Examining Preference for Energy-Related Information through a Choice Experiment

Abstract

Many studies have shown that providing information on energy consumption to a household is effective, to some extent, in encouraging its energy conservation behavior. These studies provided information free of charge. However, depending on the type of information, a household must bear costs, such as installing the necessary equipment to obtain the information. Are people willing to pay for the information? In this study, a questionnaire survey was conducted to examine willingness to pay (WTP) for energy-related information using a choice experiment. The data were analyzed using conditional logit and latent class models. Positive WTP was estimated for information on the total energy consumption amount for the entire house, detailed electricity usage amount for each major home appliance, electricity rates by time zone, and power source composition of electricity. No significant positive results were obtained for comparison with the other households, as the class that accounted for about 40% of the analyzed sample had negative WTP for this information. When electricity companies provide comparative information, it is better to carefully consider how and to whom they provide it. The results of the latent class model show that preferences vary among classes. Although some preference variations exist, some households have a positive WTP for information on energy consumption.

1. Introduction

Energy consumption in the household sector is affected by the energy efficiency of the equipment used and people’s behavior, such as how and for how long people use the equipment. Therefore, research has been conducted on encouraging energy conservation behavior in the household sector. There are studies that examine whether providing information on the amount of energy consumption can promote energy-saving behavior in the household sector (Chatzigeorgiou and Andreou [1], Buckley [2], Andor and Fels [3]). Providing information to people and leading them to reduce energy consumption is often referred to as feedback. Many of these studies on feedback have been conducted through experiments, and information on the amount of energy consumed is provided free of charge to the participants.

Currently, energy companies, such as electricity and gas companies, provide information services on energy consumption. Customers can obtain information on total energy consumption for the entire house free of charge if registered. In addition, if a home energy management system (HEMS) is installed in the house, it is possible to obtain information on the total energy consumption of the entire house and on electricity consumption for each major home appliance or room (Japan Economic Center, [4]).

If a household has installed HEMS in its house, it must pay the cost of HEMS. In addition, even if electricity or gas companies provide information on energy consumption free of charge to households, providing this information incurs costs for companies. Therefore, it cannot be said that there is no possibility that some kind of cost recovery will have to be considered if the content of the information provided becomes more sophisticated in the future. Therefore, it is important to examine what information households are willing to pay for and to what extent.

This study aims to examine consumers’ willingness to pay (WTP) for energy-related information using a choice experiment based on a questionnaire survey. The information covered in this study includes information that is in line with the deregulation policy in the electricity industry in Japan, in addition to information that has been covered in previous studies on energy-related information provisions explained in Section 2. In Japan, the retail sale of electricity to individual households had long been a regional monopoly. However, after 1 April 2016, new entrants for the retail sale of electricity to households became possible, allowing consumers to choose their power companies and rate menus from various companies. From the viewpoint of environmental issues, it has become possible to purchase electricity from companies using power generation methods that are less harmful to the environment. According to the Electricity and Gas Transaction Monitoring Committee [5], disclosure of power source composition was deemed a “desirable practice” in the “Guidelines for Retail Sales of Electric Power”. As purchasing electricity from a company that uses environmentally friendly power sources such as renewable energy is one way to reduce the burden on the environment, information on power source composition is also considered in this study.

A choice experiment was conducted using a questionnaire survey, and the data were first analyzed using a conditional logit model to estimate average preference. The results show that consumers have a positive WTP to obtain energy-related information on the total energy consumption amount for the entire house, detailed electricity usage amount for each major home appliance, electricity rates by the time zone of a day, and the power source composition of electricity. In addition to the conditional logit model, latent class models were estimated to consider preference heterogeneity. The results of the latent class model showed that the respondents could be classified into four groups based on their preferences. The class accounting for about 40% of the respondents was unwilling to pay for any information, and they had a negative WTP for comparative information with other households. The other three classes showed positive WTP for some information but zero WTP for comparative information with other households. Even when WTP is positive, the amount varies from class to class.

The remainder of this paper is organized as follows. Section 2 reviews previous studies. Section 3 describes the survey design of the choice experiment, and Section 4 describes the model. Section 5 presents the results and Section 6 provides a discussion. Finally, Section 7 concludes the paper.

2. Literature Review

Methods to reduce the environmental impact of electricity use in the household sector include changing power generation methods to more environmentally friendly ones, switching to energy-efficient appliances, and reducing the amount of time spent using electricity. Several studies have examined household preferences as WTP.

First, studies examining WTP for power source composition include Zoric and Hrovatin [6], Guo et al. [7], Akcura [8], Morita and Managi [9], Murakami et al. [10], Vecchiato and Tempesta [11], Lee and Heo [12], and Nakai et al. [13]. For a meta-analysis, see Ma et al. [14], although they excluded gas and nuclear power. In Japan, the Fukushima nuclear power plant accident increased skepticism about reliance on nuclear power. Morita and Managi [9] conducted discrete choice experiments and choice probability experiments to examine WTP for residential electricity generated by solar, wind, nuclear power, and natural gas. They showed that Japanese people, on average, have a negative WTP for the electricity generated by nuclear power. Nakai et al. [13] investigated preferences for electricity supply in the Tokyo area and showed that consumers prefer major providers with a renewable energy component in the energy mix with a stable and cheap supply. Murakami et al. [10] conducted a choice experiment which revealed that the average consumers had a negative preference for increases in nuclear power generation, while renewable energy was preferred.

Second, the studies examining WTP for energy-efficient appliances are as follows: According to Kuhn et al. [15], regarding air conditioners, those with greater environmental awareness and knowledge place more importance on energy efficiency. Jacobs and Hörisch [16], for washing machines, found that labels regarding the product’s service life were more important than energy consumption. Jain et al. [17] obtained the WTP for labels with stars for the energy efficiency of air conditioners and found that labels were positively evaluated. Jain et al. [18] compared the WTP for labels with stars for air conditioners and refrigerators. They found that those with more knowledge about labels evaluated the labels more highly for both air conditioners and refrigerators. Shen and Saijo [19] investigated the WTP for eco-labels for air conditioners and refrigerators in Shanghai.

Third, with regard to reducing electricity consumption, information can be provided to encourage energy-saving behavior. Providing information on energy consumption to each household is known as “feedback”. Experimental studies have been conducted mainly on electricity consumption to determine what kind of information and how to provide it is effective in reducing energy consumption. For a detailed review, refer to Chatzigeorgiou and Andreou [1], Buckley [2], and Andor and Fels [3]. Many studies have found that providing information on energy consumption reduces energy consumption. In many of these studies, information was provided free of charge to conduct the experiments. However, some of this information may require equipment installation or other costs to obtain the information. For example, installing a smart meter or HEMS makes it possible to visualize and understand energy use; however, installing the equipment involves both risks and benefits in addition to being costly. Therefore, we do not know whether households will introduce the equipment. When the visualization of energy consumption is accompanied with the introduction of smart home technology, there are advantages and disadvantages regarding smart home technologies. Sovacool and Del Rio [20] discussed benefits and risks based on expert interviews and site visits to retailers. Li et al. [21] investigated motivations, barriers, and risks. They found that efficient energy management is included in the motivation, whereas privacy and security issues are included in the barriers and risks.

Considering these benefits and risks, how do households accept smart home services? Sanguinetti et al. [22] demonstrated that four consumer groups are on the path to adopting smart homes. Other studies on the adoption of smart home energy technology include Ji and Chan [23], and on the acceptance of HEMS includes Pfeiffer et al. [24]. Chen et al. [25] showed that middle- and high-income households are more willing to adopt HEMS than low-income households and that the latter are not willing to adopt HEMS even if it is free. Perceived usefulness and ease-of-use influence adoption. Chen et al. [26] examined smart meter support and adoption in the U.S. Many people were willing to adopt smart meter technology, and the usefulness and risk of privacy influenced their support. However, perceived costs did not impact support. In addition to adoption, Chen et al. [25] examined WTP for different HEMS features, such as monitoring electricity use, controlling appliances, reducing community environmental impacts, telemedical, community updates, and job searches. High-income households were found to be more willing to pay for all HEMS features than other income groups. The HEMS is a new technology for households. The perception of the difficulty of using HEMS and the perception of usefulness differ among income groups.

Therefore, investigating WTP is important to consider the diffusion of such equipment. Our study can be considered in this category. The following studies examined WTP for HEMS. Chen et al. [27] investigated the factors influencing the WTP for HEMS in New York and Tokyo. People in both areas with high-cost concerns were less willing to pay a larger amount. Washizu et al. [28] investigated WTP for HEMS functions, such as money-saving, automatic control, and environmental impacts in New York and Tokyo. In New York, an automatic control function increases the demand probability, whereas in Tokyo, the environmental impact function increases the demand probability at a given price. Smart meters can also provide information on energy use. Gerpott and Paukert [29] showed that trust in the protection of smart meter data and the intention to change energy consumption are constructs strongly related to the WTP for smart meters in Germany. Kaufmann et al. [30] showed that customers are willing to pay for smart metering in Switzerland and found four segments. In terms of water, Koo et al. [31] evaluated the preference for a smart water metering and monitoring system (SWMMS) and found that the monthly public value exceeds the monthly related cost. Real-time monitoring is one of the functions of a smart home. Rihar et al. [32] examined the WTP for smart home functionalities in Slovenia. They showed that consumers prefer real-time monitoring of energy consumption in household appliances.

Information provision has the potential to change energy consumption behavior. However, as visualization involves costs, such as introducing HEMS, energy conservation behavior will not change if visualization is not introduced in the first place. In addition, even if the information is provided free of charge by a company, there may be a cost to the company involved in providing the information. If the information becomes more sophisticated, it may be necessary to recover the cost. For this reason, it is important to know the amount consumers are willing to pay for information. Thus, we examined the type of information consumers are willing to pay for. Devices that enable visualization, such as HEMS, can optimize energy consumption by automatic control if they are compatible with home appliances. Home appliances compatible with these automatic control functions are not widespread at present; therefore, visualization is the main focus. Moreover, by studying the kind of information consumers mainly want in the visualization, it may be possible not to provide unnecessary information, thereby reducing costs. Therefore, this study examines WTP by classifying information in detail.

Previous studies have included a variety of functions other than information provision for estimating WTP, but detailed categorization has not been made regarding the type of information to be provided.

Chen et al. [27] classified HEMS functions as “monitoring”, “automation”, and “social benefits”. Washizu et al. [28] used classifications of “money-saving”, “automatic control”, and “environmental impact”. Chen et al. [25] used “monitoring electricity use”, “controlling appliances”, and “reducing community environmental impacts during the pandemic”, as well as “telemedical”, “community updates and social networking”, and “job search”. Kaufmann et al. [30] include the functions of smart meters such as “Remote meter reading with accurate monthly billing”, “Real-time consumption feedback”, “Programming and steering services”, and “Home security and surveillance services with alert functions”. On the one hand, Rihar et al. [32] used many smart home functions. Among them, monitoring functions are included, such as “Real-time energy consumption monitoring of household appliances” and “Electronic (wireless) metering and monitoring of usage of various resources”. On the other hand, Gerpott and Paukert [29] considered smart meters in a lump and did not break them down into more detailed functions when examining WTP.

Therefore, Chen et al. [27], Washizu et al. [28], Chen et al. [25], Kaufmann et al. [30], and Rihar et al. [32] included various functions other than information provision but did not provide a detailed classification of the type of information provided. Our study did not evaluate functions other than the provision of information. Instead, the types of information provided are elaborated upon. Given the variety of information that can be provided, it is useful to know what type of information households seek. Our study fills this gap by focusing on information provision and subdividing the type of information provided. Our study includes the following information: total energy consumption amount for the entire house, detailed electricity usage amount for each major home appliance, electricity rates by time zone, comparative information with other households, and the power source composition of electricity. The details are explained in Section 3.

3. Survey Design

The survey was conducted between 22–26 January 2016, in the form of an internet survey of the registered monitors of a research firm. The study was a choice experiment in which participants were asked to select the most desirable information provision service among hypothetical information provision services on energy use. The respondents answered eight choice questions. The price of the information provision service and the type of information provided vary from service to service. We asked the participants to assume that the information provided could be easily viewed on a personal computer, smartphone, or cell phone. A hypothetical information provision service, consisting of a combination of information provided and price, was created using an orthogonal array. The following describes the information provided by hypothetical information provision services. The summary of the attribute levels is presented in

Table 1. Attribute levels.

Attribute	Level
Total energy consumption amount for the entire house	Not provided
	Information on electricity is provided
	Information on electricity and gas is provided
	Information on electricity and water is provided
Detailed electricity usage amount for each major home appliance	Provided
	Not provided
Electricity rates by time zone	Provided
	Not provided
Comparative information with other households	Provided
	Not provided
Power source composition of electricity	Provided
	Not provided
Monthly fee for information provision service (JPY)	100
	200
	400
	600

.

(1)

Information on total energy consumption amount for the entire house

Information on the energy consumption of an entire house is provided in an easy-to-understand form, such as numerical values and graphs that show “at what time of day you use a lot of electricity” and “when you use a lot of electricity in a month”. While most of the information provided as feedback in previous studies addresses electricity consumption, this study addresses water and gas in addition to electricity. This attribute has four levels: not provided, information on “electricity”, information on “electricity and gas”, and information on “electricity and water”.

(2)

Information on detailed electricity usage amount for each major home appliance

The system displays a summary of how much electricity is used by major home appliances and information for each home appliance usage (e.g., time of day when it is used most and day of the week when it is used most). This attribute has two levels: “provided” and “not provided”.

(3)

Information on electricity rates by time zone

The electricity rate-setting method may become more sophisticated in the future. For example, electricity rates might temporarily increase with notification; for example, the day before electricity supply shortages are likely to occur. Therefore, the electricity rates by time of day and their fluctuations are expected to become important information in the future. This attribute has two levels: “provided” and “not provided”.

(4)

Comparative information with other households

The system displays the average energy use of households with similar family structures and the households’ ranking of energy use among other similar households. This attribute has two levels: “provided” and “not provided”.

(5)

Information on the power source composition of electricity

The system indicates the type of power generation (coal-fired, oil-fired, liquefied natural gas (LNG), nuclear, hydroelectric, solar, wind, etc.) from which the electricity currently being purchased is generated. This attribute has two levels: “provided” and “not provided”.

(6)

Monthly fees for information provision services.

There is a monthly fee to subscribe to the “Information Provision Service on Energy Use”. This attribute has four levels: “JPY100”, “JPY200”, “JPY400”, and “JPY600”.

An example of the choice set is presented in

Table 2. An example of a choice set.

Information Provision Service	A	B	I do not apply for any of these services.
Total energy consumption amount for the entire house	Information on electricity and gas is provided	Information on electricity is provided
Detailed electricity usage amount for each major home appliance	Provided	Not provided
Electricity rates by time zone	Not provided	Provided
Comparative information with other households	Provided	Provided
Power source composition of electricity	Not provided	Provided
Monthly fee for information provision service	JPY 100	JPY 200

. For example, if you subscribe to Information Provision Service A, you can obtain information on total electricity and gas consumption amount for the entire house, detailed electricity usage amount for each major home appliance, and comparative information with other households. The monthly fee for this information provision service is JPY 100. If you subscribe to Information Provision Service B, you can obtain information on the total electricity consumption amount for the entire house, information on electricity rates by time zone, comparative information with other households, and information on the power source composition of electricity. The monthly fee for this subscription is JPY 200. If a respondent does not prefer either information provision service A or B, the respondent can choose “I do not apply for any of these services”. Therefore, this was a three-choice question.

4. Model

A conditional logit model was used to analyze average consumer preferences. Preference heterogeneity was considered using a latent class model. In recent applied work, Nakano [33] uses these models, and the explanations in this section are based on Nakano [33], and Kuriyama and Shoji [34].

This study uses a choice experiment. A latent class model can be used when preferences are diverse, and respondents are divided into several groups. The advantage of this approach is that the parameters of the membership function for dividing respondents into groups and the parameters of the utility function for understanding each group’s preferences can be estimated simultaneously. This approach was developed by McFadden [35,36], and using this approach makes the decision-making structure behind choice more elaborate than traditional demand theory does (Kuriyama and Shoji, [34]). Therefore, this study used a latent class model and a conditional logit model. (In this approach, WTP is estimated, but each respondent’s individual WTP is not collected in the questionnaire survey. Therefore, this study does not perform a cluster analysis and regression analysis that needs respondents’ individual WTP.)

4.1. Conditional Logit Model

Individual k receives utility $U_{ki}$ by choosing alternative i. $U_{ki}$ is decomposed into the observable part $V_{ki}$ and the unobservable part ${\epsilon }_{ki}$.

$$U_{ki}=V_{ki}+{\epsilon }_{ki}$$

(1)

$P_{ki}$ is the probability that alternative i will be chosen by individual k from choice set C.

$$\begin{gathered} P_{ki}=\mathrm{Pr}\left[U_{ki}>U_{kj},\forall j\in C,j\ne i\right] \\ =\mathrm{Pr}[V_{ki}-V_{kj}>{\epsilon }_{kj}-{\epsilon }_{ki},\forall j\in C,j\ne i] \end{gathered}$$

(2)

We assume that ${\epsilon }_{ki}$ and ${\epsilon }_{kj}$ are distributed independently and identically to a type-I extreme-value distribution. In a conditional logit model, $P_{ki}$ is expressed as (McFadden, [37])

$$P_{ki}=\frac{\exp \left(\lambda V_{ki}\right)}{{{\displaystyle \sum }}_{j\in C}\exp \left(\lambda V_{kj}\right)}$$

(3)

where $\lambda$ is the scale parameter, which is assumed to be 1. The attribute parameters were estimated using the maximum likelihood method (Train [38]). $V\left(x\right)$ was assumed to be linear.

$$V\left(x\right)={\displaystyle \sum }_n{\beta }_n{x}_n$$

(4)

$x_n$ is the vector of attributes and the Alternative Specific Constant (ASC). ${\beta }_n$ is the vector of parameters.

4.2. Latent Class Model

The latent class model was developed by McFadden [35,36] and applied to logit models by Kamakura and Russell [39] and Swait [40]. When individual k chooses alternative i, utility is described by

$$U_{ki|s}={\beta }_s^{\prime }{x}_{ki}+{\epsilon }_{ki|s}$$

(5)

S is assumed to be the number of classes in the population, and individual k is assumed to belong to one of the classes s (s = 1…S).

The choice probability in class s is

$$P_{k|s}\left(i\right)=\frac{\exp \left({\mu }_s{\beta }_s^{\prime }{x}_{ki}\right)}{{{\displaystyle \sum }}_{j\in C}\exp \left({\mu }_s{\beta }_s^{\prime }{x}_{kj}\right)}$$

(6)

where ${\beta }_s$ and ${\mu }_s$ are the class-specific parameters for class s and class-specific scale parameters for class s, respectively.

According to Swait [40] and Boxall and Adamowicz [41], individuals are classified according to their explanatory variables. When an individual k belongs to class s, the membership function M* with explanatory variables $z_k$ is

$$M_{ks}^{*}={\gamma }_s^{\prime }{Z}_k+{\varsigma }_{ks}$$

(7)

where ${\gamma }_s$ and ${\varsigma }_{ks}$ are parameter and error terms, respectively. ${\varsigma }_{ks}$ is assumed to be distributed independently and identically with a type I extreme value distribution. $P_{ks}$ is the probability that an individual k belongs to class s.

$$P_{ks}=\frac{\exp \left(\lambda {\gamma }_s^{\prime }{Z}_k\right)}{{{\displaystyle \sum }}_{s^{*}=1}^s\exp \left(\lambda {\gamma }_{s^{*}}^{\prime }{Z}_k\right)}$$

(8)

where λ is a scale parameter, s is a specific class, and s* is a class. In the estimation, the parameters of the base class are set to zero.

$P_k\left(i\right)$ is the probability that an individual k chooses alternative i.

$$P_k\left(i\right)={\displaystyle \sum }_{s=1}^s\left[\frac{\exp \left(\lambda {\gamma }_s^{\prime }{Z}_k\right)}{{{\displaystyle \sum }}_{s^{*}=1}^s\exp \left(\lambda {\gamma }_{s^{*}}^{\prime }{Z}_k\right)}\right]\left[\frac{\exp \left({\mu }_s{\beta }_s^{\prime }{x}_{ki}\right)}{{{\displaystyle \sum }}_{j\in C}\exp \left({\mu }_s{\beta }_s^{\prime }{x}_{kj}\right)}\right]$$

(9)

All scale parameters (λ and ${\mu }_s$) were set to one (Boxall and Adamowicz [41]).

5. Results

Results were obtained using NLOGIT (Econometric Software, Inc., Plainview, NY, USA).

5.1. Conditional Logit Model

Table 3. Estimation results of the conditional logit model.

	Coefficient	WTP (JPY/Month)
Total_electricity	0.4097 *** (0.1219)	86.44
Total_electricity_gas	0.6233 *** (0.1188)	131.51
Total_electricity_water	0.7452 *** (0.1211)	157.22
Appliance	0.5734 *** (0.0784)	120.97
Rates	0.4506 *** (0.0637)	95.06
Comparison	−0.0493 (0.0675)	0
Source	0.2701 *** (0.0774)	56.98
Price	−0.0047 *** (0.0002)	-
ASC3	0.8755 *** (0.1467)	-
Sample size	3328
Log-likelihood	−2894.756
McFadden Pseudo R-squared	0.0976

Standard errors are in the parenthesis. *** indicates significance at the 1% level.

presents the estimation results obtained by using the conditional logit model. Data from 416 respondents were used for the estimation. Of the total number of respondents (885), 432 correctly answered the comprehension check question to determine whether they understood the explanatory text in this survey. One reason for the low percentage of correct answers is that the comprehension check question was “Please choose the wrong answer from the following sentences”, which may have caused some respondents to give incorrect answers because of confusion. Those who chose “I do not apply for any of these services” for all eight questions in the choice experiment were asked about the reasons for their choices. Subsequently, we excluded respondents who chose “it was annoying or difficult to choose” as their reason. Consequently, 416 responses were used for estimation. Among the 416 respondents, 215 (52%) were men, and 201 (48%) were women. Regarding age, 50 (12%) were in their 20s, 97 (23%) were in their 30s, 90 (22%) were in their 40s, 75 (18%) were in their 50s, and 104 (25%) were in their 60s.

WTP for attribute $x_1$ is

$$WTP=-\frac{{\beta }_{x1}}{{\beta }_{price}}$$

(10)

where the denominator is the coefficient of the variable Price that shows the monthly fee for the information provision service.

The variable Total_electricity takes the value of 1 if the information on the total electricity consumption amount for the entire house is provided and 0 otherwise. The variable Total_electricity_gas takes the value of 1 if the information on the whole house totals for electricity and gas consumption amount is provided, and 0 otherwise. Total_electricity_water takes the value of 1 if the information on the whole house totals for electricity and water consumption amount is provided, and 0 otherwise. The coefficients of the three variables are positive and significant.

The WTP is higher when information on “electricity and gas” or “electricity and water” is provided than when information on electricity only is provided. Additionally, the WTP for “electricity and water” is higher than the WTP for “electricity and gas”.

The variable Appliance takes the value of 1 if the information on the detailed electricity usage amount for each major home appliance is provided and 0 otherwise. This variable is significantly positive, and the WTP is approximately JPY 121. This is slightly higher than the WTP of JPY 86 when information on the total electricity consumption for the entire house is provided. This is thought to be because specific information is necessary to implement energy-saving behaviors.

The variable Rates takes the value of 1 if the information on electricity rates by time zone is provided and 0 otherwise. Rates was significantly positive, suggesting that there was also a high level of interest in electricity rates.

The variable Comparison takes the value of 1 if comparative information with other households is provided and 0 otherwise. However, these results were not statistically significant. Although feedback studies often use the presentation of comparative information from other households, the results of this study indicate that consumers are unlikely to be willing to pay for this information.

The variable Source takes the value of 1 if the information on the power source composition of electricity is provided and 0 otherwise. Although Source is positive and significant, the WTP is approximately JPY 57, which is low compared with other information. Nevertheless, the fact that information unrelated to energy savings is significantly positive may reflect the respondents’ interest in environmental issues in power generation.

ASC3 is the alternative-specific constant for the third alternative.

5.2. Latent Class Models

The variables used to estimate the membership function in this study are as follows. The variable Nuclear is set to 1 if a respondent believes that the use of nuclear power should continue in the future and 0 otherwise. In Japan, the accident at the Fukushima Daiichi Nuclear Power Plant was caused by the Great East Japan Earthquake in March 2011. This incident heightened Japan’s sense of crisis regarding nuclear power generation. As reducing the use of nuclear power requires either a decrease in the total electricity demand or a shift to other power sources, it is likely that those who oppose the use of nuclear power are also interested in energy conservation.

In Japan, the deregulation of electricity retailing began in April 2016. The questionnaire survey for this study was conducted in January 2016, immediately before that date. Therefore, we included the variable Deregulation, which takes the value of 1 if the respondent knew that the deregulation of electricity retailing would begin in April 2016, and 0 if the respondent did not know. If respondents knew about deregulation, they would likely be more interested in energy issues.

We included the variable Carbon, which takes the value of 1 if the respondent has some knowledge of the differences in the amount of carbon dioxide emissions from fuel combustion among coal-, oil-, and liquefied natural gas (LNG)-fired power generation, and 0 if the respondent has no knowledge at all. Those with this knowledge are likely to be knowledgeable about environmental issues and interested in power source composition and energy consumption.

We also included the variable Income of annual income per household divided by the number of persons in the household.

The estimation results for the membership function with the above variables are presented in

Table 4. Membership function.

	Class 1	Class 2	Class 3	Class 4
Constant	0.2723 (0.6143)	−0.1271 (0.7855)	−0.3054 (0.9464)	0
Nuclear	−0.0353 (0.4058)	−0.3313 (0.4319)	0.5286 (0.5262)	0
Deregulation	1.5090 ** (0.6299)	2.1146 *** (0.7797)	1.3813 (0.9108)	0
Carbon	−0.9206 (0.5809)	−0.8279 (0.5993)	−1.4121 ** (0.6791)	0
Income	0.0006 (0.0010)	−0.0010 (0.0012)	−0.0010 (0.0016)	0

Standard errors are in the parenthesis. *** and ** indicate significance at the 1% and 5% levels, respectively.

.

Respondents were classified into four classes based on the AIC (Akaike information criterion) (AICs are 4232.8 for Class 2, 3911.3 for Class 3, and 3856.2 for Class 4). Class 4 was set as the criterion, with all coefficients set to 0. Classes 1 and 2 contain more people who know about the deregulation of electricity retailing than Class 4 because the variable of knowledge about the deregulation of electricity retailing is significantly positive. Class 3 has more people who do not know about the CO₂ emissions from fossil fuels, compared with Class 4. The shares for each class are 0.42 for Class 1, 0.31 for Class 2, 0.14 for Class 3, and 0.13 for Class 4.

Table 5. Utility function.

	Class 1	Class 2	Class 3	Class 4
Total_electricity	1.8286 (2.0670)	0.6174 * (0.3235)	−0.1987 (0.6592)	0.8391 *** (0.2772)
Total_electricity_gas	3.0067 (2.1967)	1.0186 *** (0.2932)	0.9658 * (0.5252)	1.0658 *** (0.2909)
Total_electricity_water	3.5879 (2.6452)	1.4913 *** (0.3595)	1.2257 ** (0.4933)	1.1421 *** (0.2833)
Appliance	1.8898 (1.7532)	1.1557 *** (0.2480)	0.9116 ** (0.3588)	0.5301 *** (0.1709)
Rates	0.7279 (0.6905)	1.0844 *** (0.1414)	1.0426 *** (0.3539)	0.4750 *** (0.1610)
Comparison	−1.3820 * (0.7142)	−0.0187 (0.1504)	0.5278 (0.5030)	0.2684 (0.1790)
Source	2.0303 (1.7859)	0.6261 *** (0.2239)	−0.3933 (0.3796)	0.4374 *** (0.1431)
Price	−0.0166 ** (0.0067)	−0.0097 *** (0.0008)	−0.0119 *** (0.0023)	−0.0021 *** (0.0007)
ASC3	5.5835 ** (2.2617)	0.7178 * (0.3902)	−5.4366 *** (1.5989)	−1.2641 *** (0.4457)
Sample size	3328
Log-likelihood	−1877.0961
McFadden Pseudo R-squared	0.4148

Standard errors are in the parenthesis. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

shows the preferences for each class, and

Table 6. WTP for the latent class models (JPY/month).

	Class 1	Class 2	Class 3	Class 4
Total_electricity	0	63.65	0	399.57
Total_electricity_gas	0	105.01	81.16	507.52
Total_electricity_water	0	153.74	103	543.86
Appliance	0	119.14	76.61	252.43
Rates	0	111.79	87.61	226.19
Comparison	−83.25	0	0	0
Source	0	64.55	0	208.29

calculates the WTP using the coefficients in Table 5.

Monthly fees were significantly negative for all classes. The lower the price of the information provision service, the more it is preferred.

Class 1 has no positive significant coefficients related to energy and has a significantly negative coefficient with respect to comparative information with other households. This suggests that they are not interested in information about energy use and that comparing themselves to other households is rather distressing.

Class 2 is not significantly interested in information about “comparison with other households” but is significantly interested in other information. Although similar to Class 4, the WTP was lower than that in Class 4.

Class 3 is not significant in comparison with the other households, as is the power source composition of electricity. This suggests that, although they are interested in information on energy conservation, they are not interested in the situation of other households or the power source composition of electricity. Class 4 is significantly positive regarding information, except for the comparison with other households.

Class 1 has the highest share, accounting for approximately 40% of the total. This suggests that whereas approximately 40% do not intend to pay the price for such information, the remaining 60% may agree to pay a cost for some information.

WTPs for the latent class models are shown in Table 6.

Whereas studies on HEMS have focused almost exclusively on electricity with respect to monitoring (Chen et al. [25], Chen et al. [27], Washizu et al. [28], Kaufmann et al. [30], Gerpott and Paukert [29]), this study deals with water and gas in addition to electricity in the attribute of “Total energy consumption amount for the entire house”. Regarding WTP, information on electricity and water has a higher WTP than information on electricity and gas. This is because gas is not used in all-electric households, whereas water is used in all households.

Information on electricity rates by time zone is considered of interest because it is about electricity prices and is significantly positive in Classes 2, 3, and 4. Yoshida et al. [42] examined whether consumers accept dynamic pricing and what method of dynamic pricing is preferred through a choice experiment. Dynamic pricing is intended to reduce the difference in electricity demand between peak and off-peak time zones. Among the rules on dynamic pricing, the time-of-use rate scheme is the most preferred by consumers, and this scheme has the highest WTP. Although direct comparisons cannot be made because the subjects estimating WTP are different, it is considered that households are interested in information on changes in electricity prices.

It is worth noting that no classes have a positive WTP for comparative information with other households. In particular, Class 1 has a negative WTP. Clearly, households are not interested in, or do not prefer, this information.

The WTP for information on the power source composition of electricity has not been dealt with in previous studies. This study examined WTP for the information and showed that people in classes 2 and 4 have a positive WTP. In previous studies, it has also been shown that nuclear power is not preferred in Japan (Nakai et al. [13], Morita and Managi [9], Murakami et al. [10]). However, in this study, Nuclear is not significant in the membership function; therefore, opinions about whether to continue using nuclear power do not seem to have much impact on the classification. Knowledge of the differences in the amount of carbon dioxide emissions from fuel combustion among coal-, oil-, and LNG-fired power generation is negatively significant in class 3. This means classes 1, 2, and 4 had more knowledge than class 3. However, WTP for information on power source composition varies among these classes. Therefore, knowledge about the amount of carbon dioxide emissions does not have much impact on the WTP for the power source composition of electricity.

Chen et al. [25] found a significant relationship between WTP and income. However, Table 4 shows that income was not significantly estimated in the membership function in our study. This. means that the four classes did not differ in terms of income. This means that income is not related to the difference in WTP among classes. This might be partly because the respondents who answered “I do not apply for any of these services” to all eight choice questions were excluded from the sample.

6. Discussion

The results of the latent class model revealed that preferences differed by class. In Chen et al. [27], WTP for HEMS in Japan was examined, and respondents who did not want to pay at all for monitoring electricity usage were about 40% of the total respondents. In our study, the group unwilling to pay corresponded to Class 1 (42%). This was approximately equal to that reported by Chen et al. [27]. In Chen et al. [27], visualization and monitoring focused on electricity usage, whereas our study included other information. In our study, the items related to electricity usage amount are the total electricity consumption amount for the entire house (Total_electricity) and the detailed electricity usage amount for each major home appliance (Appliance). The WTP for the sum of the two items is JPY 182.79 (USD 1.55) for Class 2 (31%), JPY76.61 (USD 0.65) for Class 3 (14%), and JPY 652 (USD 5.54) for Class 4 (13%). (Converted at the January 22, 2016 rate of USD 1 = JPY 117.77.) The shares in Chen et al. [27] are approximately 15%, 10%, and less than 10% for respondents in each price range, respectively. There is a divergence in the price range that corresponds to Class 2, but not much divergence in the others.

Washizu et al. [28] studied WTP for HEMS in Japan. As the functions in their study are not limited to visualization and monitoring, they are combined with a function to warn when there is excessive use of electricity. Therefore, direct comparison with our study is difficult. However, they showed that 60% of the respondents had a positive WTP. This is consistent with our study as approximately 60% indicated a positive WTP (sum of Class 2, Class 3, and Class 4 in the latent class model).

To summarize, the latent class model analysis indicates that approximately 60% of respondents are willing to pay for information provision services. These results are consistent with those of previous studies.

Studies have shown that households with information comparing electricity use with others reduce energy consumption (Andor and Fels, [3]). However, in this study, in Class 1, which accounts for approximately 40% of the analyzed sample, the WTP for comparative information with other households is negative. Therefore, comparisons with others may be painful for these households. Therefore, when electricity companies provide comparative information with others, it is better to carefully consider how and to whom they provide it.

Koo et al. [31] investigated WTP to introduce smart water metering and monitoring systems and found that the system is socially profitable. As a positive WTP was observed in our study, the introduction of smart metering and monitoring systems may also be considered for water.

7. Conclusions

In this study, a choice experiment was conducted to examine consumers’ WTP for energy-related information. Based on the conditional logit model, positive WTP was estimated for information on the total energy consumption amount for the entire house, detailed electricity usage amount for each major home appliance, electricity rates by time zone, and power source composition of electricity. Among these, the WTP was high for information on electricity consumption per appliance, likely to lead to savings through specific energy-saving behavior. In contrast, a low WTP was estimated for information on the power source composition of electricity, which is not directly related to savings. The results of the latent class model also revealed that preferences differed by class, and approximately 60% were willing to pay for the information provision service.

As discussed in Section 2, prior studies seeking WTP for HEMS and smart meters have also examined functions other than information provision. However, this study focused on the information provision function, and the types of information provided were elaborated upon. As mentioned in Section 5, many previous studies have focused only on electricity. However, this study includes gas and water in addition to electricity in the attribute of “total energy consumption amount for the entire house”. It has shown that providing information on gas and water is also meaningful. Although WTP for information on the power source composition of electricity was not addressed in previous studies, this study found that there are groups that had positive WTP. Regarding the comparative information with other households, the class that accounted for about 40% of the analyzed sample had negative WTP. These results indicate that comparative information can be distressing for some households. These additional contributions may be useful for considering future information provision services.

The results of this study can also be used to obtain choice probabilities regarding which hypothetical information provision services people will select for the given options. A simulation study using this method will be the subject of future research.

The limitations of this study and related future research directions are as follows: The first limitation is that some respondents seemed unfamiliar with the idea of purchasing information. This might have made the questionnaire survey difficult for the respondents to understand. One of the future research directions related to this limitation is that explanations of information provision services need to be more clarified. One way to do this is to show respondents a video showing an image of the screen of the information service. Second, the possibility that the WTP is estimated to be higher should be considered. In stated preference methods, including choice experiments, there is a hypothetical bias due to the lack of real payments (Loomis, [43]). In other words, it should be kept in mind that a higher WTP may be estimated because respondents who choose an information service do not actually have to pay for it. One future research direction related to this limitation is the implementation of an economic experiment with actual payments.