3.1. Survey Design and Data Collection: A Choice Experiment
The AMI data market has not yet been activated, and the smart energy data service provided based on the data has not yet been released. At this point, there is a limit to collecting related preference data. Therefore, the value of the smart energy data service is estimated by collecting the stated preference data of respondents through a survey. This study explained the AMI data-based energy service to the respondents. In addition, the respondents’ preferences for the service were investigated by explaining in detail the service currently being provided and the service to be provided in the future. The selected service attributes are ‘electricity information service for apartment complexes’, ‘electricity information service for individual households’, ‘household demand response service’, ‘electricity trading service’, and ‘data convergence service’.
Table 2 shows the five attributes and levels included in the CE questionnaire.
Currently, the service provides only information pertaining to the amount of power used, such as the power consumption of individual households, rates, and a comparison with the previous month’s usage. It does not provide consultations about used power or services for efficient energy use management. Therefore, by adding personalized services to the existing services, new energy data services such as an ‘electricity information service for apartment complexes’ and an ‘electricity information service for individual households’ can be added. Consumers can receive consulting services on rate contract methods that can reduce electricity charges jointly borne through the ‘electricity information service for apartment complexes’. The ‘electricity information service for individual households’ provides customized information (e.g., basic customer information, power consumption (kWh) and rates, power demand levels (kW), the current rate (KRW), time-based rates (KRW) and the predicted electricity rate for this month, access to a rate table, comparisons among neighbors, message transmission (e.g., notices, peak notifications), advance notifications when entering a progressive stage, access to standby power information) to consumers and provides efficient energy management services.
DR programs are an important strategy for controlling and optimizing energy consumption at the end-user level [
34,
35]. There should be two-way information between the DR aggregator and the consumers participating in the DR market, and the installation of AMI devices that can read power consumption in real time should precede the start of the program. The AMI-based energy data service provides notifications to the public through the platform when electricity use is concentrated or particulate matter is ‘bad’. Through AMI, it is possible to check how much electricity consumption has been reduced, and through the platform, national DR participants can receive incentives according to how much they reduced their electricity consumption [
36]. The South Korean government is making efforts to maximize public participation in DR along with the supply of AMI to manage electricity demand. For example, a ‘household demand response service’ is provided to allow consumers to directly reduce their power consumption and to provide incentives according to the level of consumer participation (approximately KRW 1300 per kWh). However, the ‘household demand response service’ has been piloted since 2016, but there is still no data to present proper performance due to low awareness. Therefore, the ‘household demand response service’ was selected as an attribute in this study to investigate consumers’ awareness of DR.
Recently, in order to improve the utilization of AMI and develop related new energy businesses, a service by which users of small amounts of electricity can trade is being planned through an AMI-based energy data platform [
37]. The energy data service is a real-time rate system based on big data that helps to match sellers and buyers in the electricity market by identifying their progress levels. By comparing and showing expected profits after a transaction, consumers can engage in profitable activities. Therefore, in order to attract consumers’ attention, this study aims to review consumers’ acceptance by adding an ‘electricity trading service’ that not only reduces electricity use but also facilitates profitable activities as an attribute.
The ‘data convergence service’ refers to the provision of a new service by combining two or more types of energy data or combining energy data and non-energy data. In this study, the ‘social safety-net service’, ‘transportation-related convergence service’, and ‘building safety management service’ were selected as major services and applied to the level of the convergence service. The data convergence services are now in the formative stage. Currently in South Korea, the ‘social safety-net service’, which detects abnormalities for the elderly living alone or for the socially underprivileged through electricity consumption, is a representative example of a convergence service [
38]. However, the ‘social safety-net service’ was selected as an attribute level to re-evaluate the value of the service due to low public awareness.
Next, the government’s plans to expand the supply of electric vehicles and the increase in consumer demand for electric vehicles are both factors that are accelerating the construction of charging station infrastructure in residential areas [
39]. Reflecting the current market, the ‘transportation-related convergence service’ was added at the attribute level to attract consumer demand by providing information such as the locations of electric vehicle charging stations and charging availability.
In an apartment complex where many people live in one building, casualties may occur in the event of a fire. Therefore, safety management to prevent accidents is very important. AMI can detect building anomalies by identifying power consumption patterns [
40]. However, there is no service that provides information by diagnosing building safety and detecting anomalies. Therefore, the ‘building safety management service’, which detects and diagnoses abnormal power consumption through AMI to check the safety of buildings, was selected as a new convergence service business and was included as an attribute level.
The last attribute, the payment vehicle, was selected as an ‘additional electricity bill’. Several existing studies of electric power services also set the payment vehicle as the electricity bill [
22,
23,
27]. Therefore, this study also considered the electricity bill as a payment vehicle. To explain this in detail, this refers to a service usage fee that is added to the existing electricity bill every month when using a smart energy service. The established attribute levels of the payment vehicle were KRW 1000, KRW 2000, and KRW 3000 to reflect the recent increase in electricity rates. In South Korea, household electricity bills increased by KRW 7.4 per kWh, representing approximately a 6% increase [
41]. The average monthly electricity bill per household is about KRW 50,000 [
42], which is an increase of KRW 3000 considering the rate of increase of 6%. After setting KRW 3000 as the maximum amount, the remaining level was set by decreasing this fee at a constant rate.
In this study, we set the number of attributes to six, considering the essential characteristics of the CE survey. While determining the appropriate number of attributes in a CE should be based on research objectives and the trade-offs between the complexity of the survey design and the quality of the data, it is generally recommended to limit the number of attributes to six or fewer [
43]. Increasing the number of attributes can result in survey fatigue and response errors. However, there are other crucial attributes that energy services based on AMI data can have. For instance, a strategy that can enhance consumer engagement in the energy service, such as gamification, is an important attribute. Recently, research has reported that gamification, which involves using game elements in non-game contexts to improve the user experience and user engagement, can increase consumer participation and effectively induce behavioral changes and energy efficiency improvements [
44,
45,
46]. In the future, it is essential to examine the effects of such consumer engagement techniques on consumer choice. The levels of all potential attributes not included in this CE survey were considered to be equal for each alternative; these were fully recognized by the respondents.
As shown in
Figure 1, alternatives consist of a combination of the attribute level and the price level. A total of 288 possible alternatives for the proposed attributes can be selected. Suggesting all 288 alternatives is time-consuming, costly, and inefficient. Therefore, it is necessary to derive a minimum set of alternatives among the 288 options. The minimum set of alternatives in each case was obtained using an orthogonal main effects design created with the SPSS 12.0 package. Finally, twenty-four selected alternatives were obtained, and the randomly mixed alternative sets were divided into a total of eight sets by merging three alternatives into one set, as shown in
Figure 1. Then, four sets were randomly assigned to each respondent. In the CE survey, respondents were shown questions consisting of three alternatives, as shown in
Figure 1, and were asked to choose their preferred alternative.
This survey was conducted for about a month in June of 2022 by a specialized survey agency (Gallup) targeting 17 major cities. An online survey was conducted by randomly selecting 1021 households nationwide. Respondents to the survey were limited to household heads or spouses of household heads between the ages of 20 and 68 with income. A total of 1021 data instances was obtained, and the representativeness of the sample was reviewed by comparing the characteristics of each respondent with the entire set of households (population).
Table 3 compares the characteristics of the respondents and the population.
3.2. Model Specifications
Consumer preference for smart energy services based on AMI data is analyzed using a discrete choice model (DCM) based on a probability utility model that assumes a decision-maker’s utility maximization behavior. Respondents select the alternative that gives them the greatest utility among a plurality of service alternatives. Because CE data have discrete characteristics intrinsically, DCM is suitable as the analysis methodology in this study.
This study analyzed CE data using the mixed logit model, a type of DCM. The mixed logit model is quite flexible and exhibits heterogeneity of preferences because the researcher can directly set the distribution of the coefficients of the attributes [
48]. In addition, the mixed logit model can relax the assumptions of the standard logit model because it allows the use of a variance–covariance matrix between the randomized taste parameters [
48].
In the mixed logit model, the utility
by which respondent
obtains from one selected alternative
in choice set
can be expressed as Equation (1) [
48,
49].
The utility (
) of each respondent can be divided into
, a deterministic term, and
, a stochastic term with uncertainty.
can be expressed as the product of a vector (
) of attributes related to the selected alternative
and a vector (
) of the coefficients of attributes. It is assumed that
follows a normal or log-normal distribution with mean
and variance
and that
is independent and follows an identity Type I extreme value distribution [
48]. In this study, the utility that respondent
feels for the alternative
for the AMI data-based smart energy service can be configured as in Equation (2).
Given that respondents choose an alternative that maximizes their utility, the probability that respondent
will choose alternative
can be expressed as Equation (3).
The mixed logit model used here was estimated by applying the Bayesian approach presented by Train [
48]. Compared to the classical approach, the Bayesian approach has been widely applied in various studies because the estimates of respondents’ part-worth values are nearly identical [
50,
51,
52,
53]. The Bayesian approach solves the problem of computational complexity and can overcome the initial value problem because it does not require function maximization calculations [
50,
54]. Consistency and efficiency can also be obtained under more flexible conditions [
48]. Therefore, the Bayesian approach is an appropriate approach for the analysis in this study.
On the other hand, the coefficient values estimated based on the mixed logit model represent the marginal contribution to the utility of each attribute with an arbitrary unit. They do not have any comparable meaning between attributes. Therefore, it is necessary to calculate the marginal willingness to pay (MWTP) for each attribute from the estimation results. The MWTP is the amount consumers are willing to pay to keep their utility unchanged when the quantity or quality of an attribute changes by one unit. This can be interpreted as the amount of change in the consumer’s compensated surplus when the attribute changes. The MWTP that respondent
has for the remaining attributes (
), except for the attributes (
) related to price, can be derived through Equation (4).
Additionally, by analyzing the relative importance (RI) of each attribute, it is possible to compare how much each attribute affects the decision-making step. The RI was calculated using the part-worth of each attribute, and the partial value of each attribute
was derived by multiplying the level difference of the attribute by the coefficient value
.