Regional Heterogeneity Analysis of Residential Electricity Consumption in Chinese Cities: Based on Spatial Measurement Models

Abstract

The share of electricity consumption by urban and rural residents in China’s overall electricity consumption is very close to that of the tertiary sector, which has become an important driver of China’s electricity consumption growth. At the same time, due to the mismatch between China’s regional resource endowments and the level of regional development, the regional supply and demand situation for electricity varies. Therefore, it is urgent to clarify the regional differences in residential electricity consumption and the factors affecting it, and accordingly adopt targeted and feasible measures to regulate residential electricity consumption. This article includes data from 285 Chinese prefecture-level cities from 2006 to 2019, and adopts a “three lines” method of region-partitioning (Qinling–Huaihe line, Huhuanyong line, and Shanhaiguan line) to divide four regions. We used spatial econometric models to examine residential electricity consumption and its influencing factors in China from the standpoint of regional heterogeneity. The results show that there is significant regional heterogeneity in residential electricity consumption in China, and the difference between the north of the Shanhaiguan line and other areas is significant. Moreover, there is a positive spatial correlation in the per capita domestic electricity consumption of urban residents, and each influencing factor has obvious regional heterogeneity, among which household appliances are the significant influencing factor. Based on the regional heterogeneity of residential electricity consumption, management measures should be formulated according to local conditions, and the supply of electricity should be ensured by strengthening multidimensional initiatives.

1. Introduction

Recent years have witnessed a green transformation from high-speed growth to high-quality economic development in China, among which ecological conservation is the top priority. At the UN General Assembly on 22 September 2022, President Xi Jinping announced China’s commitment to achieving carbon peaking by 2030 and carbon neutrality by 2060 (hereinafter “dual carbon” goals). Since then, China has undertaken relevant efforts in emissions reduction as part of its commitment to decarbonization.

The production activities of the secondary industry, as the primary source of carbon emissions, have received great public attention. A large number of policy adjustments and technological innovation studies have been carried out. However, the increasing living standard of Chinese people brought by urbanization gives rise to a dramatic increase in the energy consumption of the residential sector [1]. As one of the most widely used terminal energy sources, electric power occupies a large proportion on the consumer side. However, there is still much room for studying Chinese residential electricity consumption. Previous studies have shown that as societies and economies in developed countries improve, the increase in household income will lead to an increase in the proportion of residential electricity consumption in the total electricity consumption [2,3]. That said, since residential electricity consumption will eventually represent a significant source of carbon emissions, it is crucial to regulate residential electricity consumption. On the other hand, the government ought to cautiously formulate relevant policies as residential electricity consumption is closely related to people’s livelihood. Adopting measures tailored for industry directly to household electricity consumption cannot achieve desirable outcomes. Therefore, in order to pursue ecological conservation while meeting people’s demands for a better life, attention should also be paid to regulating residential electricity consumption while focusing on controlling the carbon emission of energy-intensive industries.

Residential electricity consumption is greatly affected by climate conditions and regional economic development [4]. Considering the vast territory of China and the difference in the natural and economic conditions of various regions, it is necessary to take into account regional differences in order to regulate the residents’ electricity consumption, so that it can not only meet the people’s needs for a better life, but also meet the requirements of carbon reduction and emission reduction. In order to highlight the distinct features of various regions in China, this paper divided China into four regions by adopting the “three lines” partitioning method, namely the Qinling–Huaihe line, Huhuanyong line, and Shanhaiguan line. First, the Qinling–Huaihe line is the division line to distinguish between northern and southern China. The northern and southern regions divided by the line differ from each other in climate, geography, and customs. Moreover, it is also the 0 °C average January temperature line, 800 mm annual precipitation line, subtropical monsoon climate, and temperate monsoon climate line. Considering the influence of geographical factors on residential electricity consumption, it is necessary to analyze the heterogeneity of prefecture-level cities on both sides of this line. Second, the Huhuanyong line, also known as the Heihe–Tengchong line, is a line put forward by the geographer Hu Huanyong that divides the area of China into two parts with contrasting population densities. It coincides with China’s ecologically sensitive zone proposed by Wang et al. [5]. Thus, this paper intends to use this line as one of the criteria for classification. Third, the east of the Shanhaiguan line represents northeastern China, which is the country’s largest old industrial base. The residential electricity consumption in northeast China is also of great significance. Therefore, this paper aims to investigate two questions by using the “three lines” partitioning method: (1) the spatial differences of urban residents’ electricity consumption and its underlying reasons, and (2) the regional heterogeneity of influencing factors of urban residents’ electricity consumption. The main marginal contribution of this paper is that it is the first time that China is divided into four regions by the “three-line method” to study the regional heterogeneity of residential electricity consumption, which not only expands the content of research on the analysis of the heterogeneity of residential electricity consumption, but also further enriches the research system of policies that take into account both carbon emissions and residents’ demand in the context of dual-carbon.

The remainder of this paper is organized as follows. Section 2 reviews previous studies’ findings of residential electricity consumption; Section 3 presents the variables and models of this paper, and explains the data sources and processing methods used; Section 4 introduces the spatial difference analysis of residential electricity consumption based on “three lines”; Section 5 shows the spatial econometric empirical analysis on the residential electricity consumption, which studies the spatial correlation of residential electricity consumption and the heterogeneity of influencing factors in different regions, and then analyzes the influencing mechanism of each factor; Section 6 provides conclusions and policy recommendations based on the results.

2. Literature Review

As China is in a period of rapid urbanization, the residential sector is responsible for a large proportion of energy consumption. The proportion of electricity consumption keeps increasing, and studies on electricity demand and its influencing factors have received considerable attention [6]. In order to explore the changing pattern and influencing factors of residential electricity consumption, previous research has shed light on the characteristics of residential electricity demand and its influencing factors by employing different approaches.

First and foremost, in terms of data, in view of the diversity and complexity of residential electricity consumption behavior, some studies chose to use micro household survey data rather than panel data which contained dynamic information [7,8,9]. On the other hand, some scholars chose to apply macro panel data to reveal the spatial and temporal change in samples while sacrificing the accuracy of residents’ behavior [10,11,12].

Secondly, a area of growing literature has focused on the impact of economic, social and family, climate, and geographical factors on residential energy consumption. Of all of these factors, economic factors have attracted the greatest attention, and research on residential electricity demand mostly focuses on electricity price and household income [13,14,15]. For example, Zhang et al. [16] and Tang and Tan [17] proved that income has a positive impact on electricity consumption in China and Malaysia. Lin and Liu [18] discussed the impact of Chinese residents’ income on household appliance use, indicating that the improvement in residents’ income has a positive effect on household appliance consumption and electricity consumption. But as the overall resident’s income rises to a certain extent, the impact of income on household appliance consumption gradually weakens. Guang et al. [15] used LMDI to study the influencing factors of the growth rate of China’s residential electricity consumption, showing that the decrease in the growth rate of electricity consumption was related to the adjustment of industrial structure and the decrease in the power intensity of residents’ income. However, given that China’s residential electricity price is regulated by the government, relevant studies showed either the poor significance of the electricity price variable or the lack of price elasticity of Chinese household electricity demand [13,14]. A number of studies also concentrated on sociological and psychological factors. Guo et al. [19] analyzed and sorted out the influencing factors of residential electricity consumption based on social psychological theories. Park and Yun [20] adopted the spatial panel model to analyze the impact of social structure and housing type on the electricity consumption of South Korean residents, revealing that there was a spatial effect on domestic electricity consumption. The impact of aging on electricity demand changed with the local tax revenue while housing has a significant impact on electricity consumption. Additionally, climate and geographical factors should also be taken into consideration. Du et al. [13] used the partial linear function coefficient panel model to study the impact of climate change on China’s residential electricity demand, showing that the impact of climate change on cooling electricity consumption was more significant than that on heating electricity consumption. Sheng et al. [21] also considered climate variables such as high-temperature days and the sultry index when studying the influencing factors of residential electricity consumption in Beijing. Wang et al. [22] explored the features of residential electricity consumption in “hot summer and cold winter areas” in China by employing the quantile regression model, but it failed to conduct a comparative analysis with other areas.

Lastly, as the ultimate goal of studying residential electricity consumption is to formulate policies to regulate residential electricity consumption, it is imperative to fully grasp the electricity consumption features and mechanisms of each influencing factor under different heterogeneous perspectives. Du et al. [13] discovered that different income levels would affect the sensitivity of residential electricity consumption to climate change. Wang et al. [12] studied the differences in the distribution and convergence of electricity consumption between urban and rural areas in China, indicating that the residential electricity consumption in rural areas has a significant positive spatial autocorrelation, while this was not the case in urban areas. In addition, the convergence rate of σ and β in rural areas was faster than that in urban areas, and rural and urban areas would eventually converge to different equilibrium levels. Lin and Wang [14] studied the electricity elasticity of Chinese prefecture-level cities’ residents, proposing that multiple pricing schemes should be implemented according to regional features in an effort to promote emission reduction in the residential sector. In order to present it more visually, we used

Table 1. Literature summary.

Thesis Title	Authors	Particular Year	Research Topics
Electricity consumption in G7 countries: A panel cointegration analysis of residential demand elasticities	Narayan et al. [23]	2007	Impact of electricity price and household income on residential electricity consumption
Residential consumption of gas and electricity in the US: The role of prices and income	Alberini et al. [7]	2011
Effects of demand side management on Chinese household electricity consumption: Empirical findings from Chinese household survey	Khanna et al. [9]	2016
Effects of demand side management on Chinese household electricity consumption: Empirical findings from Chinese household survey	Du et al. [13]	2020
Analyzing the elasticity and subsidy to reform the residential electricity tariffs in China	Lin and Wang [14]	2020
Energy efficiency improvements and industry transition: An analysis of China’s electricity consumption	Guang et al. [15]	2022
Principal component analysis of electricity consumption factors in China	Zhang et al. [16]	2012	Impact of income on electricity consumption
Exploring the nexus of electricity consumption, economic growth, energy prices and technology innovation in Malaysia	Tang and Tan [17]	2013
Impacts of income and urbanization on urban home appliance consumption	Lin and Liu [18]	2016	Impact of residential income on appliance use
Residential electricity consumption behavior: Influencing factors, related theories and intervention strategies	Guo et al. [19]	2018	Factors affecting residential electricity consumption
Social determinants of residential electricity consumption in Korea: Findings from a spatial panel model	Park and Yun [20]	2022
Climatic impact on China’s residential electricity consumption: Does the income level matter?	Du et al. [13]	2020
Heterogeneous Study of Multiple Disturbance Factors Outside Residential Electricity Consumption: A Case Study of Beijing	Sheng et al. [21]	2021
Heating demand with heterogeneity in residential households in the hot summer and cold winter climate zone in China—A quantile regression approach	Wang et al. [22]	2022	Features of residential electricity consumption

to sort the above references.

In addition to vast land territory, China spans five geographical time zones and many degrees of latitude with varied climate conditions. The different living conditions of Chinese residents make it difficult to apply a “one-size-fits-all” approach to regulate and control residential electricity consumption. In view of this, regional heterogeneity should not be ignored in the study of domestic electricity consumption in China. However, previous studies have focused on consumption at the provincial level and attached more importance to income heterogeneity and urban–rural heterogeneity. The regional heterogeneity of residential electricity consumption in China is rarely considered in research on energy consumption. That is the essential goal of this article, which aims to focus on the regional heterogeneity of China’s urban residents’ per capita daily electricity consumption and discover the distribution and differences of residential electricity consumption by applying “three lines” to divide the country into four regions. The spatial effects of important influencing factors on residential electricity consumption are analyzed, and then policy recommendations are put forward based on regional heterogeneity.

3. Model Building and Data Sources

3.1. Model Construction

Residential electricity consumption ($PRE$) is chosen as the explained variable, since this paper studies the geographical variation in residential electricity consumption in China.

Given the regulation of residential electricity prices by the Chinese government, the weak price elasticity of residential electricity demand [13,14], the rebound effect of electricity demand [24,25,26], and the availability of electricity price data for prefecture-level cities in China, the electricity price variable is not considered in the analysis model in this paper. Instead, this study will perform analysis by taking three aspects of factors into consideration, namely economic factors, social and family factors, and climate and geography factors. Then, four explanatory variables are used in this paper: income ($inc$), the average number of hot days per year ($hd$), emphasis on local education ($edu$), and average household size ($hsc$). The average household size ($hsc$) is the ratio of the local population and the number of households. The emphasis on local education ($edu$) is expressed by the proportion of local fiscal expenditure on education to the total fiscal expenditure. The per capita disposable income is used to measure income ($inc$). The number of days with an average daily temperature higher than 20 degrees Celsius is applied to calculate the average number of hot days per year ($hd$).

Based on the above variables, this paper constructs the spatial lag model (SAR), the spatial error model (SEM), and the spatial dubin model (SDM) to analyze residential electricity consumption and its influencing factors. The models are shown as follows:

$$\begin{array}{l}lnPR{E}_{it}=\rho \mathit{W}lnPR{E}_{it}+{\beta }_{1}lnin{c}_{it}+{\beta }_{2}lnh{d}_{it}+{\beta }_{3}lned{u}_{it}\\ +{\beta }_{4}lnhs{c}_{it}+{\mu }_{it}\end{array}$$

(1)

$$\begin{array}{l}lnPR{E}_{it}={\beta }_{1}lnin{c}_{it}+{\beta }_{2}lnh{d}_{it}+{\beta }_{3}lned{u}_{it}+{\beta }_{4}lnhs{c}_{it}+{\mu }_{it}\\ {\mu }_{it}=\lambda \mathit{W}{\mu }_{it}+{\epsilon }_{it}\end{array}$$

(2)

$$\begin{array}{l}lnPR{E}_{it}=\rho \mathit{W}lnPR{E}_{it}+{\beta }_{1}lnin{c}_{it}+{\beta }_{2}lnh{d}_{it}+{\beta }_{3}lned{u}_{it}\\ +{\beta }_{4}lnhs{c}_{it}+{\theta }_1\mathit{W}lnin{c}_{it}+{\theta }_2\mathit{W}lnh{d}_{it}+{\theta }_3\mathit{W}lned{u}_{it}+{\theta }_4\mathit{W}lnhs{c}_{it}\\ +{\mu }_{it}\end{array}$$

(3)

Model (1)–(3) are SAR, SEM, and SDM, respectively, where $PRE$ is the per capita residential electricity consumption, $\mathit{W}$ represents the spatial weight matrix, $inc$ refers to income, $hd$ stands for the average number of hot days per year, $edu$ expresses the emphasis on local education, $hsc$ indicates the average household size, $\mu$ and $\epsilon$ are random disturbance terms, ${\beta }_i$ is the coefficient of each explanatory variable, $\rho$ and $\lambda$ are, respectively, the spatial autoregressive coefficients of the explained variable and the random disturbance term, and ${\theta }_i$ is the spatial lag coefficient.

3.2. Data Sources

Considering the availability of data, some prefecture-level cities with serious data deficiency are excluded, and the year span of the data is set from 2006 to 2019. Among them, residential electricity consumption is represented by the per capita residential electricity consumption data from 2006 to 2019 in the China City Statistical Yearbook. The data of the registered population in the corresponding years are used to calculate the years in which only the total residential electricity consumption is counted, and the interpolation method is used to complete the data. The data on the registered population, number of households, local financial expenditure on education, total local financial expenditure, and per capita disposable income are all from the China City Statistical Yearbook. By searching the statistical yearbook of each province and city and the CEIC database, the missing part is completed with the moving average method and interpolation method. Based on the method of Chen and Zhang [27], the data of the average diurnal temperature are collected by using the product “Daily Meteorological Dataset of Basic Meteorological Elements of China National Surface Weather Station (V3.0)” in the daily value monitoring of meteorological stations provided by the “National Meteorological Information Center”, and data matching is completed by comparing the longitude and latitude of meteorological stations and prefecture-level cities. The descriptive statistical results of each variable are shown in

Table 2. Descriptive statistics results.

Variable	Unit	Obs	Mean	Std. Dev.
PRE	Kwh	3990	443.603	464.82
inc	CNY	3990	18,393.955	6961.16
edu	%	3990	18.174	4.295
hsc	Person/household	3990	3.194	0.463
hd	Days	3990	146.3	55.494

.

4. Spatial Difference Analysis of Residential Electricity Consumption

4.1. Introduction to “Three Lines” Partitioning Method

Considering the vast territory of China and the difference in natural and economic conditions of various regions, it is necessary to take into account spatial differences. Unlike other studies which divided China into three regional groups: eastern China, central China, and western China [28,29,30], 285 prefecture-level cities of China are divided into four regions in this paper by the Qinling–Huaihe line, Huhuanyong line, and Shanhaiguan line, as shown in Figure 1 and Appendix A. The rest of the paper will follow Figure 2.

4.2. Spatial Difference in Electricity Consumption

The per capita residential electricity consumption in the four regions is shown in Figure 3. It can be found that the electricity consumption of the four regions is increasing year by year. Located to the south of the Qinling–Huaihe line, the electricity consumption and its growth rate in Region 3 are significantly higher than the other three due to the rapid economic development and hot weather, while the electricity consumption to the north of the Qinling–Huaihe line is always the lowest. Although the electricity consumption of Region 1 is basically the same as that of southern China at the beginning of the sample period, its growth rate is the lowest one among the four regions from the whole sample period, which may be related to the serious population loss and the lack of economic growth momentum in Northeast China in recent years. In addition to the south of Qinling–Huaihe, the growth rate of the per capita residential electricity consumption is fastest in Region 4, whose electricity consumption is only lower than the south, which shows the remarkable achievements of the western development strategy to improve the development of the power infrastructure of western China to effectively satisfy the electricity demand of residents here.

In order to further analyze the regional differences in residential electricity consumption among different regions, Dagum’s Gini coefficient is used to calculate the differences among different regional groups. The inter-group differences and changes are shown in Figure 4. G_ij represents the inter-group Gini coefficient between Region i and Region j. It can be found that the inter-group differences of each group are all higher than 0.3, indicating that there are obvious regional differences in the “three-line” partitioning method, thus showing this paper’s research value. Among them, the curves G₁₂, G₁₃, and G₁₄ are all higher than the other three curves that do not involve Region 1, meaning that there are great differences between the both sides of the Shanhaiguan line.

Region 1 is located on the northeast side of the Shanhaiguan line with three characteristics: (1) it has long and extremely cold winters, and electricity accounts for a very small proportion of the energy used for heating in winter compared to other regions; (2) it faces serious problems such as population loss and aging, and the overall electricity consumption habits of residents differ greatly from those of the other three; (3) affected by industrial restructuring, the economic development of this region lacks drivers in recent years compared to others, and a positive relationship between the residential electricity demand and economic development has been identified by some studies [22]. Therefore, the characteristics of residential electricity consumption behavior vary greatly between both sides of the Shanhaiguan line.

5. Results of the Spatial Econometric Empirical Analysis

5.1. Spatial Autocorrelation Test

In order to fully compare the differences of residential electricity consumption and its influencing factors among the four regions, this paper uses the spatial econometric model for analysis. Before the analysis, the spatial autocorrelation test is necessary, so the Moran test is chosen in this paper. In terms of spatial weight matrix, this paper uses four spatial weight matrices, namely the geographic distance matrix (${\mathit{W}}_d$), economic distance matrix (${\mathit{W}}_e$), geoeconomic distance matrix (${\mathit{W}}_{de}$), and geoeconomic distance nested matrix (${\mathit{W}}_{dei}$), by referring to the practice of Shao et al. [31]. For the geographical distance matrix, the test statistic is calculated as the Euclidean distance calculated from the longitude and latitude data of each prefecture-level city. The test statistic in the economic distance matrix is the inverse of the absolute value of the difference value among the mean GDP values of the corresponding prefecture-level cities, which are calculated as ${\omega }_{ij}=1/\left|GD{P}_i-GD{P}_j\right|$. The test statistic of the geoeconomic distance matrix is the product of the inverse of the Euclidean distance between city i and city j and the proportion of the annual mean per capita GDP of city i in the annual mean per capita GDP of all cities. The test statistic in the geoeconomic distance nested matrix is calculated as ${\mathit{W}}_{dei}=\lambda {\mathit{W}}_d+\left(1-\lambda \right){\mathit{W}}_e$ by referring to Zhang and Zhu [32], and the analysis is simplified according to the method of Shao et al. (2016) [31]; that is, $\lambda =0.5$. The test results are shown in

Table 3. Results of the spatial autocorrelation test.

	Geographic Distance Matrix (Wd)		Economic Distance Matrix (We)		Geoeconomic Distance Matrix (Wde)		Geoeconomic Distance Nested Matrix (Wdei)
	I	P	I	P	I	P	I	P
2006	0.203	0.000	0.252	0.000	0.070	0.000	0.204	0.000
2007	0.239	0.000	0.241	0.000	0.084	0.000	0.239	0.000
2008	0.223	0.000	0.263	0.000	0.078	0.000	0.224	0.000
2009	0.199	0.000	0.296	0.000	0.069	0.000	0.200	0.000
2010	0.194	0.000	0.306	0.000	0.067	0.000	0.197	0.000
2011	0.202	0.000	0.302	0.000	0.070	0.000	0.205	0.000
2012	0.204	0.000	0.285	0.000	0.071	0.000	0.205	0.000
2013	0.224	0.000	0.307	0.000	0.077	0.000	0.225	0.000
2014	0.234	0.000	0.279	0.000	0.082	0.000	0.235	0.000
2015	0.239	0.000	0.264	0.000	0.085	0.000	0.240	0.000
2016	0.229	0.000	0.305	0.000	0.080	0.000	0.231	0.000
2017	0.208	0.000	0.317	0.000	0.073	0.000	0.209	0.000
2018	0.183	0.000	0.322	0.000	0.064	0.000	0.184	0.000
2019	0.159	0.000	0.318	0.000	0.057	0.000	0.161	0.000

.

The test results reveal that the p-values of each year under the four spatial weight matrices are all 0, which rejects the null hypothesis of no spatial correlation. The per capita residential electricity consumption shows spatial correlation through the global Moran test. Considering the subsequent robustness test, the Moran test is also conducted on the residential electricity consumption per unit area ($AreaE$), and the test results also shed light on the spatial correlation. Considering this paper’s length, only the spatial autocorrelation test results of the per capita residential electricity consumption are presented here.

5.2. LM Test

Before the estimation of model parameters, appropriate models need to be selected from the spatial lag model (SAR), spatial error model (SEM), and spatial dubin model (SDM). This paper intends to complete model selection based on the results of the LM test and robust-LM test. Firstly, the LM statistics of the SAR model and SEM model are compared, and the results show that both the SAR model and SEM model pass the LM test and the robust-LM test with high significance. In this case, this paper adopts the method used by most scholars; that is, the spatial dubin model is used. At the same time, given that the other two models show a high significance in the LM and robust-LM test, they are included in the follow-up study together to compare the results of the three models. The results are shown in

Table 4. Test results of LM and robust-LM.

Test	Statistics	p-Values
SEM
LM	92.771	0.000
Robust-LM	57.992	0.000
SAR
LM	41.377	0.000
Robust-LM	6.599	0.010

.

5.3. Result Analysis

Firstly, the spatial econometric regression analysis of the per capita residential electricity consumption at the national level is carried out to observe the influencing factors at the national level and verify the rationality of the model.

(1)

Benchmark regression

Three spatial econometric models, namely SDM, SAR, and SEM, are used here, and the above four spatial weight matrices are respectively applied, then the maximum likelihood estimation is adopted. The regression results are shown in

Table 5. National regression results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
	Wd			We			Wde			Wdei
	SDM	SAR	SEM	SDM	SAR	SEM	SDM	SAR	SEM	SDM	SAR	SEM
lninc	−0.00957	0.118 ***	0.183 ***	0.0204	0.125 ***	0.200 ***	−0.00419	0.0296 ***	0.00443	−0.00970	0.117 ***	0.181 ***
lnhd	0.0122	0.0261 **	0.0266 **	0.0269 **	0.0313 ***	0.0340 ***	0.0148	0.0206 **	0.0183 *	0.0128	0.0263 **	0.0271 **
lnedu	0.0536 ***	0.0323 **	0.0357 **	0.0391 ***	0.0275 *	0.0199	0.0574 ***	0.0486 ***	0.0590 ***	0.0528 ***	0.0324 **	0.0355 **
lnhsc	−0.112 *	−0.0877	−0.138 **	−0.0419	−0.0619	−0.0770	−0.114 *	−0.0647	−0.0916	−0.114*	−0.0877	−0.139 **
ρ	0.312 ***	0.464 ***		0.354 ***	0.476 ***		0.750 ***	0.875 ***		0.316 ***	0.470 ***
λ			0.426 ***			0.375 ***			0.945 ***			0.435 ***
wlninc	0.198 ***			0.151 ***			0.109 ***			0.198 ***
wlnhd	0.0645 *			0.0273			0.0714 *			0.0629 *
wlnedu	−0.0767 **			−0.00523			−0.0659 **			−0.0741 **
wlnhsc	0.297 **			0.00309			0.589 **			0.306 **
Individual fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	3990	3990	3990	3990	3990	3990	3990	3990	3990	3990	3990	3990
R2	0.141	0.259	0.308	0.238	0.297	0.327	0.0980	0.104	0.000	0.142	0.259	0.307

* p < 0.1, ** p < 0.05, *** p < 0.01.

.

The results are summarized in Table 5. On the basis of the aforementioned Moran test, it is further proved that the residential electricity consumption shows a positive spatial correlation at the national level, since both the spatial autocorrelation coefficient $\rho$ of per capita residential electricity consumption and the spatial autocorrelation coefficient $\lambda$ of random disturbance term in each model are significantly positive.

By comparing the SDM, SAR, and SEM models, the significance of the key variable $inc$ is above 1% only when the SAR model is used in the four spatial weight matrices, indicating that the robustness of the SAR model is better than the other two models. By further comparing the SAR model regression results in the four spatial weight matrices, it can be found that when the economic distance matrix ${\mathit{W}}_e$ is used (Column (5)), the value of R² is the largest, indicating that the explanatory variables in the model are opposite to the explained variables at this time. In view of this, the subsequent analysis of this paper is based on the application of the SAR model in the economic distance matrix.

The significance of $inc$ and $hd$ in Column (5) is strong, showing that they are the key variables influencing the residential electricity consumption, which is consistent with the assumption. Although the significance of $edu$ is less strong than that of the first two factors, it is still considerably positive. The regression coefficient of $hsc$ is negative, as expected, but not significantly. The possible explanations are: (1) This paper uses the national household registration statistics when calculating the $hsc$, but the household registration system in China cannot comprehensively reflect the actual number; (2) Different economic development and geographical and climatic conditions in different regions lead to the different impact of $hsc$ on residential electricity consumption.

(2)

Heterogeneity analysis

In order to further analyze the regional heterogeneity of the influencing factors, the regression analysis of the four regions is carried out, and the regression results are shown in

Table 6. Regional regression results.

	Region 1	Region 2	Region 3	Region 4
lninc	0.224 ***	0.206 ***	0.0984 ***	0.125 ***
lnhd	0.0782 **	0.0524 ***	0.0258 *	0.0106
lnedu	−0.0144	−0.00102	0.0121	0.0634 **
lnhsc	0.335 **	0.0292	−0.0917	−0.347 ***
ρ	0.225 ***	0.138 **	0.595 ***	0.308 ***
Individual fixed effects	Yes	Yes	Yes	Yes
N	476	1064	2072	378
R2	0.443	0.367	0.299	0.298

* p < 0.1, ** p < 0.05, *** p < 0.01.

.

First, the spatial autocorrelation coefficient $\rho$ of the explained variables in all four regions is significantly positive, consistent with the results at the national level. The reason why the largest absolute value is in Region 3 and the smallest absolute value is in Region 2 is because there are more economically-developed big cities in Region 3, which has a strong economic agglomeration effect with an average higher annual temperature. Therefore, the positive spatial correlation of the residential electricity consumption in Region 3 is stronger than that in other regions. In contrast, the weak economic agglomeration effect in Region 2 leads to the weak spatial correlation.

Secondly, in terms of explanatory variables, the coefficients of $inc$ and $hd$ are significantly positive, among which the coefficient value of Region 3 is the smallest, only 0.0984, which indicates that the impact of the income effect on Region 3 is the weakest. Since the coefficient of $hd$ in Region 4 is the only one that is not significant, the possible explanation relates to the climatic characteristics of large daily temperature differences in some regions on the west of the Huhuanyong line. The $edu$ is only significantly positive in Region 4. Considering that there is a significant gap in the local education development between here and east of the Huhuanyong line, it is tentatively concluded that the education effect is related to the local education development. The coefficient of $hsc$ is significantly positive in Region 1 and negative in Region 4, which may be related to population density and population structure. According to the above estimation results, it is preliminary found that there is heterogeneity in the influencing factors of the per capita residential electricity consumption among regions. However, due to the spatial autoregressive term, the coefficient of explanatory variables in this model cannot accurately reflect its marginal effect on the explained variables, and effect decomposition is needed to identify the actual impact of each variable. Therefore, the economic significance of the regression coefficients here is not analyzed in detail [33,34].

5.4. Effect Decomposition

In order to further study the influence of each explanatory variable on the per capita residential electricity consumption, the average direct effect and indirect effect are used to describe the influence of each explanatory variable on the explained variable by referring to the effect decomposition method proposed by LeSage and Pace [33]. The results of effect decomposition at the national level and regional level are shown in

Table 7. Results of effect decomposition among regions.

		Whole Nation	Region 1	Region 2	Region 3	Region 4
Direct effect	lninc	0.130 ***	0.228 ***	0.208 ***	0.106 ***	0.129 ***
	lnhd	0.0293 ***	0.0780 **	0.0519 ***	0.0269 *	0.0103
	lnedu	0.0298 **	−0.0113	0.00125	0.0153	0.0679 **
	lnhsc	−0.0643	0.339 **	0.0333	−0.0982	−0.361 ***
Indirect effect	lninc	0.110 ***	0.0646 **	0.0330 **	0.138 ***	0.0547 ***
	lnhd	0.0256 **	0.0218 *	0.00871	0.0383	0.00463
	lnedu	0.0263 *	−0.00315	0.000610	0.0219	0.0295 *
	lnhsc	−0.0573	0.101	0.00705	−0.138	−0.156*
Total effect	lninc	0.240 ***	0.292 ***	0.241 ***	0.244 ***	0.184 ***
	lnhd	0.0549 ***	0.0998 **	0.0606 ***	0.0652	0.0149
	lnedu	0.0561 **	−0.0145	0.00185	0.0373	0.0974 **
	lnhsc	−0.122	0.440 **	0.0403	−0.237	−0.517 ***
	N	3990	476	1064	2072	378
	R2	0.290	0.443	0.367	0.299	0.298

* p < 0.1, ** p < 0.05, *** p < 0.01.

.

(1)

Per capita disposable income

When using the economic distance matrix, both the direct and indirect effects of the key variables per capita disposable income are positive at the significance level of 1%, indicating that the increase in per capita disposable income has a positive effect on the per capita residential electricity consumption in local and neighboring areas. By comparing the effect numbers, it can be found that the direct effect of per capita disposable income in Regions 1, 2, and 4 is larger than that in Region 3, and their indirect effect is smaller than that in Region 3. The possible reasons are: (1) Region 3 is located to the south of the Qinling–Huaihe line and to the east of the Huhuanyong line, with advanced economic development and a relatively higher per capita disposable income. Under the influence of diminishing marginal effects, the direct effect of per capita disposable income is weakened. (2) The economic agglomeration level of Region 3 is higher, while economic agglomeration has a positive effect on people’s income [35], and the increase in the individual per capita disposable income in Region 3 has a positive influence on the residential electricity consumption of the adjacent cities, thus leading to the larger indirect effect in this area.

(2)

The average number of hot days per year

Compared with per capita disposable income, the absolute values of the three effects of the average number of hot days per year are smaller at both the national and regional levels. Among them, the direct effects of Regions 1, 2, and 3 are all significantly positive, which is consistent with the expectation that “the more hot days in the local area, the more electricity used by residents for cooling”. Although the direct effect of Region 4 is positive, it is not significant. There are two possible reasons: (1) Due to the vast territory, small population density, and relatively scattered urban distribution of the west side of the Huhuanyong line, the data of meteorological stations may not accurately reflect the relationship between temperature and residential electricity consumption in each city. (2) Region 4 is located on the west side of the Huhuanyong line; that is, the west side of the ecological fragile zone [5]. The difference in the environment compared to the east Huhuanyong line is apparent. Due to the large elevation span, various landforms such as plateau, desert, and basin, as well as bigger daily and seasonal temperature changes compared with other regions, the average daily temperature may not be able to effectively reflect the residential electricity consumption. Therefore, the direct effect of the average number of hot days per year on residential electricity consumption is not significant.

The significance of the spillover effect is not great. Although the results at the national level are positive at the significance level of 5%, only Region 1 is positive at the significance level of 10% at the regional level, while the results in the rest of the regions are also positive but not significant. There are two reasons for this. First, according to geographic common sense, the higher temperature in city A will lead to a higher temperature in the neighboring areas, which will also increase the corresponding electricity consumption for cooling; however, the increase in electricity consumption at this time is not caused by the high temperature in city A, but by the geographic characteristics of the neighboring areas themselves, so the spillover effect here is generally less significant. Second, since the main production areas of electricity are the northeast and the west of China, and the electricity demand gap of Regions 2 and 3 to the east of the Huhuanyong line needs to be filled by purchasing electricity from Regions 1 and 4, indicating that the electricity demand in Regions 2 and 3 is constrained by the supply in Regions 1 and 4, and the spillover effect of hot days is not significant. Meanwhile, the low overall electricity consumption in western China and the large western region being sparsely populated, in conjunction with a relatively scattered urban distribution, lead to less spillover effect of high-temperature days between neighboring cities.

(3)

Emphasis on local education

At the national level, the direct and spillover effects of emphasis on local education are significantly positive; at the regional level, the direct and spillover effects are significant only in Region 4 which is located to the west of the Huhuanyong line, but not significant in the rest of the regions. Firstly, although an emphasis on local education may reduce residential electricity consumption by enhancing residents’ awareness of energy saving, the improvement of residents’ literacy level will also increase the demand for quality of life in residents’ minds, thus increasing residential electricity consumption, and the latter effect is obviously greater from the decomposition of the effect. Secondly, education is always regarded as the cornerstone in China, so the education development in Regions 2 and 3 is generally higher, and due to the diminishing marginal effect, the variable emphasis on local education in these two regions is not significant. The emphasis on local education in Region 1 is not significant due to the serious population loss and the failure of exchange for the local talents by investment. Compared with the other three regions, the education development in Region 4 is relatively backward due to economic and geographical factors, so the marginal effect of the increase in education importance is stronger, which is also reflected in the significance of the regression coefficient of this region.

(4)

Average household size

The regional differences in the effect decomposition of average household size are also significant. In terms of direct effects, only Regions 1 and 4 are significant. Among them, the sign of the coefficients is positive in Regions 1 and 2, and negative in Regions 3 and 4. This may be because a large part of the residential electricity demand to the south of the Qinling–Huaihe line comes from the cooling electricity demand, and the increase in household size will not only have a smaller effect on the total household cooling electricity consumption, but also reduce the per capita cooling electricity consumption; while Region 4 is located to the west of the Huhuanyong line and mostly in special topographic landscapes with large temperature differences between day and night, thus having a large cooling demand compared to Regions 1 and 2, so it is the same as Region 3. In contrast, Regions 1 and 2 are both located to the north of the Qinling–Huaihe line and have a smaller cooling demand, so the original residential electricity consumption mainly comes from other demands, and when the household size increases, it may lead to an increase in the per capita residential electricity consumption due to the need to purchase new appliances. The spillover effect of the average household size has the same sign as the direct effect, but only Region 4 is significant at the level of 10%. Considering that Region 4 is located to the west of the Huhuanyong line and has a low population density, the spillover effect is significantly negative, as the increase in average household size is beneficial to improve the overall efficiency of electricity consumption in the area.

5.5. Robustness Test

To ensure the reliability of the model estimation results, robustness tests based on the benchmark regression are required. In this paper, the robustness of the models is tested in three directions: the explained variable substitution test, the estimation method substitution test, and the endogeneity test.

(1)

Substitution of the explained variables

In this paper, the explanatory variables are replaced: from the per capita residential electricity consumption to the residential electricity consumption in per 1000 square kilometers ($AreaE$). Then, the same model for maximum likelihood estimation is used, and the results are shown in

Table 8. Robustness test (AreaE).

	AreaE	Region 1	Region 2	Region 3	Region 4
lninc	0.482 ***	0.422 ***	0.960 ***	0.425 ***	0.280 ***
lnhd	0.0917 ***	0.170 **	0.205 ***	0.0314	0.0240
lnedu	0.0459 **	0.00648	−0.138 *	−0.0397	0.0702
lnhsc	0.323 ***	0.276	0.704 **	−0.0836	−0.478 ***
ρ	0.505 ***	0.258 ***	0.195 **	0.549 ***	0.448 ***
Individual fixed effects	Yes	Yes	Yes	Yes	Yes
N	3990	476	1064	2072	378
R2	0.347	0.416	0.419	0.344	0.148

* p < 0.1, ** p < 0.05, *** p < 0.01.

. Among the remaining variables, the sign of the coefficients of the other variables in each region does not change, except for the average household size at the national level and the emphasis on local education in Regions 1 and 3. Overall, the model is robust.

(2)

Substitution of estimating method

In order to further test the robustness of the model, this paper again uses the per capita residential electricity consumption, non-spatial panel fixed effects model, and OLS method for estimation, and the results are shown in

Table 9. Robustness test (OLS).

	Whole Nation	Region 1	Region 2	Region 3	Region 4
lninc	0.227 ***	0.266 ***	0.236 ***	0.228 ***	0.178 ***
lnhd	0.039 ***	0.0871 **	0.0576 **	0.0268	0.0115
lnedu	0.022 **	−0.0137	−0.00175	0.00825	0.0604 **
lnhsc	−0.078 ***	0.326 ***	0.0204	−0.0823 **	−0.392 ***
Constant	−1.979 ***	−2.823 ***	−2.278 ***	−1.878 ***	−1.099 ***
Individual fixed effects	Yes	Yes	Yes	Yes	Yes
N	3990	476	1064	2072	378
R2	0.3282	0.398	0.524	0.422	0.448

* p < 0.1, ** p < 0.05, *** p < 0.01.

. The sign of each variable is basically consistent with the previous one, so the model set is considered to have a good robustness and the estimation results have a high credibility.

(3)

Endogeneity test

Considering the possible two-way causality between each explanatory variable and per capita residential electricity consumption, referring to the method of Chen and Lin, Xie, et al., Chen et al., and Shao et al. [36,37,38,39], the parameter estimation is re-performed after taking a lag of one period for the explanatory variables on the basis of the benchmark regression analysis to alleviate the possible two-way causality endogeneity problems. The estimated results are shown in

Table 10. Results of endogeneity test.

	Whole Nation	Region 1	Region 2	Region 3	Region 4
Llninc	0.130 ***	0.234 ***	0.212 ***	0.104 ***	0.135 ***
Llnhd	0.0243 **	0.0455	0.0326	0.0202	0.0200
Llnedu	0.0323 ***	−0.0297	0.0135	0.0212 *	0.0705 ***
Llnhsc	−0.00319	0.457 ***	0.128 **	−0.00952	−0.346 ***
ρ	0.462 ***	0.242 ***	0.133 ***	0.572 ***	0.237 ***
Individual fixed effects	Yes	Yes	Yes	Yes	Yes
N	3705	442	988	1924	351
R2	0.282	0.426	0.314	0.293	0.287

* p < 0.1, ** p < 0.05, *** p < 0.01.

, and the signs of the coefficients of the variables are basically consistent with the benchmark regression, indicating that the model remains relatively robust after the endogeneity has been treated.

5.6. Mechanism Analysis

Based on the above empirical results, this section analyzes the mechanism of the factors influencing residential electricity consumption due to the characteristics of the areas divided by the “three lines” partitioning method.

(1)

Per capita disposable income

Table 6 shows that the increase in per capita disposable income has a positive effect on the per capita residential electricity consumption in local and neighboring areas, which may relate to three aspects. First, the increase in per capita disposable income relaxes the budget constraint of residential consumption, and residents tend to improve their quality of life by increasing electricity consumption, such as by purchasing new home appliances [18]. Second, China’s residential electricity prices, as an important part of ensuring people’s livelihood, are currently controlled by the government, which neither reflects the true cost of the upstream side of electricity generation nor establishes an effective price-adjustment mechanism from consumption-side information to electricity price setting, and the price elasticity of residential electricity demand is weak, so the income effect from the increase in per capita disposable income is much larger than the situation under normal supply and demand [2,14]. Third, the development of industries has a clustering effect, which leads to economic agglomeration. The economic development of a region can stimulate the development of surrounding areas. Additionally, in light of the positive effect of economic agglomeration on rising residential income [35], an increase in per capita disposable income in one region can, to a certain extent, drive an increase in per capita disposable income in nearby regions. This has a favorable impact on the per capita residential electricity consumption in nearby regions. This explanation is evidenced by the decomposition of the per capita disposable income effect in Region 3.

(2)

The average number of hot days per year

The mechanism of the average number of hot days per year on the per capita residential electricity consumption is relatively simple, mainly by influencing the demand and usage behavior of residents for domestic appliances. For Regions 1, 2, and 3 which are on the east side of the Huhuanyong line, the per capita residential electricity consumption varies in the same direction according to the increase or decrease in the average number of hot days per year. However, for Region 4 which is on the west side of the Huhuanyong line, the large diurnal temperature difference prevents the average temperature from accurately reflecting the residential electricity consumption. Therefore, under the circumstance of small diurnal variation, the average number of hot days per year has a positive effect on per capita residential electricity consumption through the increase in residential cooling demand with higher temperatures.

(3)

Emphasis on local education

The emphasis on local education affects the residential electricity demand mainly through residents’ literacy level. Such influences are two-way. On the one hand, the emphasis on local education leads to the improvement of residents’ literacy level and their awareness of energy saving and environmental protection, which will have a negative effect on per capita residential electricity consumption [40,41]. On the other hand, with the improvement of residents’ literacy level, residents’ demand for quality of life will also increase, which will have a positive effect on the per capita residential electricity consumption. In addition, the magnitude of these two effects is influenced by the development of local education. As shown in the decomposition of the effect on the east side of the Huhuanyong line, the marginal effect decreases severely due to advanced local education, then the effect of emphasis on local education on the per capita residential electricity consumption decreases.

(4)

Average household size

The average household size also influences the residential electricity demand mainly through the demand for household appliances, and this influencing mechanism contains a key node, namely the proportion of cooling demand in the electricity demand of the area. When the proportion of cooling demand is large, families tend to have a larger base of cooling appliances to ensure that the cooling capacity matches the size of the house. At that time, the increase in family size does not easily lead to an increase in the number of cooling appliances, so the impact on the total residential cooling electricity consumption is smaller, resulting in a reduction in per capita cooling electricity consumption and per capita overall electricity consumption. Regions 3 and 4 correspond to the above situation. When the proportion of cooling demand is small and residential electricity consumption is mainly composed of other demands, the increase in family size may lead to an increase in per capita residential electricity consumption due to the need to purchase new appliances. Regions 1 and 2 correspond to the above situation. In addition, when family size increases, the consequent increase in electricity costs may force residents to adopt energy efficiency measures, which may have a negative effect on per capita residential electricity consumption [42].

Based on the heterogeneity analysis of the spatial metrology area and the above mechanism analysis, Figure 5 is shown as follows.

6. Conclusions and Policy Suggestions

6.1. Conclusions

This article uses the data from 285 prefecture-level cities in China from 2006 to 2019, and the regional heterogeneity of China’s residential electricity consumption by applying “three lines” partitioning method (the Qinling–Huaihe line, Huhuanyong line, and Shanhaiguan line) to divide the country into four regions. Our findings are as follows:

(1)

Based on the “three lines” partitioning method, significant regional heterogeneity of the residential electricity consumption in China has been identified. Among them, Region 3, which is to the south of the Qinling–Huaihe line and east of the Huhuanyong line, has a higher per capita residential electricity consumption than the rest of the regions and maintains a higher growth rate due to its relatively advanced economic development and hot climate. Among the other regions, Region 4, which is to the west of the Huhuanyong line, has the highest growth rate of per capita residential electricity consumption. It has risen to the second at the end of the sample period, indicating that the living standard of residents in the west has improved significantly in recent years. However, the per capita residential electricity consumption in Region 2 which is surrounded by the Shanhaiguan line, the Huhuanyong line, and the Qinling–Huaihe line has receded to the lowest in 2019 and is similar to the value of Region 1 which is to the north of the Shanhaiguan line.

(2)

The results of the analysis of inter-group differences show that the difference in electricity consumption between Region 1 and others is larger, and further effect decomposition studies reveal that this difference stems from the difference in the marginal effects of the explanatory variables. Due to the cold climate, lack of economic development, population loss, and population aging, the direct effect of its per capita disposable income and hot days are the largest among the four regions, and it is the only region where the effect of emphasis on local education is negative and the effect of family size is negative; in terms of indirect effects, only Region 1 has the positive effect of high temperature influenced by cold weather and electricity supply and demand, and a negative education effect influenced by population loss.

(3)

In terms of the spatial econometric analysis, the per capita residential electricity consumption has a significant positive spatial correlation, and the regional heterogeneity of the influencing factors is relatively significant. Among them, income is the core influencing factor, and the empirical results are significant. The effect of income on per capita residential electricity consumption is relatively uniform in the four regions, showing positive effects through home appliances, weak price elasticity, and economic agglomeration. The regional heterogeneity of the effect of hot days is significant, with the east side of the Huhuanyong line being positively influenced by the average number of hot days. But the west side of the Huhuanyong line has the smallest and most insignificant direct effect due to the large temperature difference and sparsely populated area. The direct effect of emphasis on local education is the same as the indirect effect, which mainly affects the per capita residential electricity consumption by influencing the awareness of electricity saving and the demand for quality of life; only Region 1 is negatively affected by population loss and aging, while Region 4 is positively affected due to the relative lack of education development. The variable average family size is influenced by the ratio of electric appliances to population and shows negative/positive effects when respectively facing the large/small proportion of cooling electricity demand to total residential electricity demand.

6.2. Policy Implications

Based on the above findings, this paper puts forward the following policy implications.

(1)

Improve electricity management based on local conditions

Since China’s residential electricity consumption shows significant regional heterogeneity, electricity management should be tailored to local conditions. First, regardless of the slow pace of the growth of residential electricity consumption due to population loss, the flexibility of electricity supply in Region 1 should be ensured in view of the severe cold climate and labor return during holidays. Second, the proportion of electric heating in Region 2 should be increased, while the proportion of coal heating should be reduced, taking into account the reduction in carbon emissions and load shedding. Third, considering the rigid demand for electricity for cooling in the area, we should increase the promotion of energy-efficient domestic appliances, optimize the design of energy-efficiency labeling for domestic appliances, and combine time-of-use power price measures to reasonably guide residents to avoid peaks and save electricity in Region 3. Fourth, the promotion of distributed photovoltaic and wind power storage in Region 4 should be accelerated to make full use of local resources to achieve energy saving and emission reduction.

(2)

Strengthen multi-dimensional initiatives for electricity supply

Due to the positive effect of income improvement on residential electricity demand, combined with China’s goal of “Stable Growth”, it is projected that the future residential electricity demand will still be on the rise. Therefore, China faces unprecedented challenges in maintaining its electricity supply while ensuring a low-carbon transition. This paper suggests strengthening the electricity supply in the following four aspects, namely production, technology, market, and electricity load. First, the safety of the electricity supply should be the top priority, coal power units should be in operation at the current stage to ensure the flexibility and reliability of the electricity supply. The proportion of new energy applications should be gradually increased to promote the transformation of coal power from the main power supply to basic supply, and to realize the replacement of coal power by nuclear power and hydrogen power in the future. Second, technologies like carbon capture, utilization, and storage (CCUS) need to be introduced to make full use of existing coal power units and reduce the energy storage system pressure caused by the application of wind power and photovoltaic. Third, in an effort to accelerate the pace of electricity price reform, it is necessary to allow residential electricity prices to fluctuate with changes in the cost of electricity supply and change the residential electricity subsidy from a hidden subsidy to an explicit subsidy. Additionally, the step tariff scheme needs to be adjusted by letting affluent households with high electricity consumption subsidize low-income households with low electricity consumption. Fourth, time-of-use power prices need to be optimized to guide residents to reasonably avoid peak electricity consumption.

This paper quantitatively assesses the regional heterogeneity of residential electricity consumption and the influencing factors, but it uses macro prefecture-level city data, sacrificing the depth of the portrayal of residents’ behavior, and it does not empirically analyze the mechanism of the influencing factors. Therefore, in future research, it is an important research direction to start from micro household survey data, to analyze the mechanism of each influencing factor in more detail, and to put forward targeted policy recommendations accordingly.