How Efficient China’s Tiered Pricing Is for Household Electricity: Evidence from Survey Data

Abstract

Due to the wide coverage of first-tier electricity consumption and the small price difference between different tiers, the current tiered pricing for household electricity (TPHE) cannot give full play to the advantages of the increasing block electricity tariffs (IBTs). Based on the microscopic survey data provided by the Chinese General Social Survey (CGSS) in 2015, this paper innovatively uses the predicted average electricity price as the instrumental variable of electricity price to explore the influencing factors of household electricity consumption in order to solve the possible endogenous problems. Simultaneously, the samples are further grouped by income and electricity consumption, and the electricity consumption characteristics of different groups are discussed separately. The results show that, for low-income groups, the price elasticity of electricity consumption is relatively low because the electricity consumption of low-income households is concentrated on meeting the energy demand necessary for basic life, while the price elasticity of high-income groups is relatively high because the electricity consumption of the high-income households is mostly the energy demand generated by improving the quality of life.

1. Introduction

The power industry is one of the essential, strategic, primary industries in a country and plays a vital role in promoting the national economy and social progress. The development of China’s power industry began later than that of developed countries. In order to ensure the stable operation of the economy and society and the basic needs of people’s lives, China’s electricity prices have been deliberately lowered for a long time (Guo, Davidson et al., 2020; Hu, Ho et al., 2019) [1,2]. The long-distorted electricity price system has caused resource mismatches which are not conducive to high-quality economic development (Athukorala, Wilson et al., 2019; Du, Lin et al., 2015) [3,4]. Since the implementation of the power system reform in 2002, China’s electricity price formation mechanism has been gradually improved, and policies such as tiered pricing for household electricity (TPHE), discriminated pricing, and punitive pricing for energy-intensive sectors have been introduced. However, the current electricity prices are still dominated by government pricing. Electricity price adjustments often lag behind cost changes, making it difficult to timely and reasonably reflect electricity generation costs, market supply, demand, resource scarcity, and environmental protection expenditure (Restrepo and Morales-Pinzón 2020, Yan, Zhang et al., 2020) [5,6].

Compared to industry and commerce, the electricity consumption of residents is related to the basic living needs of households [7]. Therefore, the government’s attitude is more cautious, causing the price reform of residents’ electricity consumption to lag [8]. As the government has difficulty in accurately grasping the impact of electricity price reform on residents’ burdens, when formulating the TPHE in 2012, the National Development and Reform Commission prudently considered “covering 80% of residents” as the standard for formulating the TPHE’s first-block electricity consumption [9]. However, during the implementation process, the above standards were continuously relaxed because the government was too worried about the negative impact of electricity price adjustments. Studies have shown that the coverage of the first-block electricity users in some regions exceeds 91%, far exceeding the target of “80% coverage” initially designed by the government [10].

With the continuous development of China’s economy and the continuous improvement of people’s living standards, the total amount and proportion of residents’ electricity consumption are also increasing (Figure 1). The importance of guiding residents to use electricity reasonably is becoming more and more prominent. The existing tiered electricity price system finds it more and more difficult to meet its established policy objectives. In this context, formulating a scientific and reasonable step-by-step electricity-price plan is not only related to ensuring the fairness of the living needs of low-income groups but also to the reform of the power system and the efficiency of guiding residents to use electricity reasonably.

TPHE was gradually developed based on the Incremental Block Tariff (IBT) proposed by Boiteux (1971) [11] and Ramsey (1927) [12]. IBT sets the consumption of public utility resources such as electricity, natural gas, and water as several steps, so that pricing is carried out in tiers, and prices increase between each tier. Theoretically, the tiered pricing mechanism can better solve the fairness and efficiency problems of utility pricing. It can simultaneously meet various policy objectives, such as income redistribution, energy conservation, and emission reduction [13]. Since wealthy households consume more energy than low-income households, the incremental price structure can enable rich households to cross-subsidize low-income households, while also ensuring that poor people can obtain the energy needs necessary for life at lower costs. In the 1970s, IBT use began in the US power sector and was gradually extended to other utilities such as water resources [14,15]. Since the 21st century, the ladder price mechanism has also been rapidly promoted in other countries [16,17,18].

The existing research generally states that the core issue of achieving the goal of TPHE is to reasonably determine the first-block electricity consumption [19]. Many factors may affect the electricity demand of residents, such as household income, family size, age distribution, and living habits [20,21,22]. Additionally, differences in the electricity consumption structure of urban and rural households and meteorological conditions will also have a particular impact on household electricity consumption [23,24,25]. Therefore, the determination of TPHE’s first-block electricity consumption must be empirical research based on regional household micro-level data.

There are some shortcomings in the current study. Firstly, a suitable formulation of TPHE needs to clarify the factors affecting household electricity consumption of different income types. Using energy data from the CGSS 2015, this study empirically analyzes the influencing factors of the household electricity consumption of Chinese residents. The relevant research results can be used to improve the existing tiered power plan, to give full play to the tiered electricity price policy to guide residents’ reasonable consumption, and to ensure the fairness and improvement of the TPHE. The effect of efficiency, etc., provides a reference. Secondly, in the case of the implementation of TPHE, household electricity consumption will be affected by electricity prices, and the average electricity price will also change with changes in electricity consumption, which may bring endogenous problems and affect the accuracy of the empirical results.

China’s residential electricity price has been at a low level for a long time. For most residents, electricity price reform means that electricity prices will rise [26]. In formulating and implementing the existing TPHE, the government had considered the “fairness” of meeting the basic living needs of low-income groups more and paid less attention to the “efficiency”. The reasonable determination of the first-block electricity consumption is the key to giving full play to the advantages of the TPHE. Out of consideration for factors such as maintaining social stability and guaranteeing the living standards of residents, the coverage of China’s 2012 TPHE’s first-block electricity consumption was too high, making it difficult to achieve its intended policy objectives in improving the efficiency of residential electricity consumption and guiding residents to use electricity reasonably. On the other hand, due to the difficulty in obtaining data, existing studies are mostly analyses and empirical research is more challenging to carry out. Thus, based on the micro-survey data from the Chinese General Social Survey (CGSS) 2015, this article makes a more detailed analysis of the characteristics of household electricity consumption, especially for different income groups, and, on this basis, the redesign of the tiered-electricity-price mechanism.

The main contributions of this paper are as follows: first, a suitable formulation of the TPHE needs to clarify the factors affecting household electricity consumption of different income types. Using the energy data from the CGSS 2015, this study empirically analyzes the influencing factors of household electricity consumption in Chinese residences. The relevant research results can be used to improve the existing tiered power plan, to give full play to the tiered electricity price policy to guide residents’ reasonable consumption, and to ensure fairness and improvement. The effect of efficiency, etc., provides a reference. Second, in the case of the implementation of TPHE, household electricity consumption will be affected by electricity prices, and the average electricity price will also change with changes in electricity consumption, which may bring endogenous problems and affect the accuracy of the empirical results. In this paper, we innovatively use the predicted average price of electricity as an instrumental variable (IV) of electricity prices to solve possible endogenous problems.

The rest of the paper is organized as follows. Section 2 describes the methodology and data. Section 3 reports the empirical results. Section 4 further discusses the rationality of IV; simultaneously, the sample is divided into different groups according to income and electricity consumption. Next, the differences in electricity consumption features between different groups are discussed. Finally, the conclusion and policy implications are found in Section 5.

2. Methodology and Data

2.1. Methodology

Labandeira, Labeaga et al. (2012) [27] integrated the Household Electricity Demand model based on the theory of household production proposed and improved by Flaig (1990) [28], Filippini and Hunt (2011) [29], and Shen, Ghatikar et al. (2014) [7]. According to Labandeira, Labeaga et al. (2012) [27], a theoretical model of household electricity demand based on the prices of electricity and alternative energies was built in this paper.

There are certain substitution effects between different energy types in household energy consumption, such as the replacement of coal by electricity in heating, the replacement of oil by electricity in vehicles, and the substitution of natural gas and electricity in the most extensive cooking process in daily life. Therefore, we can regard the total household energy consumption as a composite commodity including electricity, natural gas, oil, and coal. The compound commodity N is given as follows:

$$N=N\left(E,F\right)$$

(1)

where E is the electricity consumption and F is the alternative fuel consumption. On this basis, the utility function of the family can be expressed as follows:

$$U=U\left(N,X;D,G\right)$$

(2)

where X represents the demand for numeraire goods that directly yields a utility, D is the family characteristic that affects residents’ preference, and G represents the geographic environment characteristics of the family. Household budget constraints are given by:

$$Y=P_N\times N+P_X\times X$$

(3)

where Y represents income, $P_N$ represents the price of the composite energy commodities, and $P_X$ represents the price of composite-priced items. The household production decision contains two stages of optimization. The first stage is the process of minimizing the cost of manufacturing composite-energy commodities, which can be expressed as follows:

$$Min\left(P_E\times E+P_F\times F\right)$$

$$s.t. N=N\left(E,F\right)$$

(4)

where $P_E$ and $P_F$ represent the price of electricity and the price of alternative energies, respectively. The cost function of household energy consumption can be obtained by solving Equation (4), which is given as follows:

$$C=C\left(P_E,P_F,N\right)$$

(5)

Using Shephard’s lemma, we can obtain the function for demand derived from inputs. For the household electricity consumption of residents, this is:

$$E=\frac{\partial C\left(P_E,P_F,N\right)}{\partial P_E}=E\left(P_E,P_F,N\right)$$

(6)

In the second phase, households choose the consumption of composite energy commodities and benchmark commodities to maximize their utility:

$$Ma\mathrm{x}\left(N,X;D,G\right)$$

$$s.t. Y=C\left(P_E,P_F,N\right)+P_X\times X$$

(7)

On this basis, the corresponding Lagrangian function can be constructed as follows:

$$L=U\left(\left(N,X;D,G\right)+\lambda \left(Y-C\left(P_E,P_F,N\right)-P_X\times X\right)\right)$$

(8)

By solving Equation (8), we can obtain the demand function of the composite energy commodity:

$$Q^{\prime }=Q^{\prime }\left(P_E,P_F,Y;D,G\right)$$

(9)

Using Equations (6) and (9), the residents’ electricity consumption function can be obtained as follows:

$$E=E\left(P_E,P_F,Q^{\prime }\left(P_E,P_F,Y;D,G\right)\right)$$

(10)

It can be seen from Equation (10) that the household electricity demand depends on electricity prices, alternative fuel prices, household income, household preference characteristics, and geographical environment characteristics.

2.2. Data

The micro-survey data used in this paper are mainly from the 2015 Chinese General Social Survey (CGSS). CGSS is the earliest national, comprehensive, and continuous academic survey project in China, which comprehensively collects data from multiple levels of society, communities, families, and individuals. It summarizes the trends of social changes [30]. At present, CGSS data has become one of the most critical data sources for research on Chinese society.

The 2015 CGSS survey covered 478 villages in 28 provinces of China and included 10,968 valid questionnaires. The energy module (Part E) was integrated into the project for the first time, and the sample containing the energy module accounted for about one-third of the total [31]. Therefore, the basic data used in this paper are from the energy module (Part E) of the 2015 CGSS survey project, which was mainly aimed at the household energy utilization in 2014 and had a total of 3863 samples.

The core issue of this study considers the influencing factors of the electricity consumption of residents, corresponding to the following two questions in the survey, namely: “What is the average monthly electricity consumption of your home in 2014?” and “What is the average monthly electricity expenditure of your home in 2014” (The question numbers of the two questions are e86_1 and e87_1 in the CGSS 2015, and the units of the two questions are kW·h and yuan, respectively.) Given that China’s electricity price is a regulated, non-market price, the household’s electricity expenditure can be calculated as long as we know the household’s monthly electricity consumption. Therefore, in a certain sense, the household electricity consumption and electricity consumption expenditure of Chinese residents are entirely equivalent, which also provides us with a suitable sample-validity screening mechanism. In previous studies, most electricity consumption data came from the subjective answers of the respondents rather than the objective, factual figures of the electricity meter. Since the authenticity of the data depended on the respondents’ knowledge of their household electricity consumption, the credibility of those results is often questioned.

In this paper, we used the above two homogenous questions in the questionnaire for screening to avoid a sample bias caused by the cognitive bias of the interviewee. The specific measures were as follows: first, the TPHE was matched according to its location, and the household electricity consumption was then calculated according to the monthly electricity expenditure. This was divided by the electricity consumption answer provided by the interviewee. Finally, to ensure the number of samples and eliminate biased samples, the trusted bandwidth was chosen to be from 0.7 to 1.3.

After screening, 1829 samples fell within the interval and the retained proportion was 47.35%. As is shown in Figure 2, the household electricity consumption after sample correction was more in line with the normal distribution. Data accuracy was a point overlooked by the previous use of CGSS data for household electricity research [32,33].

In terms of household electrical appliance use, the duration of electrical appliance use was selected as an independent variable to examine the impact of different appliance use on the household electrical electricity consumption, according to Pothitou, Hanna et al. (2016) [34]. Household appliances included cooking appliances (induction cookers, microwave ovens, electric kettles, and ovens), electric lights, televisions, computers, refrigerators, washing machines, water heaters, and air conditioners. The units were hours per month. In terms of household electricity consumption behavior, we selected three dimensions, including “Do you know the tiered pricing for household electricity,” “Electricity settlement method: payment before consumption (charging card) or consumption after payment (monthly settlement),” and “whether to use smart meters” to examine the differences in household electricity consumption caused by different cognitions and consumption habits.

In terms of control variables, natural gas prices, household size, urban or rural areas, and high-temperature days were selected to control the heterogeneity of household dimensions, urban–rural differences, and climate [35]. As there is currently no unified national electricity price for Chinese residents, there are differences between provinces, and some provinces have different electricity prices at different times. Therefore, we calculated the electricity expenditure based on the tiered electricity price policy of the surveyed area and the average monthly household electricity consumption. This reverses the average electricity price based on the ratio of electricity expenditure to electricity consumption. Studies have shown that consumers who use more electricity respond more to average prices [36]. Ito (2014) [37] also demonstrated that households will react to their perceived price rather than marginal price.

Natural gas is the primary energy source consumed by Chinese residents and electricity. It is also an important alternative energy source for electricity. Therefore, the price of alternative energy was represented by the price of natural gas. This data came from the average price of civil-pipeline natural gas in the wind database in 2014. The temperature was expressed in terms of low-temperature days and high-temperature days. The data from eleven cities, including Changchun, Jixi, Xuzhou, Jinan, Wuhan, Changsha, Guangzhou, Xi’an, Lanzhou, Jiuquan, and Xining came from the NOAA database, and the data from other cities were taken from the China Meteorological Network.

The descriptive statistics of the main variables used in this paper are shown in

Table 1. Descriptive statistics of main variables.

Variable	Abbreviation	Obs.	Average	Standard Deviation	Min	Max
Family size	size	1829	2.909	1.380	1.000	14.000
Whether to use smart meters	emet	1829	0.476	0.500	0.000	1.000
Electricity-bill settlement method	bill	1829	0.657	0.475	0.000	1.000
Understand the tiered electricity price	policy	1829	0.388	0.487	0.000	1.000
Electricity price	p1	1829	0.540	0.046	0.377	0.820
Natural gas prices	p2	1829	2.549	0.710	1.480	5.930
Electricity consumption	Q	1829	150.253	131.717	2.377	2519.690
Low-temperature days	cdd	1829	25.940	27.728	0.000	112.000
High-temperature days	hdd	1829	171.385	46.409	54	320
Household income	income	1829	6.267	15.188	0.000	400.000
City/Rural	rural	1829	0.353	0.478	0.000	1.000
Frequency of cooking appliances	cooking	1829	32.290	45.877	0.000	671.250
Frequency of electric lamp	lighting	1829	16.864	17.324	0.000	150.500
Frequency of refrigerator usage	fridge	1829	8.741	5.093	0.000	33.000
Frequency of washing machine	washing	1829	22.561	12.138	0.000	161.250
TV frequency	tv	1829	2.866	1.265	1.000	10.000
Computer frequency	pc	1829	2.987	1.212	0.000	17.500
Water-heater use frequency	heater	1829	5.964	30.201	0.000	630.000
Frequency of air conditioning	air	1829	3.675	6.061	0.250	66.125

.

3. Results

According to the theoretical model constructed in Section 2, household electricity demand depends on electricity prices, alternative fuel prices, income levels, and other household characteristics. Therefore, a model including household electricity consumption habits and information feedback is constructed based on Equation (10). To ensure the stability of the data and partially eliminate the effect of heteroscedasticity, the residential electricity demand model in logarithmic form has been organized as follows:

$$\begin{array}{l}ln{E}_i={\beta }_0+{\beta }_{1}ln{P}_1+{\beta }_{2}ln{P}_2+{\beta }_{3}lnincom{e}_i+{\beta }_4\times Feedbac{k}_i+{\beta }_5\times H{B}_i+{\beta }_6 lnsiz{e}_i+{\beta }_{7}rura{l}_i\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{8}lncd{d}_i+{\beta }_{8}lnhd{d}_i+{\mu }_i\end{array}$$

(11)

where $E_i$ represents the average monthly electricity consumption of household i, and $P_1$ and $P_2$ represent the prices of residential electricity and alternative energy, respectively. Additionally, $incom{e}_i$, $H{B}_i$, $Feedbac{k}_i$, $siz{e}_i$, and $rura{l}_i$ represent the income level, the characteristic variable of electricity consumption habit, the information feedback variable, the family size, and the dummy variable represents the urban-rural difference of households, respectively. The random error term is represented by ${\mu }_i, \mathrm{while} cd{d}_i$ and $hd{d}_i$ represent low-temperature days and the number of high-temperature days, respectively. Since the variables in Equation (11) adopt the logarithmic form, ${\beta }_1$, ${\beta }_2$, and ${\beta }_3$ represent the demand price elasticity, demand cross-price elasticity, and demand income elasticity, respectively.

The electricity price used in this paper is the average electricity price. According to the existing TPHE, the average electricity price will be affected by electricity consumption. At the same time, the electricity consumption of households will also be affected by the electricity price level, which will cause endogenousness due to reverse causation, and thus affect the accuracy of the estimation results.

According to Ito (2014) [37], the predicted average price of electricity is used as an instrumental variable (IV) of electricity prices to solve possible endogenous problems. The predicted average price of electricity is obtained by simulating the behavior-selection process of households through a multinomial logit model [38] rather than the data of household electricity consumption obtained from the survey. That is, using the multinomial logit model to evaluate the probability under three combinations that affect the gradient selection of the TPHE, the corresponding explained variables are socio-economic factors of the household (household income, household size, etc.), characteristics of electricity consumption habits, and information feedback variables. We then use the simulation block to find the corresponding electricity price of the province where the user is located to calculate the average simulated electricity price of the user. The socio-economic factors of the household are exogenous variables, and the characteristics of electricity consumption habits are obtained once, according to this survey, and remain unchanged during the survey period. In the short term, it is generally assumed that the amount of household electrical appliances and usage habits also remain unchanged, so these corresponding variables are exogenous.

The method of estimating the multi-value selection model of the brothers as an IV has similar applications in previous studies such as Wooldridge (2002) [39], in which the logit or probit regression results were used as the IV. Simultaneously, because many factors may affect the choice of the TPHE gradient, the direct use of actual average electricity prices may cause a weak IV. Therefore, the above estimation method can not only solve the endogenous problem but can also increase the price change under the condition that the regulation price remains unchanged. After using the predicted electricity price as an IV, regression results are shown in

Table 2. Regression results of factors influencing household electricity demand.

	IV-2SLS
	(1)	(2)	(3)
lnp1	−0.882 ***	−1.031 ***	−1.092 ***
	(−3.055)	(−3.951)	(−4.068)
lnincome	0.062 ***	0.025 ***	0.023 ***
	(5.867)	(2.917)	(2.889)
lnp2	0.328 ***	0.267 ***	0.278 ***
	(4.086)	(3.627)	(3.804)
lnsize	0.420 ***	0.310 ***	0.303 ***
	(11.178)	(8.889)	(8.737)
rural	0.358 ***	0.143 ***	0.110 ***
	(10.577)	(4.158)	(3.202)
lncdd	0.008 **	0.003	0.003
	(1.994)	(0.746)	(0.753)
lnhdd	−0.368 ***	−0.221 ***	−0.195 ***
	(−5.271)	(−3.222)	(−2.826)
lncooking		0.007 ***	0.007 ***
		(3.299)	(3.402)
lnlighting		0.023 ***	0.021 ***
		(2.782)	(2.588)
lnfridge		0.022 ***	0.022 ***
		(7.090)	(7.138)
lnwashing		−0.048 ***	−0.048 ***
		(−2.600)	(−2.585)
lntv		0.128 ***	0.112 ***
		(3.207)	(2.800)
lnpc		0.044	0.050
		(1.080)	(1.306)
lnheater		0.014 ***	0.013 ***
		(6.294)	(5.885)
lnair		0.191 ***	0.180 ***
		(10.171)	(9.569)
bill			0.050
			(1.550)
policy			0.140 ***
			(4.510)
emet			0.095 ***
			(3.219)
_cons	5.173 ***	4.566 ***	4.288 ***
	(12.836)	(10.831)	(9.759)
Obs	1829	1829	1829
Adj-R2	0.194	0.333	0.343

Note: ** p < 0.05; *** p < 0.01.

.

It can be seen from Table 2 that household electricity demand is significantly negatively related to electricity prices. The self-price elasticity of electricity is between −0.882 and −1.092; that is, for every 1% increase in electricity prices, the average monthly electricity consumption of Chinese households decreases by 0.882–1.092%. This finding is consistent with the estimation results in the existing literature. For example, Auffhammer and Mansur (2014) [40] found that the price elasticity of electricity consumption of residents in the United States at different income levels ranged from −0.24 to −1.32; Hung and Huang (2015) [41] used monthly data to estimate the price elasticity of residents ’electricity consumption in Taiwan, the results show that the short-term and long-term elasticities in the summer are −0.451 and −1.130, respectively, and −0.819 and −1.270 in other periods.

An income-elasticity coefficient of 0.025–0.062 indicates that, for every 1% increase in household income, the average household electricity consumption will increase by 0.025–0.062%, which means that an increasing income can improve the household’s electricity consumption. This conclusion is consistent with our expectations. Although the income-elasticity coefficient is relatively small, it is consistent with the results of the existing literature. For example, Dubin and McFadden (1984) [42] showed that the price elasticity of household electricity demand was between 0.006 and 0.98; Du, Guo et al. (2017) [43] used survey data (CRECS-2012) to analyze that the income elasticity of household electricity consumption was about 0.1. The cross-price elasticity coefficient of electricity and natural gas is 0.267–0.328, indicating that for every 1% increase in natural gas prices, household electricity consumption will increase by 0.267–0.328%, showing that, electricity and natural gas are substitutes for each other for household energy consumption.

The characteristics of household electricity consumption habits have a significant impact on electricity consumption. Household electricity consumption is indirectly triggered by daily needs such as cooking, lighting, temperature regulation, and entertainment. Therefore, electricity consumption is closely related to lifestyle habits. The results in Table 2 show that the coefficients of household electricity consumption habits are all significant and greater than 0, which shows that all those electricity consumption habits have a positive impact on residents’ electricity consumption.

The regression results of the information feedback variable indicate that the knowledge of relevant policies has not achieved the effect of suppressing electricity consumption. There are two possible reasons for this phenomenon: first, although residents’ understanding of TPHE will prompt them to control their consumption in the corresponding tier, residents will consume as much electricity as possible within this tier due to lower price, which will, in turn, lead to an increase in electricity consumption. Second, because the price difference between the first and second tier of China’s TPHE is tiny, so even if residents know that cross-tier consumer prices will rise, this knowledge will not be able to produce a strong binding power.

The coefficient of smart meters is significantly positive; that is, households with smart meters are associated with higher electricity consumption levels, which means that smart meters may not necessarily reduce electricity consumption. This finding is consistent with Hargreaves, Nye et al. (2013) [44] and Du, Guo et al. (2017) [43]. Smart meters help households grasp electricity consumption information. Still, when households realize that they have limited energy-saving potential, significantly lower electricity price levels make electricity expenditures less binding and their electricity consumption behavior may be more uncontrolled. The coefficient of the type of payment (bill) is also significantly positive, indicating that the type of households with post-event feedback (consumption before payment) consumes more electricity than those with pre-report (pay before consumption). This result is consistent with existing literature such as Faruqui, Sergici et al. (2010) [45].

It is worth noting that the premise of using the IV is the validity of the IV. If the IV is invalid, it may lead to inconsistent estimation or excessive variance of the estimator.

Table 3. Descriptive statistics of instrumental variables and related rationality tests.

Descriptive Statistics of Instrumental Variables
Average	0.536
Max	0.881
Min	0.377
S.D.	0.044
Underidentification test
Kleibergen–Paap rk LM statistic	537.599
p-val	0.000
Weak identification test
Kleibergen–Paap rk Wald F statistic	1132.941
Stock–Yogo weak ID test critical values	16.38 (10%)
Tests of endogeneity
Durbin–Wu–Hausman F statistic	108.933
p-val	0.000

shows the descriptive statistics of the IVs and the test results of endogenousness of explanatory variables and weak instrumental variables. Considering the possible heteroscedasticity, the Durbin–Wu–Hausman test was used to test the performed endogeneity. The test result was 128.194, rejecting the null hypothesis that all explanatory variables are exogenous at a 1% significance level; that is, that electricity prices are considered to be endogenous variables and instrumental variables are required for estimation. Furthermore, this paper uses insufficient recognition and a weak instrumental variable test to verify the rationality of the instrumental variables. The results show no problem of insufficient recognition and weak instrumental variables, indicating that the selection of IV in this paper is reasonable.

In general, the rationality of IVs in this paper can be explained from the following two aspects. First, from the perspective of the correlation of IVs, the estimated results of the multinomial logit model are positively related to the user’s real electricity consumption, which meets the essential requirement that instrumental variables and endogenous variables must be related. Secondly, IVs and disturbance terms cannot be related. As the explanatory variables of the multinomial logit model are exogenous, the estimated results are therefore not related to the disturbance terms.

4. Discussion

4.1. Examination of Price Elasticity under Income Differences

Income is an essential factor that affects residents’ electricity consumption. With the change in income distribution, household electricity consumption at different income levels shows significant differences. For example, He and Reiner (2016) [46] used CFPS data to measure the electricity demand of Chinese households. They found that when income reached a certain level, household electricity consumption would increase with income. On one hand, households with different income levels have different views on consumption, so the price elasticities of electricity consumption are also different. For example, studies have pointed out that the sensitivity of urban households to electricity prices will decrease as income increase [47,48]. On the other hand, households with different income levels also have different amounts of household appliances, leading to different characteristics of electricity consumption.

The original design of China’s TPHE was to ensure the fairness of income redistribution and the primary energy use of low-income groups. According to the relevant regulations of the National Development and Reform Commission of China, the electricity consumption of residents is divided into three tiers: the first tier should cover 80% of the residential users, mainly to protect the necessary living energy needs of low-income people; the second tier should cover 80–95% of residential users; and the third tier is to encourage residents with high power consumption to save electricity while compensating for the electricity costs not borne by users on the first tier. Although the original intention of this regulation was to charge different electricity prices to residents of different income levels, the design process did not set the electricity price level based on the relationship between electricity consumption and income. In other words, to ensure the rationality of redistribution, the design of TPHE must be based on income. Therefore, it is significant and crucial to find a tiered electricity price mechanism that matches income.

To explore the differentiated characteristics of electricity consumption by different types of households, this paper uses income as a variable to distinguish household categories, divides the sample into different income groups, and further analyzes the elastic differences of heterogeneous families of different income classes. Specifically, we divide the sample into two groups, three groups, and four groups according to income. The sample individuals in each group receive increasing income from top to bottom. The price elasticities of different groups are shown in

Table 4. Price elasticity under income heterogeneity.

		Electricity Price	Controls	Obs	Adj-R2
Dichotomous	lnP21	−0.729 *	Yes	967	0.2968
		(−1.91)
	lnP22	−1.621 ***	Yes	862	0.1829
		(−4.01)
Tertile	lnP31	−0.469	Yes	619	0.3248
		(−1.02)
	lnP32	−1.310 ***	Yes	679	0.2124
		(−3.02)
	lnP33	−1.918 ***	Yes	531	0.1616
		(−3.46)
Quartile	lnP41	−0.366	Yes	553	0.3391
		(−0.76)
	lnP42	−1.341 **	Yes	414	0.1901
		(−2.23)
	lnP43	−1.508 ***	Yes	414	0.1737
		(−2.98)
	lnP44	−1.948 ***	Yes	448	0.1515
		(−3.07)

Note: * p < 0.10; ** p < 0.05; *** p < 0.01. All control variables are included in the model, and space constraints permit only representative electricity price results.

.

The results of Table 4 show that the self-price elasticity coefficient of electricity is less than 0. It is particularly important to note that, in the low-income group, the absolute value of the elasticity coefficient is less than 1, and the self-price of electricity is inelastic and insignificant. In the middle–high income group, the absolute value of the elasticity coefficient is significantly greater than 1, and the self-price elasticity of electricity is significant and elastic, which shows that as income levels increase, households’ sensitivity to electricity prices increases. In this sense, when compared with unified pricing, tiered pricing has the unique advantage of being managed and adjusted according to different income levels.

This may be because, for residents with lower income, electricity is a necessity for meeting the basic energy needs of the family’s daily life and the potential for energy saving is small, so household electricity consumption is not significantly affected by price and income, However, for middle- and high-income families, in addition to providing basic energy needs to meet basic living needs, electricity is in more demand for improving the quality of life. Compared to basic energy needs, the latter is more susceptible to price. Moreover, the high-income group consumes a large amount of electricity, and it also has a more significant energy-saving potential.

To further verify that different income groups have different electricity consumption behaviors, the sample is divided into two groups according to income, and the details of the differences in electricity consumption behaviors of the two groups shown in Figure 3.

After dividing the sample into high-income and low-income groups according to household income, we find that the average electricity consumption of the low-income group is 115.84 kWh, which is much less than the 188.86 kWh of the high-income group. Figure 3 shows that the cooking electricity consumption, the electricity consumption of refrigerators and washing machines, and the use frequency of televisions, computers, electric lights, water heaters, and air conditioners of low-income groups are all significantly lower than those of high-income groups (data on the electricity consumption of TVs, computers, electric lights, water heaters and air conditioners is unavailable, and can only be represented by use frequency). Furthermore, the proportions of electricity consumed by necessary appliances in the low-income group is significantly higher than in the high-income group. As the electricity consumption of Chinese residents is generally low, especially considering that the electricity consumption of low-income groups is concentrated on basic living needs, the price elasticity is also relatively low.

It is worth noting that, if the model settings in the two sample groups are the same, the coefficients between the two groups can be compared. However, as there may be an intersection on the confidence interval between the two, a statistical basis to directly compare the coefficients between different subsamples is lacking (Cleary 1999) [49]. Therefore, it is necessary to conduct a statistical test on the difference between the two groups. According to Tibshirani and Efron (1993) [50] and Cleary (1999) [49], we use Fisher’s Permutation test and tests based on seemingly uncorrelated regression to examine the difference of coefficients between groups. The results are shown in

Table 5. Coefficient difference test between groups.

Variables	Coefficient Difference by Fisher’s Permutation Test	p-Value	Coefficient Difference by SUR	p-Value
lnp1	0.892	0.037	0.892	0.081
lnincome	−0.069	0	−0.069	0.085
lnp2	−0.086	0.283	−0.086	0.529
lnsize	−0.023	0.354	−0.023	0.737
rural	−0.105	0.051	−0.105	0.116
lncdd	0.004	0.296	0.004	0.587
lnhdd	0.303	0.013	0.303	0.018
lncooking	0.005	0.138	0.005	0.248
lnlighting	0.034	0.031	0.034	0.022
lnfridge	0.002	0.362	0.002	0.706
lnwashing	−0.124	0.007	−0.124	0.027
lntv	0.19	0.008	0.19	0.012
lnpc	−0.002	0.491	−0.002	0.98
lnair	0.078	0.011	0.078	0.031
lnheater	0.005	0.148	0.005	0.289
bill	0.06	0.183	0.06	0.323
policy	0.036	0.278	0.036	0.544
emet	−0.046	0.203	−0.046	0.409
_cons	−0.814	0.183	−0.814	0.332

.

4.2. Re-Examination of Price Elasticity under Income Differences

According to the above discussion, there are large differences in electricity consumption and electricity consumption characteristics between high-income groups and low-income groups. Therefore, if the average electricity consumption of the low-income group is used as the threshold of the first tier of the TPHE, it can not only protect the basic energy demand of the low-income group but also more strictly reduce the electricity demand of the high-income group to improve the quality of life, thereby balancing both fairness and efficiency. Therefore, taking the average electricity consumption of the low-income group as the threshold of electricity consumption (115.84 kWh/month), we reassessed the influencing factors of household electricity consumption. The results are shown in

Table 6. Re-estimation of price elasticity under income differences.

	Monthly Electricity Consumption ≤ 115.84 Kwh/Month			Monthly Electricity Consumption > 115.84 kWh/Month
	(4)	(5)	(6)	(7)	(8)	(9)
lnp1	−0.642 **	−0.459 *	−0.485 *	−0.440 *	−0.678 ***	−0.722 ***
	(−2.215)	(−1.679)	(−1.728)	(−1.909)	(−2.831)	(−2.929)
lnincome	0.033 ***	0.014	0.014	0.025 ***	0.013 **	0.013 **
	(3.516)	(1.574)	(1.548)	(3.195)	(2.015)	(2.118)
lnp2	0.260 ***	0.201 ***	0.215 ***	0.087	0.087	0.090
	(3.772)	(2.985)	(3.201)	(1.220)	(1.150)	(1.194)
lnsize	0.236 ***	0.143 ***	0.143 ***	0.135 ***	0.131 ***	0.131 ***
	(6.238)	(3.969)	(4.024)	(4.116)	(3.958)	(3.970)
rural	0.180 ***	0.076 **	0.053	0.048 *	−0.017	−0.015
	(5.200)	(2.256)	(1.583)	(1.704)	(−0.547)	(−0.471)
lncdd	0.013 ***	0.010 ***	0.009 **	−0.004	−0.007 **	−0.007 *
	(3.236)	(2.709)	(2.517)	(−1.200)	(−2.118)	(−1.945)
lnhdd	0.063	0.083	0.078	−0.299 ***	−0.254 ***	−0.242 ***
	(0.868)	(1.187)	(1.090)	(−4.789)	(−3.783)	(−3.524)
lncooking		0.009 ***	0.009 ***		0.000	0.000
		(4.045)	(4.131)		(0.181)	(0.175)
lnlighting		0.013	0.012		0.003	0.004
		(1.396)	(1.313)		(0.425)	(0.485)
lnfridge		0.018 ***	0.018 ***		0.002	0.002
		(6.690)	(6.622)		(0.421)	(0.411)
lnwashing		−0.087**	−0.082 **		−0.019	−0.020
		(−2.385)	(−2.251)		(−1.555)	(−1.572)
lntv		0.151 ***	0.145 ***		0.024	0.027
		(3.596)	(3.394)		(0.665)	(0.743)
lnpc		−0.012	−0.008		0.031	0.032
		(−0.501)	(−0.365)		(0.919)	(0.931)
lnheater		0.005 **	0.005 **		0.005 **	0.005 **
		(2.181)	(2.031)		(2.406)	(2.530)
lnair		0.048 ***	0.040 **		0.095 ***	0.096 ***
		(2.635)	(2.175)		(5.425)	(5.419)
bill			0.010			0.039
			(0.307)			(1.353)
policy			0.086 ***			−0.007
			(2.733)			(−0.260)
emet			0.068 **			−0.011
			(2.202)			(−0.404)
_cons	2.937 ***	3.327 ***	3.259 ***	6.323 ***	5.939 ***	5.824 ***
	(7.074)	(7.521)	(6.986)	(16.625)	(13.676)	(12.980)
Obs	948.000	948.000	948.000	881.000	881.000	881.000
r2	0.130	0.256	0.266	0.029	0.061	0.055

Note: * p < 0.10; ** p < 0.05; *** p < 0.01.

.

The results show a certain difference between the influencing factors and the electricity consumption between the two groups. Taking Model (6) and Model (9) for example, which control the usage habits of household appliances and policy information feedback, the electricity-consumption price elasticity of the group whose monthly electricity consumption is greater than the threshold is −0.722, which is significantly higher than the group whose monthly electricity consumption is less than the threshold. This indicates that the electricity consumption of households with low electricity consumption is mainly used to meet the necessary energy demand for living, so the price elasticity is relatively low. Simultaneously, the electricity consumption of the group whose monthly electricity consumption is greater than the threshold is not affected by the price of natural gas (the coefficient is not significant). This can also be explained by the fact that households with a higher electricity consumption mostly use electricity to improve their lives; therefore, natural gas and electricity as a pair of substitutes are seldom considered from the perspective of cost.

5. Conclusions and Policy Implications

Due to the wide coverage of the first-tier electricity consumption and the small price difference between the first- and second-tier electricity consumption, the current TPHE in China cannot give full play to the role of the TPHE in both fairness and efficiency effects. Based on the microscopic survey data provided by the Chinese General Social Survey, this paper explored the influencing factors of household electricity consumption. The results show that the price elasticity of electricity consumption is significantly negative. The higher the income and the greater the electricity consumption, the greater the price elasticity of electricity consumption. In contrast, the low-income and low-power consumption groups have relatively low electricity consumption price elasticity, which shows that the low-income group’s power consumption is concentrated on meeting the energy demand necessary for basic life. The electricity consumption of high-income groups is mostly due to the energy demand generated by improving the quality of life, which leads to higher price elasticity. According to the empirical results of the paper, we make the following policy recommendations:

• Reduce the coverage of the first-grade electricity of the tiered electricity price. The core of the TPHE is the first level of consumption. Due to historical reasons, China’s residents’ electricity prices have been at a relatively low level for a long time, and the government is too worried about the impact of rising electricity prices on residents’ lives. Therefore, the current first-tier electricity price policy covers a wide range of electricity. As a result, for the vast majority of residential users, electricity prices do not reflect their actual electricity cost, which is not conducive to the use of electricity prices in guiding residents to use electricity reasonably or the efficiency of website resource allocation.
• Raise the difference between the first- and second-tier electricity prices of the tiered electricity price policy. The research results in this paper show that the electricity consumption of the higher-income groups concentrates on the demand generated by improving the quality of life, and therefore has higher price elasticity. However, in the current tiered electricity price policy, the gap between the price of the first-grade electricity and the second-grade electricity is too small, which makes it difficult to raise the policy goal of guiding residents to use electricity reasonably and leads to subsidies for some users who should not be subsidized.
• Further promote the reform of China’s electricity pricing system, improve the role of the market in allocating resources, and use market instruments to improve the efficiency of the electricity market while ensuring the basic needs of low-income groups. In recent years, significant progress has been made in the reform of China’s energy-market mechanism. However, the pricing of the electricity market is still dominated by government administrative pricing, which limits the further improvement of resource allocation efficiency.