Understanding housing consumption behavior for sustainable urban development: the influence of perceived value and perceived risk on purchase intention in China's higher-tier urban housing markets

1 Introduction

The pursuit of sustainable urban development has placed housing policy at the forefront of urban governance agendas worldwide. Housing constitutes a fundamental component of urban sustainability (UN-Habitat, 2016), directly influencing social equity, economic vitality, and environmental quality in cities (SDG 11: Sustainable Cities and Communities). In China, the world's largest urbanizing society, the real estate industry has undergone remarkable growth, with residential property investment maintaining an average annual growth rate of approximately 25% from 1998 to 2013 (National Bureau of Statistics of China, 2022). However, this rapid expansion has been accompanied by escalating housing affordability challenges, widening income-based consumption disparities, and concerns over the social sustainability of urban development. Following the Central Economic Work Conference in December 2016, which articulated the principle that housing is for living in rather than for speculation, the market entered a sustained structural adjustment oriented toward social equity and sustainable consumption. As regulatory policies tightened, financial deleveraging deepened, and the COVID-19 pandemic introduced additional uncertainty, the market entered a deep adjustment phase in 2021–2022, with sales volume and prices declining in most cities, and underwent a marked shift toward a buyer's market in most cities, with consumers becoming increasingly cautious and rational in their purchasing decisions.

Against this backdrop, understanding the behavioral foundations of housing consumption has emerged as a critical imperative for sustainable urban development. While the sustainable cities literature has extensively examined green building technologies, urban planning instruments, and housing policy frameworks (Durdyev et al., 2022; Jung et al., 2022), the consumer-side psychological factors that shape housing purchase decisions have received growing but still fragmented attention, and the resulting evidence remains dispersed across diverse national and market contexts. From a behavioral science perspective, housing purchase decisions represent high-involvement, high-risk, low-frequency complex decision-making processes in which consumers must navigate information asymmetry and high uncertainty (Mitchell, 1999; Phamthi et al., 2024). Two key psychological variables—perceived value (the overall assessment of benefits received relative to sacrifices made) and perceived risk (the subjective expectation of potential loss)—have been identified as central correlates of purchase intention in high-stakes consumption contexts (Zeithaml, 1988; Bauer, 1960). Understanding how these psychological variables relate to purchase intention is essential for designing housing policies and urban development strategies that promote sustainable consumption patterns across diverse socioeconomic groups.

Existing research has extensively examined housing purchase intention, including studies grounded in perceived value, perceived risk, and the theory of planned behavior. Nevertheless, several gaps remain in the sustainable cities literature. First, while the green housing literature has examined environmental attitudes and sustainable building preferences (Masukujjaman et al., 2022; Zhao and Cao, 2025), limited research has examined the dimensional structure of perceived value and perceived risk jointly, and their associations with purchase intention, in the context of China's post-2021 market adjustment. Second, many housing studies examine perceived value at an aggregate level, and the differential associations of its dimensions with purchase intention remain underexplored in the housing context, which limits the practical utility of findings for targeted urban housing policy design. Third, few studies have integrated perceived value and perceived risk into a unified analytical framework, despite the theoretical recognition that these constructs operate jointly in shaping consumption behavior in high-stakes urban housing decisions (Syamsuar and Witarsyah, 2025). Fourth, and most critically for sustainable urban development, limited research has examined how income-based consumer heterogeneity relates to housing consumption patterns, a significant gap given the centrality of housing affordability and social inclusion to urban sustainability agendas (Jin et al., 2026). To address this gap, we hypothesize and formally test income-group moderation of the associations between the perceived value dimensions and purchase intention.

The variables examined in this study connect to sustainable urban development through three channels. First, housing market sustainability depends on the alignment between supply and genuine residential demand. When developers misread what consumers actually value, the consequences include overbuilding, elevated vacancy, and inefficient use of land, capital, and construction materials, outcomes that directly undermine the resource efficiency objectives of sustainable urbanization (Glaeser et al., 2017). Demand-side evidence on how consumers weigh quality, price, emotional, and social considerations is therefore a necessary input for calibrating housing supply to real needs. Second, the dimensions measured in this study map onto specific targets of SDG 11 rather than onto marketing concerns alone. Price value speaks to housing affordability, the core of Target 11.1, because affordability is fundamentally a judgment about whether housing expenditure leaves households sufficient resources for other necessities (Stone, 2006). Quality value concerns the adequacy, safety, and durability of the housing stock, and durable, well-built housing extends building lifespans and reduces the lifecycle consumption of materials and energy. Social value captures the community attachment and social recognition that underpin inclusive neighborhood development, and emotional value reflects the livability and quality of life that sustainable cities are expected to deliver. These mappings are conceptual rather than empirical equivalences: perceived price value is not a direct measure of housing affordability, and perceived social value is not a direct measure of social inclusion. Third, the risk dimensions connect individual psychology to governance: environmental risk, a dimension that in this study measures perceived external risks at the policy and macroeconomic level rather than ecological or physical environmental risks, links purchase decisions to the predictability of urban governance, a recognized precondition for sustainable urban investment. Framed in this way, the psychological variables analyzed below provide an indirect behavioral perspective on the affordability, adequacy, inclusion, and governance objectives that define sustainable urban development.

This study addresses these gaps by developing and testing a theoretically grounded model that examines the associations of perceived value (decomposed into four dimensions: price value, emotional value, social value, and quality value) with housing purchase intention, with perceived risk (decomposed into three dimensions: transaction subject risk, transaction object risk, and environmental risk) as a mediator and income level as a moderator whose role is formally hypothesized and tested alongside the mediation model. Drawing on survey data from 521 consumers across six major Chinese cities, we employ hierarchical multiple regression, supplemented by exploratory factor analysis and the Sobel test, to test our hypotheses. Additionally, we apply the Analytic Hierarchy Process (AHP) to construct an importance evaluation index system for consumer perceived value, which complements the regression analysis by quantifying the relative importance of value dimensions from a multi-criteria decision-making perspective. The two methods address distinct questions and draw on different populations. Regression analysis identifies statistical associations between value dimensions and purchase intention in the consumer survey data and thus constitutes demand-side behavioral evidence, whereas AHP captures the priority judgments of supply-side practitioners concerning the relative importance of value dimensions when financial and managerial resources are constrained. The regression results and the AHP weights therefore correspond to demand-side evidence and supply-side decision support respectively, and the two approaches are complementary rather than mutually validating. Our findings contribute to the sustainable cities literature in three ways. First, they clarify the behavioral correlates of housing purchase intention by examining how the perceived value and perceived risk dimensions are jointly associated with purchase intention in a rapidly urbanizing emerging economy. Second, they provide evidence that income-based heterogeneity in these associations carries implications for equitable housing policy design. Third, they provide an expert-based priority ranking that complements the survey findings and may inform resource allocation in urban housing development aligned with sustainable urban development objectives (SDG 11).

2 Literature review

2.1 Theoretical foundations

This study is anchored in two core theoretical perspectives, perceived value theory and perceived risk theory, while the theory of planned behavior (TPB) serves as a general sensitizing framework that situates these constructs within the broader study of behavioral intention. Perceived value theory, rooted in the work of Zeithaml (1988), posits that consumers' purchase decisions are driven by their overall assessment of the utility derived from a product or service relative to the perceived sacrifices. Zeithaml defined perceived value as the consumer's overall assessment of the utility of a product based on perceptions of what is received and what is given. Subsequent research has operationalized perceived value as a multidimensional construct. Sweeney and Soutar (2001) developed the PERVAL scale, identifying four key dimensions: emotional value, social value, quality value, and price value. Sheth et al. (1991)= proposed a broader taxonomy comprising functional, social, emotional, epistemic, and conditional values. This study adopts Sweeney and Soutar's (2001) four-dimensional framework, adapted to the housing consumption context.

Perceived risk theory, introduced to consumer research by Bauer (1960), contends that consumer behavior involves risk in the sense that any action will produce consequences that the consumer cannot anticipate with certainty. Cox and Rich (1964) elaborated on this concept by introducing the distinction between uncertainty and consequences. Cunningham (1967) further decomposed perceived risk into two core components: uncertainty (the consumer's subjective probability of an adverse outcome) and consequence (the severity of the potential loss). Extant literature has identified multiple dimensions of perceived risk, including financial, performance, physical, psychological, social, and time risks (Stone and Gronhaug, 1993). In the housing context, we classify perceived risk into three dimensions: transaction subject risk, transaction object risk, and environmental risk.

The theory of planned behavior (Ajzen, 1991), an extension of the theory of reasoned action, provides a comprehensive framework for understanding behavioral intention. TPB posits that intention is shaped by three factors: attitude toward the behavior, subjective norms, and perceived behavioral control. In the present study, TPB is used as a sensitizing framework rather than as a directly tested model: it indicates that behavioral intention is jointly shaped by attitudinal, social, and control-related considerations, without requiring that the constructs examined here be treated as equivalents of its three canonical antecedents. This study does not directly measure attitude, subjective norm, or perceived behavioral control. Instead, it focuses on perceived value and perceived risk, two evaluative constructs that are especially salient in high-cost, low-risk-tolerance consumption contexts such as housing purchases, in which consumers commit substantial financial resources under information asymmetry and limited reversibility. Perceived value captures the overall assessment of benefits relative to sacrifices, and perceived risk captures the subjective expectation of potential loss; both are expected to shape purchase intention alongside, rather than as substitutes for, the canonical TPB antecedents. Contemporary research has extended TPB to housing contexts; for instance, Masukujjaman et al. (2022) applied TPB to explain green home buying intention among Malaysian millennials, finding that attitude and perceived behavioral control were the strongest predictors, with perceived risk serving as an important boundary condition. Furthermore, Zhou et al. (2025) integrated TPB with the norm activation model to explain prefabricated housing purchase intention in China, demonstrating that both egoistic and altruistic motivations significantly drive purchase decisions. Positioned in this way, the present study constitutes an applied integration of perceived value and perceived risk theories in the housing consumption context, rather than an extension of TPB or of perceived value theory.

2.2 Perceived value and purchase intention

The positive relationship between perceived value and purchase intention has been extensively documented across diverse consumption contexts. Dodds et al. (1991) demonstrated that perceived value significantly influences buyers' product evaluations and willingness to buy. Grewal et al. (1998) found that comparative advertising affects purchase intention through its impact on perceived acquisition value and transaction value. Chiu et al. (2009) examined the determinants of repurchase intention in B2C e-commerce, finding that both utilitarian and hedonic values significantly influence consumers' repeat purchase intentions.

In the real estate domain, empirical evidence supports the positive impact of perceived value on housing purchase intention. Yao et al. (2024) investigated Chinese residents' house purchase intention during the COVID-19 pandemic and found that perceived economic value, health safety value, and emotional comfort value significantly influenced purchase intention, with perceived value serving as a key mediator between external housing attributes and behavioral intention. Hoe et al. (2018) investigated the Malaysian residential property market and found that functional, social, relational, situational, and cognitive values all exert significant positive effects on purchase intention, with emotional value demonstrating a relatively weaker influence. Guo (2018) identified four perceived value dimensions (external functional value, internal functional value, psychological value, and emotional value) in the Cambodian context, each exerting differential effects on purchase intention. In the domain of green housing consumption, Zhao and Cao (2025) employed a dual-stage SEM-ANN approach to examine Chinese residents' green housing purchase intention, revealing that perceived functional value, environmental value, and emotional value were significant predictors of purchase intention, while price value exhibited a relatively weaker effect—findings that resonate with the hierarchical importance of value dimensions identified in this study. Similarly, Kumar et al. (2025) integrated TPB with environmental awareness and perceived risk to investigate green housing purchase behavior in India, finding that environmental awareness significantly influenced attitudes and purchase intention, while perceived risk had an insignificant impact on TPB constructs. Their findings highlight the predominant role of value-laden cognitions over risk perceptions in driving sustainable housing purchase intentions. Feng and Zhang (2011) argued that assessing potential customers' perceived value enables sales professionals to identify customer needs accurately and tailor marketing communications accordingly.

The four dimensions of perceived value—price value, emotional value, social value, and quality value—capture distinct aspects of consumer value perception (Sweeney and Soutar, 2001). Price value reflects the economic benefits derived from the price-quality trade-off. Emotional value captures the affective responses elicited by the product or service. Social value represents the extent to which the product enhances the consumer's social self-concept and status. Quality value encompasses the consumer's objective evaluation of the product's overall performance and reliability. Understanding the differential impact of these dimensions on purchase intention is essential for developing targeted marketing strategies in the real estate sector. Empirical support for the differential effects of these value dimensions on purchase intention has been documented by Lin and Dong (2023), who found that functional value and price value significantly influenced purchase intention toward energy-efficient home appliances, while emotional and social values showed context-dependent effects, a finding that confirms the context-dependent nature of value dimension salience. Furthermore, understanding the differential impact of these value dimensions is essential not only for marketing strategy but also for designing urban housing policies that promote sustainable consumption patterns across diverse population segments (Chanda et al., 2025).

2.3 Perceived risk and purchase intention

Perceived risk has been identified as a critical inhibitor of purchase intention across various consumption contexts. Mitchell (1999) provided a comprehensive review of perceived risk conceptualizations and models, demonstrating that consumers employ various risk-reduction strategies (information search, brand loyalty, word-of-mouth reliance) when facing high perceived risk. Stone and Gronhaug (1993) validated six risk dimensions—financial, performance, physical, psychological, social, and time—and confirmed their negative effects on purchase intention. Featherman and Pavlou (2003) extended perceived risk research to the e-services context, finding that perceived risk significantly impedes online service adoption. In the specific context of mortgage-driven housing purchases, Reddy and Thanigan (2023) examined the interplay between customer knowledge, information search, perceived risk reduction, and purchase intention among 554 mortgage buyers. Their findings revealed that active customer involvement in information search significantly reduced perceived risk and enhanced purchase intention, with satisfaction serving as a critical mediator. This study underscores the dynamic nature of perceived risk in housing decisions and the efficacy of risk-reduction strategies through information empowerment.

In the housing context, the importance of perceived risk is particularly pronounced due to the unique characteristics of real estate transactions: large monetary commitment, complex transaction procedures, high information asymmetry, and limited reversibility (Dong et al., 2005). Dong et al. (2005) identified four dimensions of perceived risk in online shopping contexts: privacy risk, counterfeit risk, retailer service risk, and delivery risk. Yang (2018) examined the relationship between perceived value and perceived risk among housing consumers in Quanzhou, China, finding a significant negative correlation and substantial heterogeneity across demographic segments. Liu (2013) investigated perceived risk and purchase intention among ordinary residential property consumers in Baoding, Hebei Province, revealing that property rights risk exhibited the strongest negative association with purchase intention.

In the context of sustainable urban development, perceived risk in housing consumption extends beyond individual financial concerns to encompass broader anxieties about policy stability, environmental quality, and neighborhood sustainability, factors that are central to the environmental risk dimension examined in this study (Njo and Sugeng, 2023). Contemporary research has continued to refine the dimensional structure of perceived risk and its behavioral consequences. Koay et al. (2023) developed and validated a comprehensive six-dimensional perceived risk scale encompassing aesthetic risk, functional risk, financial risk, psychological risk, sanitary risk, and social risk. Their multigroup analysis demonstrated that the relative importance of risk dimensions varied significantly between experienced and novice consumers, with functional and financial risks exhibiting stronger inhibitory effects among non-consumers. These findings highlight the need for context-specific risk dimension operationalization in housing purchase research. Similarly, Setiawan et al. (2024) investigated the effects of multiple perceived risk dimensions on behavioral intention, finding that financial risk exerted a significant negative direct effect on behavioral intention, while overall risk perception mediated the relationship between other risk dimensions and intention. A contemporary systematic review by Phamthi et al. (2024) synthesized 133 studies and confirmed that financial, performance, and psychological risks remain the most consistently significant risk dimensions inhibiting purchase intention across diverse consumption contexts, highlighting the mediating role of trust and the moderating role of consumer knowledge in the risk-intention relationship.

We classify housing purchase perceived risk into three dimensions: transaction subject risk (risks associated with the counterparty, including contract risk, capital risk, time risk, and information asymmetry risk); transaction object risk (risks associated with the property itself, including quality defects, property rights disputes, design flaws, and safety hazards); and environmental risk (macro-level risks including policy changes, economic downturns, and social environment deterioration). This classification follows the object structure of a housing transaction, in which potential losses originate from three analytically distinct sources: the counterparty with whom the transaction is conducted (who), the property that constitutes the subject matter of the transaction (what), and the macro-environment in which the transaction is embedded (context). Because the three dimensions correspond, respectively to who the consumer transacts with, what is transacted, and the context of the transaction, they are designed to be comprehensive in coverage and mutually exclusive in attribution: any source of perceived loss in a housing purchase can be assigned to one of the three dimensions, and each dimension corresponds to exactly one source. This three-dimensional classification captures the multifaceted nature of perceived risk in housing consumption while maintaining parsimony for empirical analysis. The item sources and adaptation history of the three-dimensional scale are documented in Section 3.2.

2.4 The mediating role of perceived risk

The mediating role of perceived risk in the relationship between perceived value and purchase intention has received limited but promising attention in the literature. From a theoretical perspective, perceived value and perceived risk are inherently interrelated: higher perceived value implies greater benefits relative to sacrifices, which should reduce the consumer's subjective risk assessment (Chen and Chang, 2012). Conversely, elevated perceived risk necessitates increased cognitive effort and information search costs, thereby diminishing the perceived net value of the transaction (Chen and Tsai, 2007). The complementary roles of perceived value and perceived risk in shaping consumer purchase intention have been documented by Syamsuar and Witarsyah (2025), who found that higher perceived value attenuated the negative impact of perceived risk on purchase intention, confirming that the two constructs operate as countervailing forces in consumer decision-making processes. Because value and risk perceptions may influence each other, the direction of the proposed mediation requires explicit justification. Cross-sectional data cannot statistically discriminate between reciprocal specifications. This study therefore adopts the direction that predominates in the theoretical literature, in which value appraisal precedes risk assessment: consumers first form an overall judgment of benefits relative to sacrifices, and this judgment subsequently conditions their subjective expectation of potential loss (Chen and Chang, 2012; Syamsuar and Witarsyah, 2025). The alternative direction, in which perceived risk shapes perceived value, cannot be excluded with the present design and is acknowledged in Section 7.3.

In the housing context, the mediating mechanism can be elaborated through three pathways. The three risk dimensions are specified as mediators rather than as ordinary predictors because they constitute the evaluative channel through which value appraisal is transmitted to intention: a higher perceived value reduces the subjective expectation of loss from a specific source, and this reduced expectation of loss in turn weakens the inhibition of purchase intention. First, transaction subject risk mediates the influence of perceived value on purchase intention: when consumers perceive high value in developer brand reputation, intermediary service quality, and contract fairness, their subjective expectation of counterparty default, fraud, or service failure is attenuated, and this reduced concern in turn facilitates purchase intention. Second, transaction object risk serves as a mediator: when consumers perceive high value in housing quality, property rights clarity, and layout design, their anticipated losses from physical defects and legal disputes diminish, which in turn supports the formation of purchase intention. Third, environmental risk mediates the relationship: when consumers perceive high value in location amenities and policy stability, their subjective assessment of macro-environmental uncertainty declines, and this reduced anxiety in turn strengthens purchase intention (Zhou et al., 2022).

The final component of the framework concerns income-based heterogeneity, which can be derived from the logic of budget constraints. Housing expenditure competes with all other household consumption under a finite budget, and the tightness of this constraint determines the weight assigned to each value dimension (Stone, 2006). For low-income consumers, the budget constraint is binding: housing expenditure absorbs a large share of household resources, the marginal weight of price considerations in the overall value judgment rises, and price value is expected to become the dominant criterion in the purchase decision. For high-income consumers, the budget constraint is relatively slack: basic affordability is less pressing, the marginal weight of quality and residential attributes rises, and quality value is expected to become the dominant criterion. This expectation is consistent with evidence that housing demand composition and value prioritization vary systematically with household resources in China's higher-tier cities (Glaeser et al., 2017; Jin et al., 2026). Income group is therefore specified as a moderator of the associations between the perceived value dimensions and purchase intention.

Based on the theoretical analysis and literature review, we propose the following hypotheses:

• H1: Perceived value positively influences housing purchase intention.
• H1a: Perceived quality value positively influences housing purchase intention.
• H1b: Perceived emotional value positively influences housing purchase intention.
• H1c: Perceived price value positively influences housing purchase intention.
• H1d: Perceived social value positively influences housing purchase intention.
• H2: Perceived risk mediates the relationship between perceived value and housing purchase intention.
• H2a: Transaction object risk mediates the relationship between perceived value and housing purchase intention.
• H2b: Transaction subject risk mediates the relationship between perceived value and housing purchase intention.
• H2c: Environmental risk mediates the relationship between perceived value and housing purchase intention.
• H3: Income group moderates the association between perceived value dimensions and housing purchase intention.
• H3a: The positive association between price value and purchase intention is stronger among low-income consumers.
• H3b: The positive association between quality value and purchase intention is stronger among high-income consumers.

Figure 1 presents the conceptual framework of this study. Perceived value (comprising four dimensions) serves as the independent variable, perceived risk (comprising three dimensions) as the mediator, income group as the moderator of the associations between the value dimensions and purchase intention, and purchase intention as the dependent variable. Demographic variables (gender, age, education, marital status, monthly income, homeownership status, and purchase purpose) are included as control variables. It should be noted that purchase intention is examined as the proximate behavioral antecedent rather than as behavior itself, because housing acquisition is additionally constrained by mortgage access, household wealth, policy eligibility, and market timing; stated purchase intention therefore does not translate mechanically into completed transactions.

3 Research design

3.1 Participants and procedure

This study targeted Chinese consumers with recent housing purchase experience or explicit purchase intentions. Eligibility was defined operationally and verified through screening questions administered at the beginning of the questionnaire: recent housing purchase experience referred to the completion of a housing transaction within the past 3 years, and explicit purchase intention referred to a definite plan to purchase housing within the next 12 months accompanied by active property viewing. Respondents who met neither criterion were excluded at the screening stage. A mixed-mode sampling strategy combining online and offline data collection was employed. Online surveys were distributed through the Wenjuanxing platform and disseminated via social media (WeChat Moments, WeChat groups) and real estate forums (Anjuke, Fangtianxia). Offline surveys were conducted through intercept interviews at real estate sales centers, property agency offices, and commercial complexes in multiple cities (Beijing, Shanghai, Guangzhou, Shenzhen, Hangzhou, and Chengdu). All interviewers received standardized training to ensure consistent survey administration. The 521 valid respondents were distributed geographically as follows: Guangzhou (n = 150, 28.8%), Shenzhen (n = 80, 15.4%), Beijing (n = 66, 12.7%), Shanghai (n = 62, 11.9%), Hangzhou (n = 51, 9.8%), and Chengdu (n = 44, 8.4%), with the remaining respondents (n = 68, 13.0%) residing in other cities reached through the online platform. The sample is thus concentrated in China's first-tier and new first-tier cities (87.0%), a distribution that should be borne in mind when assessing the generalizability of the findings to lower-tier urban markets. Before the formal analyses, robustness checks with group membership (recent purchasers vs. prospective purchasers) entered as a covariate confirmed that the directions of the main associations were consistent across the two groups, and the two groups were therefore pooled for analysis. In addition, comparisons using survey mode (online vs. offline) as the grouping variable showed no substantive differences in the means of the main constructs, with negligible effect sizes, so responses collected through the two modes were combined.

We employed a combination of convenience sampling and snowball sampling. Convenience sampling enabled efficient data collection within a limited timeframe, appropriate for exploratory research (Saunders et al., 2019). Snowball sampling facilitated access to latent consumer segments through respondents' social networks. While non-probability sampling limits generalizability, it is appropriate for examining whether the observed associations are consistent with the proposed theoretical model, although the estimates cannot be treated as population parameters or causal effects (Hair Jr et al., 2014).

Sample size was determined following Green's (1991) recommendation that the minimum sample size for multiple regression should exceed 104 + m (where m is the number of predictor variables). With approximately 15 predictor variables in the full model (four value dimensions, three risk dimensions, and control variables), the minimum required sample size was approximately 120. Accounting for an anticipated invalid response rate of 5%−10% (based on comparable studies), we targeted 550–600 distributed questionnaires.

Data collection followed a rigorous multi-stage procedure: First, a pilot study (n = 30) was conducted to refine questionnaire wording and assess scale reliability through cognitive interviewing. Second, formal data collection was conducted over 4 weeks (October-November 2022), yielding 544 distributed questionnaires. Third, stringent screening criteria were applied: (1) responses completed in less than 3 min or more than 30 min were excluded; (2) responses exhibiting patterned answering (e.g., selecting the same option for all items) were excluded; (3) questionnaires with missing values exceeding three items were excluded; (4) responses with logical inconsistencies between positively and negatively worded items were excluded. Questionnaires containing one to three missing items were retained after person-mean substitution, in which each missing response was imputed using the respondent's mean score on the corresponding subscale. After screening, 521 valid responses were retained, representing a usable-response rate of 95.8% among the 544 completed questionnaires; because the online surveys were disseminated through open links, a conventional response rate cannot be computed, and this figure denotes the retention rate of valid questionnaires among those returned, satisfying the sample size requirements for the regression-based analytical approach employed in this study. The study protocol was reviewed and approved by the Ethics Committee of Guangzhou Urban Planning and Design Company Limited prior to data collection (approval no. GZUPD-2022-0417, approved 18 October 2022). All participants provided informed consent.

3.2 Measurement

This study employed a structured questionnaire comprising four sections: perceived value scale, perceived risk scale, purchase intention scale, and demographic variables. All scales used a 5-point Likert format ranging from 1 (strongly disagree) to 5 (strongly agree). The questionnaire was adapted to the Chinese housing context through a standardized procedure. All adapted items were translated into Chinese and back-translated into English, and two bilingual scholars independently verified the semantic equivalence of the two versions. The pilot study (n = 30) described above incorporated cognitive interviews to confirm that respondents understood the items as intended. Throughout the questionnaire, the phrase this residential property referred to the most recently purchased dwelling for respondents with recent purchase experience and to the target property under most serious consideration for prospective purchasers; both groups answered identical scales with the same item wording.

Perceived value was measured using a 20-item scale adapted from Sweeney and Soutar's (2001) PERVAL scale, modified for the housing consumption context. The scale comprises four dimensions: price value (4 items; e.g., “This residential property offers good value for money”), emotional value (7 items; e.g., “This residential property gives me a pleasant and delightful feeling”), social value (5 items; e.g., “This residential property enhances my social status”), and quality value (4 items; e.g., “This residential property offers better quality than its price suggests”). The Cronbach's alpha for the overall scale was 0.969.

Perceived risk was measured using a 17-item scale developed for this study by adapting items from established perceived-risk measures (Featherman and Pavlou, 2003) and contextualizing them to housing transactions, with the dimension structure derived from the source classification of Dong et al. (2005). The scale comprises three dimensions: transaction subject risk (6 items; e.g., “There is a risk of capital loss during the home purchase process”), transaction object risk (5 items; e.g., “The purchased property may have quality defects”), and environmental risk (6 items; e.g., “The government may introduce stricter housing policies”). The Cronbach's alpha for the overall scale was 0.926.

Purchase intention was measured using a 3-item scale adapted from Dodds et al. (1991), including items such as “I would consider purchasing this residential property,” “The probability that I will purchase this residential property is high,” and “My intention to purchase this residential property is very strong.” The Cronbach's alpha was 0.814.

3.3 Analytical strategy

Data analysis was conducted using SPSS 19.0 and proceeded through the following stages: First, descriptive statistics were computed for all demographic and research variables to characterize the sample and assess data distribution properties. Second, reliability was assessed using Cronbach's alpha coefficients, with values above 0.70 deemed acceptable. Construct validity was evaluated through exploratory factor analysis (EFA), requiring KMO values above 0.70 and significant Bartlett's sphericity tests (p < 0.05), with factor loadings exceeding 0.50 and no substantial cross-loadings. Harman's single-factor test was conducted as a preliminary diagnostic for common method variance. The first unrotated factor accounted for 32.6% of the total variance, below the 40% benchmark, suggesting that common method bias is unlikely to account for the observed associations (Podsakoff et al., 2003). Third, Pearson correlation coefficients were computed to examine bivariate relationships among variables and provide preliminary evidence for subsequent regression analyses. Effect sizes were interpreted following Cohen's (1988) guidelines: 0.10–0.30 (weak), 0.30–0.50 (moderate), and above 0.50 (strong). Fourth, hierarchical multiple regression was employed to test the main effects of perceived value on purchase intention and the mediating role of perceived risk, following the four-step procedure: (1) regressing the dependent variable on the independent variable; (2) regressing the mediator on the independent variable; (3) regressing the dependent variable on the mediator; and (4) regressing the dependent variable on both the independent variable and mediator to assess attenuation of the direct effect. To provide a more robust assessment of indirect effects, the SPSS PROCESS macro (Hayes, 2018) was employed as the primary mediation analysis tool. Model 4 (simple mediation) was specified with 5,000 bootstrap samples and bias-corrected and 95% confidence intervals. The Sobel test and the hierarchical four-step procedure were retained as complementary approaches to enable comparison across methods and to ensure transparency in the analytical process. The bootstrap method was prioritized for hypothesis testing because it does not assume normality of the sampling distribution and offers superior statistical power for detecting indirect effects, particularly when effect sizes are small (Preacher and Hayes, 2004; Hayes, 2018). Categorical control variables (gender, education, marital status, income, homeownership status, and purchase purpose) were entered into the regression models as dummy variables, with female, junior college or below, unmarried, monthly income less than 5,000 CNY, no homeownership, and purchase for personal residence serving as the respective reference categories.

Fifth, income-based heterogeneity was examined through a formal moderation analysis rather than a side-by-side comparison of subgroup coefficients. Following established recommendations for testing moderation in multiple regression (Dawson, 2014), monthly income was dichotomized at 10,000 CNY into a binary moderator (0 = low income; 1 = high income), the four perceived value dimensions were mean-centered, and four interaction terms between the value dimensions and the income-group indicator were constructed. Hierarchical regression on the full sample (N = 521) was then conducted, entering the control variables, the four value dimensions, and the moderator's main effect in the first step and the four interaction terms in the second step. A statistically significant interaction term indicates that the corresponding regression slope differs significantly across income groups. The 10,000 CNY threshold was retained because it approximates the median income in China's first- and second-tier cities and separates budget-constrained consumers from financially comfortable ones. The resulting subgroup sizes were unequal (low-income: n = 368, 70.6%; high-income: n = 153, 29.4%), reflecting the skewed income distribution in China's urban housing market, and the smaller high-income subgroup reduces statistical power within that stratum. Supplementary subgroup regressions were estimated for descriptive purposes only, to illustrate the direction and approximate magnitude of the significant interactions.

Terminological note. Given the cross-sectional, observational design, all relational terms used in the empirical sections below (including predict, effect, and association) denote statistical associations estimated within the specified models and should not be read as causal effects. The hypotheses express theoretical propositions; the analyses evaluate whether the observed associations are consistent with them.

4 Results

4.1 Descriptive statistics

Table 1 presents the demographic characteristics of the sample. The gender distribution was reasonably balanced (53.7% male, 46.3% female), with a slight male predominance consistent with traditional gender roles in housing purchase decisions. The age distribution was concentrated in the 26–45 range (73.4%), reflecting the demographic profile of China's primary housing consumers. Educational attainment was predominantly at the bachelor's level or below (82.6%), consistent with the general education distribution among urban Chinese housing consumers. The majority of respondents were married (68.2%), and most purchased for owner-occupation (97.1%). Income distribution was right-skewed, with 70.6% earning below 10,000 CNY monthly, consistent with urban income distribution patterns in China.

Table 1. Demographic characteristics of the sample.

Descriptive statistics for the study variables are provided below. Perceived value demonstrated a mean of 3.624 (SD = 0.821), indicating a moderately high level of value perception. Purchase intention showed a mean of 3.438 (SD = 0.887), suggesting moderate purchase intention consistent with current market conditions. Perceived risk demonstrated a mean of 2.856 (SD = 0.768).

4.2 Reliability and validity

This study assessed the Cronbach's alpha coefficients for all scales. The overall perceived value scale achieved an alpha of 0.969, with subscale alphas of 0.917 (price value), 0.907 (emotional value), 0.932 (social value), and 0.865 (quality value). The perceived risk scale achieved an overall alpha of 0.926, with subscale alphas ranging from 0.887 to 0.889. The purchase intention scale achieved an alpha of 0.814. All alpha values exceeded the 0.80 threshold, demonstrating satisfactory internal consistency reliability.

Construct validity was assessed through exploratory factor analysis. The perceived value scale yielded a KMO value of 0.910 and a significant Bartlett's test (chi-square = 15,752.820, df = 190, p < 0.001). The perceived risk scale yielded a KMO of 0.850 and significant Bartlett's test (chi-square = 13,562.820, df = 122, p < 0.001). The purchase intention scale yielded a KMO of 0.748 and significant Bartlett's test (chi-square = 762.767, df = 3, p < 0.001). All KMO values exceeded 0.70, and Bartlett's tests were significant, supporting the appropriateness of factor analysis. Factors were extracted using principal axis factoring with promax rotation, an oblique rotation selected because the value and risk dimensions were expected to be correlated, and the number of retained factors was determined by the eigenvalues-greater-than-one criterion in combination with inspection of the scree plot. All primary factor loadings exceeded 0.50 (range: 0.512–0.891), with no cross-loadings exceeding 0.35, providing preliminary evidence of dimensional separation at the indicator level (Hair Jr et al., 2014). Cumulative variance explained exceeded 60% for all scales (perceived value: 62.3%; perceived risk: 64.7%; purchase intention: 71.2%). These results provide evidence of adequate construct validity within the EFA framework.

Discriminant validity was further assessed through the inter-dimensional correlation matrices (Tables 2, 3). Within the perceived value construct, the four dimensions exhibited positive intercorrelations ranging from 0.313 to 0.502 (mean r = 0.424), with the emotional value–price value correlation (r = 0.502) approaching the threshold that Cohen (1988) classifies as strong. This correlation magnitude indicates that consumers who associate high emotional value with a housing purchase also tend to perceive favorable price terms, a pattern consistent with the overall evaluative nature of the perceived value construct in which positive assessments across dimensions tend to cohere. Nevertheless, the squared correlation (r² = 0.252) indicates that these two dimensions share only approximately 25% of their variance, leaving 75% unique variance unaccounted for by their linear association. The remaining inter-dimensional correlations fell within the moderate range (r = 0.313–0.441), with the weakest association observed between emotional value and social value (r = 0.313). Within the perceived risk construct, the three dimensions showed weaker positive intercorrelations (r = 0.198–0.361, mean r = 0.293), supporting their conceptual separation. The average variance extracted (AVE) for each dimension was computed from the EFA-derived standardized loadings as the mean of the squared loadings of the items belonging to that dimension. In addition, the square root of the AVE (presented in Tables 2, 3) exceeded all inter-dimensional correlations, satisfying the Fornell and Larcker (1981) criterion for discriminant validity. These results suggest that the four perceived value dimensions and three perceived risk dimensions are theoretically related and empirically correlated yet retain sufficient unique variance to be treated as separable facets rather than redundant indicators.

Table 2. Pearson correlation matrix for the aggregate variables.

^**p < 0.01. The value in parentheses represents the square root of the average variance extracted (AVE) for the corresponding construct.

Table 3. Dimensional-level correlation matrix.

PQV, Perceived quality value; PEV, Perceived emotional value; PPV, Perceived price value; PSV, Perceived social value; TOR, Transaction object risk; TSR, Transaction subject risk; ER, Environmental risk; HPI, Housing purchase intention. ^*p < 0.05; ^**p < 0.01. The value in parentheses represents the square root of the average variance extracted (AVE) for the corresponding construct.

4.3 Correlation analysis

Table 2 presents the Pearson correlation matrix for the aggregate variables. Perceived value demonstrated a significant positive correlation with purchase intention (r = 0.386, p < 0.01), providing preliminary support for H1. Perceived value showed a significant negative correlation with perceived risk (r = −0.239, p < 0.01), indicating that higher perceived value is associated with lower perceived risk. Perceived risk exhibited a significant negative correlation with purchase intention (r = −0.177, p < 0.01), confirming that higher perceived risk is associated with lower purchase intention. These bivariate associations provide preliminary evidence supporting the hypothesized mediation mechanism.

Table 3 presents the dimensional-level correlation matrix. All four perceived value dimensions demonstrated significant positive correlations with purchase intention: quality value (r = 0.302, p < 0.05), emotional value (r = 0.252, p < 0.01), price value (r = 0.231, p < 0.01), and social value (r = 0.142, p < 0.01). Notably, quality value displayed the strongest bivariate correlation, indicating that objective attributes such as structural integrity, construction materials, seismic performance, energy efficiency, and building system readiness (e.g., plumbing, electrical, HVAC) are central to consumers' preliminary value judgments in residential property decisions. Among perceived risk dimensions, transaction object risk showed the strongest negative correlation with purchase intention (r = −0.225, p < 0.01), followed by environmental risk (r = −0.153, p < 0.01). All significance markers in Tables 2, 3 were regenerated from verified SPSS output and cross-checked against the reported p-value thresholds.

4.4 Regression analysis

Table 4 presents the hierarchical regression results for the main effect of perceived value on purchase intention. Model 1, including only control variables, yielded a non-significant model (R² = 0.011, p > 0.05). Model 2, adding the aggregate perceived value variable, showed a significant positive association with purchase intention (β = 0.432, p < 0.01), though the model explained a modest 15.3% of variance (R² = 0.153, p < 0.01), providing tentative support for H1. Model 3 decomposed perceived value into its four dimensions, revealing differential associations: quality value carried the relatively largest standardized coefficient among the four dimensions (β = 0.574, p < 0.01), followed by emotional value (β = 0.395, p < 0.01), price value (β = 0.377, p < 0.01), and social value (β = 0.303, p < 0.01). All four dimensions were significantly and positively associated with purchase intention within the estimated model, with the standardized coefficients ordered as quality value > emotional value > price value > social value, a pattern consistent with H1a through H1d. However, given the modest R-squared values, these findings should be interpreted cautiously: perceived value explains only a limited portion of the variance in purchase intention, and other unmeasured factors (e.g., market conditions, policy changes, family circumstances) likely play substantial roles. The relative ranking of value dimensions suggests that quality and emotional considerations carry more weight than price and social in shaping purchase intention, though the absolute effect sizes remain modest. To quantify the substantive magnitude of this association, we computed Cohen's (1988) local effect size f² for the incremental contribution of perceived value. Adding aggregate perceived value to the control-only model yielded f² = 0.168, which corresponds to a medium effect by conventional benchmarks (Cohen, 1988). The explanatory power of the psychological model is therefore statistically robust yet realistically modest.

Table 4. Hierarchical regression results of perceived value on purchase intention.

^**p < 0.01.

Table 5 presents the regression of perceived value on perceived risk. Model 4 demonstrated that aggregate perceived value was negatively associated with perceived risk (β = −0.319, p < 0.01), with the model explaining 16.2% of variance (R² = 0.162, p < 0.01). Model 5 decomposed the four value dimensions, revealing that quality value (β = −0.501, p < 0.01), emotional value (β = −0.422, p < 0.01), price value (β = −0.237, p < 0.01), and social value (β = −0.209, p < 0.01) were all negatively associated with perceived risk, with quality value carrying the relatively largest negative standardized coefficient.

Table 5. Hierarchical regression results of perceived value on perceived risk.

^**p < 0.01.

Table 6 presents the regression of perceived risk on purchase intention. Model 6 demonstrated that aggregate perceived risk was negatively associated with purchase intention (β = −0.397, p < 0.01), with the model explaining 18.7% of variance (R² = 0.187, p < 0.01). Model 7 decomposed the three risk dimensions (R² = 0.291, p < 0.01), and its higher explained variance relative to Model 6 is expected because decomposing the aggregate construct adds explanatory terms to the model. Model 7 showed that transaction object risk (β = −0.357, p < 0.01), transaction subject risk (β = −0.451, p < 0.01), and environmental risk (β = −0.233, p < 0.01) were all negatively associated with purchase intention, with transaction subject risk carrying the relatively largest negative standardized coefficient.

Table 6. Hierarchical regression results of perceived risk on purchase intention.

^**p < 0.01.

Table 7 presents the joint regression models testing the mediation hypothesis. Model 8 simultaneously included perceived value and perceived risk, revealing that the coefficient for perceived value on purchase intention attenuated from 0.432 (M2) to 0.324 (p < 0.01) while perceived risk remained significant (β = −0.340, p < 0.01), a pattern consistent with partial mediation under the proposed model. Model 9 simultaneously included the four perceived value dimensions and aggregate perceived risk, showing that all value dimensions retained directional consistency with attenuated coefficients after controlling for perceived risk. The results indicated that quality value (β = 0.312, p < 0.01), emotional value (β = 0.276, p < 0.01), price value (β = 0.174, p < 0.01), and social value (β = 0.148, p < 0.01) retained significant positive associations. Model 10 simultaneously included all value and risk dimensions, showing that environmental risk (β = −0.264, p < 0.01), transaction subject risk (β = −0.193, p < 0.01), and transaction object risk (β = −0.162, p < 0.01) were negatively associated with purchase intention. These results provide preliminary support for H2, H2a, H2b, and H2c. A more rigorous test using the PROCESS macro (Hayes, 2018; Model 4) with 5,000 bootstrap resamples yielded a statistically significant indirect association of 0.108 [95% CI (0.050, 0.167), shown in Table 8], a result consistent with the proposed mediation model and thereby supporting H2. The non-overlap of the confidence interval with zero satisfies the contemporary criterion for a nonzero indirect association (Preacher and Hayes, 2008). Dimension-level analyses (Table 9) were conducted within a single parallel multiple-mediator model (PROCESS Model 4) in which the three risk dimensions were entered simultaneously, so that the indirect association of each dimension was estimated net of the other two dimensions. Each dimension showed a statistically significant indirect association in this parallel multiple-mediator model: environmental risk [Ind = 0.028, 95% CI (0.008, 0.048)], transaction object risk [Ind = 0.048, 95% CI (0.009, 0.087)], and transaction subject risk [Ind = 0.032, 95% CI (0.013, 0.052)], providing convergent support for H2c, H2a, and H2b, respectively. The incremental contribution of perceived risk in Model 8, relative to Model 2, was f² = 0.060, a small effect approaching the medium threshold, which is consistent with the modest magnitude of the bootstrap indirect effects reported above.

Table 7. Analysis of mediating effects.

^**p < 0.01.

Table 8. Analysis of mediating effects via process (at the aggregate level).

The symbols denote the following significance levels: ^**p < 0.01; ^***p < 0.001.

Table 9. Analysis of mediating effects via process (at the risk dimension level).

^*p < 0.05; ^**p < 0.01.

Two features of the regression results warrant explicit discussion. First, the explained variance is modest: Model 3 accounts for 18.7% of the variance in purchase intention, leaving the majority of variance to factors outside the model. Such values are nevertheless typical of consumer behavior research, in which R² values around 0.20 are commonly regarded as substantial because behavioral intentions are shaped by a wide array of psychological, social, and situational forces (Hair et al., 2019). Housing purchase intention is additionally constrained by structural conditions that a psychological model is not designed to capture, including mortgage credit policy, down payment requirements, household wealth, hukou status, and expectations about market cycles. The present model is intended to test theorized mechanisms rather than to maximize predictive coverage, and its coefficients identify the relative contribution of value and risk perceptions net of demographic controls. Second, modest individual-level effects can carry considerable aggregate significance in a market of China's scale: even small shifts in the purchase intentions of urban households translate, at the city level, into substantial changes in effective housing demand, with implications for inventory absorption and market stability. The findings should thus be read as identifying meaningful behavioral levers rather than as a complete account of housing purchase decisions. Finally, because standardized regression coefficients depend on model specification, the correlations among predictors, and measurement properties, the ordering of coefficients reported above identifies only the relative strength of association within the estimated model and should not be read as a direct hierarchy of substantive importance.

4.5 Income-based moderation analysis

To examine whether the effects of the perceived value dimensions differ by income level, we conducted a formal moderation analysis on the full sample (N = 521) following the procedures recommended by Dawson (2014). Monthly income was dichotomized at 10,000 CNY (0 = low income, n = 368, 70.6%; 1 = high income, n = 153, 29.4%), the four perceived value dimensions were mean-centered, and four interaction terms between the dimensions and the income-group indicator were entered in the second step of a hierarchical regression, after the control variables, the four value dimensions, and the moderator's main effect. This approach tests coefficient differences between income groups directly: a significant interaction term indicates that the corresponding regression slope differs significantly across groups, an inference that cannot be drawn from comparing standardized coefficients estimated in separate subgroup models. The moderation results are reported in Table 10.

Table 10. Hierarchical regression results testing the moderating effect of income group on the perceived value–purchase intention relationship (N = 521).

^*p < 0.05; ^**p < 0.01. Standardized coefficients are reported. The four perceived value dimensions were mean-centered before computing the interaction terms. The ordinal monthly income control was replaced by the binary income-group indicator (0 = low income; 1 = high income) to avoid conceptual redundancy with the moderator. Simple slopes for the high-income group equal the corresponding main-effect coefficient plus the interaction coefficient.

As shown in Table 10, the four interaction terms jointly explained a statistically significant but modest increment of variance in purchase intention (ΔR² = 0.018, ΔF = 2.98, p < 0.05), indicating that income level moderates the value–intention relationship with limited incremental explanatory power. Two interactions reached statistical significance. The price value × income group interaction was negative and significant (β = −0.259, p < 0.01), indicating that the positive association between price value and purchase intention weakens significantly as income rises: simple slope estimates showed that price value was strongly and positively associated with purchase intention in the low-income group (β = 0.416, p < 0.01) but only modestly in the high-income group (β = 0.157, p < 0.05). Conversely, the quality value × income group interaction was positive and significant (β = 0.184, p < 0.05): quality value was positively associated with purchase intention in both groups, but the association was significantly stronger among high-income consumers (β = 0.573, p < 0.01) than among low-income consumers (β = 0.389, p < 0.01). The interactions for emotional value (β = 0.021, p > 0.05) and social value (β = −0.104, p > 0.05) were not significant (high-income simple slopes: emotional value β = 0.376; social value β = 0.187), indicating that the associations of these two dimensions with purchase intention do not differ reliably across income groups in the present sample.

For descriptive purposes, separate subgroup regressions were also estimated (Table 11). These models reproduce the pattern revealed by the formal test: price value ranked first among the four dimensions in the low-income group (β = 0.449, p < 0.01) but last in the high-income group (β = 0.182, p < 0.01), whereas quality value carried the largest standardized coefficient in the high-income group (β = 0.612, p < 0.01) and ranked second in the low-income group (β = 0.415, p < 0.01). Because these subgroup models were estimated on unequal samples (n = 368 vs. n = 153), their coefficients are reported solely to illustrate the direction and approximate magnitude of the significant interactions identified in Table 10; all inferential claims regarding income-based differences rest on the interaction-term analysis.

Table 11. Income-based subgroup regression results (descriptive).

^*p < 0.05; ^**p < 0.01.

Although the incremental explanatory power of the interaction terms is modest (ΔR² = 0.018), the significant interactions are consistent with the theoretical expectation that budget constraints shape value prioritization. For low-income consumers, affordability pressures may elevate the salience of price considerations, although the nearly equal weight assigned to quality value (β = 0.415) indicates that economic constraints do not override quality aspirations; these consumers appear to seek an optimal balance between cost and construction standards. For high-income consumers, quality attributes constitute the primary decision criterion, whereas price operates as a secondary consideration that retains statistical significance. The significant social value association in both groups, with a descriptively larger coefficient in the low-income group (β = 0.346 vs. 0.238), may reflect the greater importance of school district access, family proximity, and community embeddedness for households that depend more heavily on localized support networks, although the corresponding interaction did not reach significance and this interpretation requires confirmation in future research. Emotional value showed the smallest subgroup difference, suggesting that affective considerations carry relatively consistent importance across income levels.

Two methodological qualifications should be noted. First, dichotomizing income at 10,000 CNY simplifies a continuous variable, a practice known to sacrifice measurement information and statistical power relative to a fully continuous specification (MacCallum et al., 2002); the threshold was retained because it corresponds to a substantively meaningful divide between budget-constrained and financially comfortable consumers in China's first- and second-tier cities. Second, the smaller high-income subgroup (n = 153) limits power for detecting effects within that stratum, and the significant interactions reported here should be replicated with probability-sampled subgroups before strong conclusions are drawn.

5 Construction of customer perceived value importance evaluation index system

Building upon the empirical findings, we employed the Analytic Hierarchy Process (AHP) to construct an importance evaluation index system for consumer perceived value. AHP, developed by Saaty (1980), is a multi-criteria decision-making method that integrates qualitative and quantitative analysis through pairwise comparisons, weight vector computation, and consistency testing. This approach enables systematic analysis of complex decision problems involving multiple criteria and hierarchical structures. The AHP module serves a decision-support purpose: it informs how enterprises should allocate resources across perceived value dimensions, a class of decisions that currently rests on the judgments of industry practitioners. Eliciting those judgments through a structured expert panel makes the priorities underlying routine allocation decisions explicit and allows them to be examined alongside the survey evidence. The AHP complements the regression analysis by providing a structured, expert-informed assessment of the relative importance of perceived value dimensions, offering practical guidance for resource allocation that extends beyond the statistical relationships identified in the survey data. The relationship between the two analyses is one of complementarity rather than cross-validation: the regression coefficients estimate statistical associations in consumer survey data, whereas the AHP weights represent expert judgments about criterion importance. When the expert-derived priority order and the regression-derived effect sizes converge, as they do for quality value, the two perspectives reinforce one another as complementary evidence; when they diverge, the discrepancy signals domains where statistical significance and practical importance may not align, warranting deeper investigation.

Recent methodological advancements have extended the classical AHP framework to address uncertainty and vagueness in expert judgments in urban contexts. Durdyev et al. (2022) developed a cybernetic fuzzy AHP approach to identify the key drivers of clients' decisions to purchase green buildings in Malaysia, revealing that energy cost savings, environmental responsibility, and health benefits were the highest-priority drivers. Jung et al. (2022) employed AHP to analyze housing consumer preferences in Dubai, identifying housing price and rent, investment value, safety, and transportation convenience as the top-weighted determinants of housing choice, with preference weights varying significantly across demographic segments. These applications demonstrate the efficacy of AHP for capturing the nuanced trade-offs involved in urban housing consumption decisions and inform the present analysis.

5.1 Evaluation index design

Drawing upon Woodruff's (1997) customer value hierarchy model and integrating our empirical results, we developed a three-level evaluation index system. The first level (Level A) represents the overarching goal: comprehensive evaluation of consumer perceived value. The second level (Level B) comprises four criteria dimensions: price value (B1), emotional value (B2), social value (B3), and quality value (B4). The third level (Level C) encompasses specific evaluation indicators nested within each dimension. Index selection followed four principles: systematicity, representativeness, scientific simplicity, and comparability/operability.

The price value dimension (B1) comprises four tertiary indicators: economy (C11, overall price level), price reasonableness (C12, alignment between price and quality), cost-effectiveness (C13, comparative advantage of performance relative to price), and appreciation potential (C14, value preservation and growth prospects). The emotional value dimension (B2) comprises seven indicators: satisfaction with floor area and layout (C21), overall design satisfaction (C22), location satisfaction (C23), salesperson professionalism (C24), sales environment atmosphere (C25), purchase experience (C26), and show flat quality (C27). The social value dimension (B3) comprises seven indicators: brand awareness (C31), social status enhancement (C32), identity recognition (C33), social acceptance (C34), family support (C35), resale priority (C36), and school district housing priority (C37). The quality value dimension (B4) comprises four indicators: living convenience (C41), quality-price superiority (C42), brand trust (C43), and quality of life improvement (C44).

The assignment of tertiary indicators to criteria dimensions followed the conceptual definitions of the four value dimensions rather than arbitrary assumptions. Quality-price superiority was classified under quality value because it captures the quality performance delivered per unit of price, which is fundamentally a judgment about quality rather than about price level itself. Location satisfaction was classified under emotional value because it measures the satisfaction evoked by the residential context, an affective response that falls within the definition of emotional value. Brand trust was classified under quality value because it operates as a quality signal that reduces uncertainty about otherwise unobservable product quality. These allocations constitute a theory-based classification principle grounded in the construct definitions of the perceived value framework (Sweeney and Soutar, 2001).

5.2 Computational model

The AHP computational procedure proceeded through three stages: First, pairwise comparison matrices were constructed using the 1–9 importance comparison scale (Table 12). Expert Panel Selection and Judgment Aggregation: Twenty experienced real estate professionals were purposively selected based on three criteria: (a) minimum 5 years of direct industry experience in real estate marketing, customer research, or land investment analysis; (b) current employment at recognized real estate firms or government urban planning agencies in China's first- and second-tier cities; and (c) demonstrated expertise in residential property market assessment. The panel composition included marketing specialists (n = 8), customer research experts (n = 6), and land investment analysts (n = 6), ensuring disciplinary diversity that captures multiple professional perspectives on perceived value. This composition is intentionally supply-side: because the AHP index system is designed to inform enterprise resource allocation, practitioners who routinely make such allocation decisions constitute the most relevant judgment source. The implications of this professional framing for the interpretation of the weight results are discussed together with the weight findings. Experts were invited through professional networks and industry associations, and participation was voluntary with no compensation, reducing incentives for perfunctory responses. Each expert independently completed pairwise comparison matrices for both criteria-level and indicator-level comparisons using the 1–9 importance comparison scale (Table 12). To aggregate the 20 individual judgment matrices into a single group matrix, the geometric mean method was applied to each matrix cell: a_ij(group) = (Product of a_ij(k) for k=1 to 20)⁽1/20), where a_ij(k) represents expert k's comparison of element i to element j. The geometric mean is the standard aggregation technique in AHP because it preserves the reciprocal property of pairwise matrices and accommodates divergent expert opinions without extreme values dominating the consensus (Saaty, 1980). All aggregated matrices satisfied the reciprocity condition (a_ij = 1/a_ji for i not equal to j; a_ii = 1).

Table 12. AHP pairwise comparison rating scale.

Second, weight vectors were computed using the geometric mean method: (1) computing the product of elements in each row (W_i = product of a_ij); (2) taking the n-th root of W_i to obtain approximate eigenvector values; and (3) normalizing the eigenvector to obtain criterion weights (w_i) (Equations 1–3).

Third, consistency was assessed using the consistency ratio (CR = CI/RI), where CI = (λ_max - n)/(n - 1) is the consistency index and RI is the random index (Table 13), with RI values taken from Saaty (1980). A CR value below 0.10 indicates acceptable consistency; otherwise, the matrix requires revision.

Table 13. RI values for random matrices of orders 1 through 7.

Source: Saaty (1980).

W_i=∏j=1naij,i=1,2,…,n(1)

Approximate eigenvector values:W_i=(∏j=1naij)1n,

i=1,2,…,n(2)

Criterion weights:W_i=(∏j=1naij)1n∑k=1n(∏j=1naij)1n,

i=1,2,…,n(3)

5.3 Weight results

Table 14 presents the criteria-level comparison matrix and consistency test results. The expert judgment matrix yielded a maximum eigenvalue of 4.1195; accordingly, CI = (4.1195 – 4)/(4 – 1) = 0.0398, and CR = CI/RI = 0.0398/0.89 = 0.045, below the 0.10 threshold, confirming acceptable consistency. The dimension-level weights were: price value (0.0962), emotional value (0.2533), social value (0.1791), and quality value (0.4714). Quality value commanded the highest weight (47.14%), indicating that the expert panel prioritized product quality as the most important criterion of consumer perceived value. Emotional value ranked second (25.33%), suggesting that the panel also assigned substantial weight to emotional experience and location amenities. Social value ranked third (17.91%), and price value exhibited the lowest weight (9.62%), confirming that price, while relevant, was not the primary consideration in the expert panel's judgments and that the expert panel judged quality-related attributes to carry the greatest decision weight.

Table 14. Criteria-level comparison matrix and consistency test results.

Table 15 presents the consistency test results for all five pairwise comparison matrices. The criteria-level matrix (PPV, PEV, PSV, PQV) and all four indicator-level matrices achieved CR values below the 0.10 threshold. The highest CR was observed for the quality value indicator matrix (0.060), which nonetheless remained comfortably within the acceptable range. This somewhat elevated value is substantively interpretable. Several quality value indicators, most notably living convenience, bundle heterogeneous attributes such as transportation access, educational resources, and commercial amenities, and experts reasonably differed in how they weighed these components. Judgments involving brand trust and quality-price superiority likewise require implicit trade-offs between reputational and performance considerations, on which marketing specialists and land investment analysts brought different professional frames: the former tended to privilege brand-related market signals, whereas the latter emphasized locational and product fundamentals. Because the geometric mean aggregation preserves this diversity of judgment while maintaining group-level consistency (Escobar et al., 2004), the slightly higher CR is best read as evidence of genuine and legitimate disagreement on multifaceted indicators rather than of careless responding. At the individual expert level, all 20 experts produced internally consistent judgment matrices (CR range: 0.008–0.087; median = 0.041), confirming that the aggregation procedure combined individually sound judgments into a collectively consistent consensus matrix. These results provide robust evidence that the AHP-derived weights are reliable and that the expert panel's evaluations were systematically coherent. As a robustness check, we recomputed the criteria-level weights after removing each expert in turn; the rank order of the four criteria remained unchanged across all 20 leave-one-out iterations.

Table 15. Consistency test results for all pairwise comparison matrices.

Table 16 presents the comprehensive weights and importance rankings for the indicator-level indices, computed as: comprehensive weight = indicator weight within dimension × dimension weight at the criteria level. The top five indicators by comprehensive weight were: brand trust (0.1809, Rank 1), quality-price superiority (0.1635, Rank 2), location satisfaction (0.0680, Rank 3), living convenience (0.0668, Rank 4), and satisfaction with floor area and layout (0.0616, Rank 5). These results demonstrate that the expert panel prioritized brand trust as the most influential factor in consumer perceived value, reflecting the importance of developer reputation in housing market competition from the supply side. Quality-price superiority ranked second, reinforcing the primacy of product quality over price discounts in the experts' judgments. Location satisfaction, living convenience, and satisfaction with floor area and layout rounded out the top five, highlighting the significance of locational amenities (education, healthcare, transportation) and residential spatial design in shaping value perception.

Table 16. Comprehensive weights and importance rankings for the indicator-level indices.

The evaluation index system and weight results provide quantifiable guidance for real estate enterprises seeking to optimize product design and resource allocation. Consistent with the priorities identified by the expert panel, firms should prioritize investments in brand trust cultivation, product quality enhancement, location amenity development, and layout design optimization to maximize consumer perceived value within constrained resource environments.

It is instructive to compare the AHP-derived importance weights with the regression-derived standardized effect sizes, while recognizing that the two sets of quantities derive from different populations and estimate different things: the regression coefficients capture statistical associations in the consumer survey data, whereas the AHP weights represent the expert panel's judgments about criterion importance. Agreement between their rankings should therefore be read as complementary evidence rather than as statistical cross-validation. At the criteria level, both methods consistently identify quality value as the most influential dimension (AHP weight = 0.4714; regression standardized β = 0.574), followed closely by emotional value (AHP = 0.2533; β = 0.395). This concordance in the top-two rankings across methodologically distinct approaches lends complementary support to the prioritization framework. In contrast, price value exhibits a notable divergence: regression yields a statistically significant yet modest positive effect (β = 0.377, p < 0.01), whereas AHP assigns it the lowest relative weight (0.0962). This discrepancy is theoretically interpretable: although price exerts a detectable aggregate influence on purchase intention, domain experts grounded in strategic resource-allocation practice judge its actionable importance to be comparatively limited. Consequently, price appears to function less as a competitive differentiator and more as a threshold condition or baseline expectation within China's urban housing market. Thus, the AHP and regression results are best understood as complementary: their convergence on the leading dimensions provides mutually reinforcing evidence from distinct perspectives, while their divergence on price value critically refines the interpretation of the statistical results, revealing how empirical associations map onto managerial relevance and contextual nuance.

The supply-side composition of the expert panel deserves explicit consideration when interpreting these weights. Because the 20 experts were real estate practitioners responsible for marketing, customer research, and land investment decisions, their judgments naturally reflect a developer-oriented perspective, and the top ranking of brand trust and quality-price superiority is consistent with that perspective: from the supply side, brand reputation and demonstrable product quality are the principal instruments of market competition. The consumer survey tells a more differentiated story. In the regression results, price value exerted a statistically significant effect on purchase intention, particularly among low-income consumers, yet the expert panel assigned price value the lowest priority. This divergence is not a defect of the design but an informative finding of the complementarity analysis: it suggests that industry practitioners may systematically underestimate the weight that affordability considerations carry in consumers' actual decisions, a gap between supply-side assumptions and demand-side realities that can contribute to pricing mismatches and unsold inventory (Glaeser et al., 2017). At the same time, the supply-side lens is appropriate for the purpose of the AHP module, which is to support enterprise resource allocation; the judgments of practitioners are precisely the judgments on which such decisions currently rest, and making their implicit priorities explicit allows them to be confronted with demand-side evidence. Future research should nevertheless complement this design with multi-stakeholder panels that include consumers, urban planners, and public officials, so that the sensitivity of the weight structure to panel composition can be assessed.

6 Discussion

The findings of this study can be integrated into a coherent associational account of housing purchase intention among consumers in China's higher-tier cities. Within the estimated model, all four perceived value dimensions were positively associated with purchase intention, and quality value carried the largest standardized coefficient, followed by emotional, price, and social value. Perceived risk was negatively associated with purchase intention, and the bootstrap analysis indicated a statistically significant indirect association consistent with the proposed mediation model, although the magnitude of this indirect association was limited and the direct association remained predominant. The moderation analysis further indicated that the association between price value and purchase intention was stronger among low-income consumers, whereas the association between quality value and purchase intention was stronger among high-income consumers. Because the interaction terms added only limited incremental explanatory power (ΔR² = 0.018), these differences are best understood as variation in group-level value prioritization rather than as evidence of strong income-based consumption stratification.

These findings both converge with and extend prior research. The positive value-intention association is consistent with the foundational account of perceived value proposed by Zeithaml (1988) and with the multidimensional operationalization developed by Sweeney and Soutar (2001), and it aligns with evidence that perceived value predicts willingness to buy (Dodds et al., 1991; Grewal et al., 1998) and housing purchase intention in particular (Hoe et al., 2018; Yao et al., 2024). The negative risk-intention association accords with established accounts of perceived risk in high-involvement decisions (Mitchell, 1999; Featherman and Pavlou, 2003; Phamthi et al., 2024) and with recent evidence that the salience of risk dimensions varies across consumer segments (Koay et al., 2023), while the countervailing pattern of value and risk echoes Syamsuar and Witarsyah (2025). Three aspects go beyond replication: the study decomposes value and risk structures in a conventional housing context, documents income-group differences consistent with Jin et al. (2026), and estimates the value-risk indirect channel jointly with dimension-level value associations, a combination rarely examined in prior housing research.

Theoretically, this study is best characterized as an applied integration of perceived value theory and perceived risk theory in the housing consumption context, rather than as an extension of either framework. The theory of planned behavior served as a background framework that motivated the focus on behavioral intention; the canonical antecedents of that theory were not directly measured, and we therefore do not claim to test, refine, or extend it, nor do we propose modifications to value or risk theory. Within this applied scope, the exploratory factor analysis provided preliminary support for the four-dimensional perceived value structure adapted from Sweeney and Soutar (2001) and for the three-dimensional perceived risk structure specific to housing transactions. Because this measurement evidence derives from exploratory procedures applied to a single non-probability sample, it should be regarded as initial. Confirmatory factor analysis or structural equation modeling on independent, preferably probability-based samples is required before these dimensional structures can be considered validated, a point developed further in Section 7.3.

The regression and AHP components estimate different quantities from different populations and should be read as complementary rather than as mutual validation. The regression models characterize statistical associations between value perceptions and purchase intention among 521 consumers, a demand-side source, whereas the AHP weights encode the priority judgments of 20 real estate professionals, a supply-side source. Where the two rankings converge, notably the leading positions of quality and emotional value, each source provides complementary corroboration for the broad ordering of the value dimensions. The divergence on price value is more informative: price value significantly predicted purchase intention in the consumer sample, particularly among low-income consumers, yet it received the lowest expert weight. This pattern suggests that supply-side practitioners may systematically underweight payment-capacity considerations relative to their role in demand-side decisions, a gap between professional assumptions and consumer realities. We therefore interpret the comparison as an informative contrast between two perspectives on housing value, not as statistical cross-validation of either result.

Finally, the contribution of this study to sustainable cities research is indirect and behavioral in nature. Purchase intention data provide a behavioral window on housing demand rather than direct measurements of sustainability outcomes. Price value is conceptually related to perceived affordability but is not a measure of actual housing affordability, which requires income, expenditure, or residual-income indicators (Stone, 2006). Social value captures status and recognition perceptions associated with social identity, not measured social inclusion. Within these limits, the findings offer a behavioral reference point for housing policy discussions under SDG 11: they indicate which value and risk perceptions are associated with purchase intention across income groups in higher-tier Chinese cities, and this evidence can inform how demand-side responses to housing provision are anticipated. Whether these perceptual associations translate into affordability, inclusion, or market-stability outcomes remains an empirical question for future research that measures such outcomes directly.

7 Conclusions and future directions

7.1 Conclusions

This study examined the association between perceived value and housing purchase intention, with perceived risk as a mediator and income level as a potential subgroup differentiator, in the context of the higher-tier cities covered in this study within China's urban residential market. Using survey data from 521 consumers with recent housing purchase experience or intentions, and applying hierarchical multiple regression and AHP methodologies, we arrived at the following conclusions:

First, perceived value is positively associated with housing purchase intention, with variation in the strength of association across the four value dimensions. Within the estimated model, quality value carried the largest standardized coefficient (β = 0.574, p < 0.01), followed by emotional value (β = 0.395, p < 0.01), price value (β = 0.377, p < 0.01), and social value (β = 0.303, p < 0.01); because the predictors are correlated, these coefficients indicate relative associations within this model specification rather than a direct hierarchy of substantive importance. It is important to note that these beta coefficients indicate only modest absolute effects, and the overall model explains a limited portion of purchase intention variance. This finding offers preliminary insights for sustainable urban housing development, as product quality attributes (construction quality, brand reputation, design standards) appear to show relatively larger associations with urban residents' purchase intentions, suggesting that sustainable housing initiatives may benefit from considering quality assurance among multiple improvement areas. Emotional value's second-ranked position reflects the growing premium urban consumers place on livability factors such as purchase experience, community amenities, and lifestyle enhancement, dimensions that are central to the quality-of-life objectives of sustainable urban development (Lin and Dong, 2023).

Second, perceived risk showed a statistically significant indirect association consistent with partial mediation within the proposed model, as supported by PROCESS bootstrap results [indirect effect = 0.108, 95% CI (0.050, 0.167)], with differential indirect associations across the three risk dimensions. Transaction object risk demonstrated the relatively most pronounced indirect association among the three risk dimensions [PROCESS indirect effect = 0.048, 95% CI (0.009, 0.087)], followed by transaction subject risk [indirect effect = 0.032, 95% CI (0.013, 0.052)], and environmental risk [indirect effect = 0.028, 95% CI (0.008, 0.048)]. The absolute magnitude of these indirect effects remains modest, suggesting that the indirect pathway accounts for only a limited portion of the association between perceived value and purchase intention. This finding warrants cautious interpretation given the cross-sectional design, but it is particularly salient for sustainable urban governance, as macro-level policy uncertainty (regulatory changes, credit tightening, property tax prospects) represents the foremost risk concern in relation to urban residents' housing purchase intention. Given the cross-sectional and non-probability nature of the data, municipal governments may wish to consider enhancing policy transparency, engaging in consistent regulatory communication, and mitigating environmental perceived risk through predictable governance frameworks (Phamthi et al., 2024), pending corroboration from longitudinal and probability-sampled studies.

Third, the formal moderation analysis indicates that income group significantly moderates two of the four value-intention associations. The association between price value and purchase intention was significantly stronger among low-income consumers (interaction β = −0.259, p < 0.01; simple slopes: β = 0.416 for the low-income group vs. β = 0.157 for the high-income group), whereas the association between quality value and purchase intention was significantly stronger among high-income consumers (interaction β = 0.184, p < 0.05; simple slopes: β = 0.389 for the low-income group vs. β = 0.573 for the high-income group). The associations of emotional value and social value with purchase intention did not differ significantly across income groups. It should be noted that the four interaction terms jointly added limited incremental explanatory power (ΔR² = 0.018); the results therefore indicate income-group differences in the relative weight of price and quality considerations rather than evidence of strong income-based consumption stratification. These patterns are consistent with the theoretical expectation that budget constraints shape value prioritization, although the specific affordability mechanisms through which such constraints operate were not directly measured. This pattern is also consistent with need-based accounts of consumption (Maslow, 1943), although this interpretation is offered post-hoc and should be treated as heuristic rather than as a tested derivation of the model. Jin et al. (2026) provided related evidence from China's first-tier cities, finding that family income was associated with differential patterns in the relationship between perceived behavioral control and purchase intention. For sustainable cities, these findings suggest the potential value of differentiated housing supply strategies that respond to the distinct consideration profiles of different income groups; whether such strategies would promote social inclusion or reduce urban socioeconomic stratification remains a policy hypothesis that the present study did not test (see SDG 11.1 on adequate, safe, and affordable housing).

Fourth, the AHP-based evaluation index system quantified the relative importance of consumer perceived value components as judged by the expert panel. At the criteria level, the importance ranking was: quality value (0.4714) > emotional value (0.2533) > social value (0.1791) > price value (0.0962). At the indicator level, the top five factors were brand trust (0.1809), quality-price superiority (0.1635), location satisfaction (0.0680), living convenience (0.0668), and satisfaction with floor area and layout (0.0616). These quantified importance rankings provide supply-side expert guidance for resource allocation and strategic prioritization. Notably, the AHP priority order converges with the regression effect-size ranking for the top two dimensions (quality value and emotional value rank first and second in both methods), which indicates a degree of convergence between the consumer-survey and expert-judgment perspectives; because the two methods draw on different populations and estimate different quantities, this agreement is interpreted as complementarity rather than mutual validation. The divergence for price value, statistically significant in regression but lowest-ranked in AHP, further illustrates that statistical significance does not automatically translate into strategic priority, a nuance that would be missed if either method were used alone.

7.2 Contributions

This study offers several contributions to the sustainable cities literature, and these contributions are delimited by the cross-sectional, non-probability design. First, it extends consumer behavior theory to the domain of urban housing consumption, providing exploratory support for the four-dimensional structure of perceived value in a high-stakes housing setting and identifying quality value as the dimension with the largest standardized coefficient in the estimated model. By situating housing purchase intention within the broader framework of sustainable urban development, this study offers a preliminary connection between the consumer psychology literature and the sustainable cities agenda, while recognizing that purchase intention data provide an indirect behavioral perspective on sustainable urban development rather than direct evidence of sustainability outcomes. Second, it examines the indirect association of perceived risk, particularly environmental risk related to policy uncertainty and macroeconomic instability, within the proposed model linking perceived value and purchase intention. This finding highlights a potentially important consideration of governance quality and policy predictability in fostering sustainable housing markets, although the cross-sectional design does not permit causal or temporal claims. Third, it demonstrates through formal interaction testing that income group significantly moderates the associations of price value and quality value with purchase intention, a pattern that is consistent with consumption stratification accounts in urban housing contexts; given the limited incremental explanatory power of the interaction terms (ΔR² = 0.018), this evidence is preliminary rather than confirmatory. These findings offer preliminary insights that may inform inclusive urban housing policies that ensure housing affordability across diverse socioeconomic strata, though causal conclusions about income effects should be deferred pending experimental or longitudinal validation. Fourth, this study combines multi-criteria decision analysis (AHP) with regression-based behavioral modeling as complementary demand-side and supply-side perspectives: the regression results reflect associations in consumer survey data, whereas the AHP weights reflect the structured judgments of real estate professionals. Because these quantities derive from different populations and represent different constructs, agreement between the two rankings is interpreted as complementarity rather than cross-validation. This integration addresses the disconnection between behavioral research that identifies what consumers value and decision-support tools that inform how resources might be allocated. Taken together, the strongest defensible contribution of this study is evidence concerning associations among multidimensional perceived value, perceived risk, purchase intention, and income-group heterogeneity within a predominantly higher-tier Chinese urban sample, supplemented by supply-side expert prioritization.

Recognizing the limitations of the cross-sectional, non-probability design, and acknowledging that this study measures consumer perceptions and purchase intention rather than sustainable urban development outcomes directly, the following implications should be treated as preliminary and subject to empirical validation through longitudinal and experimental research. The implications are directed toward distinct actors operating at different institutional levels. For developers, the findings suggest that product design, construction quality, brand trust cultivation, and location-related amenities figure prominently in consumers' value perceptions. The AHP results indicate that brand trust cultivation, product quality enhancement, and location amenity development were assigned the highest priority by the expert panel; these weights reflect expert judgments about relative importance and do not establish causal investment returns. For municipal authorities and urban planners, the salience of location satisfaction and living convenience points to the potential value of continued provision of transportation, education, healthcare, and related public services and facilities in residential areas. For national policymakers, the negative association between policy-related environmental risk and purchase intention suggests that policy predictability and consistent regulatory communication warrant further investigation; the present findings support additional research on regulatory certainty but cannot establish that increasing policy transparency will causally increase purchase intention. The income-group differences further suggest plausible directions for differentiated housing strategies: for lower-income segments, expanded affordable housing supply, price advantages, and flexible payment terms; for higher-income segments, quality-oriented developments that emphasize livability and community amenities. These directions are suggested by the pattern of associations observed in this study and were not directly tested, and whether such differentiated strategies would reduce socioeconomic stratification is a policy hypothesis rather than an observed result. Two interpretive cautions apply to all of the preceding implications. Perceived price value is not a direct measure of housing affordability, which typically requires indicators such as household income, price-to-income ratios, or residual income, and the social value scale reflects social recognition and status-related perceptions rather than a direct measure of urban social inclusion. Policy interpretations concerning affordability and inclusion should therefore be made with corresponding caution.

7.3 Limitations and future research

This study has several limitations that suggest directions for future research. First, the non-probability sampling approach (convenience and snowball sampling), while yielding a substantial sample, may limit generalizability. The sample was concentrated in first-tier and new first-tier cities (87.0% of respondents resided in Guangzhou, Shenzhen, Beijing, Shanghai, Hangzhou, or Chengdu), where housing prices, market liquidity, and policy environments differ markedly from those in third- and fourth-tier cities; the findings should therefore be interpreted as characterizing higher-tier urban markets. Future research should employ stratified random sampling with broader geographic coverage, including lower-tier cities and county-level markets, to enhance external validity. Second, the cross-sectional design precludes strong causal inferences due to the inability to establish temporal precedence. Consequently, the observed relationships should be interpreted as associational rather than causal. Future studies should adopt longitudinal designs to track the same cohort across multiple time points, enabling more rigorous testing of causal relationships among perceived value, perceived risk, and purchase intention. Third, the explanatory power of the models is modest (Model 3 explains 18.7% of the variance in purchase intention), and some bivariate correlations are small, indicating that the relationships, while statistically significant, account for a limited share of the variance. As discussed in the Results section, such values are common in consumer behavior research (Hair et al., 2019), and the structural determinants of housing demand, including credit policy, household wealth, and market expectations, lie outside the present psychological framework. Readers should therefore treat the findings as identifying meaningful but partial behavioral mechanisms rather than a comprehensive account of housing purchase decisions. Fourth, income was dichotomized at 10,000 CNY for the moderation analysis, which simplifies a continuous variable (MacCallum et al., 2002) and, together with the smaller high-income subgroup (n = 153), limits statistical power. Although the formal interaction test addresses the inferential limitations of a simple subgroup comparison, the significant interactions should be replicated with continuous income measures, including tests of nonlinear moderation, and probability-sampled subgroups. Fifth, although this study employed the PROCESS macro with 5,000 bootstrap samples to assess mediation, the indirect effects remain modest in magnitude and the cross-sectional design precludes causal claims about the temporal ordering of the perceived value–perceived risk–purchase intention sequence. Future research should employ longitudinal designs to establish temporal precedence and cross-validate the mediation pathways identified here. Sixth, the AHP expert panel consisted exclusively of supply-side real estate practitioners. Although this composition matches the decision-support purpose of the index system, the resulting weights reflect a developer-oriented perspective and may differ from the priorities that consumers, urban planners, or public officials would assign; multi-stakeholder panels should be used in future research to test the robustness of the weight structure. Seventh, perceived value, perceived risk, and purchase intention were measured with the same questionnaire, administered to the same respondents at the same time, so common-method variance cannot be ruled out. Harman's single-factor test, reported as a preliminary diagnostic, indicated that the first unrotated factor accounted for 32.6% of the total variance, below the conventional 40% benchmark; however, procedural remedies in the present design were limited, and future research should combine procedural and statistical controls for common-method bias (Podsakoff et al., 2003). Eighth, data were collected through a mixed-mode design combining online recruitment through WeChat and housing forums with offline interception at real estate sales centers and commercial complexes, and no formal assessment of mode effects was conducted; in addition, the analysis included no city-level controls or multilevel specification, although the sampled cities differ substantially in housing prices, regulatory environments, and mortgage conditions. Ninth, the data were collected in late 2022, and the observed associations reflect the market and policy conditions prevailing at that time; given the substantial adjustment that China's housing market has undergone since 2021 and 2022, temporal extrapolation of the present findings should be undertaken with caution. Tenth, the study measured purchase intention rather than completed transactions and therefore cannot establish whether stated intentions translated into actual purchases; this intention-behavior gap is particularly consequential in a high-cost and institutionally constrained market such as housing, and future research should link survey responses to subsequent transactional outcomes.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Dante Di Matteo, University of eCampus, Italy

Reviewed by: Ilkim Markoc, Yıldız Technical University, Türkiye

Khavid Normasyhuri, Lampung University, Indonesia

Junxiao Wang, Nanjing University of Finance and Economics, China