Sage Journals: Discover world-class research

Abstract

Given China’s pressure to reduce energy use and the rise of Chinese residents’ energy consumption in the last decades, it is of significance to investigate the energy consumption behaviors of Chinese households. This paper examines the impacts of energy efficiency labeling on Chinese households’ use of domestic appliances, using the data from a national household survey (i.e., the Chinese General Social Survey) in 2015. Applying the propensity score matching (PSM) approach, we find a positive relationship between energy efficiency label and the use of domestic appliances, indicating the possible existence of rebound effect. This finding is significantly robust across a series of appliances and a battery of regression methods. The heterogeneity analysis yields that the effect of energy efficiency label on appliance use is heterogeneous in residential places, income level and head’s gender of households. The findings shed light on how to better apply energy efficiency labeling programs.

JEL Classification: Q48, Energy: Government Policy; C13, Estimation: General; C54, Quantitative Policy Modeling

Keywords

China energy efficiency label household energy consumption behavior propensity score matching rebound effect

1. Introduction

Residential electricity consumption in China is growing at an extraordinary pace driven by rapid economic development, population growth and urbanization. According to China’s National Energy Administration, Chinese households consumed 1,094.9 billion kWh electricity in 2020, accounting for about 14.6% of total electricity consumption in China.¹ Compared with the residential electricity consumption in 2019, this number had increased by 6.9 percentage points.² Figure 1 presents the overall trend of residential electricity consumption in China, which reveals that electricity consumption of Chinese households kept growing in the last decades. Residential electricity consumption in China will keep this ascending trend in the coming years with the increase in income giving rise to increasing electric appliances adoption (Menezes and Zheng 2018).

Figure 1.

Residential electricity consumption in China.

The large base and rapid growth rate of China’s end-use energy imply that there is a substantial potential for energy savings from Chinese households. To improve energy efficiency and to reduce the growth of energy used by electric appliances, China has implemented a number of policy measures such as increasing block pricing for electricity and energy efficiency labeling. In particular, energy efficiency labeling – which provides clear and simple information on the performance index of domestic appliances’ energy efficiency rating – is a well-known alternative. Energy efficiency labeling has been widely introduced to many countries as the information provided by labels helps consumers make better decisions in both purchase and use (Davis and Metcalf 2016; Sammer and Wüstenhagen 2006; Stadelmann and Schubert 2018).

The aim of the paper is to examine the impacts of energy efficiency labeling on Chinese household’s use of domestic appliances. It is expected that higher energy efficiency leads to energy consumption reduction and hence domestic appliances labeled with higher energy efficiency will contribute to energy savings; however, it is argued that an increase in energy consumption from improved efficiency may undermine the energy savings from the technological improvement (Greening et al. 2000; Sorrell and Dimitropoulos 2008). To illustrate, following the technical efficiency improvement of delivering energy services, the consumption of energy service usually increases because of behavioral or other systemic responses. When the increased energy consumption is sufficiently large and offsets the energy savings achieved by improve energy efficiency (i.e., if the rebound is more than 100%), it is called as back-fire rebound effect.

From the perspective of mechanisms for rebound effect, three types of rebound effects have been theoretically identified in the literature (Herring 2006): direct rebound effect, indirect effect, and tertiary or general-equilibrium effect. Specifically, direct rebound effect results from increased energy consumption due to the more affordable energy services led by energy efficiency improvement in energy services; indirect rebound effect results from the saving in expenses from the reduction in energy-related costs, which allows the consumer to spend more on other goods and services (including energy); tertiary or general-equilibrium effect is the result of adjustments in supply and demand involving all producers and consumers in all sectors.

As can be found in the next section on literature review, a large number of current studies have shown empirical evidence of rebound effect related to energy efficiency labeling (e.g., Tiefenbeck et al. 2013; Waechter et al. 2015b). One typical reason for the existence of rebound effect is consumers’ misunderstanding of the information on energy efficiency labels. To be more specific, the energy consumption of an appliance depends on both energy efficiency class and the size of this product. For example, it is possible that a large product (such as a television with large screen size) labeled with better energy efficiency could use more energy than a small product (such as a television with small screen size) with lower energy efficiency class. Due to this misinterpretation of energy label, overall energy consumption may increase despite enhanced energy efficiency.

In the prior literature on energy efficiency labels, some studies have investigated the impacts of China’s energy efficiency label on consumer behaviors. For instance, Shen and Saijo (2009) and Wang et al. (2019) examined whether the CEL had altered consumer’s purchasing intentions and decisions of energy efficient appliances. Feng et al. (2010) analyzed the barriers for the effectiveness of energy efficiency labeling in China, and Shen (2012) studied the determinants of consumers’ willingness to pay (WTP) for products with the CEL. Yu et al. (2013) and Zhang et al. (2020) have estimated the rebound effects of energy consumption related to energy efficiency labeling.³ In a couple of studies which reviewed China’s energy efficiency policies, such as Price et al. (2011) and Lo (2014), the CEL policy has also been discussed.

Overall, the literature focusing on China’s energy efficiency label is very limited. With the findings from this small number of studies, the actual effectiveness of the CEL has not been well depicted. In particular, it is hard to know to what extent the current findings of the current empirical studies can be generalized to consumers in other places, as the sample size of the surveys in these studies is relatively small and the samples are collected from just one city or province (e.g., Beijing, Shanghai, and Liaoning).⁴ Moreover, the simple multiple regression model employed in these studies cannot prudently identify the causal relationship between households’ energy use and energy efficiency labeling. In contrast to the studies on other policies aiming to reduce energy consumption and those on other countries’ energy efficiency labeling, the impacts of China’s energy efficiency labeling have received too little attention in the current literature.⁵

Therefore, we focus on the effects of China’s energy efficiency labeling on Chinese households’ energy use in this study to bridge the wide research gap. The study is of novelty and significance for energy savings goals and climate change mitigation in the presence of the increasing share of residential electricity consumption in China. China’s aims to have CO₂ emissions peak before 2030 and to achieve carbon neutrality before 2060 reinforce the evaluation of energy efficiency label’s effects.

This study estimates the impacts of energy efficiency labeling policy on Chinese households’ energy consumption behavior based on the data of Chinese General Social Survey (CGSS) in 2015. The CGSS is one of representative national survey projects run by reputable Chinese academic institutions.⁶ The identification method applied in this paper is the propensity score matching (PSM), a counterfactual research method that matches treated (i.e., the households with an energy efficiency label on the electric appliance In question) and untreated observations (i.e., the households without energy efficiency label on their electric appliance in question) on the estimated probability of being treated (i.e., propensity score).

We find that China’s energy efficiency labeling policy leads to more use of domestic appliances, in terms of use frequency and use time per use, indicating the possible existence of rebound effect. This finding stays across different types of appliances, including washing machine, air conditioner, and water heater. The positive relationship between energy use and energy efficiency labeling remains robust in a number of robustness checks, which consider alternatives of matching algorithms, regression approaches, and dependent variable. The empirical results also show the effect is heterogeneous across households’ residential places, their income level and the gender of household head.

The remainder of this paper proceeds as follows. Section 2 provides information about China’s energy efficiency label. Section 3 reviews related studies. Section 4 presents the data and methodology. Section 5 reports the estimation results of the baseline model. Section 6 conducts robustness checks. Section 7 provides heterogeneity analysis, and Section 8 draws main conclusions and provides some policy implications.

2. Background of Chinese Energy Efficiency Label Policy

In China, the mandatory energy efficiency label policy, that is, China Energy Label (CEL), was implemented in March 2005. Prior to the mandatory policy, a voluntary endorsement label (i.e., China Energy Conservation Label) similar to the USA’s Energy Star Label was launched in 1999.⁷

With the CEL, manufacturers of domestic appliances are obligated to attach an energy efficiency label to their products. The labels should contain information about energy efficiency classes and other performance indicators. The energy efficiency class is ranked from one to three or five, with the lower number indicating higher energy efficiency. Regarding other performance indicators, one typical example is the information for energy consumption. The left label in Figure 2 presents the typical outlook of a CEL. From Figure 2, we can also observe that China’s energy label is similar with the European energy efficiency label (which is given on the right of Figure 2).

Figure 2.

Examples of energy efficiency label in China and Europe: (a) China and (b) Europe.

In the initial stage, only domestic refrigerators and air conditioners are listed in the CEL product catalogue, yet the types of products covered by the CEL have been widely broadened afterwards. For instance, in 2008, the labeling program was expanded to include induction cookers, electric heaters, and personal computer monitors (Feng et al. 2010). By 2020, more than 42 groups of products (including electric appliances, office equipment and many other products) manufactured by over 15,000 enterprises had been covered by China’s energy efficiency label policy. In the product catalogue, the frequently used domestic appliances such as refrigerators, air conditioners, washing machines, and water kettles are all included.

3. Literature Review

3.1. Impacts of Energy Efficiency Labeling

This paper relates to the broad literature on energy efficiency labeling and energy consumption. Many studies have examined the effects of energy efficiency labeling on consumers’ purchase decision of domestic appliances. The resulted energy savings from energy efficiency labeling have also been extensively assessed by the existing studies, of which many have found evidence of rebound effect of energy consumption.

3.1.1. Impacts on Consumer’s Purchase Decision

There is abundant empirical evidence, both in field and experiments, that consumers tend to be influenced by energy labels when making purchase decision of household appliances. For instance, Stadelmann and Schubert (2018) assessed the impact of the EU Energy Label and a newly designed monetary lifetime-oriented energy label with the assistance from a large online retailer in Switzerland. They found that the display of both labels can significantly increase the proportion of energy efficient appliances relative to the absence of any energy label. Newell and Siikamäki (2014) found that simple information on the monetary value of energy savings was the most important element guiding cost-efficient energy efficiency investments, with information on physical energy use and carbon dioxide emissions having additional but lesser importance. From a stated-choice experiment with 5,000 German households, Andor et al. (2020) indicated that displaying operating cost information promotes the choice of energy efficient durables. Andor et al. (2019) also confirmed that lifetime-cost information considerably increases consumers’ WTP for energy efficiency. However, in a real-stakes randomized controlled trial, Andor et al. (2019) pointed out that the EU energy label does not increase demand for energy efficient products, which is different from the findings by Sammer and Wüstenhagen (2006) and Stadelmann and Schubert (2018).⁸

Although energy efficiency labeling is regarded as a good option for energy efficiency improvement, a number of studies have found that these labels failed to nudge energy efficiency. Existing explanations for this failure include the presentation format and information content on the label. For instance, Davis and Metcalf (2016) argued that the information provided by labels is often too coarse to allow consumers to make efficient decisions. In addition, consumers tend to take the information in labels as given, without careful analysis (Davis and Metcalf, 2016). Waechter et al. (2015a) have given similar explanations why energy label’s effect on consumers’ actual product choices seems to be rather low.

3.1.2. Impacts on Energy Savings and Rebound Effect

A parallel literature investigates the role of energy efficiency labeling in energy savings. It is intuitive that energy efficiency labeling contributes to reducing energy consumption and resulted greenhouse gas emissions, as labels make it easier for consumers to make purchase decisions for energy efficient appliances by providing a clear and simple indication of the energy efficiency and other key features of electric appliances (Waechter et al. 2015a). Sanchez et al. (2008) and Banerjee and Solomon (2003) have confirmed the positive effect of energy efficiency labeling on energy conservation and emission reduction.

Meanwhile, another strand of literature has provided evidence for the existence of rebound effect in the energy use of household appliances. That is, rather than induce energy savings, energy label may in fact enhance energy consumption. For instance, Tiefenbeck et al. (2013) argued that the promotion of energy efficiency and the energy efficiency rating on the energy label critically neglect the role of actual energy consumption. One explanation of the rebound effect is that, because of behavioral factors, the label may mislead consumers to overestimate the role of energy efficiency. After all, the effectiveness of its measurement depends on consumers’ use and interpretation of the information provided (Waechter et al. 2015a). It is also possible that consumers focus excessively on energy efficiency information and thus neglect actual energy consumption (Waechter et al. 2015b).

3.2. Investigation Over China’s Energy Efficiency Labeling

In line with the large number of general literature on energy efficiency labeling, the line of analysis specifically concentrating on China’s energy efficiency labeling has also examined its effects on households’ purchase decisions (e.g., Shen and Saijo 2009; Wang et al. 2019) and energy consumption behavior (e.g., Feng et al. 2010). The determinants of consumers’ WTP for energy efficient domestic appliances and associated energy consumption rebound effect have also received some attention (e.g., Shen 2012; Yu et al. 2013).

From the perspective of consumers’ purchase decision, Shen and Saijo (2009) examined whether the CEL influences consumers’ choice of air conditioners and refrigerators via a hypothetical choice experiment conducted in Shanghai. They found that the energy label induced more purchase of energy efficient air conditioners and refrigerators, as the explicitly presented information about energy efficiency raised consumers environmental awareness. Wang et al. (2019) also studied how the CEL affects consumers’ purchase decision of energy efficient appliances applying the theory of planned behavior. Based on a survey conducted in Beijing in 2018, they confirmed the effectiveness of the CEL and pointed out the important role of individuals’ environmental awareness and subjective norms on the purchase of energy efficiency appliances.

Regarding household energy consumption behavior, Feng et al. (2010) found that, among the 600 observations collected in Liaoning province, 55 percent stated an intention to save electricity and 43 percent reported their use of more efficient light bulbs. Moreover, a quarter of the sample indicated that the energy efficiency label cannot affect their purchase decision at all. They further investigated the barriers to energy efficiency, showing that only a small fraction of the sample believe that the energy efficiency label can accurately describe the actual performance of products.

As for the influencing factors of the CEL’ effectiveness, Shen (2012) explored the determinants of consumers’ WTP for seven types of products with the CEL based on the data collected online. The study found that consumers’ environmental awareness and related experience in purchasing eco-labeled products have positive impact on consumers’ WTP for products with the CEL. Socio-demographic characteristics such as gender, age, education, and household income are also important. In addition, consumers’ WTP differs across various types of products.

Despite the positive impact of the CEL on households’ choice of energy efficient appliances and their energy saving behavior, the existing literature has also proved the rebound effect of energy consumption related the CEL. One example is the study by Yu et al. (2013) that evaluated the direct and indirect rebound effects in household energy consumption taking Beijing as a case.⁹ They show that increases in energy efficiency of air conditioners, clothes washers, microwave ovens and cars leads to more energy consumption, whereas there is no evidence of rebound effect (either direct or indirect effect) for refrigerators, electric fans, gas showers, televisions, and private computers.

There are also a number of studies conducting a thorough survey of related energy efficiency policies in China, in which energy efficiency labeling is included; for instance, the work by Price et al. (2011) and Lo (2014). However, all these studies focused on the description of appliance standards and energy efficiency labeling policy, without much in-depth analysis. In particular, the labeling policies reviewed by Lo (2014) are limited to the measures applied in the transportation sector and buildings.

It can be seen that, while there is an extensive literature on the effects of energy efficiency labeling, China’s energy efficiency labeling remains an under-researched issue. The research on the CEL’s impacts on households’ energy use behavior is even less. The study by Feng et al. (2010) that assessed households electricity savings and consumer behavior is an exception. However, they simply demonstrated the descriptive outcomes, without any in depth discussion based on econometric models.

Hence, this paper attempts to examine the CEL’s effects on Chinese households’ energy consumption behaviors. This paper contributes to the current literature in three ways. First, this paper extends the previous research on energy efficiency label. To the best of our knowledge, it is the first attempt to examine the CEL’s impacts on Chinese households’ energy use behavior. Second, we provide solid empirical evidence using the authoritative national micro-data, unlike the existing studies that are conducted based on limited observations. Third, we carefully identify the causal relationship between the CEL and households’ energy consumption behavior with a number of reliable approaches.

4. Data and Methodology

4.1. Data Description

This section briefly describes the sources of data and the construction of variables. Our data are mainly from the CGSS 2015, which includes an energy modular in addition to the key geographical modular and other modules (such as those for law and institution).¹⁰ The variable, $Temperature$ , which indicates the number of days where the average daily temperature of a province or municipality is greater than or equal to 28°C, is from the China Meteorological Administration.

4.1.1. Dependent Variable

According to the China National Bureau of Statistics, 96 percent of Chinese households had got laundry facilities by the end of 2019, indicating that washing machines have become a necessity for households in China. In addition, there is usually just one washing machine in a household; however, the number of refrigerates, air conditioners, or heaters tends to be more than one.¹¹ Hence, in the main part of this study, we concentrate on the analysis of households’ energy consumption behavior of washing machine. In practice, we look at this consumption behavior from three perspectives, including the use frequency (i.e., $Frequency)$ , length of time per use (i.e., $Time$ ) and total use (which is equal to the interaction of the first two variables, i.e., $Frequency \times Time$ ) of washing machines. The use frequency of washing machine is derived from the question “How often do you use washing machine” in the questionnaire, and the use time of it is obtained from the question “How long do you use it for each time?” The specific assignments to these questions are given in more details in Table 1.¹²

Table 1.

Variables and Questions in the Questionnaire.

Variables	Questions in the questionnaire
Dependent variables
Frequency: use frequency	Never = 0; less than once per month = 1; two to three times per month = 2; one to three times per week = 3; fourth to six times per week = 4; once per day = 5; twice per day = 6; over three times per day = 7.
Time: time taken per use	Less than fifteen minutes = 1; fifteen to twenty-nine minutes = 2; thirty to forty-four minutes = 3; forty-five to fifty-nine minutes = 4; one to two hours = 5; two to four hours = 6; four to six hours = 7; more than six hours = 8.
Total use: interaction of frequency and time	Total use = Frequency × Time
Independent variables
Label: energy efficiency label	With energy efficiency label = 1; 0, otherwise.
Control variables (demographic)
Education	Years of education
Marital status	Married = 1; 0, otherwise.
Income	Logarithm of income
Age	Square of age
Household scale	Number of family members
Political status	Communist party member = 1; 0, otherwise.
Hukou: census register	Urban = 1; 0, otherwise.
Control variables (housing characteristics)
Ownership: housing ownership	Not own property rights = 1; 0, otherwise.
Housing area	Square of residential housing area
Control variables (appliance characteristics)
Quantity: number of washing machines	Number of washing machines in the residential place
Volume: volume of washing machine	Less than 3 kg=1; 3–5 kg = 2; 5–7 kg = 3; over 7 kg = 4.
Power: washing machine power	Less than 300 W = 1; 300–499 W = 2; 500–699 W=3; 700–999 W = 4; 1,000–1,499 W = 5; over 1,500 W = 6.
Control variables (electricity pricing)
ToU: time of use	Time of use tariff = 1; 0, otherwise.
Control variables (climate factor)
Temperature: temperature	Number of days when the average temperature of a city in 2015 was over 28°C

4.1.2. Independent Variable

The key independent variable (i.e., $Lable$ ) is a dummy variable, indicating whether the washing machine has an energy efficiency label or not. It equals to one if the washing machine has an energy efficiency label; otherwise, zero. There are six options for the question about the energy efficiency label of washing machine (or other appliances). The first option indicates that there is no label, and the other five show that the washing machine has a label, with two to five revealing that the energy efficiency of the washing machine ranks at level one to level five. The information looks like as follows:

● Question: How about the energy efficiency label of this washing machine?

● Answer list:

(1). no label; (2). ranks as class 1; (3). ranks as class 2; (4). ranks as class 3; (5). ranks as class 4; (6). ranks as class 5.

To make the identification strategy feasible, we set the variable $Lable$ equal to zero if the observation selected the first option, and set the value as one if the observation selected any of the left five alternatives, combining them together.

4.1.3. Control Variables

The existing literature (e.g., Ward et al. 2011) showed that, in addition to energy efficiency labeling, households’ energy consumption behaviors are also affected by factors such as demographic factors, characters of housing and appliances, and so on. Therefore, in this study we avoid the omitted variables bias by relying on five types of control variables, including: (i) demographic variables (i.e., $Education$ , $Marital status$ , $Income$ , and $Scale$ of household, $Political status$ and $Hukou$ ), (ii) housing characteristics variables (i.e., $Ownership$ of housing and $Housing area$ ), (iii) appliance characteristics variables (i.e., $Quantity$ , $Volume$ and $Power$ of the type of appliance in question), (iv) whether the electricity pricing applies time of use tariffs or not (i.e., $ToU$ ), and (v) climatic variable (i.e., $Temperature$ of the household’s residential place). More details of these variables can be found in Table 1.

In the process of data clearing, we remove the observations with missing values for the variable regarding energy efficiency label. For other variables, we have deleted the observations responding with “no idea,”“not applicable,”“indifferent,” or “refuse to answer.” In addition, we have also eliminated the observations with outlier. Finally we totally have 2330 observations in our sample. Table 2 shows the main variables and descriptive statistics of the data used in this study.

Table 2.

Descriptive Statistics.

Variables	Obs.	Mean	Std. Err	Min.	Max.
Frequency	2,323	3.380	1.282	0	7
Time	2,284	3.395	1.127	1	8
Total use (=Frequency × Time)	2,280	11.430	5.228	0	40
Label	2,330	0.557	0.497	0	1
Education	2,328	9.481	4.579	0	18
Marital status	2,330	0.797	0.402	0	1
Income (10 ³ RMB)	2,161	78.125	195.407	0	5,000
Age (years)	2,330	55.485	16.574	24	99
Household scale (persons)	2,325	2.949	1.371	1	13
Political status	2,320	0.125	0.331	0	1
Hukou	2,330	0.523	0.450	0	1
Housing ownership	2,330	0.119	0.324	0	1
Housing area (m²)	2,330	116.911	82.667	5	1,000
Quantity of washing machine	2,330	1.021	0.151	1	3
Volume of washing machine	2,263	2.481	0.798	1	4
Power of washing machine	1,780	2.842	1.303	1	6
Time of Using (ToU)	2,330	0.297	0.457	0	1
Temperature (days)	2,330	106.806	37.822	13	208

We can see that the average use frequency of washing machine is between one to six times per week, and the average time length per use is in between thirty and fifty-nine minutes. The mean of the label variable indicates that over half of the households have an energy efficiency label on their washing machine, reflecting the enforcement of the energy efficiency policy launched in 2005.¹³ The mean of educational time is 9.48 years, a little bit longer than the compulsory education time in China (which is nine years). Regarding the political status, most observations are not communist party member, consistent with the level of the population.¹⁴ Overall, there are three persons in each household on average, in consistence with the one-child policy implemented in China for decades. Only a very small fraction of households live in a rented house, in line with the level suggested by the China Household Finance Survey (CHFS) in 2017.¹⁵ It can also be checked that the statistics of other variables, such as income, age, urbanization, and housing area, align with the level shown in other data sources.¹⁶

In the following regression, we take the logarithm of income to make it more approximately normal. In addition, we use the square of age and housing area considering the potential inverted U-shape relationship between them and consumers’ energy use.

4.2. Identification Methodology: PSM

In our study, whether a household owns an electric appliance with energy efficiency label or not depends on individual characteristics such as income, age, and educational level. This implies that the treatment (i.e., having energy efficiency labels on their appliances) is not randomly assigned and self-selection bias occurs in ordinary least squares (OLS) estimation. Therefore, we apply the propensity score matching (PSM), a quasi-experimental technique designed to mimic some of the particular characteristics of a randomized controlled trial to address the non-random assignment and self-selection bias (based on observed characteristics). The PSM approach estimates the effect of a treatment by accounting for the covariates that predict the treatment being received.

To identify the treatment effect by PSM, we estimate the average treatment effect (ATT) by calculating the average difference between the treatment group and the control group. The ATT is denoted as

ATT = E (Y_{1 i} - Y_{0 i} | Labe l_{i} = 1)

where $Y_{i}$ denotes household $i$ ’s energy consumption behavior, which will be specifically measured by $Frequency$ , $Time$ , and $Total use$ . $Labe l_{i}$ is a binary variable indicating whether the household $i$ is treated ( $Labe l_{i} = 1$ ) or untreated ( $Labe l_{i} = 0$ ). $E (Y_{1 i} | Labe l_{i} = 1)$ represents household $i$ ’s energy consumption behavior with energy efficiency label, which is observable, whereas $E (Y_{0 i} | Labe l_{i} = 1$ ) is unobservable. As $E (Y_{0 i} | Labe l_{i} = 1$ ) is not observed, we address this issue by matching treated and untreated observations based on the estimated probability of treatment assignment conditional on observed baseline characteristics (i.e., propensity score).

The propensity score, conditional probability of treatment given the covariates, is calculated as $P (X_{i}) = ⪻ (Labe l_{i} = 1 | X_{i})$ , where $Labe l_{i} = 1$ indicates that household $i$ has an energy efficiency label on its electric appliance in question (which is washing machine in the baseline analysis and becomes air conditioner and water heater in robustness checks), and $X_{i}$ denotes the vector of covariates for household $i$ . Whether a household’s appliance has the energy efficiency label needs to be independent of their energy consumption behavior conditional on $X_{i}$ , that is, $(Y_{1 i}, Y_{0 i}) ⊥ ⊥ Labe l_{i} | X_{i}$ , there should be no unobserved outcomes affecting the treatment. It is almost impossible to control for all possible predictors, yet we try our best to reduce selection and endogeneity biases by considering as many variables as possible in our covariate set, following Eriksen and Wiese (2019).¹⁷

To identify the treatment effect by PSM, the propensity score should be balancing; that is, conditional on the propensity score, the distribution of observed baseline covariates should be similar between treated and control groups. Hence, based on the estimated propensity score, we test the balance and common support assumption to check the property of the propensity score, before estimating the treatment effect of energy efficiency labeling on households’ energy consumption behavior.

5. Estimation Results

5.1. Propensity Score Estimation

The first stage of the PSM approach is to estimate the propensity score. We use a logit model to calculate the propensity score (i.e., the predicted probabilities of having energy efficiency label) for each observation, following Rosenbaum and Rubin (1985). In the logit model, an energy efficiency label dummy is on the left-hand side and the determinants of it are on the right-hand side as follows

Logit (Labe l_{i} = 1) = β_{0} + β_{1} X_{i} + ε_{i}

(1)

where $Labe l_{i}$ is a dummy variable equal to 1 if an observation has energy efficiency label for washing machine and 0 otherwise. $X_{i}$ is a set of factors that may affect an individual’s choice of a washing machine with an energy efficiency label, including demographic characteristics (i.e., education, political status, hukou, household scale, marital status, age), housing characteristics (i.e., housing ownership and housing area), appliance characteristics (i.e., number, volume, and power of washing machines) and the application of time-of-using electricity pricing, and $ε_{i}$ is the error term.

5.2. Balancing Test

In the second stage, we use the estimated propensity score to match the treatment and control groups. A one-to-four matching approach with replacement was adopted while using the nearest neighborhood matching strategy. In other words, we choose four households from the ones without energy efficiency label for washing machine as a match for a treatment household regarding their closest propensity score. To construct the covariate vector used for matching, we consider the observable information provided by education, marital status, income, age, household scale, political status, Hukou, housing ownership, housing area, application of ToU pricing, and number, volume, and power of washing machine. The number of observations becomes 1,619 after the PSM as the observations out of the common support have been dropped from the sample.

To confirm that the matching procedure balances the two groups successfully, we implement balancing tests on the differences in baseline characteristics between treatment and control before and after matching. Table 3 provides the balancing test results for the PSM. The differences in covariate means between the treatment and control groups are sufficiently small, and the t test shows that there is no statistically significant differences between the two groups. Hence, the balancing property of the propensity score is satisfied and the matched result is desirable.

Table 3.

Balancing Property Test of PSM.

Variables	Matched or not	Means		Std. Bias (%)	Bias red. (%)	t value	p > \|t\|
Variables	Matched or not	Treated	Control	Std. Bias (%)	Bias red. (%)	t value	p > \|t\|
Education	U	9.827	9.058	16.8		3.35	.001
Education	M	9.804	9.819	−0.3	98.1	−0.07	.942
Marital status	U	0.819	0.801	4.5		0.90	.370
Marital status	M	0.818	0.825	−1.6	63.6	−0.37	.711
Income	U	10.784	10.585	19.3		3.87	.000
Income	M	10.775	10.792	−1.7	91.4	−0.37	.713
Age	U	3,223.4	3,421	−10.9		−2.15	.032
Age	M	3,230.2	3,233	−0.2	98.6	−0.03	.973
Household scale	U	2.964	2.901	4.8		0.96	.339
Household scale	M	2.965	2.947	1.4	71.7	0.31	.756
Political status	U	0.125	0.112	4.2		0.82	.410
Political status	M	0.127	0.118	2.7	35.1	0.59	.556
Hukou	U	0.535	0.488	9.5		1.88	.061
Hukou	M	0.536	0.537	−0.2	97.8	−0.05	.964
Ownership	U	0.096	0.121	−8.1		−1.62	.105
Ownership	M	0.097	0.093	1.3	83.6	0.31	.757
Housing area	U	23,357	18,836	9.8		1.84	.066
Housing area	M	20,678	20,165	1.1	88.6	0.33	.742
Quantity	U	1.014	1.028	−8.7		−1.79	.073
Quantity	M	1.014	1.012	1.9	78.6	0.55	.582
Volume	U	2.563	2.301	33.5		6.62	.000
Volume	M	2.547	2.536	1.4	95.9	0.31	.760
Power	U	2.986	2.630	27.6		5.49	.000
Power	M	2.962	2.947	1.2	95.7	0.25	.800
ToU	U	0.296	0.292	0.9		0.17	.863
ToU	M	0.299	0.293	1.4	−61.9	0.31	.756
Temperature	U	104.96	105.09	−0.4		−0.07	.942
Temperature	M	105.57	106.91	−3.9	−945	−0.83	.409

5.3. Test of Common Support Assumption

The inferences about treatment effects cannot be made for a treated individual if there is not a comparison individual with a similar propensity score. Therefore, it is necessary to ensure that there is overlap in the range of propensity scores across treatment and comparison groups (called “common support”) once a propensity score has been calculated for each observation. Besides overlapping, the propensity score should have a similar distribution (“balance”) in the treated and comparison groups. The overlap of the distribution of the propensity scores regarding frequency across treated and comparison groups is displayed in Figure 3. The propensity score distribution of use time and total use across the two groups are given in Figures A1 and B1.

Figure 3.

Distribution of propensity scores by treated and control groups (frequency).

In Figure 3, we compare the distribution of propensity scores of treated group and control group, with the results before and after matching in the two sub-figures, respectively. The horizontal axis in the sub-figures show the propensity scores, while on the vertical axis we have the proportion of observations in each distribution bin. The solid line shows the distribution of the treatment group, whereas the dotted line indicates the distribution of control group. The similar kernel density of propensity score with observations in the two groups suggest that the extent of overlap is satisfactory.

5.4. Treatment Effects

After creating a balanced propensity score, we examine the average treatment effect of energy efficiency labeling on households’ use of electric appliance considering washing machine as example, including the use frequency, use time, and total use (i.e., the interaction terms of use frequency and use time) of them. We calculate the impact by comparing the means of outcomes across the treatment observation and their matched pairs. The results are present in Table 4.

Table 4.

Average Treatment Effect With PSM.

Variables	Matching method	ATT	Std. Err	t value
Frequency	k-Nearest (k = 4)	0.197***	0.071	2.79
Time	k-Nearest (k = 4)	−0.058	0.067	−0.86
Total use	k-Nearest (k = 4)	0.749***	0.294	2.55

Note. We applied bootstrap method and repeated sampling 400 times.

p < .1. **p < .05. ***p < .01.

The results in Table 4 indicate that energy efficiency labeling has significantly increased households’ use frequency and total use of washing machine, whereas the impact on use time is statistically insignificant across all the matching strategies. This outcome is reasonable, as the use time of washing machine is relatively fixed given the defaulted setting for specific purposes. The odds ratio (i.e., the odds that an outcome occurs given a particular exposure, compared to the odds of the outcome occurring in the absence of that exposure) for use frequency is 0.197, and that for total use is 0.749, indicating that energy efficiency labeling leads to more frequent use and therefore more total use of washing machine. If the increased energy consumption is sufficiently large and offsets the energy savings achieved by improve energy efficiency, the energy efficiency labeling induces a rebound effect on households’ energy use.

6. Robustness Checks

We test the robustness of the baseline estimates with a battery of checks in this section. First, we conduct the PSM using different matching algorithms. Second, we examine the estimation results obtained with alternative regression strategies. Third, we tackle potential endogenous problem with instrumental variable model. Fourth, we examine the effect of energy efficiency labeling on households’ energy use behavior taking air conditioner and water heater as examples.

6.1. Applying Alternative Matching Algorithms

In addition to the estimation outcomes derived with the above-mentioned nearest neighborhood matching, we also present the results obtained with other three alternative matching methods, including radius matching, kernel matching, and Mahalanobis matching.

With radius matching, all untreated observations within the specified radius of the treated observation are used, and they all receive the same weight (regardless of how close they are to the treated observations value). With kernel matching, the closer the treated and untreated observations are based on the propensity score, the larger weight is given to the untreated observation. Thus, the more “similar” the untreated observations are to the treated observations, the more weight they are given. Mahalanobis distance matching (MDM) is different from the PSM that works by pairing units that have similar propensity scores, as the MDM works by pairing units that are close based on a distance called the Mahalanobis distance, which is like a scale-free Euclidean distance. For two observations to have a Mahalanobis distance of 0, they must have identical covariate values. The more different the covariate values are, the larger the Mahalanobis distance is. The idea is that if you find control units close to the treated units on the Mahalanobis distance, each pair will have similar covariate values, and the distribution of the covariates in the treatment groups in the matched sample will be similar.¹⁸

The results in Table 5 show that the main findings remain robust across all the matching strategies. Specifically, the odds ratio for use frequency is 0.239, 0.226, and 0.193 under the three alternative matching approaches, respectively, compared to 0.197 estimated with nearest neighbor method. Similarly, the odds ratio for total use is 0.692, 0.729, and 0.631, respectively, not vary much from the result with nearest neighbor algorithm. The mean of odds ratio for use frequency is 0.214, and the average odds ratio for total use is 0.7, confirming that energy efficiency labeling leads to more frequent use and therefore more total use of washing machine.

Table 5.

Average Treatment Effect With PSM – Alternative Matching Methods.

Variables	Matching method	ATT	Std. Err	t value
Frequency	k-Nearest (k = 4)	0.197***	0.071	2.79
	Radius (radius = 0.01)	0.239***	0.067	3.59
	Kernal	0.226***	0.066	3.45
	Mahalanobis	0.193***	0.068	2.82
	Means	0.214
Time	k-Nearest (k = 4)	−0.058	0.067	−0.86
	Radius (radius = 0.01)	−0.041	0.063	−0.65
	Kernal	−0.028	0.062	−0.45
	Mahalanobis	−0.001	0.066	−0.01
	Means	−0.032
Total use	k-Nearest (k = 4)	0.749**	0.294	2.55
	Radius (radius = 0.01)	0.692**	0.280	2.47
	Kernal	0.729***	0.275	2.65
	Mahalanobis	0.631**	0.315	2.00
	Means	0.700

Note. We applied bootstrap method and repeated sampling 400 times.

p < .1. **p < .05. ***p < .01.

6.2. Examining Energy Efficiency Labels at Different Ranks

Considering that there are several energy efficiency ranks on the label (from 1 to 5), and that when the label attached indicates that the appliance is energy-consuming, the consumers will not necessarily use this appliance more, in this part we conduct separate estimation over the effect of labels at different ranks.

The estimation results are presented in Table 6. It shows that when the label informs that the energy efficiency of a domestic appliance is at the highest level (i.e., ranking at the first class), consumers will use this appliance more intensively. Although the impact of energy efficiency label on use time per use is insignificant, the total use of energy is increasing with energy efficiency. Possibly, this is because consumers overestimate the importance of energy efficiency, as the existing literature has found (e.g., Waechter et al. 2015a). On contrary, the labels indicating a lower energy efficiency of appliance have negligible impact on consumers’ energy consumption behavior. That is, only the label of sufficiently high energy efficiency could possibly induce more energy consumption and thus lead to rebound effect.

Table 6.

Effect of Labels at Different Ranks.

Variables	Frequency		Time		Total use
Label at rank 1	0.237**(0.097)	0.282***(0.097)	−0.112(0.084)	−0.080(0.093)	1.003**(0.417)	0.832**(0.413)
Label at rank 2	0.167**(0.084)	0.103(0.090)	−0.054(0.080)	−0.025(0.088)	0.657*(0.382)	0.398(0.392)
Label at rank 3	0.126(0.103)	0.061(0.108)	0.039(0.119)	0.035(0.131)	0.727(0.512)	0.370(0.526)
Label at rank 4	−0.224(0.228)	−0.192(0.262)	−0.190(0.308)	−0.206(0.299)	−1.150(1.248)	−0.878(1.135)
Label at rank 5	0.128(0.360)	−0.001(0.358)	−0.083(0.361)	0.082(0.314)	3.046(1.896)	2.976**(1.505)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effects	No	Yes	No	Yes	No	Yes
Observations	1,457	1,457	1,448	1,448	1,439	1,439

Note. In parenthesis are robust standard error.

p < .1. **p < .05. ***p < .01.

6.3. Using Alternative Regression Approaches

We also check the estimation results using alternative regression approaches. Considering that use time per use is not significantly affected by energy efficiency labeling, we focus our analysis on use frequency and total use time here. Regarding use frequency, we apply ordered logit model for regression as the dependent variables have more than two categories and the values of each category have a meaningful sequential order (where a value is indeed “higher” than the previous one). For total use, we look at the OLS estimator. We use the weights from PSM when conducting the ordered logit and OLS regressions.

The results in Table 7 indicates that the energy efficiency labeling of washing machine induces more energy use, from the perspective of use frequency and total use, consistent with the findings with PSM approach. With the coefficients of labeling being 0.169 and 0.737 for use frequency and total use, respectively, the magnitudes of impact are also similar with the findings above.

Table 7.

Estimation Results With Alternative Regression Methods.

Variables	Frequency		Total use
Label	.169***(.063)	.146**(.067)	.738**(.289)	.506*(.295)
Education	.003(.012)	.010(.011)	−.018(.050)	−.039(.046)
Political status	−.062(.112)	−.019(.123)	−.564(.503)	−.585(.493)
HuKou	.282***(.083)	.269***(.091)	1.065***(.406)	1.016**(.416)
Household scale	.085***(.029)	.130***(.031)	.358***(.138)	.439***(.136)
Marital status	.237**(.094)	.294***(.085)	.794**(.402)	.858**(.373)
Age	−.000**(.000)	−.000***(.000)	−.000***(.000)	−.000***(.000)
Income	.042(.054)	.064(.050)	.212(.228)	.233(.186)
Housing area	.000**(.000)	.000(.000)	.000*(.000)	.000(.000)
Housing ownership	.268***(.104)	.296**(.117)	.157(.567)	.050(.574)
Washing machine quantity	.285(.206)	.320(.259)	2.977**(1.194)	2.811***(.983)
Washing machine volume	−.056(.043)	.004(.047)	.240(.190)	.419**(.198)
Washing machine power	.078***(.026)	.077***(.029)	.381***(.118)	.438***(.126)
Tou pricing	.213***(.081)	.075(.086)	.759**(.356)	.706*(.366)
Temperature	.008***(.001)	.054***(.011)	.009**(.004)	.146***(.049)
City fixed effects	No	Yes	No	Yes
Observations	1,457	1,457	1,439	1,439

Note. In parenthesis are robust standard error.

p < .1. **p < .05. ***p < .01.

We can also find that Hukou, household scale, marital status, washing machine power, time of use pricing and temperature have significantly positive impacts on both the use frequency and total use of washing machine, whereas the impact of the age of household head is negative. These outcomes are intuitive as the living standard of urban residents is usually higher than that of rural residents, which implies that the need for laundry of households in urban areas is more. It is also straightforward that when the household scale is larger, the use of washing machine is more. Likewise, compared with singles, married households need to do more laundry and hence occur more use of washing machine. When the time-of-use electricity pricing is applied, households may find that it is cheaper to use the washing machine at specific times; hence, they might increase the overall use of the washing machine. Regarding temperature, it is reasonable to wash clothes more frequently and use washing machine more in areas where the average temperature is higher. The negative coefficient of age can be explained by the fact that the elders tend to be more thrift than younger generations and hence are less likely to use washing machines. The above-mentioned explanation can be generally applied to other electric appliances beyond washing machine.

6.4. Estimation With Instrumental Variable

The PSM approach reduces selection bias (introduced by the selection of individuals) by equating groups based on the covariates. However, there might be endogeneity caused by reverse causality and unobservable variables. For instance, individuals using more energy might be more willing to purchase electric appliances with energy efficiency labels. Individual’s environmental concern affects both individuals’ energy use behavior and the selection of electric appliances, yet this variable is unobservable in our data.

To deal with the potential endogenous problem, we apply instrumental variable (IV) model and run the regressions again. An instrumental variable needs to be related to the endogenous explanatory variable (i.e., energy efficiency label of electric appliances here) yet not relate to the error term and independent variable (i.e., households’ energy use behavior here). In this study, we employ the mean value of energy efficiency label of other households in the same community, which is denoted as $AveLabel$ . On one hand, the average level of this variable in a community is strongly related to that of the household of interest, as households in the same community have a lot in common; on the other, the average level of energy efficiency label of other households in the community has negligible impact on the energy use of a household.

Specifically, we apply the two-stage least squares (2SLS) regression as follows:

Labe l_{i} = α_{0} + α_{1} Avelabe l_{i} + α_{2} X_{i} + μ_{i}

(2)

Us e_{i} = β_{0} + β_{1} Labe l_{i} + β_{2} Z_{i} + μ_{i}

(3)

where $i$ represents individual $i$ . $X_{i}$ and $Z_{i}$ are the vectors of control variables in the two equations, respectively. Equations (2) and (3) denote the regression in the first and the second stage.

Before using the instrumental variable method for subsequent analysis, we test the validity of the instrumental variable and present the test results at the bottom of Table 8. We first assess the strength of identification based on a Langrange Multiplier (LM) test for under-identification using the Kleibergen and Paap (2006) rk statistic. The statistics are 16.009 and 20.927 for $Frequency$ and $Total Use$ , respectively and the p value is 0, indicating that we can reject the null hypothesis that the instrumental variable cannot be identified. We second use the Cragg-Donald Wald statistic based on Stock and Yogo (2005) to test whether the instrumental variable is weak or not. The statistics are 20.434 and 25.390, respectively, larger than the critical value at the 10% level (which is 16.38); hence, we reject the null hypothesis that the instrumental variable is weak at least at the 10% level.

Table 8.

Estimation Results With Instrumental Variable.

Variables	Frequency		Total use
	First stage	Second stage	First stage	Second stage
	(1)	(2)	(5)	(6)
Avelabel	0.224***(0.055)		0.248***(0.053)
Label		1.213*(0.735)		5.637**(2.619)
Constant	Yes	Yes	Yes	Yes
Provincial fixed effects	Yes	Yes	Yes	Yes
Kleibergen-Paaprk LM statistic		16.009		20.927
LM p-value		.000		.000
Cragg-Donald Wald statistic		20.434		25.390
Observations	1,442	1,442	1,424	1,424

p < .1. **p < .05. ***p < .01.

The estimation results with instrumental variable are also presented in Table 8.¹⁹ Specifically, columns (1) and (2) demonstrate the results with frequency as dependent variable and the last two columns show the results with total use (i.e., the interaction of frequency and use time per use) as the explained variable. In columns (1) and (3), we have the estimation outcomes in the first stage of regression and in columns (2) and (4) we present the estimates of the second stage estimation. We can see that the selection of energy efficiency label of washing machine in a household positively relates to the average level in the community. In addition, the above finding that the energy efficiency labeling leads to more energy use, in terms of both use frequency and total use, is confirmed.

In addition, we conduct an auxiliary regression including IV as a control to examine whether the IV would affect the dependent variable through other channels except for energy efficiency labels. In this regression, we also control city fixed effect, considering that the mean value of energy efficiency labels of other households in the same community might be correlated with the random error if the regional fixed effects are not controlled for. The estimates results are shown in Table 9. The coefficients of instrumental variable are insignificant, further indicating that our instrument variable is valid.

Table 9.

Auxiliary Regression for the Instrumental Variable Method.

Variables	Frequency	Total Use
Label	.141**(.067)	.493*(.296)
Avelabel	.723(.493)	2.191(2.020)
Control variables	Yes	Yes
City fixed effects	Yes	Yes
Observations	1,457	1,439

p < .1. **p < .05. ***p < .01.

6.5. Estimation With Endogenous Switching Regression Method

We also conduct endogenous switching regression (ESR) method proposed by Lokshin and Sajaia (2004) to further deal with endogeneity problem. The ESR models are natural extensions of classical experimental designs that allow tests of assumptions about the exogeneity of effects from survey data. Since the energy efficiency label is one of the significant appliance attributes affecting households’ choices and that households would determine whether to buy an application with energy efficiency labels and how much energy is consumed, the model is set as follows:

I_{i} = 1 if γ Z_{i} + μ_{i} > 0

(4)

I_{i} = 0 if γ Z_{i} + μ_{i} \leq 0

(5)

Us e_{1 i} = β_{1} X_{1 i} + ε_{1 i} if I_{i} = 1

(6)

Us e_{2 i} = β_{2} X_{2 i} + ε_{2 i} if I_{i} = 0

(7)

where $I_{i}$ is the criterion function that determines whether a household chooses a washing machine that has energy efficiency label on it or not (with $I_{i} = 1$ if chooses; $I_{i} = 0$ , otherwise).²⁰ Equations (6) and (7) are regression equations; $Us e_{ji}$ are the dependent variables (i.e., households’ energy use behavior); $Z_{i}$ is a vector of variables affecting households’ choice of appliances (with or without energy efficiency label); $X_{ji}$ are vectors of (weakly exogenous) variables influencing households’ energy consumption behaviors; $γ$ , $β_{1}$ , and $β_{2}$ are vectors of parameters; and $μ_{i}$ and $ε_{ji}$ are residual error terms.

Table 10 reports the estimates of the endogenous switching regression model estimated by full information maximum likelihood. Columns (1) presents the estimated coefficients of the selection equation on choosing energy efficiency labeled appliance or not. We can see education, the volume and power of washing machine are the main drivers of households’ decision to choose an appliance with energy efficiency label. In contrast, the quantity of washing machine one household owns, as well as the pricing of electricity play significantly negative role in their selection of appliances with energy efficiency label.

Table 10.

Estimation Results With Endogenous Switching Regression Model.

Variables	Frequency			Total use
	(1)	(2)	(3)	(4)	(5)	(6)
	Selection	With-label	Without-label	Selection	With-label	Without-label
Education	.018*(.010)	.025**(.012)	−.010(.014)	.023**(.010)	.047(.054)	−.092(.068)
Political status	.030(.112)	−.248**(.118)	−.030(.147)	.053(.110)	−.346(.530)	−.599(.761)
HuKou	.025(.087)	.149(.091)	.327***(.115)	.039(.085)	.262(.413)	.680(.584)
Household scale	−.016(.029)	.081***(.031)	.056(.037)	−.007(.028)	.320**(.139)	.384**(.193)
Marital status	.054(.088)	.162*(.096)	.135(.116)	.055(.087)	.624(.435)	.521(.594)
Age	−.000(.000)	−.000(.000)	−.000**(.000)	.000(.000)	−.000(.000)	−.000***(.000)
Income	.066(.040)	.087*(.046)	.060(.051)	.067*(.040)	.403*(.211)	−.188(.264)
Housing area	.000(.000)	.000*(.000)	.000(.000)	.000(.000)	−.000(.000)	−.000(.000)
Housing ownership	−.129(.113)	.138(.129)	.155(.144)	−.129(.110)	−.122(.583)	.884(.744)
Washing machine quantity	−.406*(.233)	.592*(.307)	.026(.258)	−.405*(.215)	2.546*(1.386)	4.120***(1.360)
Washing machine volume	.250***(.045)	−.055(.060)	−.039(.072)	.235***(.045)	.451(.280)	−.719**(.317)
Washing machine power	.112***(.028)	.088**(.035)	.044(.043)	.101***(.028)	.478***(.156)	−.306(.192)
ToU pricing	−.152*(.081)	.054(.091)	.064(.108)	−.167**(.080)	.380(.419)	1.403***(.541)
Temperature	.005*(.003)	.007**(.003)	.015***(.005)	.004(.003)	.005(.013)	−.012(.021)
Constant	−2.089***(.573)	.307(.826)	1.026(.715)	−1.888***(.555)	−1.578(3.816)	6.716*(3.686)
ρ1		−.060(.262)			−.084(.292)
ρ2			−.126(.257)			−.126(.023)
Province fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Log likelihood	−3,407.0			−5697.7
Observations	1,621	1,621	1,621	1,597	1,597	1,597

p < .1. **p < .05. ***p < .01.

Columns (2) and (3) demonstrate the estimates of the coefficients of correlation between the random errors in the system of equations. The estimated coefficient of correlation between the selection equation and the label-households’ use frequency function, $ρ_{1}$ , is not significantly different from zero, implying that the hypothesis of absence of sample selectivity bias may not be rejected in this case. This is the same case with the estimated coefficient of correlation between the selection equation and the non-label-households’ use frequency function $ρ_{2}$ . Columns (5) and (6) give the estimates regarding total use, and the insignificant result of $ρ_{1}$ and $ρ_{2}$ show that the sample selectivity bias can be neglected.

Table 11 presents the average treatment effects of energy efficiency label on households’ energy use behavior. In the counterfactual case (i.e., label-households choose appliances without energy efficiency label), households who actually choose labeled appliances would have use appliances less intensively if they choose appliances without energy efficiency label. This is also the case regarding the total use of energy. These results imply that the selection of energy efficiency labeled appliances significantly increase consumers’ energy use in terms of both use frequency and total use.

Table 11.

Average Treatment Effect With Endogenous Switching Regression Model.

Variables	Frequency	Total use
ATT	.390	10.978
Std. Err	.026	.089
t value	15.077	122.716

6.6. Using Alternative Measurement of Dependent Variables

In the above analysis, we used ordered categorical dependent variable (i.e., 0, 1, 2, 3, 4, 5, 6, 7) to measure the use frequency as well as total use of electronic appliance. Since the observations are obtained with a form of censored data in the questionnaire, we now use the median of each interval of censored data for further robustness check. The regression results with alternative measurement of dependent variable are given in Table 12.

Table 12.

Average Treatment Effect With Alternative Dependent Variables.

Variables	Matching method	ATT	Std. Err	t value
Frequency	k-Nearest (k = 4)	25.475***	9.269	2.75
	Radius (radius = 0.01)	29.126***	8.739	3.33
	Kernel	28.336***	8.599	3.30
	Mhalanobis	25.109***	9.242	2.72
	Mean	27.012
Time	k-Nearest (k = 4)	−0.050	0.040	−1.24
	Radius (radius = 0.01)	−0.042	0.037	−1.12
	Kernel	−0.034	0.037	−0.94
	Mhalanobis	−0.024	0.039	−0.62
	Mean	−0.038
Total use	k-Nearest (k = 4)	23.282**	9.544	2.44
	Radius (radius = 0.01)	21.642**	9.035	2.40
	Kernel	21.927**	8.879	2.47
	Mhalanobis	13.838	11.969	1.16
	Mean	20.172

Note. We applied bootstrap method and repeated sampling 400 times.

p < .1. **p < .05. ***p < .01.

The impact of energy efficiency labeling on use frequency and total use of washing machine is significantly positive at the 1% level, yet statistically insignificant for use time of per use. Overall, these findings are in line with the conclusions withdrawn from the baseline estimation. The magnitude of odds ratio is over 25 when the nearest neighborhood matching, radius matching, and kernel matching approaches are applied, much larger than those found in the baseline model.²¹

The estimation results reveal that, compared with the households whose washing machine does not have any energy efficiency label, the households having energy efficiency labels for their washing machine use washing machines more than twenty times per year on average. When the use time per use is considered, the households that have energy efficiency label for their washing machines use this electric appliance more than twenty-five hours per year.²²

6.7. Examination Over Alternative Electric Appliances

The energy efficiency label is mandatory for a wide range of household devices. We also examine the impacts of energy efficiency labeling on other domestic electric appliances, such as air conditioner and water heater to further ensure the above findings regarding household energy use. The estimation results over air conditioner is presented in Table 13.²³ We can see that households’ use frequency and total use of air conditioner are affected by the energy efficiency labeling, using more when there is an energy efficiency label relative to those without it. In contrast, the effect of energy efficiency labeling on use time per use is not statistically significant, like that for washing machine.

Table 13.

Average Treatment Effect With PSM – Air Conditioner.

Variables	Matching method	ATT	Std. Err	t value
Frequency	k-Nearest (k = 4)	0.246***	0.095	2.60
	Radius (radius = 0.01)	0.253***	0.085	2.96
	Kernel	0.221***	0.079	2.81
	Mhalanobis	0.349***	0.079	4.40
	Mean	0.267
Time	k-Nearest (k = 4)	0.170	0.174	0.98
	Radius (radius = 0.01)	0.165	0.167	0.99
	Kernel	0.173	0.154	1.13
	Mhalanobis	0.116	0.154	0.75
	Mean	0.156
Total use	k-Nearest (k = 4)	1.497**	0.746	2.01
	Radius (radius = 0.01)	1.427**	0.696	2.05
	Kernel	1.326**	0.653	2.03
	Mhalanobis	1.631***	0.626	2.61
	Mean	1.470

Note. We applied bootstrap method and repeated sampling 400 times.

p < .1. **p < .05. ***p < .01.

The impacts of energy efficiency labeling on water heater are slightly different from those on washing machine and air conditioner. As shown in the Table C1, if there is an energy efficiency label on water heater, households tend to use a longer time, inducing to higher total use of energy compared to those without energy efficiency label for their water heater. Anyway, the rebound effect of energy use induced by energy efficiency labeling is also verified through the examination of air conditioner and water heater.

7. Heterogeneity Analysis

In this section, we examine the heterogeneity of the treatment effects of energy efficiency labeling from the perspective of households’ residential places, income level and the gender of household head.

7.1. Heterogeneity in Residential Place

Considering that households living in urban areas and rural areas may have different habit in using electric appliances due to the differences in economic development and supply infrastructure of electricity and water, we conduct heterogeneity analysis in residential places. We classify all observations into two groups according to the places they live in the past five years, one from urban area and the other from rural area. The associated estimation results are presented in Table 14.

Table 14.

Heterogeneity Analysis in Households’ Residential Places.

		ATT
Variables	Matching method	Urban	Rural
Frequency	k-Nearest (k = 4)	0.236**(0.105)	0.179(0.118)
	Radius (radius = 0.01)	0.223**(0.092)	0.188*(0.101)
	Kernel	0.226***(0.088)	0.196**(0.098)
	Mhalanobis	0.224***(0.082)	0.167(0.110)
Total use	k-Nearest (k = 4)	0.807*(0.427)	0.514(0.484)
	Radius (radius = 0.02)	0.785*(0.404)	0.479(0.412)
	Kernel	0.786**(0.384)	0.519(0.397)
	Mhalanobis	0.885**(0.386)	0.272(0.526)

Note. In parenthesis are robust standard error.

p < .1. **p < .05. ***p < .01.

We find that the effect of energy efficiency labeling on urban households’ energy consumption behaviors is statistically significant, across various matching methods. That is, in urban areas, energy efficiency labeling leads to more energy consumption, in terms of both use frequency and total use of energy, indicating the possible existence of obvious rebound effect. In contrast, the impact of energy efficiency labeling on rural household’s energy use frequency is not robust across different matching strategies. In addition, the effect on rural households’ total energy use is statistically insignificant. This outcome can be attributed to the averagely better amenity, higher income and living standard in urban areas.

7.2. Heterogeneity in Income Level

We also examine the heterogeneity of effect in households’ income level given the important role of income in consumers’ energy consumption behavior as well as in electric appliance purchasing decision. The full sample is divided into three groups. Households with income ranking at the top 20 percent are clustered in the “upper” income group, those whose income level locates at the last 20 percent are in the “lower” income group, and the remainder are considered as the “middle” income group. The heterogeneous results are demonstrated in Table 15.

Table 15.

Heterogeneity Analysis in Households’ Income.

		ATT
Variables	Matching method	Lower	Middle	Upper
Frequency	k-Nearest (k = 4)	0.136(0.172)	0.274***(0.105)	0.297*(0.155)
	Radius (radius = 0.01)	0.145(0.164)	0.221**(0.090)	0.338**(0.153)
	Kernel	0.094(0.145)	0.218**(0.087)	0.262*(0.140)
	Mhalanobis	0.135(0.145)	0.229**(0.090)	0.222(0.147)
Total use	k-Nearest (k = 4)	0.931(0.656)	0.677(0.427)	1.023(0.679)
	Radius (radius = 0.02)	0.916(0.610)	0.430(0.389)	1.133*(0.653)
	Kernel	0.807(0.566)	0.513(0.372)	1.033*(0.615)
	Mhalanobis	0.751(0.565)	0.544(0.419)	0.963(0.739)

Note. In parenthesis are robust standard error.

p < .1. **p < .05. ***p < .01.

It shows that, for the middle income households the display of energy efficiency labels increases the use frequency of domestic appliances, whereas the treatment effect is statistically insignificant for the households with lower income. At the same time, we observe that the inﬂuence of the energy efficiency labels is not robust across all the matching approaches for the richest group. The insignificant impact for the poorest group can be explained by the sensitivity for money of the poor, as electricity bill might account for a relatively larger fraction of their expenditure. Similarly, the ambiguous treatment effect for the upper income households can partly be explained by their insensitivity to electricity bill.

7.3. Heterogeneity in Household Head’s Gender

Li et al. (2019) found that gender affects Chinese households’ green consumption behavior, as gender differences, in terms educational level, employment and characteristics in nature, lead to different decision making. Therefore, we also analyze the heterogeneous effect of energy efficiency labeling on households’ energy consumption behavior from the perspective of gender. We classify all observations into two groups based on the gender of household heads. The associated estimation results are reported in Table 16.

Table 16.

Heterogeneity Analysis in Household Head’s Gender.

		ATT
Variables	Matching method	Male	Female
Frequency	k-Nearest (k = 4)	0.273**(0.120)	0.129(0.105)
	Radius (radius = 0.01)	0.303***(0.098)	0.134(0.096)
	Kernel	0.294***(0.095)	0.155*(0.092)
	Mhalanobis	0.332***(0.103)	0.126(0.092)
Total use	k-Nearest (k = 4)	0.831*(0.499)	0.534(0.432)
	Radius (radius = 0.02)	0.710*(0.416)	0.682*(0.388)
	Kernel	0.855**(0.409)	0.590(0.375)
	Mhalanobis	1.144**(0.466)	0.561(0.413)

Note. In parenthesis are robust standard error.

p < .1. **p < .05. ***p < .01.

It yields that when the head of household is male, the impacts of energy efficiency label on use frequency and total use of electric appliances are statistically significant, whereas the effect is negligible when female plays a stronger role in a household. These results can be explained by the findings by Andreoni and Vesterlund (2001) and Li et al. (2019), which suggested that females are more altruism and tend to be more environmentally friendly than males.

8. Concluding Remarks

Based on the data of CGSS 2015, we provide plausible evidence for the effect of China energy efficiency labeling on households’ energy consumption behavior, employing the PSM approach. Using washing machine as an example, we find that the energy efficiency labeling has greatly induced higher use frequency and more total use of electric appliances. This finding indicates that China’s enforcement of energy efficiency labeling has possibly exerted rebound effect on household energy consumption, extending the current literature examining rebound effect and China’s energy efficiency labeling.

To ensure the validity of findings, we conduct various robustness checks, including applying alternative matching algorithms, using alternative regression approaches (e.g., Logit model), addressing potential endogeneity problem (with instrumental variable model and switching regression method), changing the measurement of dependent variable and examining other types of appliances. The main findings remain robust across these checks, providing further evidence for the positive relationship between energy efficiency label and households’ energy consumption. Additional analysis yields that the impacts are substantially heterogeneous in households’ residential place, income level and gender of head. Specifically, the effect of energy efficiency labeling is statistically significant for the households living in urban areas, for those enjoying higher income and for those having a male household head, yet the display of energy efficiency label does not significantly affect the energy consumption behavior of other households.

In the presence of possible rebound effect, supportive measures such as education to raise consumers’ environmental awareness can be applied. The impact heterogeneity found above can shed some light on the targeted households. Considering that monetary cost is crucial in guiding energy consumption behavior, consistent with the finding by (Newell and Siikamäki 2014), the information currently provided by China energy efficiency label might be too simple to stimulate sufficient energy savings. In this sense, additional information such as monetary values and carbon dioxide emissions of energy savings can be presented on future labels.

Footnotes

Appendix A

Appendix B

Appendix C

Table C1.

Average Treatment Effect With PSM – Water Heater.

Variables	Matching method	ATT	Std. Err	t value
Frequency	k-Nearest (k = 4)	−0.132*	0.083	−1.79
	Radius (radius = 0.01)	−0.131*	0.073	−1.78
	Kernel	−0.105	0.067	−1.56
	Mhalanobis	−0.110	0.070	−1.57
	Mean	−0.120
Time	k-Nearest (k = 4)	0.313***	0.080	4.44
	Radius (radius = 0.02)	0.311***	0.070	4.42
	Kernel	0.290***	0.064	4.49
	Mhalanobis	0.288***	0.072	3.99
	Mean	0.301
Total use	k-Nearest (k = 4)	1.243**	0.458	2.71
	Radius (radius = 0.02)	1.243***	0.391	3.17
	Kernel	1.225***	0.359	3.41
	Mhalanobis	1.123***	0.384	2.93
	Mean	1.209

Note. We applied bootstrap method and repeated sampling 400 times.

p < .1. **p < .05. ***p < .01.

Acknowledgements

We would like to express our sincere gratitude to the Editor-in-Chief for the insightful guidance and constructive feedback throughout the review process. Additionally, we extend our appreciation to the anonymous referees for their thorough review and valuable suggestions, which significantly improved the quality of this paper.

ORCID iD

Rabindra Nepal

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We acknowledge the financial support from the National Natural Science Foundation of China (Grant No. 72003156) and the Social Science Foundation of Sichuan Province (Grant No. SCJJ23ND159).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

1

See

2

One of the reasons for the substantial increase of residential electricity consumption was the outbreak of the COVID-19 pandemic, which induces residents to spend much more time at home.

3

Rebound effect of energy efficiency means that energy consumption is increased, rather than reduced, in the presence of energy efficiency improvement, as consumers may use appliances more when they expect that technological improvement leads to energy savings ().

4

There are 600, 1000, and 775 observations in the study by Feng et al. (2010), , and Yu et al. (2013), respectively. Moreover, the observations in Shen (2012) were collected online and the quality of sample cannot be fully guaranteed.

5

For instance, a large number of studies (e.g., Lin and Zhu 2021; Liu and Lin 2020; Wang et al. 2020; Zou et al. 2020) have investigated the impacts of increasing block pricing of electricity on households’ electricity consumption in China. There is also extensive literature examining the energy efficiency labeling out of China, such as Brécard (2014), Davis and Metcalf (2016), and .

6

Note the CGSS is a continuous survey started from 2003. We use the data in 2015 rather than a panel data from 2003 to 2017 because only in the survey in 2015 there is a modular with energy.

7

More details about the previous policy can be found in .

8

studied how EU energy label affects consumers’ purchasing decisions for washing machines based on the data of a survey involving a total of 151 choice-based conjoint interviews conducted in Switzerland in Spring 2004.

9

The rebound effect has two components. The first is direct rebound, which is the percentage of energy savings from efficiency that are offset by increased use. Higher efficiency makes an energy consuming technology less expensive to use, so people use it more often. The other component is indirect rebound. This results from how an individual spends the money he/she saves.

10

Since 2003, the CGSS has three different sampling designs and has used three set of sampling frames. They are 2003–2006, 2008, and 2010–present. 2003–2006 sampling design is multi-stage stratified design and uses China’s fifth census of 2,000 as sampling frame. The sample is composed of 5,900 urban households and 4,100 rural households. The urban sample is over sampled because it is hypothesized that the heterogeneity in urban area is higher than in rural area. In 2008, the CGSS program used an experimental sampling design, taking 1% of national population survey data in 2005 as sampling frame. As it is just an experimental design, the 2008 CGSS sample size is only 6,000. The sampling design made in 2010 for the CGSS is still multi-stage stratified design and used 2009 national population data as sampling frame. More details about the sampling design and geographical distribution of samples can be found at .

11

For instance, in our sample, over 98% of the observations have only one washing machine, which implies that less than 2% of them have multiple washing machines. In contrast, over 68% of the households have at least two air conditioners and over 38% of them have multiple heaters at home.

12

Note the dependent variable refers to the use behavior of the type of appliance in question in the robustness checks where other types of domestic appliances are concerned. It is likewise for the independent variables.

13

Note that electric appliances are durables; hence, in many households there is no energy efficiency label on their appliances as these products were produced prior to 2005.

14

The population size of China is 1.4 billion, of which about 90,000 thousand of people are communist party member.

15

According to the CHFS, in 2017 over 92.8% of Chinese households own at least one house or apartment, a bit higher than the level in 2015. See https://chfs.swufe.edu.cn/info/1321/2591.htm. The significant ownership of housing in China can partly be explained by institutional and cultural factors, as shown by .

16

For instance, according to the China National Bureau of Statistics, the urbanization rate, reflected by Hukou, was around 56% in 2015, similar with the mean of the sample in this study.

17

As the non-linear terms of the predictor variables can still be possible omitted variables, we have considered the quadratic term of several variables, such as age and housing area.

18

More details can refer to .

19

We applied the 2SLS method combined with the PSM. Considering that this method is complicated due to numerous restrictive conditions, as suggested by , we conducted this estimation as one of our robustness tests, rather than as the benchmark analysis.

20

$γ Z_{i} + μ_{i} > 0$ corresponds to the net benefit of choosing an appliance with energy efficiency label exceeding that of not choosing labeled appliances; hence, under this condition the household decides to choose an appliance having energy efficiency label attached.

21

This outcome is not unexpected, as the median of each interval of censored data is larger than the ordered categorical value used above.

22

This is reflected by the coefficients of total use (i.e., the interaction of use frequency and use time per use).

23

Not that in the regressions examining air conditioner, we have considered the same appliance features, including the quantity, volume and power of air conditioner. This is the same case in the regressions for water heater shown in .

References

Andor

M. A.

Gerster

Götte

2019. “How Effective Is the European Union Energy Label? Evidence from a Real-Stakes Experiment.” Environmental Research Letters 14 (4): 044001.

Andor

M. A.

Gerster

Sommer

2020. “Consumer Inattention, Heuristic Thinking and the Role of Energy Labels.” The Energy Journal 41 (1): 83–112.

Andreoni

Vesterlund

2001. “Which Is the Fair Sex? Gender Differences in Altruism.” The Quarterly Journal of Economics 116 (1): 293–312.

Banerjee

Solomon

B. D.

2003. “Eco-Labeling for Energy Efficiency and Sustainability: A Meta-Evaluation of US Programs.” Energy Policy 31 (2): 109–123.

Bilicka

K. A.

2019. “Comparing UK Tax Returns of Foreign Multinationals to Matched Domestic Firms.” American Economic Review 109 (8): 2921–53.

Brécard

2014. “Consumer Confusion over the Profusion of Eco-Labels: Lessons from a Double Differentiation Model.” Resource and Energy Economics 37: 64–84.

Davis

L. W.

Metcalf

G. E.

2016. “Does Better Information Lead to Better Choices? Evidence from Energy-Efficiency Labels.” Journal of the Association of Environmental and Resource Economists 3 (3): 589–625.

Eriksen

Wiese

2019. “Policy Induced Increases in Private Healthcare Financing Provide Short-Term Relief of Total Healthcare Expenditure Growth: Evidence from OECD Countries.” European Journal of Political Economy 59: 71–82.

Feng

Sovacool

B. K.

K. M.

2010. “The Barriers to Energy Efficiency in China: Assessing Household Electricity Savings and Consumer Behavior in Liaoning Province.” Energy Policy 38 (2): 1202–209.

10.

Greening

L. A.

Greene

D. L.

Difiglio

2000. “Energy Efficiency and Consumption – The Rebound Effect – A Survey.” Energy Policy 28 (6–7): 389–401.

11.

Herring

2006. “Energy Efficiency – A Critical View.” Energy 31 (1): 10–20.

12.

Huang

Clark

W. A.

2020. “Multiple Home Ownership in Chinese Cities: An Institutional and Cultural Perspective.” Cities 97: 102518.

13.

Ichimura

Taber

2001. “Propensity-Score Matching with Instrumental Variables.” American Economic Review 91 (2): 119–24.

14.

Kleibergen

Paap

2006. “Generalized Reduced Rank Tests Using the Singular Value Decomposition.” Journal of Econometrics 133 (1): 97–126.

15.

Zhang

2019. “Does Gender Inequality Affect Household Green Consumption Behaviour in China?” Energy Policy 135: 111071.

16.

Lin

Zhu

2021. “Has Increasing Block Pricing Policy Been Perceived in China? Evidence from Residential Electricity Use.” Energy Economics 94: 105076.

17.

Liu

Lin

2020. “Is Increasing-Block Electricity Pricing Effectively Carried Out in China? A Case Study in Shanghai and Shenzhen.” Energy Policy 138: 111278.

18.

2014. “A Critical Review of China’s Rapidly Developing Renewable Energy and Energy Efficiency Policies.” Renewable and Sustainable Energy Reviews 29: 508–516.

19.

Lokshin

Sajaia

2004. “Maximum Likelihood Estimation of Endogenous Switching Regression Models.” The Stata Journal 4 (3): 282–89.

20.

Menezes

F. M.

Zheng

2018. “Regulatory Incentives for a Low-Carbon Electricity Sector in China.” Journal of Cleaner Production 195: 919–31.

21.

Newell

R. G.

Siikamäki

2014. “Nudging Energy Efficiency Behavior: The Role of Information Labels.” Journal of the Association of Environmental and Resource Economists 1 (4): 555–98.

22.

Park

J. Y.

2017. “Is There a Price Premium for Energy Efficiency Labels? Evidence from the Introduction of a Label in Korea.” Energy Economics 62: 240–47.

23.

Price

Levine

M. D.

Zhou

, et al. 2011. “Assessment of China’s Energy-Saving and Emission-Reduction Accomplishments and Opportunities during the 11th Five Year Plan.” Energy Policy 39 (4): 2165–78.

24.

Rosenbaum

P. R.

Rubin

D. B.

1985. “Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score.” The American Statistician 39 (1): 33–38.

25.

Sammer

Wüstenhagen

2006. “The Inﬂuence of Eco-Labelling on Consumer Behaviour–Results of a Discrete Choice Analysis for Washing Machines.” Business Strategy and the Environment 15 (3): 185–99.

26.

Sanchez

M. C.

Brown

R. E.

Webber

Homan

G. K.

2008. “Savings Estimates for the United States Environmental Protection Agency’s ENERGY STAR Voluntary Product Labeling Program.” Energy Policy 36 (6): 2098–108.

27.

Shen

2012. “Understanding the Determinants of Consumers’ Willingness to Pay for Eco-Labeled Products: An Empirical Analysis of the China Environmental Label.” Journal of Service and Management 5: 87–94.

28.

Shen

Saijo

2009. “Does an Energy Efficiency Label Alter Consumers’ Purchasing Decisions? A Latent Class Approach Based on a Stated Choice Experiment in Shanghai.” Journal of Environmental Management 90 (11): 3561–73.

29.

Sorrell

Dimitropoulos

2008. “The Rebound Effect: Microeconomic Definitions, Limitations and Extensions.” Ecological Economics 65 (3): 636–49.

30.

Sorrell

Dimitropoulos

Sommerville

2009. “Empirical Estimates of the Direct Rebound Effect: A Review.” Energy Policy 37 (4): 1356–71.

31.

Stadelmann

Schubert

2018. “How Do Different Designs of Energy Labels Inﬂuence Purchases of Household Appliances? A Field Study in Switzerland.” Ecological Economics 144: 112–23.

32.

Stock

J. H.

Yogo

2005. “Testing for Weak Instruments in Linear IV Regression.” In Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, edited by Andrews

D. W. K.

Stock

J. H.

Cambridge University Press.

33.

Tiefenbeck

Staake

Roth

Sachs

2013. “For Better or for Worse? Empirical Evidence of Moral Licensing in a Behavioral Energy Conservation Campaign.” Energy Policy 57: 160–71.

34.

Waechter

Sütterlin

Siegrist

2015a. “Desired and Undesired Effects of Energy Labels – An Eye-Tracking Study.” PLoS One 10 (7): 1–26.

35.

Waechter

Sütterlin

Siegrist

2015b. “The Misleading Effect of Energy Efficiency Information on Perceived Energy Friendliness of Electric Goods.” Journal of Cleaner Production 93: 193–202.

36.

Wang

Sun

Wang

Zhang

2019. “Purchasing Intentions of Chinese Consumers on Energy-Efficient Appliances: Is the Energy Efficiency Label Effective?” Journal of Cleaner Production 238: 117896.

37.

Wang

Sun

Wang

2020. “Policy Cognition is More Effective Than Step Tariff in Promoting Electricity Saving Behaviour of Residents.” Energy Policy 139: 111338.

38.

Ward

D. O.

Clark

C. D.

Jensen

K. L.

Yen

S. T.

Russell

C. S.

2011. “Factors Inﬂuencing Willingness-to-Pay for the Energy Star^® Label.”Energy Policy 39 (3): 1450–58.

39.

Zhang

Fujiwara

2013. “Evaluating the Direct and Indirect Rebound Effects in Household Energy Consumption Behavior: A Case Study of Beijing.” Energy Policy 57: 441–53.

40.

Zhang

Xiao

Zhou

2020. “Willingness to Pay a Price Premium for Energy-Saving Appliances: Role of Perceived Value and Energy Efficiency Labeling.” Journal of Cleaner Production 242: 118555.

41.

Zou

Luan

Zheng

Huang

2020. “The Effect of Increasing Block Pricing on Urban Households’ Electricity Consumption: Evidence from Difference-in-Differences Models.” Journal of Cleaner Production 257: 120498.

Energy Efficiency Labeling and Household Energy Consumption Behavior in China: A Propensity Score Matching Approach

Abstract

Keywords

1. Introduction

2. Background of Chinese Energy Efficiency Label Policy

3. Literature Review

3.1. Impacts of Energy Efficiency Labeling

3.1.1. Impacts on Consumer’s Purchase Decision

3.1.2. Impacts on Energy Savings and Rebound Effect

3.2. Investigation Over China’s Energy Efficiency Labeling

4. Data and Methodology

4.1. Data Description

4.1.1. Dependent Variable

4.1.2. Independent Variable

4.1.3. Control Variables

4.2. Identification Methodology: PSM

5. Estimation Results

5.1. Propensity Score Estimation

5.2. Balancing Test

5.3. Test of Common Support Assumption

5.4. Treatment Effects

6. Robustness Checks

6.1. Applying Alternative Matching Algorithms

6.2. Examining Energy Efficiency Labels at Different Ranks

6.3. Using Alternative Regression Approaches

6.4. Estimation With Instrumental Variable

6.5. Estimation With Endogenous Switching Regression Method

6.6. Using Alternative Measurement of Dependent Variables

6.7. Examination Over Alternative Electric Appliances

7. Heterogeneity Analysis

7.1. Heterogeneity in Residential Place

7.2. Heterogeneity in Income Level

7.3. Heterogeneity in Household Head’s Gender

8. Concluding Remarks

Footnotes

Appendix A

Appendix B

Appendix C

Acknowledgements

ORCID iD

Funding

Declaration of Conflicting Interests

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

References