Community solar PV adoption in residential apartment buildings

The potential benefits of adopting PV (photovoltaic) in residential apartments in China, such as reducing peak demand and electricity transmission issues, have been overlooked. Community PV is more applicable for most Chinese cities residents living in apartment buildings. However, existing studies failed to provide comprehensive insight regarding factors influencing community PV adoption and adoption decision strategies. This study utilized a discrete choice experiment performed in Wuhan to examine individuals ’ community PV adoption preferences and analyze the factors influencing their decisions. The study found that respondents who were well-educated, older than 40, lived in residential buildings constructed earlier than 2006, had a living area > 120 m 2 , or used AC for more months in the summer were more likely to adopt PV. Three decision strategies (classes) with different PV product preferences were identified. Peer effect is identified to significantly influence the community PV adoption as almost 65% of the respondents are more likely to adopt PV if there are high neighboring PV adoption or installation agreement rates. Cost was not a determining factor for a significant proportion of respondents, who instead focused on revenue. Subsidy policies also had limited influence, affecting only about 43% of respondents. Policy implications are provided based on the research results.


Introduction
The transition towards decarbonized and sustainable energy systems plays a significant role in mitigating unequivocal global warming and improving human well-being [1].Solar energy has long been recognized as a promising renewable energy source with enormous potential to replace traditional fossil fuels globally [2].The growth of solar PV has been rapid [3], with PV capacity exceeding 50 times the International Energy Agency's 2002 forecast [4].This rapid expansion can contribute to recalibrating climate prospects and more ambitious climate targets [4].Moreover, developing PV also brings co-benefits, including alleviating energy poverty and generating local employment opportunities [5].For instance, the Chinese government initiated the photovoltaic poverty alleviation project (PPAP) in rural areas to address both environmental pollution and regional poverty [6,7].Studies indicate that the project significantly promoted economic growth in targeted areas and delivered multiple social benefits.[8][9][10].
The Chinese government set the goal of achieving carbon neutrality before 2060 [1] to reduce environmental pollution, build a sustainable ecological civilization, and prevent energy crises [11].Developing PV is crucial to realizing this goal, and China is one of the leading countries in the solar industry, with the world's largest installed PV capacity in 2022 [12].However, most PV projects are large-scale concentrated PV plants in northern China, where electricity demand is significantly less than in eastern China [13].Meanwhile, distributed PV systems in high-load cities can lower peak electricity demand and reduce transmission issues [14].Besides, the potential to mitigate peak electricity demand in rural areas is less significant than in urban areas due to lower population density [14].The economic analysis of PV adoption suggests that household PV is financially viable in 86% of Chinese cities, even without subsidies [15].The internal rates of return (IRR) of investing in PV are higher than 8% in about 66% of all cities in China [16].Additionally, in about 75 Chinese cities, solar generation electricity prices can compete with desulfurized coal benchmark electricity prices [17].Furthermore, the National Development and Reform Commission proposed that developing distributed energy sources such as rooftop PV in cities is necessary [18].Yet the distributed PV resources of building rooftops have been overlooked [14].The adoption of household PV in China accounts for only about 1 ‰ [19], despite the advantages, feasibilities, and government encouragement.
The diffusion of PV is contingent upon individual-level residents' decision-making.The adoption of individual-level PV systems is influenced not only by economic and technical feasibility but also by environmental attitudes, personal features, and social influences [1,20,21].Previous studies on PV adoption and diffusion in China have focused on PV power potential, technical and economic feasibility, and environmental impact.These studies suggest that PV is economically feasible [16,17] and can contribute to fulfilling a net-zero electricity system if the installed capacity grows fast in the near future [15,22].
There have been a growing number of studies investigating people's adoption behavior and adoption willingness in China.Most of these studies consider the context of rural China, such as [23,24].Studies conducted in urban contexts often approach this matter using qualitative methods for identifying barriers and motivators, such as [25][26][27], except for the study conducted by Tan et al. [28] that used a discrete choice model to investigate the impact of economic factors, participation rate, and environmental impact on people's willingness to participate in shared solar programs.These studies investigated distributed PV adoption on an individual level, despite the fact that community PV has the potential to increase PV adoption in urban areas and avoid long-distance transmission problems [28].Moreover, in Chinese urban areas, most people live in apartment buildings, and community PV has greater feasibility compared to individual PV systems, which can be hindered by the issue of unclear property rights in apartment buildings [28].These studies do not offer a comprehensive evaluation of the factors that might affect future community PV adoption and identify adoption decision strategies.
Therefore, this study aims to identify and evaluate the factors affecting community PV adoption.Potentially influential factors were derived from the literature, and their impacts on community PV adoption were evaluated using a stated choice experiment.PV product preferences and decision-making strategies were analyzed and compared to provide insight into people's choices.Finally, policy measures to encourage adoption were presented after the analysis.This paper is organized as follows: the following section discusses existing literature on PV adoption and community PV projects.Section 3 presents the design of the discrete choice experiment and the analysis method.Section 4 showcases the results and discusses their implications, and the paper concludes with a summary of the findings and their implications in section 5.

Background
There have been successful community photovoltaic (PV) projects implemented worldwide.In the UK, community PV projects experienced rapid growth from 2007 to 2017 [29].Researchers have compiled a comprehensive database encompassing 61 community PV projects in the United States [30].Examples of community PV projects around the world can be found in Mah [31].These case projects offer valuable insights into the institutional frameworks [32], planning and assessment methods [33], and profit-sharing tools [34] employed in community PV initiatives.Community PV originated from community energy, and the community can be based on both geographic proximity and social relations [28].Community PV is gaining more attention in the cities due to cost reduction, flexibility, the potential for achieving ambitious climate and energy targets, and other social benefits [31].
These international projects contribute to a collective understanding of the experiences surrounding PV adoption.Financial barriers such as high costs and extended payback periods are commonly identified as obstacles to PV adoption [35,36].Conversely, environmental motivations, economic incentives, and peer effects motivate PV adoption [37][38][39][40].Individuals who adopt PV systems often demonstrate a sense of responsibility towards energy and the environment [1,41].Peer effect, which refers to the influence of observed behaviors within a peer group on an individual's behavior, is a widely recognized driver of PV adoption [42].
Policy interventions often play a crucial role in low-carbon energy adoption and diffusion [43].Common incentives for PV adoption include Feed-in-Tariffs (FiTs), Investment Tax Credits (ITC), and investments in PV research and development departments [44].Subsidies dominate the landscape among financial incentives globally [43].This trend is also observed in China, where subsidies are the most prevalent incentive policy for PV [45,46].However, the Chinese central government discontinued subsidies in 2013 and FiTs in 2021 [46,47].
The cost of PV systems is identified as a primary barrier to adoption in rural China [48], aligning with international experiences, and products with relatively low prices are more acceptable [49].In rural China, opinion leaders [24,50] and peer effects [51] have a significant impact on individuals' adoption behavior.Active communication proves to be more effective than passive observation of PV units in encouraging PV adoption intentions among rural Chinese residents [23].Several studies emphasized the importance of perceived information, including perceived cost-benefit [51], perceived risks [48], and perceived social pressure [23], in PV adoption in rural China.Furthermore, the impact of subsidy interventions is found to be considerably weaker than social influences due to the reduction of solar subsidies [23].Wang et al. [19] investigated how household PV subsidy policy influences consumers' purchase intentions in the context of China's urban-rural divide.The results show that rural residents are more attracted to subsidy policies, and their urban counterparts have lower adoption intention due to installation condition limitations, such as unclear roof property rights [14].This study aims to conduct a comprehensive investigation into the factors influencing the adoption of community PV in urban areas.

Methodology
The discrete choice experiment (DCE) was employed to investigate community PV adoption preferences and decision-making strategies in Wuhan.Wuhan was selected as the survey location for two primary reasons: its sizable population and its climate characterized by hot summers and cold winters.The hot summers and cold winters climate zone, primarily located in the central and southern parts of China, covers approximately 25% of the country's land area, including both urban and rural regions, and 55% of the nation's population resides in this climatic region [52] with almost 65% of them living in cities [53].In 2022, the average summer and winter temperatures in Wuhan were approximately 30 • C and 6 • C, respectively, while the national averages stood at 22.3 • C and − 3.2 • C [54,55].The National Energy Administration has categorized China into four levels of solar irradiation, with the hot summer and cold winter region falling into the third level [56].This classification suggests that cities in this region share relatively similar PV power generation potential [57].Given these climatic characteristics, Wuhan emerges as a suitable candidate to represent a large number of Chinese cities situated in the hot summers and cold winters climate zone when H. Du et al. investigating community PV adoption.
The DCE methodology is rooted in consumer choice theory [58] and random utility theory [59], assuming that participants of the experiment choose the option that yields the highest utility among the given alternatives.Several factors informed the selection of DCE for this study.Firstly, the scarcity of community residential PV projects in Chinese urban areas necessitated the adoption of a hypothetical adoption experiment.Secondly, discrete choice experiments enable the examination of factors that are often not easily observed [60].
Community solar energy is a highly dynamic concept [31], as 'community' is often based on geographical proximity but can also be communities of interest and social relations [61,62].In this study, community solar PV adoption is defined by physical proximity, where a community comprises residents living in one apartment building or multiple buildings within the same neighborhood.

Survey design
The design of the attributes in the stated choice experiment is based on the theory of planned behavior [63] and innovation diffusion theory [64].The theory of planned behavior links beliefs and behaviors through attitude, subjective norms, and perceived behavioural control.Innovation diffusion theory identified four main elements in innovation diffusion processes: innovation, communication channel, time, and the social system.These two theories were studied to gain a general framework of the factor types that should be included in the experiment: technology features, attitudes towards sustainability, and social influences.More specifically, the technology feature is based on the innovation element in innovation diffusion theory; the attitude factor originates from the attitude element in the theory of planned behavior, while the social influence factor is designed based on the communication channel element and the subjective norm from the two theories, respectively.
Furthermore, peer effects and social norms were included in the DCM to compare their impact on community PV adoption.Then, other studies were considered for the detailed design of the survey.The structure and questions in the survey are described as follows.
The survey contains two parts.The first part intends to capture the features of respondents, their energy consumption behavior, and their housing situation.Respondents' features (socio-demographics) include gender, age, income, and education.Housing situation refers to the age of the building, the annual energy bill, and the house size.There are two questions about environmental attitude and four questions about energy consumption behavior based on previous studies [65][66][67][68].Questions about environmental attitudes use a five-point Likert scale and ask respondents to evaluate the severe level of global warming and natural resource depletion.The energy usage habits include the following questions: 1) Do you consciously buy energy-saving light bulbs or appliances? 2) Do you consciously reduce your gas use? 3) Do you deliberately use less electricity?4) Do you choose public transportation, cycling or walking, rather than private cars or taxis?These questions are labeled ESB1, ESB2, ESB3, and ESB4, respectively.Explanations of these questions were also given in the questionnaires, as shown in Appendix A1.The answers to these questions include: no; yes, for the environment; and yes, for saving money.
The second part aims to investigate the influences of PV adoption, assuming the adoptions are community decisions.The community PV adoption model was chosen for several reasons.The most important reason is that it is the more applicable method for most Chinese city residents who want to adopt PV panels.Besides, community PV has some advantages.Wüstenhagen et al. [69] found that community adoption increased PV adoption rate, local environmental awareness, and social acceptance.Studies also suggest community PV contributes to a more just energy transition than household PV [70,71].
Respondents of the experiments choose the products with the highest utility value among hypothetical products with various levels of attributes.The design of the DCE followed the experimental design process proposed by Hensher et al. [72].A short description is given at the beginning of the DCE to explain the setting of the choice experiment.Besides the description, an explanation table was given, considering that most people may not be familiar with the concepts or the meaning of some attributes and levels.Both the description and explanation table are shown in Appendix A2.A DCE survey question example is also shown in Appendix A2.An overview of the attributes and levels of the experiment is shown in Table 1.It is assumed that each community for the PV adoption includes 50 households that will share the cost and revenue equally.The cost and revenue shown in Table 1 are after equal distribution.
There are two types of attributes: product features (rows 2 to 7 in Table 1) and external factors attached to the products, such as subsidy or acceptance level (rows 7 to 12 in Table 1).
The installation package (including cost), the payback period, and the cost of home storage attributes are calculated according to market information [74].The cost and benefit are in Chinese Yuan (CNY).The financial feasibility of PV panels is built upon the extended performance warranty, which also increases cost [75].Hence, a PV performance warranty can influence people's adoption behavior and adoption preferences.Subsidies influence the adoption of PV due to the high cost of PV systems [76], as subsidy incentives help trigger the adoption behavior by reaching the threshold [77].The levels for the subsidy attribute are 5% and 10%.The levels were chosen considering the decreasing financial incentives for PV projects in China, and these incentives are only limited to several cities.In this study, we want to investigate whether the subsidy still influences people's decisions in a community PV adoption.The energy service company attribute refers to the third-party ownership business model for PV.This business model can be referred to by different names, such as dealer financing [78], leasing [79] or third-party ownership [80].In this business model, the site owner lends the place suitable for PV installation (such as the building rooftop) to energy service companies that will install and maintain the PV [81].The involvement of energy service companies (the third party) can resolve the challenges of operating and maintaining the PV system [82] and high upfront costs [83].Moreover, Chen and Chen [15] suggest that the rental model can address the problem of rooftop property for urban high-rise buildings.Three levels for this attribute include no involvement of energy service companies and obtaining all the benefits, benefits decrease of CNY20 per household per year, and benefits decrease of CNY40 per household per year.
Adopting a product is sometimes influenced by adopters' peers or the change in society [84].The definition of peer group varies in studies; it can be based on spatial or geographic proximity or social network connections [42].Mundaca and Samahita [38] suggest that peer effect is a major factor that drives the adoption likelihood.However, Alipour et al. [35] found that the influence of the peer effect is limited and cannot be categorized as a driver.The difference can be because the peer effect does not trigger interest but decreases uncertainties and shortens the decision period [42].The peer effects include interpersonal communication and persuasion and normative social influences [42].The underlying mechanism of recommendation and negative comments is word-of-mouth communication.This type of peer effect encourages potential adopters in two ways.It raises awareness of the new technology and reduces uncertainty [42].Moreover, potential adopters can get information such as reliable contractors, available incentives, and the application process [85].The negative comments are included as Wang et al. [86] suggest that peers' rumors and deliberate discrediting of residential PV damage PV diffusion.According to innovation diffusion theory, the visibility of the innovation affects its diffusion significantly [64].The visibility of the innovation is related to observational learning, which describes people's imitation behavior after observing [42].Studies suggest that the visibility of PV panels can influence the diffusion of rooftop PV and lead to spatial clustering [87,88].Researchers developed and tested the theory of normative social behavior to demonstrate that perceived norms impact behaviors from three paths: social approval, outcome expectancies, and group identity [89].
This study investigated both communication peer effects and normative social influences.The first (positive and negative comments) is interpersonal communication, while the last two (the residents' agreement rate and installation rate in the neighboring community) are social norm influences for investigating individuals' normative behavior in society.The residents' agreement rate of the PV installation is included, considering the most dwelling type in Chinese urban areas.Most urban residents in China live in apartments, and the adoption of PV will be a group decision of the community as the roof and/or the façade of the building will be changed, which are public areas owned by the community or residents' group in the building.Hence, the attribute of residents' agreement rate of the installation can determine the actual PV adoption in practice.
The orthogonal fractional factorial design was used in the study, and an uncorrelated design of all attributes and levels with 36 profiles (alternatives) was constructed from a standard orthogonal array.Then, the 36 alternatives were randomly selected into 18 groups.This selection process was repeated once to increase the randomness and reliability of the research results.Every 18 groups of alternatives were assigned to 6 blocks.Each respondent needs to finish the task of one block; in other words, every respondent needs to make three choices.
The survey was conducted online and offline in Wuhan from 11th September to 19th November 2021.The questionnaire's readability and understandability were tested before the survey.Data collected from the test survey was not included in the results.However, the answering time during the test was recorded and later used as a selection criterion.The questionnaire was improved according to suggestions and comments from the test panel.Several measures were adopted to account for common concerns for DCEs [90].First, we conducted the survey both online and offline for better coverage and representation of the population.We also presented and explained the imaginary circumstance and the decision situation they were facing before the questions.

Mixed logit model and Latent class model with attribute nonattendance
The discrete choice model is based on the random utility theory [91].The utility functions and the probability function are shown in eqs.( 1), (2), and (3).Eqs. ( 1) and ( 2) describe the utility perceived by the participant n choosing alternative j in the choice situation s.U nsj is the utilities perceived by individuals, while V nsj and ε nj are the observed and unobserved components, respectively.β is the parameter vector.
The mixed logit (ML) model was selected for its flexibility in comparison to other models, including allowing the error term to combine different statistical distributions and allowing random taste variation parameters [92,93].The latent Class model (LC) model, which does not require assumptions about the parameter distributions, was also employed.These two models were chosen to serve different research purposes, as outlined below.The ML model with interactions is used to determine whether people's social demographics and housing factors will influence the adoption preference for PV panels.The ML model is also called the random parameter model as it assumes that some of the parameters are random and follow certain probability distributions over the sampled population.In the ML models, we used 100 Halton draws, as Bhat [94] and Melbourne et al. [95] suggested.
The LC model is introduced in the study to approach the heterogeneity preference among individuals [96].Individuals are categorized into classes based on preferences, and the probability of each class is yielded in the LC model [97].The preference difference can be caused by both observable attributes and factors that the analysts do not observe.LC models assume the individuals belong to Q classes; the parameter vectors β for different classes are different, while they remain consistent within the classes.
Studies about processing heuristics originated to explore whether individuals use simple strategies to make choices [98].Swait and Adamowicz [99] found that individuals use different complexity levels of decision strategies in choice experiments.Moreover, these decision strategies vary with respect to whether they are attribute-based or alternative-based [100].Hensher et al. [72] summarized eleven different decision strategies.In this study, the process heuristic of attribute non-attendance (ANA) is introduced in the model.ANA describes the alternative processing rule of ignoring a subset of attributes in the DCE [101].One of the advantages of using ANA is that researchers do not need to know the incidence of attribute non-attendance from supplementary questions [101].However, some studies argue that the phenomenon of ANA can be explained as taste heterogeneity in some cases [102,103].LC model was used in this study to investigate people's different preferences and decision strategies when making community PV adoption decisions.All presented discrete choice modelling analyses were conducted in Nlogit 6. Analysis of variance (ANOVA) is used as an extra analysis of LC model results to identify class features, which can be the causes of the preference difference.

H. Du et al.
The orthogonal polynomial coding scheme and effect coding are shown in Table 2. Orthogonal polynomial coding was used for PV attributes.In this study, there are only three and four-level to describe a PV attribute.The first variable (X 1 ) represents a linear transformation of the orthogonal variable, while the second (X 2 ) is a quadratic transformation, and the third (X 3 ) is a cubic transformation.The variables' significances suggest the attribute's significance and relationships among the levels of the attribute.For example, it suggests a linear relationship between the levels when only the first variable shows significance in a multi-level attribute scenario.The orthogonal polynomial coding scheme is generally recommended as it does not introduce correlations between the constructed variables (X 1 , X 2 ), while dummy and effect coding schemes do [104].The effect coding was used for representing individuals' features in the ML model, considering the interpretability of the interaction terms.
The survey was conducted both online and offline.The offline was conducted in public places, including major shopping malls, parks, and subway stations on weekends, while the online survey was posted in virtual communities and sent to community WeChat groups in Wuhan.In total, 494 surveys were received and then were screened and selected based on the following rules: 1) the survey should be completely finished; 2) the finish time of the survey should be longer than the minimum time needed in the test survey; 3) the survey was seriously finished (do not have the same answer across various questions or answers are not recognizable).There are 356 surveys left after the selection.According to the rule of thumb proposed by Johnson and Orme [105], the sample size required for the designed choice experiment in this study is 336.In addition, G*Power [106] was used to calculate the sample size with standard Cohen guidelines (d = 0.2, power = 0.8, p = .05).The result of the estimated required sample size is 191.The estimated statistical power of 336 samples is 0.88.Hence the sample size requirement was met.The descriptive results, including sociodemographics, housing situations, and energy-related habits, are shown in Table 3.

PV adoption preference
The ML model with interaction terms was used to determine if sociodemographics, dwelling features, and energy-related habits influence people's preference to adopt PV.Factors that might influence adoption behavior were categorized into the aforementioned three groups (sociodemographics, dwelling features, and energy-related habits).Hence, there are three ML models with interactions for the three categories of influences, as shown in Table 4, and each interaction model includes one group of factors.These factors were introduced in the model with the interaction with the variable for the adoption of one of the PV alternatives or preferred status quo.The McFadden Pseudo R 2 and AIC of the base ML model (without interactions) are 0. 2384 and 1881.1.The coefficients in Table 4 were multiplied by the variable (− 1) to represent respondents choosing one of the PV packages for interpretation.The results in Table 4 only show the coefficients for the selected variables; the full results of the three mixed logit models are shown in Appendix B.
The interaction models show that respondents with an education degree from high school or vocational education or lower are less likely to adopt PV.People younger than 32, living in a group of two, or living with more than four people also tend not to adopt PV.People who live in dwellings built before 2006 are more willing to adopt PV than their counterparts who live in relatively new buildings (built after 2014).The dwelling size also influences PV adoption preference, as people in a dwelling with a size larger than 120 m 2 are more likely to adopt PV, while people have reversed tendencies when their dwelling sizes are between 90 and 120 m 2 .People who reported deliberately reducing gas usage are more likely to adopt PV, while people who reported using more public transport are less likely to adopt PV.Besides, people who had no access to residential energy-related public service advertisements or only got the information from the internet are less likely to adopt PV.Respondents who use cooling for a long time are also more likely to adopt PV.

Latent class model
The results of the latent class model with ANA are shown in Table 5. Models with different numbers of classes are tested, and the shown results in Table 5 have the best model fit (McFadden Pseudo R 2 and AIC of the models were compared).The results show that there are three decision strategies (classes).Class 2 and class 3 have different nonattendance attributes: class 2 ignores the social-influence-related attributes except recommendation, while class 3 ignores part of package

Table 2
Orthogonal polynomial coding and effect coding schemes.

Level
Note: Two coding strategies are used for the analysis.For an attribute with k levels, k-1 variable(s) will be created in both coding schemes.For example, to use the orthogonal polynomial coding strategy for a three-level attribute, there will be two variables (X 1 , X 2 ) for representing the attribute.The first level will be coded as: X 1 = − 1, X 2 = 1, the second level: X 1 = 0, X 2 = − 2, and the third level: X 1 = 1, X 2 = 1.More details about the coding strategies can be found in [72].

features.
The results in Table 5 show that most respondents (42.7%) belong to class 3, and the least belong to class 1 (22.0%).The constants in class 2 and 3 is positive and significant, suggesting that respondent in these two classes (about 80% of the total) tend to choose PV compared with the stats-quo.Almost 80% of the respondents have heuristics and tend to ignore some product features when choosing the PV, and only one-fifth of the respondents have the decision strategy of all attribute attendance.About one-third of the respondents belong to class 2 and ignore most social influence in decision-making.The attribute home storage does not show significance in any of the classes, which suggests that this attribute does not influence people's PV choice decisions.
Compared with classes 2 and 3, respondents in class 1 are thorough decision-makers as they consider all attributes.Six attributes show the significance in class 1: PV package, warranty, energy service, resident's agreement rate, negative comments, and PV installation rate in the neighborhood.Respondents dislike a middle-level package and costs.It can be interpreted as that people either like a low-cost or a large package.The influences of warranty, agreement rate, and PV installation rate are linear: respondents prefer products with a longer warranty.In the meantime, they are more likely to adopt PV when the agreement and installation rates are high.Respondents in class 1 also prefer no energy service company involved in the adoption; however, when there is energy service provided, respondents prefer a higher cost level.The greatest gap was found in the negative comment attribute, which suggests that this attribute has the greatest influence on the adoption decision.This is also the only class considering negative comments in the decision process.The results show that all negative comments will decrease the likelihood of adopting PV and that online information has a greater impact.
The attributes not considered by respondents in class 2 are social influences, including the installation agreement rate, negative comments, and PV installation rate in the neighborhood.The decisions of PV adoption for respondents in class 2 are significantly influenced by attributes of the package, warranty, energy service, and product recommendation.First, respondents in this class prefer the middle-level package, opposite to respondents in class 1.Compared with class 1, class 2 also shows the disfavor of short warranty years; however, there is no linear correlation between the utility and the level.Class 2 prefers to rent out the rooftop and building façade and let the energy service company take care of the PV.They also prefer a lower cost for the service.The results of the recommendation attribute suggest that positive product comments from communities, coworkers, or online can increase the likelihood of PV adoption.
Class 3 is assigned as ignoring product-related attributes, including the PV system's package and price, warranty, and battery storage.Class 3 can be identified as the revenue-focused class, as respondents in this class prefer higher revenues and shorter payback periods.Moreover, they are unwilling to give up the revenue for convenience when including energy service companies.Class 3 is the only class that shows significance in subsidy and prefers the highest subsidy level.Besides, three more attributes show linear correlations: agreement rate and neighborhood PV installation rate.Respondents are more willing to adopt PV when the agreement and installation rates are high.Unlike class 2, class 3 prefers recommendations from family and friends.In contrast, recommendations from an online source may decrease the possibility of PV adoption.The ANOVA test shows that the environmental attitude and transportation choices significantly differ among the three classes.Classes 2 and 3 have more concern over resource depletion, while class 1 has the least.The results also show that class 3 is more likely to use public transport to save money than the other two groups.

Discussion and implications
This study identified three different decision strategies, including decision heuristics.Decision heuristics refer to individuals' behavior of consciously or unconsciously disregarding information and making decisions based on a limited amount of information [103].The findings indicate that most individuals exhibit heuristic decision-making processes in the context of community photovoltaic (PV) adoption, as many individuals demonstrate the attribute non-attendance heuristic.
Most previous studies investigating PV adoption found that cost is one of the major barriers, with income playing a substantial role [107,108], aligning with policymakers' use of financial incentives to promote PV adoption.However, in the community PV scenarios, the study results do not suggest that cost is an adoption barrier, although it does influence the decisions of some individuals.Additionally, the results indicate that household income does not impact adoption intentions, and non-economic factors play a more significant role, which is consistent with the findings of Mundaca et al. [109].Furthermore, the results demonstrate that higher subsidy levels in the experimental setting have limited influence on the decision-making process for adoption.Even among respondents in Class 3, who prioritize financial aspects, their focus lies more on revenue rather than cost.It is believed that the decrease in costs (as shown in Table 1, section 3.1) is the primary reason for the negligible impact of income on adoption decisions.These findings also suggest that community PV initiatives contribute to energy justice during the energy transition.
This study also identifies the significant role of peer effect in community PV adoption, which is the same in individual PV adoption as also suggested by previous studies [20,39,42,110].Peer effects based on both spatial distance and social network were investigated in this study.This study confirms the influence of peer effects and quantifies the impact of peer effects on people's adoption decisions.The results confirm that the peer effects exist in a large group of respondents (about 65%).Results suggest the positive amplifying effects as people are more willing to adopt the PV product accepted or adopted by the majority.The social network peer effects included in this study are recommendation information (positive information) and negative information.Results show that about 22% of the respondents can be significantly influenced by negative comments, while the rest are impacted by recommendations (positive comments).The results show that positive or negative comments have a greater influence than the neighboring PV installation rate on PV adoption decisions.This comparison result suggests that active peer effects (communications) are more important than passive effects (seeing PV installation), which was also found by Graziano et al. [40] and Palm [111].
The context of this study pertains to community adoption in urban areas.The study was conducted in Wuhan, which is located in the hot summer and cold winter region of China and is characterized by a substantial number of residential apartment buildings.Hence, the implications of this study are primarily applicable to urban areas where a substantial portion of the population resides in apartment buildings and experiences similar climatic conditions to Wuhan.Strategies aimed at promoting PV adoption and facilitating the transition to sustainable energy sources should encompass the following three aspects.
First, the promotion can start with people with the following features: 1) well-educated, 2) older than 40, 3) living in residential buildings constructed earlier than 2006, 4) living area is larger than 120 m 2 , 5) uses AC for more months in summer.The research results show that these people are more willing to adopt PV than their counterparts.Their adoption behavior or agreement of PV adoption can increase others' acceptance of the PV product and accelerate the PV diffusion, considering the amplifying feature of peer effects.
Three measures regarding information are necessary.Firstly, it is important to screen and clarify misguiding information about PV to control its negative impact on PV adoption and diffusion.The second measure is providing general information about renewable energy through TV, flyers, and billboards.This measure increases people's awareness of the environment and new technologies, which may increase the willingness to adopt PV.According to the research results, this H.Du et al. measure may also encourage people to actively communicate about PV adoption or adoption plans, which significantly influences potential adopters.However, this information should not be provided as general online information as it significantly decreases the likelihood of adopting PV.Lastly, a clear and thorough procedure and overview regarding standards for residential PV installation should be provided to guide potential adopters.Installations that violated regional planning or regulations happened and were reported by the news due to the absence of clear regulations and people's lack of knowledge regarding installation standards.Studies suggest that a lack of knowledge and information on PV is one of the major PV diffusion barriers in China [26,112].A thorough guide can play a crucial role in expediting the knowledge and information transaction.Preferably, the guide should also include financial incentives and information about reliable PV contractors or reliable third parties to reduce uncertainties.
Last, the policies should focus on promoting larger PV packages with batteries for faster diffusion.The results show that compared with upfront costs, respondents value other factors more, and there is no clear resistant tendency to larger PV packages.Chinese government promotes the self-consumption of distributed PV [46].Study results show that respondents do not mind paying extra for PV batteries.It suggests that PV contractors can consider providing more PV packages with batteries.There have been some practices of third-party financing PV adoption models and cases in China [83], but it is not common except in Zhejiang Province [15].It is also suggested to encourage state grid corporations to start providing energy services or start companies for this service.

Conclusion
This study utilizes a discrete choice experiment to identify the impact factors and decision strategies influencing community photovoltaic (PV) adoption in hot summer and cold winter urban areas.Community PV adoption is investigated due to its advantages and feasibility compared with individual PV in urban areas.The contribution of this study includes providing a comprehensive evaluation of influencing factors, identifying different types of decision strategies, and comparing the impact of various social influences on community PV adoption.Overall, the findings indicate a higher willingness among individuals to adopt community PV compared to the current status quo.Choices regarding PV adoption vary based on product features, social influences, and subsidies.Notably, different decision strategies for community PV adoption were analyzed.The majority of respondents expressed a willingness to adopt PV installations, according to the analysis results.Moreover, a strong peer effect was identified for approximately 65% of the respondents in relation to community PV adoption.About half of the respondents preferred to have energy services for convenience, while the remaining respondents were unwilling to forgo revenue.It is worth noting that there is an increasing number of studies highlighting the effects of non-economic factors on PV adoption behavior, which aligns with the findings of this study as discussed above.In practical terms, it should be acknowledged that PV installations in apartment buildings may sometimes be rejected due to a lack of information, resulting in resource wastage.It is important to adhere to installation rules, such as height limitations for structures, even though the installation does not need permission from the municipality.However, residents' permission is needed if community-owned areas (including roof and façade) are used for the installation.
The study results can be applied to cities with similar markets and interventions as Wuhan, where the community PV diffusion is still in its early stages, and most of the residential buildings are apartments.The results of this study should be used cautiously, considering possible common biases in the stated choice experiment, including unknown selection and hypothetical biases.The standard mixed logit model is used in this study.There are alternative models, such as mixed logit with unobserved inter-and intra-individual heterogeneity and mixed binary logit model with random effects, that can potentially improve the predictive accuracy of the results [104][105][106].Nevertheless, these alternative models require significantly longer computation times.Furthermore, while there are PV adoption cases in Wuhan, it is uncommon for a group of residents in a building or community to adopt PV collectively.Hence, it is believed that the results will be more reliable if people have more experience or knowledge about PV.Future studies can study profit distribution methods as this can be a major barrier to community PV adoption.Various new business models are developed to encourage PV adoption; the inclusion of new business models is another opportunity as they are not included in this study and may have significant influences.Additionally, future studies can further investigate how to utilize the peer effect to accelerate the diffusion of PV in urban areas, considering its significant impact.

Declaration of Competing Interest
We have no conflicts of interest to disclose.

A.2. Discrete choice experiment example
There are three PV panel choice questions in the following.It is assumed that the community or the building you live in is considering whether to install PV panels and install which category of PV.It is also assumed that the installation costs and benefits (the data shown in the questions) of solar panels are equally shared by 50 households.Each question has two different PV panel choices, and each product have different features.You can choose one of the products or choose 'do not like either of the alternatives' if you are unsatisfied with the options or unwilling to adopt PV panels.Product attributes that you may not be familiar with are shown in the table below.

Attribute Explanation
Home storage percentage The price of buying electricity is higher than selling electricity.Investing in batteries will add extra initial costs and increase benefits, which are labeled as 'battery' in the questions.

Subsidy
For example, if the price of the product is CNY100, then a 5% subsidy will decrease the price to CNY95.Energy service company and annual cost (CNY) It can be viewed as communities renting out the roof and/or the façade to the company.The company will pay and take charge of the PV systems' purchase, installation, and maintenance and there will be service fees, and labeled as ESC (energy service cost) in the questions.
Which of the following is a better choice for you?

Table 1
Descriptions and levels of the chosen attributes.
Note: CNY is Chinese Yuan, and the average exchange rate for CNY and US dollar in 2022 was $1 = CNY 6.730[73].H.Du et al.

Table 3
Descriptive results.

Table 4
Mixed logit model with interactions.
Note: *** indicates significance at the 1% level.**indicatessignificance at the 5% level.*indicatessignificance at the 10% level.1: the levels for the education variable are high school or vocational education degrees or lower (level 1), professional education degrees (level 2), and graduates (level 3).

Table 5
Latent class model with ANA.
Note: *** indicates significance at the 1% level.**indicatessignificance at the 5% level.*indicatessignificance at the 10% level.The blank cells are attributes that are assigned to be attribute non-attendance (ANA).H.Du et al.

Mixed Logit model results Table B1
Results of Mixed Logit model with interactions with socio-demographics.

Table B1 (
continued ) Note: *** indicates significance at the 1% level.** indicates significance at the 5% level.* indicates significance at the 10% level.Results of Mixed Logit model with interactions with dwelling features.Note: *** indicates significance at the 1% level.** indicates significance at the 5% level.* indicates significance at the 10% level.Results of Mixed Logit model with interactions with energy-related habits.