilizes Sen’s feasibility theory to create a multi-dimensional energy poverty index (MEPI) by considering five dimensions: cooking/equipment, space cooling/heating, living energy, entertainment/education, and communication. The study employs micro-data from the China Household Panel Survey (CFPS) to assess the multi-dimensional energy poverty characteristics of Chinese households between 1997 and 2015. Additionally, it examines the influence of multi-dimensional energy poverty on household expenditure, income, education, and health. This research also investigates the dynamics and regional disparities related to energy poverty. The main contributions are outlined below:

(1)

Based on micro-individual data, this paper conducts a multi-dimensional assessment of energy poverty. This study takes into account the variations in energy service demand among households from different regions and income levels. Five dimensions related to energy aspects are chosen for evaluation. As a result, a comprehensive multi-dimensional energy poverty index is formulated to reflect the extent of energy deprivation among Chinese households.

(2)

This study uses panel data and a two-stage least squares approach combined with a multilayer logit model to examine how energy poverty affects household well-being. By addressing endogeneity concerns, the paper provides precise estimates of how electricity accessibility impacts household expenditure and income. It also explores how electricity accessibility and cooking fuel types affect individual education and health, while testing the validity of instrumental variables.

(3)

This paper focuses on a comprehensive analysis of energy poverty at both regional and expenditure quantile levels. The analysis includes distinct regression studies for the eastern, central, and western regions, revealing disparities in energy poverty among these areas. Additionally, the study utilizes quantile regression methods to investigate the impact of electricity accessibility on household expenditure across different expenditure tiers.

3. Data and Methods

3.1. Empirical Strategies

In accordance with the utility function theory introduced by Singh and Bridge [42,43], this paper initiates an examination of the welfare implications of energy poverty by initially scrutinizing the influence of energy accessibility (or electricity accessibility) on income and expenditure. This study considers that electricity accessibility is endogenous, given its susceptibility to external influences, like government policies, infrastructure advancement, geographical conditions, and other unobserved factors. Therefore, it may yield biased outcomes if the ordinary least squares (OLS) method is used directly to estimate the influence of electricity accessibility on the multi-dimensional energy poverty index and rural development. To tackle endogeneity issues, this paper utilizes instrumental variable approaches, along with the two-stage least squares (2SLS) methodology. This paper takes community altitude difference as an instrumental variable for gauging electricity accessibility and concludes that it has a strong correlation with electricity availability but no direct effect on the multi-dimensional energy poverty index (MEPI) as well as rural development. The following equations are used for 2SLS estimation in this study:

$$1\mathrm{st} \mathrm{stage} \hspace{1em}\hspace{1em}\hspace{1em}{E}_i={\alpha }_0+{\alpha }_1{A}_i+{\alpha }_2{X}_i+{\epsilon }_i$$

(1)

$$2\mathrm{nd} \mathrm{stage} \hspace{1em}\hspace{1em}\hspace{1em}{Y}_i={\beta }_0+{\beta }_1{E}_i+{\beta }_2{X}_i+{\mu }_i$$

(2)

where Y_i refers to the income (expenditure) of household i, E_i indicates the power accessibility of household i, A_i refers to the community altitude difference, and X_i is another control variable affecting household income (expenditure).

In the long term, education and energy poverty may be determined at the same time, but energy exerts an indirect impact on education; that is, the accessibility of electricity and clean cooking fuels do not have an immediate impact on education. This effect takes a long time and may occur in the next generation. Therefore, when the impact of energy use on education is measured, the endogenous problem generally does not appear in cross-sectional data. Whether an individual has completed junior high school education or not is a standard for education, and self-rated ill health is a standard in terms of health.

$$P(y=1|x{)}_i=P(y=1|x_1,x_2\cdots ,x_k)$$

(3)

where i denotes the individual and y = 1 indicates that the individual has completed junior school education (or that the individual’s self-rated health is unhealthy). x₁, x₂…, x_k are the explanatory variables. Residential factors may play a strong role in the unobservable factors affecting education or health. As a result, this paper adopts a two-level logit model of individuals and communities:

$$P(y=1|x{)}_i=F({\xi }_{j\left[i\right]}+\eta X_i+{\mu }_i)$$

(4)

$${\xi }_{j\left[i\right]}=a+b{\kappa }_j+{\tau }_j$$

(5)

where j[i] represents individual i residing in community j. X_i and κ_j represent control variables at individual and community levels. μ_i and τ_j denote independent error items at individual and community levels.

3.2. Quantile Regression

This paper employs quantile regression to estimate the impact of electricity accessibility on rural expenditure and income, taking into account the nonlinear nature of the outcome variables and analyzing heterogeneous effects across different expenditure quantiles. Quantile regression is a commonly used method for addressing nonlinear or heterogeneous issues. Unlike ordinary least squares (OLS), quantile regression does not estimate the conditional mean of the outcome variable but rather estimates the conditional median or other location parameters in different quantiles of the outcome variable. This approach provides a better reflection of the variation in the outcome variable across the entire distribution, rather than just around the mean. Additionally, quantile regression can capture the heterogeneous effects among different expenditure quantiles, indicating that electricity accessibility has varying impacts on households with different expenditure levels.

There are several reasons explaining the inclination toward quantile regression instead of other methods. Firstly, this study reveals distinct nonlinear characteristics in the dependent variables (expenditure and income). As the accessibility of electricity increases, the relationship among expenditure, income, and accessibility does not follow a simple or monotonic pattern. Instead, there are varying trends across different intervals. Consequently, employing ordinary least squares (OLS) may overlook these nonlinear features, leading to biased or erroneous estimation results. Secondly, this study identifies significant heterogeneity effects among different expenditure quantiles, indicating that the impact of electricity accessibility on households with varying expenditure levels differs. For instance, in low-income households, electricity accessibility exerts a more pronounced positive influence on expenditure and income, whereas in high-income households, its effect is relatively smaller or negative. Utilizing ordinary least squares (OLS) may disregard this heterogeneity effect and result in biased or erroneous estimations. Lastly, this paper discovers that quantile regression exhibits a high degree of fitness and robustness and yields results consistent with or similar to those of other methods, including 2SLS and DID.

3.3. Multi-Dimensional Energy Poverty Index (MEPI)

The multi-dimensional energy poverty index is derived from the multi-dimensional poverty index, which was first proposed by Alkire and Foster in 2007 [33]. The construction method of the multi-dimensional poverty index (A-F method) is mainly influenced by the study by Sen on deprivation and capacity pioneering work [34]. The central argument is that human poverty should be regarded as lack of opportunity and choice. The MEPI uses the “double boundary” method to define multi-dimensional energy poverty. The first step is to set the dimension of multi-dimensional poverty. The second step is to set the threshold value of judging the dimension of poverty, that is, double boundaries. The first level of boundary determines whether the sample is deprived in each dimension. The second level of boundary identifies whether the sample is multi-dimensionally energy-impoverished based on the number of dimensions in which it is deprived. Each latitude value is set as follows: Let M^n,d represent the matrix of the n × d dimension; let matrix element Y ∈ M^n,d represent the value obtained by n individuals at d different latitudes. Any element of y denotes the value taken by individual i at latitude j,i = 1, 2, …, N; j = 1, 2, …, d. Deprivation matrix: Let z = (z₁, z₂, …, z_d) be the deprivation critical value matrix. Weight: Let w = (w₁, w₂, …, w_d) be a weight matrix, and w_j denotes the weight of dimensionj in the multi-dimensional poverty measurement. Deprivation count: c_i = (c₁, c₂, …, c_n) denotes the deprivation count. It reflects the extent to which an individual is deprived. (i = 1, 2, …, n) indicates the number of dimensions of deprivation that individual i has experienced. The formulas for calculating the A-F multi-dimensional poverty index (MEPI) and the average deprivation share are as follows:

$$MEPI=\sum _{i=1}^n{c}_i\left(b\right)/nd$$

(6)

$$A=\sum _{i=1}^n{c}_i\left(b\right)/qd$$

(7)

where n denotes the number of individuals, q refers to the population of multi-dimensional poverty, and c_i(b) represents the value of c_i when the dimension poverty line is b. The incidence of poverty is H = q/n, and A means the average share of deprivation.

Based on the idea of Nussbaumer [10], this paper selects the five dimensions of home cooking fuel, lighting, household appliance services, entertainment/education, and communication in the construction of the MEPI. The corresponding indicators, variables, and deprivation values are shown in

Table 1. Dimensional variables and critical values of the MEPI.

Dimension	Indicator	Variable	Deprivation Threshold (i.e., If Poverty)
Home cooking fuel	Modern cooking fuels	Fuel type	Using any fuel other than electricity, liquefied petroleum gas, natural gas, or biogas
Lighting	Electricity accessibility	Electricity supply	Not using electricity as the main energy source for lighting
Household electric appliance services	Household appliances	Refrigerator	Not having a refrigerator
Entertainment/ education	Educational/recreational equipment	TV or computer	Not having a TV or computer
Telecommunication	Telecommunication facilities	Mobile phone or landline	Not having a mobile phone or landline

Note: Mekonnen believes that households are defined as poor if they use traditional energy sources such as wood, charcoal, feces, and crop residues as cooking fuels [44]. Modern energy sources include electricity, kerosene, liquefied petroleum gas, and natural gas.

.

Considering the availability of data, and the consistency and completeness of the dimensions in each year, the MEPI index in this section is calculated based on the CHNS survey from 1997 to 2015 as the data source.

Table 2. Single-dimensional poverty in energy poverty, 1989–2015 (%).

		1989	1991	1993	1997	2000	2004	2006	2009	2011	2015
Fuel poverty	All	87	85	81	64	60	59	49	33	22	14
	Urban	74	67	56	38	33	31	23	13	6	5
	Rural	95	93	92	78	72	72	61	42	33	21
	East	77	78	74	56	47	47	37	24	16	11
	Central	86	84	83	67	66	64	55	38	33	21
	West	98	95	87	68	65	64	53	33	18	12
Lighting poverty	All	10	4	2	3	2	0.6	0.9	0.9	1	3.8
	Urban	5	0.3	0.5	2	2	0.7	0.5	0.5	0.5	2
	Rural	12	5.8	2.7	3	2	0.5	1	1	1	4.9
	East	8	2	1	3	1	0.4	0.6	0.2	0.5	4
	Central	13	7	3	2.8	2	0.6	1	1	1	5
	West	7	1	1	4	3	0.6	0.5	1	1	2
TV poverty	All	87	83	79	67	60	56	51	36	20	9
	Urban	69	61	55	43	35	33	29	18	9	3
	Rural	96	93	89	79	73	68	61	44	27	12
	East	82	76	72	55	52	47	40	25	12	5
	Central	88	83	79	68	60	56	52	36	24	9
	West	94	91	87	77	73	68	63	51	24	12
Entertainment poverty	All	50	47	45	53	64	71	76	76	61	55
	Urban	41	45	55	66	74	73	70	59	43	44
	Rural	54	48	41	46	58	71	79	84	72	63
	East	45	44	47	53	67	69	72	68	51	49
	Central	49	46	40	51	59	69	77	79	71	60
	West	56	52	51	56	67	78	82	80	60	56
Telecommunication poverty	All	/	/	/	72	52	25	20	10	6	24
	Urban	/	/	/	55	32	14	11	4	3	24
	Rural	/	/	/	80	62	30	24	13	8	25
	East	/	/	/	63	39	13	10	5	3	19
	Central	/	/	/	70	53	27	20	10	8	31
	West	/	/	/	83	69	38	35	18	6	23

Note: The investigation of whether there were fixed telephones or not began in 1997; whether there were mobile phones or not, in 2004; and whether there were fixed telephones or mobile phones, in 2004–2015. Whether there were televisions or not was investigated from 1989 to 2009, and whether there were computers or not has been investigated since 1997. That is, investigations were conducted on whether there were televisions or mobile phones from 1997 to 2006.

shows the single-dimensional poverty situation of the whole sample of energy poverty; the urban–rural sub-sample; and the sub-sample of eastern, central, and western regions from 1989 to 2015.

From the perspective of single-dimensional fuel poverty, the total sample of fuel poverty declined significantly, from 87% in 1989 to 14% in 2015. The performance of urban and rural samples is similar to that of the total sample. For example, urban fuel poverty declined from 74% in 1989 (coal is also classified as fuel poverty in this paper) to 5% in 2015, while rural fuel poverty decreased from 95% in 1989 to 21% in 2015. From a comparative perspective, the percentages of fuel poverty in the eastern, central, and western regions were 77%, 86%, and 98%, respectively, in 1989. In 2015, they were 11%, 21%, and 12%, respectively. At the beginning and end of the period, fuel poverty in the eastern region was the lowest, and fuel poverty in the central region was higher than that in the central region at the end of the period. From 1989 to 2015, lighting poverty showed a declining trend (except in 2015). The temporal variations in lighting poverty in the urban, rural, eastern, central, and western parts of the sample group were basically consistent with the performance of all the samples. Lighting poverty in rural areas was higher than that in urban areas, and that in eastern areas was lower than that in central and western areas. Household appliances (refrigerators) poverty in both the total sample and the grouped samples showed a steady decline from 1989 to 2015. The poverty of household appliances in urban areas was far lower than that in rural areas. The poverty of household appliances in eastern areas was lower than that in central areas, and that in central areas was lower than that in western areas. In terms of entertainment and education poverty, if the investigation had only been conducted on whether there was a TV or whether there was a computer, then the poverty of both the total sample and the grouped samples would have presented a declining trend, such as TV poverty in 1989–1993 and computer poverty in 2009–2015. However, during the period from 1997 to 2006, when the question of whether there was a TV or computer was investigated, poverty showed an upward trend. Telecommunication poverty of both the total sample and grouped samples in 1989–2015 basically showed a downward trend (except in 2015). Rural communication poverty was higher than that in urban areas, and communication poverty in the eastern region was lower than that in the central and western regions.

Figure 1 draws a trend map of the multi-dimensional energy poverty index in 1997–2015 for all samples, including urban and rural sub-groups, and eastern, central, and western sub-groups, when the equal weight of dimensions and the cut-off point of multi-dimensional energy poverty were 30%.

Figure 1 shows that the multi-dimensional energy poverty index presents a downward trend in both the total sample and the grouped samples. For example, in 1997, the MEPI was 0.488, while in 2015, the index dropped to 0.123. The rural multi-dimensional energy poverty index was higher than the urban one; that in the eastern region was significantly lower than that in the central and western regions. The central region’s MEPI was lower than that of the western region, with the exception of the years after 2011.

Figure 2 shows the changes in MEPI, H, and A from 1997 to 2015. According to the figure, the average share of energy deprivation (A) and the incidence of energy poverty steadily declined from 1997 to 2015.

To better illustrate the composition of and change in the multi-dimensional energy poverty index (MEPI), this paper uses the contribution rate to measure the impact of each dimension on energy poverty. The contribution rate is the proportion of each dimension in the multi-dimensional energy poverty index, which reflects the influence of each dimension on energy poverty. The method for calculating the contribution rate of each dimension is the following: First, determine whether each sample is deprived in that dimension based on the weight and deprivation value of each dimension; second, calculate the number and proportion of deprivations in all dimensions for each sample; finally, add up the proportions of deprivations in a certain dimension for all samples and then divide the total by the total number of samples to obtain the contribution rate of that dimension. Figure 3 shows the changes in the contribution rates of household cooking fuel, lighting, household appliance services, entertainment/education, and communication from 1997 to 2015.

The contribution rate of household cooking fuel declined steadily from 26% in 1997 to 20% in 2015. The contribution rate of lighting was less than 1% in each year. The contribution rate of household refrigerator ownership declined steadily from 28% in 1997 to 14% in 2015. The contribution rate of the entertainment/education dimension was V-shaped. For example, the contribution rate of this dimension decreased from 29% in 1997 to 7.8% in 2011 and increased to 22% in 2015. This period witnessed a steady rise in the contribution rate of the communication dimension, with an increase from 16% in 1997 to 42% in 2015, which was much higher than that of other dimensions. In terms of the urban and rural samples, their performance trend was basically the same as that of all samples, while the contribution rate of each dimension was different. For example, the average contribution rates of household cooking fuel, lighting, household appliance services, entertainment/education, and communication in rural areas from 1997 to 2015 were 27.6%, 0.5%, 27.3%, 17.2%, and 27.3%, respectively. By contrast, the average contribution rates of these dimensions were 21.2%, 0.6%, 25.9%, 20.3%, and 32.1%, respectively, in urban areas. That is to say, the contribution rate of rural cooking fuel was higher than that of urban cooking fuel, and the contribution rate of rural household appliance services was higher than that of the urban one. The contribution rates of urban entertainment/education and communication were higher than those of rural entertainment/education and communication.

3.4. Determinants of Multi-Dimensional Energy Poverty

This section aims to determine the energy poverty status of households. The measurement of the multi-dimensional energy poverty index provides a yardstick for determining energy poverty, which is consistent with the existing literature. In the second step, families are classified into energy-poor and non-energy-poor. This paper uses the logit model to analyze multi-dimensional energy poverty. In the logit model, the dependent variable is the multi-dimensional energy poverty index, which is converted into a binary choice by using the specific deprivation cut-off point of the energy poverty index. That is, if the index is greater than 0.3, the family is considered to be a multi-dimensionally energy-poor family. The logit model of multi-dimensional energy poverty includes householder’s characteristics, household characteristics, and residential area as variables.

Table 3. Determinants of multi-dimensional energy poverty.

	Odds Ratio	Elasticity	p > z
Family size	−0.081	−0.2093	0.001
Families headed by women	−0.437	−0.424	0
Householder ages	−0.012	−0.5619	0
Marriage	−0.611	−0.4396	0
Junior school	−0.539	−0.1655	0
High school	−1.092	−0.2402	0
College school	−1.888	−0.2232	0
Rural areas	0.6494	0.81955	0
Central areas	0.7334	0.18682	0
Western areas	0.5683	0.1512	0

Note: The basic group of household education level is primary school graduation or below; the basic group of the region is the eastern region.

shows the logit regression results for 2015.

The regression results show that when the family size increases by one member, the probability of the family falling into multi-dimensional energy poverty decreases by 0.21 and is significant at the statistical level of 1%. This may reflect that a large family size means more potential income earners or indicates that energy use has the characteristics of mass economy. Families headed by women are less likely to fall into energy poverty. The reason may be that such families need to accumulate social capital to build social networks to cope with more contingency in production and life, thus stimulating household productivity to break away from multi-dimensional energy poverty. On the other hand, Sánchez-Guevara argued that female household heads are more vulnerable in energy poverty [45], which contradicts the conclusion and the literature in this paper. The possible reason consists in the different analytical perspectives and focuses. This paper analyzes the impact of energy accessibility on individual health and education outcomes from a macro-perspective and analyzes the impact of energy use on household income and expenditure from an economic perspective, while Sánchez-Guevara analyzed the problems and challenges faced by female household heads in energy poverty from a micro-perspective and a sociological perspective and analyzed the social capital of and psychological pressure on female household heads in energy poverty.

As householders age, the probability of falling into energy poverty decreases. The degree of multi-dimensional energy deprivation in married families is significantly reduced, possibly because the resources of both spouses are more than those of the single spouse. Moreover, the other spouse can also assume the responsibility of caring for the family or improving the family’s living conditions. Marriage is beneficial to the welfare of the family, including the improvement in energy welfare. Compared with families whose householders are primary school graduates, the probability of multi-dimensional energy poverty decreases significantly with the increase in the degree of householders’ education and is significant at the 1% statistical level. This is in line with expectations, because the higher the degree of householders’ education, the higher the income of households, and the less likely they fall into multi-dimensional energy poverty. In addition, compared with households with primary school education or less, household heads who have completed junior middle school education, high school education, and college education decrease by 0.17, 0.24, and 0.22, respectively. This also manifests the importance of education to increase income, and education serves as an important determinant of multi-dimensional energy poverty. Kanagawa, for example, based on a data analysis of the relationship between energy access and the improvement in socio-economic conditions in rural areas in India, found that education was the most important component of poverty reduction and argued that due to economic constraints, poor families could not afford the required educational expenditure and thus could not complete secondary education [46]. Therefore, lower education levels can hinder household income growth and lead to energy poverty by making them unable to afford modern energy services. Compared with urban families, the degree of multi-dimensional energy deprivation of rural households increases significantly. This is partly attributed to the reason that households in urban areas can get better jobs and thus higher incomes, so that they can afford energy expenditure. On the other hand, rural households perform a slower transformation on the energy ladder, and the resources rural households acquire are less efficient than clean-energy resources. Compared with the eastern region, families living in the central and western regions are more vulnerable to multi-dimensional energy deprivation. These conclusions are consistent with the above descriptive analysis.

4. Empirical Analysis

The data used in the empirical analysis in this section come from the China Family Follow-Up Survey (CFPS) in 2014. The survey carried out tests in 2008 and 2009 and conducted national investigations in 2010, 2011, 2012, 2014, and 2016. The sampling design of the CFPS focuses on the representativeness of the first-visit sample and adopts the implicit stratified, multi-stage, multi-level probability sampling method (PPS) proportional to population size. The sample covers 25 provinces, except Hong Kong Special Administrative Region, Macao Special Administrative Region, Taiwan Province, Xinjiang Uygur Autonomous Region, Qinghai Province, Inner Mongolia Autonomous Region, Ningxia Hui Autonomous Region, and Hainan Province. CFPS questionnaires are divided into three levels: individual; the close circumstance of individual life—family; and the close environment of the family—village residence. Therefore, three kinds of questionnaires are formed: individual questionnaires, family questionnaires, and village residence questionnaires. In addition to the lack of investigation on power accessibility, the community data in the 2016 survey were also kept secret. As a result, out of the timeline, this paper takes the rural survey from 2014 as the basis for empirical analysis.

This paper chooses the answer to the question “how about the access to electricity at your home” in the CFPS survey, considering lack of access to electricity and frequent power failure to be the proxy indicators of power inaccessibility, in which the proportion of households without access to electricity is 0.25%, the proportion of frequent power failure is 3.13%, and the total of the two amounts is 3.28%.

Table 4. Descriptive statistics.

Variable	Variable Meaning	Mean	Standard Deviation	Minimum Value	Maximum Value
consumper	Per capita household expenditure	11,241	14,669	0	302,400
incomeper	Per capita household income	12,590	24,179	0.25	1,340,613
healthnot	Individuals who rated their health as unhealthy	0.1683	0.3742	0	1
eduind	Individuals above 16 years old who has completed junior school education	0.1102	0.3131	0	1
electr	Electricity accessibility	0.9672	0.1781	0	1
firewood	Firewood as cooking fuel	0.4129	0.4924	0	1
hhsize	Family size	3.906	1.9173	1	17
agehead	Householder’s age	49.865	14.096	16	95
eduhead2	Education level of householder is junior school	0.2615	0.4395	0	1
eduhead3	Education level of householder is high school	0.0857	0.28	0	1
eduhead4	Education level of householder is college school	0.0441	0.2054	0	1
area2	Living in the central region	0.2742	0.4461	0	1
area3	Living in western region	0.3567	0.4791	0	1
ageind	Individual’s age	45.012	17.465	16	104
genderind	Male	0.5059	0.5	0	1
altitude	Community altitude difference	206.42	335.4	1	2400

shows that the proportion of rural households with firewood as cooking fuel accounts for about 40%, which indicates that the problem of non-clean cooking fuel in rural areas remains serious. The analysis continues with the impact of electricity accessibility on rural expenditure and income quantification.

Table 5. Electricity accessibility and 2SLS regression of income and expenditure.

	Consumper(ln)		Electr		Incomeper(ln)		Electr
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	3.9787 ***	0.9909			9.5871 ***	1.873
hhsize	−0.077 ***	0.01	−0.006 ***	0.0013	0.0056	0.0185	−0.006 ***	0.0013
eduhead2	0.0977 ***	0.0377	0.0164 ***	0.0056	0.1181	0.0736	0.0178 ***	0.0056
eduhead3	0.2464 ***	0.0531	0.0137	0.0087	0.3453 ***	0.1023	0.0133 ***	0.0087
eduhead4	0.5273 ***	0.1186	0.0195	0.0198	0.6574 ***	0.2292	0.0188	0.0198
agehead	−0.003	0.007	0.0028 **	0.0011	0.0211	0.0133	0.0027 **	0.0011
age2head	−9 × 10−5	7 × 10−5	−3 × 10−5 ***	1 × 10−5	−3 × 10−4 **	0.0001	−3 × 10−5 ***	1 × 10−5
area2	−0.094 ***	0.0351	0.0017	0.006	−0.08	0.0666	−0.002	0.0059
area3	0.0133	0.0353	−0.005	0.0058	−0.07	0.0683	−0.007	0.0057
altitude(ln)			−0.006 ***	0.001			−0.006 ***	0.001
_cons	5.6961 ***	0.9258	0.9434 ***	0.0283	−0.79	1.754	0.9452 ***	0.028
Cragg–Donald Wald F-statistic	34.837				36.622
Anderson canon (p-value)	0				0

Note: Householders whose education level is lower than primary school and households in the eastern area are set as control groups. *** means significant at 1% level; ** means significant at 5% level.

shows the impact of electricity accessibility on rural expenditure and income quantiles. As shown in the table, electricity accessibility has a significantly positive impact on rural consumption and income, but this impact decreases as the order of magnitude of consumption increases. It is worth pointing out that the CFPS survey does not involve information about the head of the household. This paper defines a virtual “head of household”, which regards the financial respondent in the family as the head of household [47].

4.1. Electricity Accessibility, and Income and Expenditure

This section will discuss the impact of electricity accessibility on income and expenditure and conduct an analysis of its mechanism. Table 5 reports the regression results obtained using the 2SLS method. From the Cragg–Donald Wald F-statistics and Anderson canon (p-value) results in the last two rows of Table 5, it can be seen that the use of the community altitude difference as a tool variable for power accessibility is more effective (This paper uses the Cragg–Donald Wald F-statistic and the Anderson canon statistic to test the validity of instrumental variables. The Cragg–Donald Wald F-statistic is designed to test whether instrumental variables are weakly correlated, that is, whether instrumental variables have sufficient strong correlation with endogenous variables; the Anderson canon statistic is used to test whether instrumental variables are exogenous, that is, whether instrumental variables only affect outcome variables by affecting endogenous variables, rather than directly affecting or being correlated with other unobserved factors).

The results obtained by using instrumental variables show that the expenditure of households with access to electricity is 398% of that of those without electricity and the income of the former is 959% of the latter. These values are the regression results after incorporating other variables, demonstrating that modern energy has remarkably changed people’s lifestyle. The increase in household education increases the per capita expenditure of households. The increase in family size results in a decrease in per capita household expenditure, due to the fact that more people consume the equivalent resources. The expenditure in the central region is lower than that in the eastern region, while the expenditure in the western region is not significantly different from that in the eastern region. The increase in household education dramatically increases the per capita income of families. Table 5 proves that electricity accessibility has a significantly positive impact on per capita expenditure, but it does not provide a picture of how electricity affects expenditure at all levels of wealth distribution. For this reason, this paper also conducts a quantile regression of expenditure, the results of which are shown in

Table 6. Quantile regression of expenditure.

	10%		25%		50%		75%		95%
Consumper(ln)	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.024	0.0899	0.0346	0.0541	−0.033	0.035	−0.027	0.0681	0.2244 ***	0.0789
hhsize	−0.096 ***	0.0084	−0.102 ***	0.0059	−0.121 ***	0.0057	−0.125 ***	0.0065	−0.134 ***	0.0137
eduhead2	0.1409 ***	0.0336	0.1753 ***	0.0188	0.199 ***	0.0275	0.1843 ***	0.0197	0.2367 ***	0.056
eduhead3	0.2864 ***	0.0287	0.3144 ***	0.0345	0.3138 ***	0.0282	0.3522 ***	0.0411	0.2379 **	0.109
eduhead4	0.4456 ***	0.1584	0.5414 ***	0.0866	0.6024 ***	0.0667	0.6013 ***	0.057	0.5847 ***	0.1098
agehead	0.0008	0.007	−0.002	0.004	−0.001	0.0041	−0.008	0.0051	−0.012	0.011
age2head	−2 × 10−4 ***	7 × 10−5	−1 × 10−4 ***	4 × 10−5	−1 × 10−4 ***	4 × 10−5	−4 × 10−5	5 × 10−5	4 × 10−5	0.0001
area2	−0.004	0.0325	−0.095 ***	0.0303	−0.131 ***	0.0256	−0.14 ***	0.0294	−0.077	0.0585
area3	−0.118 ***	0.0325	−0.145 ***	0.028	−0.061 **	0.0253	−0.029	0.0195	0.0011	0.0394
_cons	8.7382 ***	0.1844	9.216 ***	0.1032	9.7872 ***	0.1072	10.409 ***	0.1257	10.983 ***	0.2735

Note: *** means significant at 1% level; ** means significant at 5% level.

.

The results of the expenditure quantile in Table 6 show that firstly, in the 95% expenditure quantile, power accessibility has no significant impact on expenditure, while above the 95% expenditure quantile, power accessibility makes expenditure increase significantly. This can be explained by the level of human capital. Electricity is often used as an enhancement factor in productive capacity. Families with lower living standards have lower educational level and productive capacity. Electricity, as an enhancement factor in productive capacity, cannot play its due role in the poor. At the high end of income distribution, the human capital of families increases. Maximizing the use of electricity has a compound impact on the level of expenditure. Interestingly, household size makes expenditure decrease significantly at any expenditure level, and the educational level coefficient of household heads plays a significantly positive role in promoting expenditure at each expenditure level, which shows that even the poorest households can benefit from an increase in educational level.

The above analysis points out that power accessibility plays a positive role in promoting income growth. Next, this paper examines the underlying mechanism and divides income into wage income and operational income. Electricity accessibility may positively impact both income types, because on the one hand, it makes the production process more efficient; for instance, the introduction of electric pumps in irrigation leads to increased wages, labor demand, or both. On the other hand, power accessibility may stimulate entrepreneurship and induce households to buy small equipment and start home workshop production.

Table 7. Impact of electricity accessibility on wages and operational income.

	Logarithm of Wage Income Per Capita		Logarithm of Per Capita Operating Income
	Coef.	Std. Err.	Coef.	Std. Err.
electr	7.7993 ***	2.3735	5.9255 **	2.3751
hhsize	−0.062 ***	0.0204	−0.1534 ***	0.0193
eduhead2	0.035	0.067	0.1428 *	0.0795
eduhead3	0.197 **	0.0924	0.3214 ***	0.1019
eduhead4	0.6922 ***	0.1927	−0.1496	0.2970
agehead	−0.024 *	0.0123	0.0244	0.0192
age2head	0.0002	0.0001	−0.0004 **	0.0002
area2	−0.159 **	0.0649	0.0767	0.0683
area3	−0.202 ***	0.0601	−0.0143	0.0782
_cons	2.2037	2.3253	1.8676	2.0878
Cragg–Donald Wald F-statistic	17.15		16.507
Anderson canon (p-value)	0		0

Note: *** means significant at 1% level; ** means significant at 5% level; * means significant at 10% level.

reports the impact of electricity availability on wages and operational income.

According to the results in Table 7, electricity accessibility promotes more “entrepreneurship” activities, and it also contributes to a substantial increase in per capita wages. As mentioned above, this may be due to the fact that the accessibility of electricity frees people from tedious household work (especially women) and thus promotes the increase in labor supply. It may also be attributed to the fact that the labor supply does not change, but the accessibility of electricity increases the productivity of labor and increases wages. It may also be due to both factors. The coefficient of operational income is lower than that of wage income, which may indicate that there are fewer risk-taking families engaged in “entrepreneurship” activities in rural areas and that rural families are more inclined to engage in activities where risks are less numerous and people work to earn more stable wages.

4.2. Energy Poverty, and Education and Health

To evaluate the health effects of energy poverty, it is useful to examine factors like firewood expenditure for cooking, daily indoor exposure to burning firewood, instances of eye issues, and the frequency and length of heart and respiratory diseases. This examination could provide greater insight into the health consequences linked with energy poverty. The CFPS is only designed to consider self-rated health. This study employs a multi-level logit model, with the self-rated health status being represented as an “unhealthy” dummy variable. The model incorporates individuals’ characteristics including gender, age and its square, household education level, and residential area. The regression results of the effects of energy poverty on education and health are shown in

Table 8. Multilayer logit model of the impact of energy poverty on education and health.

	Individual above 16 Years Old Has Completed Junior School Education				Individuals Rated Their Health as Unhealthy
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.0699	0.1314	0.0875	0.1447	−0.302 ***	0.0915	−0.281 ***	0.0998
firewood	−0.641 ***	0.0501	−0.252 ***	0.0545	0.5796 ***	0.0383	0.3772 ***	0.0421
genderind			0.5994 ***	0.0455			−0.566 ***	0.0359
ageind			0.0168 **	0.0081			0.1538 ***	0.007
age2ind			−6 × 10−4 ***	9 × 10−5			−9 × 10−4 ***	6 × 10−5
eduhead2			0.3058 ***	0.0573			−0.347 ***	0.045
eduhead3			2.8145 ***	0.0596			−0.531 ***	0.0734
eduhead4			2.3108 ***	0.0913			−0.618 ***	0.1341
area2			−0.031	0.0752			0.0247	0.0748
area3			−0.286 ***	0.0721			0.1669 **	0.069
_cons	−2.007 ***	0.1331	−2.604 ***	0.2192	−1.6637 ***	0.0941	−6.381 ***	0.216
comm	0.3072 ***	0.0388	0.1937 ***	0.0293	0.1854 ***	0.0222	0.2396 ***	0.0278

Note: *** means significant at 1% level; ** means significant at 5% level. _cons represents the constant term.

.

Table 8 reflects that although power accessibility has a positive impact on the education level of individuals, it does not show statistical significance. Electricity accessibility reduces the unhealthy level significantly. According to Table 8, the probability of males completing compulsory education is significantly positive and higher than that of females, which is consistent with the traditional idea of preferring sons over daughters in rural areas. The improvement in householders’ education significantly promotes the probability of individuals completing compulsory education in the family, which also shows that there is a phenomenon of “humbled families are less likely to cultivate distinguished children” in rural areas. There is no significant difference in the completion of compulsory education between the central region and the eastern region, while the probability of completion of compulsory education in the western region is significantly lower than that in the eastern region. Unclean cooking energy in households decreases individual education and increases the probability of reporting one’s health as unhealthy. In both the health model and the compulsory education model, the coefficients of gender, age, household head’s education level, and regional variables all display significance. However, with the exception of an individual’s age (which holds the same level of significance in both models), the signs of the significance for gender, household head’s education level, and regional variables differ between the two models.

4.3. Heterogeneity Analysis

Table 9. Impact of electricity accessibility on expenditure and income in eastern, central, and western regions.

	Consumper(ln)
	Eastern Region		Central Region		Western Region
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	−0.576	1.2928	−12.57	10.764	10.642 ***	2.6801
_cons	10.374 ***	1.3786	20.759 **	9.813	0.288	2.2903
Cragg–Donald Wald F-statistic	12.345		1.538		18.082
Anderson canon (p-value)	0.000		0.214		0.000

Note: *** means significant at 1% level; ** means significant at 5% level.

reports the impact of electricity accessibility on expenditure and income in the eastern, central, and western regions. From Table 9, it can be seen that the impact of electricity accessibility on western households is more significant than that on eastern and central households, as the coefficients of electricity accessibility are larger and more positive for the western region in both the expenditure and income models.

The regression results in

Table 10. Quantile regression of electricity accessibility with respect to expenditure in eastern, central, and western regions.

	Eastern Region
	10%		25%		50%		75%		95%
Consumper(ln)	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.03	0.2392	−0.14	0.1155	−0.159 *	0.0868	−0.043	0.0954	0.2124	0.1914
_cons	9.057 ***	0.3853	9.4508 ***	0.1945	10.17 ***	0.2458	10.617 ***	0.2152	11.636 ***	0.4342
	Central Region
	10%		25%		50%		75%		95%
Consumper(ln)	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	−0.005	0.2143	0.0744	0.0565	−0.022	0.1147	−0.028	0.0816	0.2507	0.2366
_cons	8.356 ***	0.3374	8.8821 ***	0.2885	9.5327 ***	0.1523	10.091 ***	0.3157	10.934 ***	0.5018
	Western Region
	10%		25%		50%		75%		95%
Consumper(ln)	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	−0.026	0.099	0.0396	0.1096	−0.049	0.0887	−0.019	0.1064	0.1843 **	0.0854
_cons	8.7404 ***	0.2972	9.2874 ***	0.2441	9.6428 ***	0.2229	10.294 ***	0.2073	10.51 ***	0.4415

Note: *** means significant at 1% level; ** means significant at 5% level; * means significant at 10% level.

show that firstly, in the 95% expenditure quantile, power accessibility has no significant impact on per capita household expenditure in the eastern, central, and western regions. Above the 95% expenditure quantile, per capita household expenditure in the western region is significantly increased by power accessibility.

Table 11. Impact of electricity accessibility on wage income and operational income in Eastern, Midwestern, and Western China.

	Eastern Region
	Logarithm of Wage Income Per Capita		Logarithm of Per Capita Operating Income
	Coef.	Std. Err.	Coef.	Std. Err.
electr	1.7084	2.0601	9.187 *	4.866
_cons	−0.072 ***	0.0185	−0.177 ***	0.0438
Cragg–Donald Wald F-statistic	9.215		6.967
Anderson canon (p-value)	0		0
	Central Region
	Logarithm of Wage Income Per Capita		Logarithm of Per Capita Operating Income
	Coef.	Std. Err.	Coef.	Std. Err.
electr	−1.099	11.014	7.8064	6.5878
_cons	10.513	10.614	−0.737	6.1331
Cragg–Donald Wald F-statistic	0.426		2.653
Anderson canon (p-value)	0.513		0.102
	Western Region
	Logarithm of Wage Income Per Capita		Logarithm of Per Capita Operating Income
	Coef.	Std. Err.	Coef.	Std. Err.
electr	15.424 **	6.2489	1.9986	3.2587
_cons	−4.371	5.7605	4.7192 **	2.1811
Cragg–Donald Wald F-statistic	6.857		4.622
Anderson canon (p-value)	0.008		0.031

Note: *** means significant at 1% level; ** means significant at 5% level; * means significant at 10% level.

shows that the impact of electricity accessibility on wage income and operational income in the eastern, central, and western regions is intriguing. Electricity accessibility has no significant impact on wage income and operational income in the central region. Electricity accessibility increases wage income in the western region but has no effect on operational income in the western region. Electricity accessibility has no effect on wage income in the eastern region, but it promotes the growth of operational income in the eastern region. This may reflect that electricity accessibility stimulates the adventurous spirit of rural households in the eastern region and enables these non-conformist households to engage in non-agricultural self-employment or “enterprises” with employees. The availability of electricity makes rural households in the western region engaged in employment, thus promoting the increase in the per capita wage income of households.

Is there heterogeneity in the impact of energy poverty on education and health in the eastern, central, and western regions?

Table 12. Multilayer logit model of the impact of energy poverty on education and health in the eastern, central, and western regions.

	Eastern Region
	Individual above 16 Years Old Has Completed Junior School Education				Individuals Rated Their Health as Unhealthy
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	−0.293	0.218	−0.15	0.2454	−0.4770 ***	0.1702	−0.437 **	0.1853
_cons	−1.484 ***	0.2203	−2.188 ***	0.3581	−1.576 ***	0.1745	−6.058 ***	0.3768
	Central Region
	Individual above 16 Years Old Has Completed Junior School Education				Individuals Rated Their Health as Unhealthy
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.2257	0.31	0.1156	0.252	−0.439 ***	0.1673	−0.383 **	0.1835
_cons	0.2297 ***	−8.52	−2.561 ***	0.3874	−1.564 ***	0.1722	−7.309 ***	0.4437
	Western Region
	Individual above 16 Years Old Has Completed Junior School Education				Individuals Rated Their Health as Unhealthy
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.3612	0.2449	0.231	0.2607	−0.089	0.1427	−0.114	0.1557
_cons	−2.536 ***	0.2465	−3.336 ***	0.3837	−1.74 ***	0.1476	−5.962	0.3179

Note: *** means significant at 1% level; ** means significant at 5% level.

reports the impact of energy poverty on education and health in the east, central, and western regions estimated with the multi-level logit model.

Table 12 shows that electricity accessibility has no significant impact on whether individuals over 16 years old complete junior high school education in the eastern, central, and western regions. Electricity accessibility reduces the probability of ill-health self-rating in rural households in the eastern and central regions but has no impact on health in the western region. The use of household non-clean cooking energy reduces the educational level of all regions and increases the probability of self-rated ill health. This kind of energy poverty does not present regional heterogeneity in terms of education or health. It also shows that the promotion of household clean cooking energy throughout the country will undoubtedly be of great practical significance for the improvement in family welfare.

4.4. Robustness Analysis

To test the robustness of the results on the impact of energy poverty on education and health, in this section, this paper uses household monthly electricity bills as an alternative measure of electricity accessibility and uses the same model to test the robustness of the impact of energy poverty on education and health (see

Table 13. Robustness test of the impact of energy poverty on education and health.

	Individual above 16 Years Old Has Completed Junior School Education				Individuals Rated Their Health as Unhealthy
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
electr	0.0005 ***	0.0001	0.0002	0.0002	−0.002 ***	0.0002	−7 × 10−4 ***	0.0002
firewood	−0.6 ***	0.051	−0.244 ***	0.0555	0.5078 ***	0.0394	0.3551 ***	0.043
genderind			0.608 ***	0.0461			−0.569 ***	0.0362
ageind			0.0188 **	0.0082			0.1533 ***	0.0071
age2ind			−6 × 10−4 ***	1 × 10−4			−9 × 10−4 ***	6 × 10−5
eduhead2			0.328 ***	0.058			−0.336 ***	0.0454
eduhead3			2.7997 ***	0.0605			−0.501 ***	0.0738
eduhead4			2.294 ***	0.0934			−0.605 ***	0.1352
area2			−0.037	0.0764			0.0145	0.0748
area3			−0.296 ***	0.0733			0.1581 **	0.0691
_cons	−2.0147 ***	0.0399	−2.591 ***	0.1712	−1.786 ***	0.0399	−6.564 ***	0.1971
comm	0.2965 ***	0.0382	0.1956 ***	0.0298	0.1797 ***	0.0219	0.2346 ***	0.0275

Note: *** means significant at 1% level; ** means significant at 5% level.

).

Table 13 shows that although electric power accessibility has a positive impact on education without considering other variables, the impact of electric power accessibility on education is no longer significant when other control variables are included. Electric power accessibility still plays a significant role in reducing the self-evaluation of unhealthy status. Electricity accessibility shortens the medical service delivery time for who demands medical services and increases the efficiency of diagnosis and operation of suppliers. Unclean household cooking fuels increase the probability of unhealthy individuals and reduce their educational level. The symbols of other control variables are basically unchanged, which shows that the conclusions in this paper are robust.

5. Conclusions and Policy Implications

This paper selects five dimensions, household cooking fuel, lighting, household appliance services, entertainment/education, and communication, to construct a multi-dimensional energy poverty index and investigates the determinants of multi-dimensional energy poverty. The measurement results from micro-survey data indicate that both the overall sample and the sub-group samples show a general downward trend in the multi-dimensional energy poverty index. The rural multi-dimensional energy poverty index is higher than the urban index, and the index in the eastern region is significantly lower than that in the central and western regions. The contribution rate of household cooking fuel steadily decreases, while the contribution rate of lighting is consistently below 1% across various years. The contribution rate of household appliance services (measured according to the presence of refrigerators) steadily declines, with the communication dimension contributing significantly more than other dimensions. Variables such as household size, household head’s characteristics, and residential area have a significant impact on multi-dimensional energy poverty. After addressing endogeneity with a two-stage least squares model, the findings highlight a significantly positive influence of electricity accessibility on enhancing expenditure and income levels in rural regions. The stratified logit model indicates that while electricity accessibility positively influences the educational attainment of rural individuals, it lacks statistical significance. Nevertheless, electricity accessibility leads to a significant reduction in rural health issues. The use of unclean cooking energy sources in rural households leads to reduced educational attainment for individuals and an increased likelihood of health problems.

There are still several challenges and obstacles to achieving universal electrification and energy security in rural areas. These include the high costs of extending the power grid, problems with electricity supply quality and reliability, lack of awareness and affordability of clean cooking fuels, and unequal distribution of energy resources across regions. Therefore, this paper suggests policy recommendations drawn from empirical findings.

Firstly, the government should persist in funding rural electrification projects, especially in remote and mountainous areas where connecting to the main power grid is challenging or economically impractical. Independent renewable energy setups, like solar photovoltaic, wind power generation, micro-hydro, and biogas, can reliably supply clean electricity to rural homes. Additionally, the government should provide subsidies, incentives, and technical assistance to promote the adoption and maintenance of these systems.

Secondly, the government needs to enhance electricity supply quality and reliability in rural regions. This entails upgrading infrastructure, improving management, and reinforcing oversight of the rural power grid. Frequent power outages and voltage fluctuations can damage electrical appliances, disrupting production and the daily activities of residents. Furthermore, the government should enforce standards and regulations for rural electricity services, including metering, billing, pricing, and addressing customer complaints.

Thirdly, it is important for the government to encourage the transition from using traditional biomass to adopting cleaner cooking fuels in rural regions. Fuels could include options such as liquefied petroleum gas (LPG), natural gas, electricity, or biogas. The utilization of solid fuels like firewood in cooking results in indoor air pollution, health problems, environmental deterioration, and greenhouse-gas emissions. To facilitate this transition, the government ought to offer subsidies or vouchers to reduce the upfront and ongoing expenses associated with clean cooking fuels. Moreover, it is essential to enhance understanding and offer education on the benefits and safety of these options for rural inhabitants.

Lastly, it is essential for the government to tackle regional disparities concerning energy access and poverty alleviation. Tailored and differentiated policies should be implemented based on local conditions and needs. While the eastern regions have achieved relatively higher levels of electrification and energy expenditure, the central and western regions still grapple with significant energy poverty and deprivation. The government ought to allocate more resources and support to underdeveloped areas while fostering collaborative and integrated development across regions.