Purchase decision process and information acquisition of zero-energy houses in Japan

ABSTRACT Compared with traditional houses, zero-energy houses (ZEHs) offer efficient and preferable living environments, e.g., reduced greenhouse gas emissions and lower health risks. Currently in Japan, such houses are not as popular as anticipated and sales do not meet the national government target. Accordingly, household buying process should be investigated to develop policies to encourage the spreading of ZEHs. Therefore, we investigated which factors influenced purchasers’ intentions and behaviors. We based our purchase process modeling on the unified theory of acceptance and use of technology, which includes six constructs, i.e., use behavior, behavioral intention, performance expectancy, effort expectancy, social influence, and facilitating conditions. Our model also considered the effects of the information content buyers obtained and the channels they used on performance expectancy. In our estimation, we used Bayesian structural equation modeling and response from 297 Japanese households. It was found that certain information content and channel combinations, e.g., health aspect information obtained from salespersons effectively enhanced performance expectancy. Although performance expectancy did not significantly facilitate the use intention, social influence and facilitating conditions effectively promoted intention leading to purchase. Our findings contribute to more appropriate information provision strategies and supporting policies to promote the spread of these houses.


Introduction
After the adoption of the Paris Agreement in 2015 (Rogelj et al. 2016;United Nations 2015), it has been expected of global society, including national governments and all other stakeholders, to institute further global warming mitigation measures. Global energyrelated CO 2 emissions amounted to approximately 32 gigaton in 2016, which are predicted to increase to 36 gigaton in 2040 (International Energy Agency 2018). Constructing zero-energy houses (ZEHs) is one of the measures to promote decarbonization. These residential buildings greatly reduce energy needs through efficiency gains, such that the balance of energy needs can be supplied with renewable technologies (Stefanović, Bojić, and Gordić 2014). The adoption of such environment-friendly living conditions is expected to spread worldwide (Farhar and Coburn 2008).

Zero-energy houses in Japan
In Japan, a ZEH is defined broadly as a house of which the annual energy consumption balance is zero. This is achieved by installing renewable energy-generation devices, improving the insulation of the envelope, and installing highly energyefficient household equipment, all of which bring about superior indoor environmental quality. More specifically, a ZEH is defined as a house satisfying the following three criteria (Ministry of Economy, Trade and Industry 2015; Oki et al. 2019): (1) It meets the criteria for strengthened outer walls and U A value (the amount of heat that escapes from the inside of the house to the outside through the floors, outer walls, roof, and windows, divided by the area of the outer skin); (2) It reduces the primary energy consumption, excluding renewable energy, by 20% or more from the standard primary energy consumption (energy consumption for heating, cooling, ventilation, hot water supply, and lighting); (3) It reduces primary energy consumption, including renewable energy, by 100% or more from the standard primary energy consumption.
The necessary equipment to meet such criteria includes heat-insulation walls, photovoltaic (PV) panels, and lightemitting diode-based (LED) lamps, with the option to include high-efficiency air conditioners, floor heating, energy-efficient water heaters, home energymanagement systems (HEMS), vehicle-to-home (V2H) communication, and the like.
Japanese society also needs to deploy ZEHs from both a societal and an individual viewpoint. From a social perspective, the household sector needs to urgently reduce its energy consumption, which has been increasing substantially and at a more rapid rate than that of other sectors. 1 In this regard, an advantage of ZEHs is their high thermal-insulation performance that realizes energy savings for the household sector. ZEHs also potentially contribute to the balanced energy mix of society as they increase the proportion of renewable energy.
From an individual perspective, ZEHs can reduce energy costs. After the Fukushima Daiichi nuclear disaster induced by the 2011 Tohoku earthquake and tsunami, the electricity price in Japan has fluctuated (Ministry of Economy, Trade and Industry 2018a) owing to nuclear power plants not being operational and the increasing cost of generating thermal power because of the soaring cost of fuel. Therefore, both the energy saving and generation realized by ZEHs could bring about economic benefits to individual households. In addition, ZEHs provide several non-energy benefits. First, they function effectively during emergencies (e.g., natural disasters) because of electricity selfconsumption (Miller 2015;Young Jr, 2009). A case in point is the blackout after the occurrence of the 2018 Hokkaido Eastern Iburi earthquake that affected three million households. All power outage was resolved two days after the earthquake; however, 85% of households having PV panels, which are necessary for ZEHs, used their self-generated electricity effectively (Japan Photovoltaic Energy Association (JPEA) 2018). Furthermore, some households making use of rechargeable batteries were unaffected by the outage and were even able to provide electricity to their neighbors (Monthly Smart House 2018a). Second, they likely provide health benefits to residents (Monthly Smart House 2018b; Oki et al. 2019). The high heat-insulation performance reduces the temperature gap among rooms, including toilets and bathrooms and, therefore, the risk of heat shock (i.e., a sudden change in blood pressure owing to rapid temperature change, which can cause a stroke or myocardial infarction) is potentially reduced (Monthly Smart House 2018b; Oki et al. 2019).
However, a gap exists between the current prevalence of ZEHs and the target set by the national government. The goal is that the majority of detached houses built by contractors would be ZEHs by 2020 and all newly constructed houses would be ZEHs by 2030 (Ministry of Economy, Trade and Industry 2018b). However, on average, ZEHs represented only 10% of detached houses built in Japan in 2018 (Sustainable open Innovation Initiative 2018). Accordingly, in an effort to promote the sale of ZEHs to reach the government goal, the process and decision of households to purchase such houses needed to be researched (See Appendix A for additional details of ZEHs in Japan).

Literature review
Although several previous studies have focused on the supplier aspect of ZEHs, such as construction companies and designers (Attia et al., 2013;Farhar and Coburn 2008;Persson and Grönkvist 2015;Shi et al. 2020;Zhao, Pan, and Chen 2018), research on the consumer aspect is scarce. Analyses have already been conducted on consumer choices relevant to various products commonly installed in ZEHs, such as energy-efficient water heaters (Goto, Goto, and Sueyoshi 2011;Ma, Yu, and Urban 2018;Michelsen and Madlener 2012;Ofuji and Nishio 2013), PV panels (Bollinger and Gillingham 2012;Graziano and Gillingham 2014;Noll, Dawes, and Rai 2014;Yamaguchi et al. 2010), LED lamps (Khorasanizadeh et al. 2016), and HEMS (Park et al. 2017). The detailed purchase process of such devices has also been analyzed (Khorasanizadeh et al. 2016;Park et al. 2017), namely, how households perceive the performance and ease of using the products, how such perception affected the intention to purchase, and how the intention promoted the actual purchasing behavior. However, the purchase process of ZEHs, not energy-saving or energy-generating products, has not been investigated.
Information generally plays a significant role in the process of adopting new technologies, as, during the initial stages, prospective buyers become familiar with the products and their functions (Rogers 2003). In studies conducted on energyrelated technologies, the researchers examined to what extent the views of neighbors or other people influenced the choice of consumers of energyefficient water heaters (Goto, Goto, and Sueyoshi 2011) and PV panels (Bollinger and Gillingham 2012;Graziano and Gillingham 2014;Noll, Dawes, and Rai 2014). This implies that the research focus was mainly on the effect of the channels of information on the adoption. For instance, recommendations from home builders and/or sales representatives of appliances (Goto, Goto, and Sueyoshi 2011), as well as suggestions from neighbors (Bollinger and Gillingham 2012;Graziano and Gillingham 2014), were found to affect the buyers' choices. However, the element lacking here is that the role of channels depends on the content of the information . In other words, how the recipients interpret the information can depend on the type of information content they obtain and from whom it is obtained. As regards ZEHs, in particular, the information would have to be associated with the various types of advantages (e.g., economic, environmental, and health benefits (Monthly Smart House 2018b; Oki et al. 2019), disaster risk reduction (Miller 2015;Young Jr, 2009)), and the different energy-saving and energy-generating equipment. Therefore, in analyzing the effect of information on promoting the sale of ZEHs, combining various information contents and channels should be considered.

Objectives
The aim of this study was to model the household purchase process of ZEHs and investigate how the intention to purchase these houses as well as the actual purchasing behavior were promoted, considering the effects of various combinations of information content and channels. We targeted households that purchased ZEHs in Japan and we conducted statistical analysis based on Bayesian structural equation modeling and the dataset containing the response received from 297 households. Our analysis revealed the effective information contents according to the information channels and the factors that could positively influence the intention and actions to purchase ZEHs. Our findings could aid in assessing and improving the current information provision strategies and supporting policies aimed at promoting the spread of ZEHs. Since concepts similar to Japanese ZEHs are found in other countries, 2 our findings could also be relevant to such countries.
The remainder of this paper is organized as follows: Section 2 describes the framework of our analysis; Section 3 shows the statistical method and data used in the analysis; Section 4 reports the estimation results; Section 5 discusses the results and limitations; and Section 6 concludes the study, describing policy implications.

Unified theory of acceptance and use of technology
We used the unified theory of acceptance and use of technology (UTAUT) (Curtale, Liao, and van der Waerden 2021;Hartwich et al. 2019;Rajapakse 2011;Sovacool 2017;Venkatesh et al. 2003) to model the household purchase process of ZEHs. UTAUT, which helps to understand the drivers of acceptance by users of new technologies, has been developed by integrating the elements of eight prominent models, including (1) the theory of rational action (TRA), (2) technology acceptance model (TAM/TAM2), (3) motivational model (MM), (4) theory of planned behavior (TPB), (5) model agreement between TAM and TPB (combined TAM-TPB) , (6) model of personal computer utilization (MPCU), (7) innovation diffusion theory (IDT), and (8) social cognitive theory (SCT). The UTAUT is believed to be more robust in evaluating and predicting technology acceptance than the other technology acceptance models (Taiwo and Downe 2013). Whereas UTAUT was developed originally to model the adoption process of information technology, it has been applied to this process in a wide variety of technologies, including energy-saving and energycreating technologies (e.g., LED lamps (Khorasanizadeh et al. 2016) and renewable energy sources (Rezaei and Ghofranfarid 2018)).
The UTAUT model considers "use behavior," which represents the acceptance and use of technology, and "behavioral intention," which represents the degree of intention to adopt the technology. Behavioral intention was initially described in the TRA, which claims that the effectiveness of a certain behavior is a consequence of whether or not the individual intends to perform the behavior. The UTAUT uses four core determinants of use behavior and behavioral intention, namely, (1) performance expectancy, (2) effort expectancy, (3) social influence, and (4) facilitating conditions. Details on these four determinants follow (Rajapakse 2011;Sovacool 2017;Venkatesh et al. 2003).
(1) Performance expectancy: performance expectancy is defined as the degree to which the user expects that using the system will help him or her attain gains in performance. This construct finds its roots in perceived usefulness from TAM/TAM2 and combined TAM-TPB, extrinsic motivation from MM, relative advantage from IDT, and outcome expectation from SCT. More broadly, performance expectancy has come to mean the degree to which users expect that technology will be beneficial in performing particular tasks.
(2) Effort expectancy: effort expectancy is the degree of ease associated with consumers' use of the technology. This construct finds its roots in concepts such as perceived ease of use from TAM/TAM2, complexity from MPCU, and ease of use from IDT. (3) Social influence: social influence refers to the degree to which an individual perceives that important others (e.g., family and friends) believe that he or she should use the new system. It finds its roots in concepts such as subjective norms (i.e., the individual's perception that most people who are important to him/ her think he/she should use the system) from TRA, TAM2, TPB, and C-TAM-TPB, social factors (i. e., the individual's internalization of the reference group's culture) from MPCU, and image (i. e., the degree to which the use of an innovation is perceived to enhance one's social image or status) from IDT. (4) Facilitating conditions: facilitating conditions are defined as the degree to which an individual believes that an organizational and technical infrastructure exists to perform a task or adopt a new system. This construct embodies perceived behavioral control from TPB, facilitating conditions from MPCU, and compatibility from IDT.
As regards the relationships between the above six constructs, namely, (1) use behavior, (2) behavioral intention, (3) performance expectancy, (4) effort expectancy, (5) social influence, and (6) facilitating conditions, we hypothesized the connections between six ovals, as shown in Figure 1, based on Venkatesh et al. (2003) and Khorasanizadeh et al. (2016). More specifically, we assumed that the behavioral intention to purchase ZEHs was facilitated by performance expectancy, effort expectancy, social influence, and facilitating conditions, and the use behavior was promoted by the behavioral intention and facilitating conditions.

Information acquisition
We extended the UTAUT model, adding the effect of acquiring information about ZEHs (rectangle in Figure 1). Although each construct may be affected to a greater or lesser degree by the content of information obtained and from whom it was obtained, we focused on their effect on the performance expectancy. The reason was that the content of information that we targeted was mainly related to the benefits of ZEHs, which will be shown below, and we avoided complicating the model. We considered the following four content types, including the multiple benefits of ZEHs: (1) Economic information: information on economic issues, such as the price of energysaving and energy-generating equipment of ZEHs and the utility cost reduction realized by ZEHs; (2) Technical information: information on the functions and mechanism of energy-saving and energy-generating equipment installed in ZEHs; (3) Environmental information: information on environmental issues, such as greenhouse gas reduction and risk reduction during emergencies (e.g., disasters); (4) Health information: information on health issues, such as the positive effect of the high heatinsulation performance on the residents' health. As regards the channels of information, the following four types were considered: (1) Salespersons: channels such as face-to-face communication with the salespersons of home builders; (2) Friends: channels such as word-of-mouth communication with friends, colleagues, and family members; (3) Advertisement: channels such as advertisements provided by home builders and the government; (4) Strangers: channels such as online word-ofmouth and reviews posted by people other than friends. Rogers (2003) categorizes communication channels as (i) interpersonal versus mass media and (ii) localite versus cosmopolite. Interpersonal channels involve the face-to-face exchange between two or more individuals. Mass media channels transmit messages through media such as the radio, television, and newspapers, which facilitate a source reaching an audience of millions. Cosmopolite communication channels are those linking an individual to sources outside the social system under study. Interpersonal channels can be either local or cosmopolite, while mass media channels are almost entirely cosmopolite.
The salespersons considered in this study are interpreted as interpersonal and cosmopolite channels; friends are interpersonal and localite; advertisement is considered mass media channels. The online wordof-mouth and review by strangers, which are categorized neither into mass media nor as interpersonal, have recently been found to be effective (Chevalier and Mayzlin 2006;Duan, Gu, and Whinston 2008;Ye et al. 2011) and, therefore, we also considered the channels of strangers.

Proposed model
Overall, we considered the structure shown in Figure 1. That is, the study assumed, as presented in Subsection 2.2, that the combination of the four information contents and the four channels (i.e., 16 combinations) affected the performance expectancy of ZEHs. Moreover, as illustrated in Subsection 2.1, not only performance expectancy but also effort expectancy, social influence, and facilitating conditions were assumed to facilitate behavioral intention to purchase ZEHs. It was also assumed that behavioral intention and facilitating conditions promoted use behavior. Based on the proposed structure, we examined (1) how the performance expectancy of ZEHs was affected by the combination of the information contents and the channels and (2) how the behavioral intention and actual behavior to purchase ZEHs were facilitated by the constructs of UTAUT.

Bayesian structural equation modeling
To estimate the structure mentioned in the previous section, we employed structural equation modeling (SEM), which is statistical modeling technique that can estimate the relationships between variables, including not only observed but also latent variables. Each of the six constructs of UTAUT in our structure (i. e., the ovals in Figure 1, namely, (1) use behavior, (2) behavioral intention, (3) performance expectancy, (4) effort expectancy, (5) social influence, and (6) facilitating conditions) were considered difficult to be represented by a single variable (Khorasanizadeh et al. 2016) and, therefore, we treated these six constructs as latent variables. We denoted each latent variable of household i 2 1; . . . ; n f g as ω B;i , ω I;i , ω PE;i , ω EE;i , ω SI;i , and ω FC;i , respectively, 3 and we denoted the vectors of measurement variables for each latent variable as z B;i , z I;i , z PE;i , z EE;i , z SI;i , and z FC;i , respectively. SEM is generally composed of (1) measurement models that examine the relationships between observed variables and latent variables and (2) a structural model that analyzes the interrelationships among latent variables.

Measurement models
In our measurement models, as the observed variable vector z B;i was measured in the yes/no format and z I;i , z PE;i , z EE;i , z SI;i , and z FC;i were measured on a five-point Likert scale using the survey as shown in Subsection 3.2, we incorporated the probit link functions suitable to model discrete outcome variables. Let ω i express the latent variable vectors as ω i ¼ More specifically, "j 2 1; . . . ; n B f g, z B;i;j is a dummy variable having either 0 or 1 and determined by y B;i;j , as follows: (1) z B;i;j ¼ 1 otherwise: (2) Furthermore, "j 2 1; . . . ; n I f g, z I;i;j , an element of z I;i , is an ordered categorical variable with a five-point Likert scale from 1 to 5 and defined by y I;i;j , as follows: (3) . . .
where α I;j;1 , . . . , α I;j;4 are unknown threshold parameters. As elements of z PE;i , z EE;i , z SI;i , and z FC;i are also measured on a five-point Likert scale, they are defined in the same way. The measurement models associate (1) y B;i , y I;i , y PE;i , y EE;i , y SI;i , and y FC;i with (2) latent variable vector ω i . For use behavior, the model is defined as follows: where a vector of intercepts where the random variables ε I;i;j , ε PE;i;j , ε EE;i;j , ε SI;i;j , and respectively. The vectors of error terms ε B;i , ε I;i , ε PE;i , ε EE;i , ε SI;i , and ε FC;i do not correlate with each other, and each vector of error terms is independent of ω i .
For model identification, some conditions are imposed on the model. One of the factor loadings for each latent variable is fixed as 1: . Threshold parameters are fixed as "k 2 Lee, Song, and Cai 2010;Lee and Song 2012). F � ð Þ is the cumulative distribution function of standardized normal distribution N 0; 1 ð Þ, f k;j;1 is the frequency of the first category (i.e., ), and f k;j;4 is the cumulative frequency of the categories that are less than five (i. e., f k;j;4 ¼ P In our analysis, we focused on the coefficients of paths from latent variables to observed ones (factor loadings)-Λ B , Λ I , Λ PE , Λ EE , Λ SI , and Λ FC .

Structural model
Whether household i obtained economic information from (1)  To formulate the model, the latent variable vector ω i is divided into η i and � i . η i is an outcome latent variable vector affected by other latent variables and/or d i ; � i is an explanatory latent variable vector. In this study, The interrelationships among latent variables are formulated by the following structural model: where C, �, and Γ are matrices of unknown coefficients. More specifically, is a random variable vector, expressed by A . δ i is independent of � i and ε i . In our analysis, we focused on the coefficients of (1) paths from d i to ω PE;i and (2) paths among latent variables: C, �, and Γ.

Bayesian estimation
We estimated path coefficients, i.e., parameters, based on the household data of the observed variable vectors, z B;i , z I;i , z PE;i , z SI;i , z EE;i , and z FC;i , and the variable vector on information acquisition, d i . Whereas many studies follow a classical frequentist approach, namely, the maximum likelihood method (Hox 1998;Kotani, Honda, and Sugitani 2019) for the estimation, we employed a Bayesian approach (Assaf, Tsionas, and Oh 2018;Lee and Song 2012;Levy and Mislevy 2017;Lu et al. 2020;Merkle and Wang 2018;Song and Lee 2012). To the best of our knowledge, this study is the first to apply Bayesian SEM to the model based on the UTAUT in building and energy research. The Bayesian estimation method treats parameters as random variables. Drawing on the Bayes' theorem, the prior probability distribution of unknown parameters, i.e., prior distribution, is updated, given the data obtained, to posterior distribution (Gelman et al. 2013;Lee and Wagenmakers 2014;Levy and Mislevy 2017). That is, p θjD ð Þ / p Djθ ð Þp θ ð Þ, where θ is an unknown parameter vector, D is data, p θ ð Þ is a prior distribution of the parameters, p Djθ ð Þ is a likelihood, and p θjD ð Þ is a posterior distribution. In most instances, obtaining the posterior distribution is done by simulation, using the so-called Markov chain Monte Carlo (MCMC) methods. The posterior distribution simulated via MCMC expresses the uncertainty of the parameters. The sampling-based Bayesian methods depend less on asymptotic theory and, therefore, have the potential to produce results that are more reliable, even with small samples, compared with those obtained by the maximum likelihood method (de Schoot et al. 2017(de Schoot et al. , 2014Lee andSong 2012, 2004). Generally, in the initial stages of the acceptance of technology, as was our target, it is difficult to obtain large samples and the Bayesian method is therefore considered suitable. Furthermore, the Bayesian method is more flexible with complex datasets, as it treats the raw data of dummy and five-point Likert scale variables more easily without approximating them to continuous variables (Lee, Song, and Cai 2010;Lee and Song 2012).

Data collection
To collect data, we conducted a questionnaire survey, as shown in Table 1. We focused on households that purchased custom-built detached ZEHs 4 and received the subsidies related to purchasing ZEHs. The survey was answered by one of the family members in each household who primarily negotiated the designs and prices with home builders or who understood the purchase process well. Through the survey, we collected n ¼ 297 sample data. The sample characteristics are presented in Appendix B.
For the observed variable vectors z B;i , z I;i , z PE;i , z SI;i , z EE;i , and z FC;i , we asked n B ¼ 7, n I ¼ 2, n PE ¼ 7, n EE ¼ 4, n SI ¼ 3, and n FC ¼ 8 questions, respectively.
The questions, answer format, and results of the answers are presented in Table 2 and Table 3. The questions were produced based on previous studies, e.g., Venkatesh et al. (2003) and Khorasanizadeh et al. (2016).
For the use behavior (Table 2), we asked whether the houses had energy-saving and energy-generating equipment (e.g., high-efficiency air conditioners, floor heating, energy-efficient water heater, and the like; however, we excluded the equipment necessary for ZEHs such as heat-insulation walls, PV panels, and LED lamps). The answers were evaluated as yes or no and transformed to dummy variables taking either 1 or 0. Therefore, the use behavior in this study means the degree of installation of additional energy-saving andcreating equipment that improves the performance of ZEHs.
As regards other constructs (Table 3), answers were evaluated on a five-point Likert scale. As regards the behavioral intention, we asked about the intention to purchase ZEHs at the time the respondents started to consider the purchase. As regards performance expectancy and effort expectancy, we asked about the opinions on the performance, functions, and maintenance of ZEHs and related equipment at the time when the respondents started to become aware of ZEHs. As regards social influence, we asked about the opinions of family members, acquaintances, and neighbors on ZEHs, and the impression of respondents of these houses, before the respondents started considering such purchase. As regards the facilitating conditions, we asked about the environmental surroundings and personal circumstances of the respondent before the purchase.
For the elements of vectors d Econ;i , d Tech;i , d Envi;i , and d Health;i regarding information acquisition, the questions, answer format, and the result of answers are presented in Table 4.

Estimation results
The posterior distribution for the unknown parameters in the Bayesian SEM demonstrated in Subsection 3.1 was obtained via MCMC simulation, given the data introduced in Subsection 3.2. The Bayesian results were sampled for 150,000 iterations following burn-in 5000 iterations for each of three chains by the Just Another Gibbs Sampler (JAGS) program (Plummer and others, 2003) using R2jags (Su and Yajima 2015). Every fifth iteration was saved for each chain. That is, this study drew 87,000 (= (150,000-5000) � 3 � 5) samples for each parameter, based on which estimation results were obtained. Model convergence was assessed via the Gelman-Rubin statistic (Gelman et al., 1992). All parameters achieved statistical values of less than 1.1.
The estimation results for the factor loadings of the measurement models (i.e., Λ B , Λ I , Λ PE , Λ EE , Λ SI , and Λ FC ) are shown in Subsection 4.1, whereas those for coefficients of (1) paths from d i to ω PE;i and (2) paths among latent variables (i.e., C, �, and Γ) are shown in Subsection 4.2. In these subsections, the results are presented in tables that include posterior mean  (sometimes referred to as "Bayesian estimate" (Lee 2007; Song and Lee 2012)), standard deviation (SD), and highest density interval (HDI). The α% HDI summarizes the distribution by specifying an interval spanning most of the distribution (i.e., α% of it) such that every point inside the interval has higher credibility than any point outside it (Kruschke 2014;Meredith and Kruschke 2016). In the tables for the structural model, we also included the probability that a parameter exceeds 0 (i.e., ò 1 0 pðθjDÞdθ, where θ is a parameter and D is data), denoted by Pr þ . Sometimes, it is also desirable to transform Bayesian estimates to a completely standardized (CS) solution such that both observed and latent variables are standardized (Lee 2007). Thus, we also list the CS solution for the factor loadings and coefficients of the paths among latent variables. The estimation results for all parameters are listed in Appendix C. For model checking, we conducted posterior predictive checking (Gelman et al. 2013) (verifying whether the estimated model fitted the obtained data appropriately) and, consequently, the simulated data from the estimated model were close to the obtained data from the survey (the details are presented in Appendix D). We also conducted sensitivity analysis of the prior distribution for parameters of paths between latent variables (details are presented in Appendix E) and, accordingly, the estimation results were almost insensitive to the prior inputs. Table 3. Questions, answer format, and results of the response to constructs other than use behavior.

Measurement model estimation
The estimation results of the measurement models -Λ B , Λ I , Λ PE , Λ EE , Λ SI , and Λ FC -are presented in Table 5. Regarding use behavior, the row described as "z B;i;2 ω B;i ," which presents the result of the coefficient of the path from the latent variable ω B;i to the observed one z B;i;2 (i.e, factor loading λ B;2 ), demonstrated a positive posterior mean (i.e., 0.480) and 95% HDI excluding 0 (i.e., [0.023, 1.001]). 5 Similarly, all posterior means of other factor loadings regarding use behavior (i.e., λ B;3 to λ B;7 ) were positive and all the 95% HDIs excluded 0. Therefore, every observed variable was likely positively correlated with the use behavior. The other latent variables (behavioral intention, performance expectancy, effort expectancy, social influence, and facilitating conditions) indicated a similar tendency (i.e., significant positive correlations with the corresponding observed variables) and showed relatively large CS solutions. The CS solution illustrated the observed variables that were closely linked with the corresponding latent variables; for example, as regards the facilitating conditions, the CS solutions for z FC;i;2 ,z FC;i;4 ,z FC;i;5 ,z FC;i;7 ,and z FC;i;8 (i.e.,parameters λ FC;2 ,λ FC;4 ,λ FC;5 ,λ FC;7 ,and λ FC;8 ) were larger than the other observed variables. Accordingly, it is implied that z FC;i;2 ,z FC;i;4 ,z FC;i;5 ,z FC;i;7 ,and z FC;i;8 are connected more strongly with facilitating conditions.

Structural model estimation
The estimation results of the structural model are discussed in this subsection. First, the results of the coefficients of paths from information acquisition (i. e., d i ) to performance expectancy (i.e., ω PE;i )-Care presented in Table 6. Some variables of d i were found to strongly influence ω PE;i . The coefficient of d Health;Sales;i (i.e., health information obtained from salespersons) (i.e., parameter c 3;13 ) had a larger mean (i.e., 0.327) and the 95% HDI (i.e., [0.087, 0.571]) excluded 0, implying that households that received such information have a higher value of ω PE;i . On the other hand, the coefficients of d Tech;Friends;i and d Envi;Sales;i (i.e., c 3;6 and c 3;9 ) had a larger mean (i.e., 0.449 and 0.245, respectively) and their 90% HDIs (i.e., [0.064, 0.824] and [0.021, 0.467], respectively) excluded 0 despite the 95% Table 4. Questions, answer format, and results of the response to information acquisition. Second, the results of the coefficients of paths between latent variables-� and Γ-are presented in Table 7. Among the paths to behavioral intention, the  path from social influence (i.e., γ 22 ) had the largest posterior mean (i.e., 1.409) and CS solution (i.e., 0.957), 95% HDI (i.e., [0.941, 1.887]) non-overlapping 0, and Pr þ indicating 100.0%. Accordingly, social influence is considered to have a large positive effect on behavioral intention. As regards the facilitating conditions (i.e., γ 23 ), the posterior mean (i.e., 0.239) was positive and the 90% HDI (i.e., [0.008, 0.484]) did not include 0. Therefore, the facilitating conditions are likely to positively influence the behavioral intention. In addition, each path coefficient of the performance expectancy and effort expectancy (i.e., π 23 and γ 21 ) had a small posterior mean and the 90% HDI overlapped 0. This means that the two latent variables are unlikely to affect the behavioral intention. The paths from the behavioral intention and facilitating conditions to the use behavior (i.e., π 12 and γ 13 ) had positive (i.e., 0.254) and negative (i.e., −0.365) posterior means, respectively, and the 95% HDIs ([0.090, 0.429] and [−0.674, −0.088], respectively) excluded 0. Accordingly, the behavioral intention and facilitating conditions are considered to definitively have positive and negative effects, respectively, on the use behavior.

Discussion
We modeled the purchasing process of ZEHs ( Figure 1) by households, including the effects of combining content and channels of information. We conducted statistical analysis targeting Japanese households that purchased custom-built detached ZEHs. As ZEHs are not common in Japan (Sustainable open Innovation Initiative 2018), the process needed to be investigated to consider and develop strategies that would be more effective in promoting the spreading of such houses.
Our estimation results, presented in Table 6-7, are summarized in Figure 2. In this figure, the paths between UTAUT constructs are listed with the CS solutions demonstrated in Table 7; as regards information acquisition, the combinations of information content and channels having significant effects on the performance expectancy are listed with their posterior means in Table 6.
Based on our results, we discuss (1) the effective combining of information content and channels in Subsection 5.1 and (2) factors that effectively promote the behavioral intention and use behavior in Subsection 5.2. Several limitations of this study and future avenues for study are described in Subsection 5.3.

Effective combination of information content and channels
The results of Table 6 showed that the following three combinations of information content and channels likely had positive effects on the performance expectancy, namely, (1) health information obtained from salespersons, (2) environmental information obtained from salespersons, and (3) technical information obtained from friends ( Figure 2).
The associated health and environmental issues are science-based and/or long-term effects, and households could have difficulty in evaluating such issues, whereas salespersons would interpret and convey them objectively and accurately. Accordingly, we consider that health and environmental information on ZEHs transmitted by salespersons had effectively enhanced the performance expectancy. On the other hand, technical information, e.g., performance, mechanism, and maintenance of equipment installed in ZEHs, was likely understood better after the users started using it. Friends can give useful advice (e.g., what equipment is suitable) according to the context, leading to the receivers understanding the technical issues better. Accordingly, technical information from friends is considered to have raised the performance expectancy.
The above findings imply that scientific information, including on health and the environment, should be transmitted through objective channels, whereas technical information should be spread through flexible channels. This implication is consistent with the finding of another study  focusing on such combination in a different context (not adoption of technology, but reconstruction of houses); consequently, our findings are considered more convincing.

Effective factors in promoting intention formation and behavior
As illustrated in Table 7, we found that the intention to purchase a ZEH (i.e., behavioral intention) was promoted primarily by the social influence and facilitating conditions and not by the performance expectancy or effort expectancy ( Figure 2). This result is consistent with the results of previous studies that examined the adoption process of an energy-saving product based on the UTAUT (Khorasanizadeh et al. 2016). We found that the effect of social influence was the most significant. The  Table 5) indicated that the social influence (i.e., ω SI;i ) was linked strongly with "I believed that to live in a ZEH would be socially preferable" (i.e., z SI;i;2 ) and "I believed that to live in a ZEH would be an advanced lifestyle" (i.e., z SI;i;3 ). This implies that the intention was facilitated considerably by the subjective norm regarding ZEHs (e.g., to live in ZEHs is socially preferable) and a positive image (e.g., ZEHs are equipped with advanced technologies). This finding is consistent with a previous study that stresses the importance of subjective norms (Schepers and Wetzels 2007). Our survey included an open-ended question on concerns before the purchase and the deciding factors for the purchase. One response was, "I wanted to live in an environmentally friendly house even if it was expensive." Another reported, " . . . despite the high initial investment cost, I decided to purchase a ZEH as I did not want to live in an out-of-date house." These answers also likely validate our findings on the significant effect of social influence. The estimation results of the measurement models (CS solution in Table 5) indicated that the facilitating conditions were linked strongly with the following: "I was aware of the subsidies for ZEHs provided by the national and/or the local government" (i.e., z FC;i;2 ); "I was aware of the after-sales service or guarantees of a ZEH or its equipment should problems arise" (i.e., z FC;i;4 ); "I was aware of the campaigns launched by home builders, such as gift vouchers and zero-interest rate on loans" (i.e., z FC;i;5 ); and "Where I live, the local municipality and community implemented activities aimed at zero energy consumption" (i.e.,z FC;i;8 ). Accordingly, the support systems provided by various stakeholders, such as national and local governments, home builders, and local communities are also considered triggering factors in the intention to purchase. This inference is also evident from the following answers to the open-ended question, " . . . campaigns and subsidies for ZEHs were useful" and " . . . the subsidies let me decide to purchase a ZEH." Overall, the implication was that the intention to purchase a ZEH was enhanced primarily by the perception of the normative and institutional aspects of ZEHs (i.e., social influence and facilitating conditions), rather than the perception of the technical functions and detailed benefits of ZEHs (i.e., performance expectancy and effort expectancy). One of the reasons why social influence was more influential than performance expectancy or effort expectancy could be the characteristics of the adopters (buyers). As mentioned in Section 1, Japanese society is still in the initial stages of widespread acceptance of such houses. According to Rogers (2003), adopters in the initial stages-early adopters-tend to be respected by their peers, and to continue to earn this esteem, they must make judicious decisions to adopt the innovation; therefore, they are probably quite susceptible to social influence. Another reason for performance expectancy or effort expectancy not being influential was probably that the government and private business operators had just started their promotional activities and therefore had insufficient time to inform prospective buyers of the benefits of ZEHs so that the intention to purchase could be activated.
As presented in Table 7, we also found that the use behavior and behavioral intention were correlated positively ( Figure 2). In other words, the behavior intention likely led households to install equipment to improve the performance of ZEHs. On the other hand, against our expectations, the facilitating conditions were found to have a negative effect on the use behavior ( Figure 2). As mentioned in Subsection 3.2, the use behavior represents the degree of installation of additional energy-saving and energy-creating equipment, which improves the performance of ZEHs. The subsidies from the national and local governments, one of the elements of the facilitating conditions, could have induced behavior other than the behavior to install the additional equipment. This behavior could be to upgrade the necessary equipment (e.g., Figure 2. Summary of estimation results: the number next to each arrow indicates the CS solution (as regards information acquisition, posterior mean is presented); ** indicates that 95% Bayesian credible interval does not include 0; * indicates that 90% credible interval does not include 0.
PV panels, heat insulation walls, and LED lamps) for ZEHs. One household also responded to the open-ended question, "I upgraded the PV panels because of the subsidies." Overall, possibly, the facilitating conditions and use behavior were negatively correlated owing to the following hypothesis, "the performance expectancy promoted the behavior to upgrade the necessary equipment for ZEHs but offset it, terminating the installation of optional equipment." This is one of the possible hypotheses that requires further analysis in the future. Future work should also include: (1) We should improve the observed variables to measure the use behavior since some of the CS solutions of use behavior, such as those for z B;i;2 , z B;i;4 , and z B;i;6 , were small (Table 5) (e.g., it would be important to consider the performance or quality of each equipment, as well as include the necessary equipment). (2) We should explore the multifaceted effects of the facilitating conditions on the purchase behavior of households.

Limitations and future directions
Including topics that we have already described, the current study has some limitations, and future work should investigate various aspects. (1) We focused on the effect of information acquisition exclusively on the performance expectancy. As described in Subsection 2.2, this was because the content of information we targeted was mainly related to the benefits of ZEHs, and the parsimonious model eased the estimation and interpretation. However, for example, information through SNS or other channels might enhance the ZEHs' positive image. To explore further details, the effects on other constructs (e.g., social influence) should be investigated in the future. (2) The data used in this study were collected after purchase and from households that voluntarily answered all questions, which may include cognitive biases to justify their purchase as well as sampling bias; therefore, analysis based on random sampling data collected since before purchase will add further evidence. (3) We limited the samples to the households that purchased ZEHs. Future work should collect samples of households that purchased non-ZEHs and compare the differences between the two. (4) Our study focused only on the initial stages of adoption, whereas the characteristics of adopters generally differ according to the stages (Rogers 2003); therefore, our discussion are considered valid at most for the present and near future. To discuss longer-term policies, followup studies are needed. (5) The present study focused exclusively on custom-built detached houses, but it is also important to target renovated houses (Oki et al. 2019) and zero-energy apartments (ZEH-M) (Sustainable open Innovation Initiative 2018).

Conclusion and policy implications
Targeting Japanese households that purchased custombuilt detached ZEHs, this study aimed to explore how intention and behavior to purchase were facilitated by a range of factors, including both the information content and the channels buyers used. Our conceptual model was constructed based on the UTAUT (Figure 1) and estimated by means of Bayesian SEM, which is suitable to treat small samples and discrete variables, with data collected from approximately 300 households. As already mentioned in Section 5, our main results are shown in Figure 2. The first prominent result is that performance expectancy, i.e., the perception of the usefulness and benefits of ZEHs was enhanced significantly by certain combinations of information content and channels. These are (1) health information obtained from salespersons, (2) environmental information obtained from salespersons, and (3) technical information obtained from friends. The second, and the most prominent, result is that behavioral intention was facilitated more significantly by the social influence and facilitating conditions (i.e., perception of social image of ZEHs and support provided by various stakeholders, e.g., national and local governments, home builders, and local communities) than by the performance expectancy or effort expectancy (i.e., the perception of performance and benefits of ZEHs and installed equipment). The third result is that the behavioral intention positively affected the use behavior, i.e., the degree of installation of additional equipment to improve the performance of ZEHs. In sum, our study revealed a series of possible processes conducted by households to purchase ZEHs: • The above-mentioned three combinations of information content and channels substantially promoted the performance expectancy of ZEHs. • The performance expectancy, however, was uncertain to have a significant effect on the intention to purchase. • The intention was significantly facilitated by the normative and institutional aspects of ZEHs, which, consequently, led to further installation of equipment to improve the quality of ZEHs.
Based on the above results, we propose several policy implications. The above-mentioned three combinations should be exploited for effective information provision strategies. For example, it could be effective for salespersons of home builders to provide consumers with health information (e.g., ZEHs can contribute to improving blood pressure and reducing the risk of heat shock) and environmental information (e.g., ZEHs can contribute to greenhouse gas reduction, as well as disaster risk reduction because of using PV panels and rechargeable batteries). Economic information was not clearly effective, regardless of channels, but it showed positive effects with relatively high probability if provided by salespersons and advertisements (Table 6). Therefore, to enhance the effect of economic information, it would be appropriate to focus on these channels. Through these strategies, the performance expectancy of ZEHs could be enhanced, so that the intention to purchase a ZEH is stimulated, which, probably, would affect the purchase decision. On the other hand, as mentioned previously, the intention was facilitated significantly by the normative and institutional aspects (i.e., the social influence and facilitating conditions) and, therefore, it could be concluded that society as a whole should take a positive position to improve acceptance. For example, the promotion of public discourse saying that ZEHs are up-to-date and socially preferable houses, the branding strategies implemented by the government and private sectors, and the support from various stakeholders (e.g., after-sales service from private business operators and activities aimed at zero energy consumption implemented by local municipalities and communities) were effective and should be continued. Although our analysis has some limitations, the aforementioned introductory findings and implications could contribute to assessing and improving the policies and activities to ensure widespread acceptance of ZEHs.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This study was supported partly by the Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (KAKENHI) (Grant No. 18K13845) and the Maeda Engineering Foundation.

Notes on contributors
Hitomu Kotani is an assistant professor at Kyoto University, Japan. He obtained his Ph.D. in Engineering in 2016 from Kyoto University. His research interests are urban and regional planning as well as social network analysis.
Kazuyoshi Nakano is a researcher at Central Research Institute of Electric Power Industry, Japan. He received his Ph.D. in Informatics in 2011 from Kyoto University, and his specialized fields are consumer behavior and energy demand and supply in Japan.

Availability of data
Data will be made available on request. • Reduces the primary energy consumption by 20% or more, excluding renewable energy, and by 100% or more, including renewable energy, compared with the standard primary energy consumption. Nearly ZEH • Reduces the primary energy consumption by 20% or more, excluding renewable energy, and by 75% or more but less than 100%, including renewable energy, compared with the standard primary energy consumption.
• Applicable in three climate zones, namely, cool, low solar-radiation, and heavy snow.
(Continued) • Reduces the primary energy consumption by 25% or more, excluding renewable energy, and by 100% or more, including renewable energy, compared with the standard primary energy consumption.
• Meets two or more of the following conditions: (1) strengthened envelope quality, (2) installation of HEMS, and (3) installation of charging facilities for EV.
Nearly ZEH+ • Meets the requirements of ZEH+, but the renewable energy offset is only more than 75% of the energy consumption.
ZEH Oriented • Achieves the same level as ZEH in energy efficiency.
However, reducing primary energy consumption, including renewable energy, is not required.
• Applicable in narrow land areas in urban regions. 13 4.4 7 and more 3 1.0 Household annual income Less than 2 million yen 2 0.7 2 million yen or more and less than 3 million yen 8 2.7 3 million yen or more and less than 4 million yen 9 3.0 4 million yen or more and less than 5 million yen 28 9.4 (Continued)