How Does the Built Environment Shape Low-Carbon Consumption in an Energy-Based City? A GIS–SEM Study of Ordos, China

Abstract

Energy-based cities often develop resource-dependent spatial structures that reinforce carbon-intensive daily routines, yet the mechanisms linking neighborhood form to low-carbon consumption remain unclear. This study investigates the core urban area of Ordos, China, by integrating geographic information system (GIS)-derived 5D built-environment indicators with questionnaire data from 825 residents and estimating a structural equation model (SEM) with bootstrap mediation tests. The results show clear dimension-specific effects. Density, land-use mix, and street connectivity have significant positive total effects on low-carbon consumption behavior and retain significant direct effects after the mediators are introduced, indicating partial mediation. By contrast, distance to transit and shopping accessibility operate mainly through the perceived built environment and psychological factors, with non-significant residual direct effects, indicating full mediation. Psychological factors show the strongest direct association with behavior (β = 0.545, p < 0.001), and the perceived built environment also exerts an indirect effect through psychological factors. Overall, the findings indicate that low-carbon transition in energy-based cities depends not only on spatial upgrading, but also on neighborhood environments that enhance perceived convenience and behavioral readiness.

1. Introduction

1.1. Background and Problem Statement

China’s goals of carbon peaking and carbon neutrality have pushed urban research to pay closer attention to the demand side of decarbonization. Low-carbon consumption behavior refers to routine choices—such as travel modes, energy-saving purchases, waste reduction practices, and the use of green services—that collectively shape urban emissions.

The built environment structures the opportunity set for these choices by shaping the cost, convenience, and everyday experience of low-carbon options. Evidence from compact, transit-oriented metropolitan areas suggests that neighborhoods with higher density, stronger land-use mix, and better street design are more conducive to low-carbon lifestyles. Whether these relationships operate in the same way in energy-based cities remains unclear. Resource-dependent cities often exhibit dispersed urban form, long travel distances, and high car ownership, which can reinforce carbon lock-in and amplify differences between core districts and peripheral settlements. Clarifying how the built environment affects low-carbon consumption in such contexts is therefore important for translating spatial planning into effective demand-side mitigation.

1.2. Research Objectives

To address this gap, this study investigates the core urban area of Ordos as a representative energy-based city and focuses on three questions. First, do 5D built-environment indicators affect residents’ low-carbon consumption behavior, and do different dimensions operate through distinct pathways? Second, what roles do perceived built environment and psychological factors (attitude, subjective norm, and behavioral intention) play, and is there evidence for a chained mechanism from perception to psychology? Third, which 5D components exhibit partial versus full mediation, and what do these patterns imply for spatial governance in energy-based cities? By integrating objective GIS metrics with survey data and applying bootstrap-based SEM, the study aims to clarify the built environment–perception/psychology–behavior mechanism in resource-dependent urban settings.

This study contributes to building and urban-environment research in three ways. First, it clarifies how different dimensions of the 5D built environment operate through distinct behavioral pathways in resource-dependent cities. Second, it demonstrates that objective spatial structure and subjective environmental perception form a serial mechanism shaping low-carbon consumption. Third, it provides spatially differentiated governance implications that inform building-scale and neighborhood-scale design interventions in energy-based cities.

2. Literature Review, Theoretical Framework, and Research Gap

2.1. Built Environment, Perception, and Behavioral Translation

Built environment–behavior research is often interpreted through a quasi-deterministic lens: higher density, stronger land-use mix, and better accessibility are expected to translate directly into more walking and transit use and, by extension, lower emissions [1,2,3,4,5]. A large body of evidence further suggests that land-use and neighborhood form are linked to health and activity outcomes, implying multiple behavioral pathways through which the built environment can shape everyday practices [6,7,8]. In transport research, neighborhood satisfaction and perceived environmental quality are also shown to condition travel behavior responses, indicating that “objective conditions → subjective evaluation → behavior” is a plausible translation process [9].

However, objective spatial attributes do not influence behavior mechanically. Residents respond to environments through subjective perceptions—such as convenience, safety, comfort, and legibility—which shape motivation and the willingness to adopt low-carbon options [10,11,12,13,14]. In particular, micro-scale streetscape quality and pedestrian-oriented design can amplify (or dampen) the behavioral influence of density and mix by affecting perceived walkability and safety [10,11,12,13]. The legibility of urban space and residents’ mental images of their neighborhoods can further influence wayfinding confidence and perceived effort, thereby affecting everyday choice sets [14]. Context also matters in cold climates and long-heating-season environments, thermal comfort constraints and seasonal barriers can raise the perceived cost of outdoor travel and low-carbon mobility, reinforcing car dependence [15,16].

These perception filters are especially salient in energy-based cities. Resource-dependent development can lock in carbon-intensive lifestyles and infrastructures, creating path dependence that makes low-carbon choices feel less feasible or less attractive [17,18]. In coal-energy or resource-oriented cities, spatial structure, service imbalance, and development trajectories may jointly constrain sustainable transitions, requiring both “hard” spatial upgrades and “soft” behavioral governance [19]. Correspondingly, recent empirical work in China has begun to connect built-environment attributes with low-carbon cognition and emission outcomes, including evidence on residential low-carbon cognition, commuting carbon emissions using big data, and multimode public transport emissions considering built-environment factors [20,21,22]. Beyond transport, household time-use patterns also imply that daily routines can meaningfully shift carbon footprints, highlighting the importance of routine consumption decisions as an emissions driver [23].

On the behavioral side, low-carbon consumption behavior refers to routine choices—such as travel modes, energy-saving purchases, waste reduction, and the use of green services—that collectively shape urban emissions [24,25]. A common psychological explanation is the Theory of Planned Behavior (TPB), which posits that attitudes, subjective norms, and perceived control (often operationalized via intention) are proximal determinants of action [26]. Extensions also show that identity and related self-conceptions can strengthen TPB-based explanations for environmental action, while broader pro-environmental behavior is linked with subjective well-being in some contexts [27,28]. In travel and consumption domains, habit and stable context can constrain deliberation, meaning that even strong intentions may not translate into sustained low-carbon practice if habitual patterns dominate [29,30]. Car dependence can therefore reflect both structural constraints and motivational dynamics—simultaneously a “must” driven by urban form and a “lust” reinforced by comfort and status meanings [31]. Social interaction and embeddedness additionally shape both habitual and occasional low-carbon consumption, implying that norms and peer contexts can sustain (or undermine) behavioral change over time [32,33]. Place-based factors such as place attachment can also matter, as stronger local attachment may support participation and engagement in low-carbon community development [34].

Accordingly, this study integrates two complementary perspectives. The first emphasizes the perceived built environment, proposing that objective attributes influence behavior partly through perceived neighborhood quality and convenience [9,10,11,12,13,14]. The second draws on TPB and related pro-environmental behavior research, focusing on psychological determinants of action [24,25,26,27,28]. At the same time, emerging evidence suggests that interventions beyond traditional planning also influence low-carbon practice: environmental education can shape low-carbon behavior through chained psychological mechanisms, virtual or gamified environments can affect environmental awareness and behavior, and digital platforms (e.g., Ant Forest) may nudge low-carbon consumption through app-based stimuli and intelligent technologies [35,36,37,38]. Complementary frameworks such as the (extended) Norm Activation Model have also been validated with SEM and machine-learning cross-checks, reinforcing the role of moral norms and activation processes in predicting low-carbon practice [39]. More broadly, cognitive constraints and “compensation” reasoning can explain why people still harm the environment despite pro-environmental intentions, aligning with the need to model psychological frictions explicitly [40]. Related perspectives on distributed cognition further underscore that cognition is situated and can be shaped by environmental artifacts and contexts, offering conceptual support for perception-based mediation mechanisms [41]. At the urban-system level, land use/cover change and ecosystem service valuation research also highlights that built environment changes have sustainability consequences beyond behavior alone, providing an extended ecological context for spatial governance [42]. Finally, recent scholarship has mapped the broader low-carbon behavior research landscape and examined emerging instruments such as personal transportation carbon trading, while policy levers such as transit fare design can shift commuter behavior—each of which supports a more integrated understanding of “space–policy–psychology–behavior” linkages [43,44,45].

Most existing studies have examined the built environment either as a direct predictor of travel-related behavior or as a broad background condition for sustainable lifestyles. Much less is known about whether different 5D dimensions operate through the same behavioral pathway, especially in energy-based cities where spatial inequality, service imbalance, and carbon-intensive routines often coexist. This study addresses that gap by testing whether structurally advantageous dimensions and accessibility-disadvantage dimensions translate into low-carbon consumption through different combinations of direct, perceptual, and psychological effects. The analysis therefore conceptualizes a chained mediation mechanism in which objective 5D built environment indicators influence low-carbon consumption behavior through perceived built environment and psychological factors.

In this study, spatial attributes refer to objective, GIS-derived indicators of neighborhood conditions, including density, land-use mix, street connectivity, distance to transit, and shopping accessibility. Perceived built environment refers to residents’ subjective evaluation of neighborhood quality, convenience, greenery, accessibility, and environmental support for daily life. Psychological factors refer to a TPB-related set of behavioral predispositions, including attitudinal support for low-carbon action, social encouragement, and behavioral readiness. Low-carbon consumption behavior refers to routine household and daily life practices such as low-carbon travel, energy-saving purchases, resource conservation, and green service use. Mediation is defined here as an explanatory process in which objective neighborhood conditions influence behavior through intermediate perceptual and psychological pathways. Accordingly, the chained pathway examined in this study is objective built environment → perceived built environment → psychological factors → low-carbon consumption behavior.

2.2. Why Mechanisms Differ in Energy-Based Cities

Energy-based cities often face carbon lock-in alongside spatial imbalance [17,18,19]. Resource-oriented economic structures can normalize carbon-intensive lifestyles, while low-density and fragmented development increases reliance on private cars and reinforces car-dependent routines [31]. In cold and long-heating-season climates, the comfort and time costs of outdoor travel further discourage low-carbon options and can strengthen perceived barriers to walking, cycling, or transit use [15,16]. Meanwhile, public services and commercial facilities tend to be unevenly distributed between core districts and peripheral towns; in such settings, transit-service design and access constraints can interact with built-environment form and shape behavioral responses [22,45].

Under these conditions, different 5D dimensions may not operate symmetrically. Some dimensions reflect structural advantages concentrated in core areas, including higher density, richer land-use/POI mix, and stronger street connectivity. These attributes can reduce the time–space costs of low-carbon options (e.g., walking, accessing local services, and trip chaining) and improve perceived convenience and comfort, thereby strengthening psychological motivation [1,2,10,11,12,13,20,21]. These dimensions are therefore expected to show both direct and mediated effects, consistent with partial mediation.

Other dimensions capture accessibility disadvantages, notably longer distance to transit and poorer shopping accessibility. In car-dependent settings, such barriers may primarily operate by lowering perceived convenience and feasibility, thereby weakening intention and subsequent behavior [7,8,9,22,45]. After accounting for perceptions and psychological factors, the residual direct effect of objective distance may be limited, suggesting a full-mediation pattern.

2.3. Research Gap and Hypotheses

Based on the above theorizing, the study proposes the following hypotheses (conceptual model in Figure 1).

H1 (Structural-advantage hypothesis).

Density, diversity, and design have positive effects on low-carbon consumption behavior, with both direct effects and indirect effects through perceived built environment and psychological factors (partial mediation).

H2 (Accessibility-disadvantage hypothesis).

Distance to transit and shopping accessibility affect low-carbon consumption behavior primarily through perceived built environment and psychological factors; after mediators are included, the direct effect becomes non-significant (full mediation).

H3 (Dominant-path hypothesis).

Psychological factors exert the strongest direct effect on low-carbon consumption behavior, consistent with the TPB. A synthesis of the literature and these hypotheses is provided in 

Table 1. Synthesis of literature and research hypotheses.

Theme	5D Indicator (Typical)	Expected Sign on Behavior	Mediating Channel Cited	Context/Notes	Key References (Examples)
Structural advantages → Partial mediation	Density/Diversity/Design	+	Perceived neighborhood quality → Psychology (attitude/norm/intent)	Direct opportunities + improved readability/safety/comfort	Ajzen [26]; Cervero & Kockelman [1]; Ewing & Cervero [2]
Accessibility disadvantages → Full mediation	Distance to transit/Shopping time	− (direct), via chain	Negative perceived convenience → Psychology resistance	Direct path often weak/non-sig; chain (Perceived → Psychology) salient	Handy et al. [7]; McCormack & Shiell [8]
Psychology → Behavior strongest path	—	+	Attitude/subjective norms/behavioral intention	TPB core linkage, robust across contexts	Ajzen [26]; Bamberg & Möser [25]

Note: “+” indicates a positive expected effect; “−” indicates a negative expected effect.

.

3. Study Area and Data

3.1. Study Area: Ordos as a Theoretically Relevant Energy-Based City

The core urban area of Ordos (Figure 2) was selected because it provides a theoretically relevant case for examining the built environment–behavior relationship under energy-based urban conditions. First, Ordos is a coal- and resource-dependent city whose development trajectory has been strongly shaped by the energy economy, making it a policy-relevant case for studying low-carbon transition on the demand side. Second, the study area contains substantial intra-urban heterogeneity, including well-serviced central neighborhoods as well as lower-density and less accessible peripheral settlements. Third, this internal variation creates meaningful differences across the 5D dimensions, which makes the area particularly suitable for testing whether built-environment effects operate through distinct direct and mediated pathways. The case is therefore used not as an ideal example of pro-environmental design, but as a representative energy-based urban setting in which spatial inequality and carbon-intensive routines coexist.

3.2. Data Sources, Sample, and GIS–Survey Fusion

Spatial units. The study area is divided into 22 subdistrict/township units (K = 22), which serve as the aggregation level for objective built-environment indicators.

GIS–survey linkage. A GIS–survey fusion design was adopted to combine objective 5D indicators with individual-level questionnaire responses. Survey quotas were allocated to reflect the spatial distribution of housing stock. Each respondent reported their residence location via a dedicated item in the questionnaire, which allowed assignment to one of the 22 units. The corresponding unit-level 5D values were then linked to each individual for SEM estimation.

The final valid sample consisted of 825 respondents. This sample size is adequate for structural equation modeling and is comparatively robust for a study that simultaneously estimates a measurement model, a multi-path structural model, and bootstrap-based mediation effects. It exceeds conventional SEM adequacy thresholds and is comparable to or larger than sample sizes commonly reported in recent built-environment and low-carbon behavior studies. The spatial distribution of respondents across the 22 subdistrict/township units is reported in the Supplementary Materials.

Table 2. Demographic and household profile of respondents (N = 825).

Percentage	n	Category	Characteristic
52.0	429	Male	Gender
48.0	396	Female
5.7	47	Under 18	Age group
40.4	333	18–24
34.5	285	25–34
14.1	116	35–44
5.3	44	45 and above
13.6	112	Junior high school or below	Education
42.2	348	High school/technical secondary school
24.2	200	Junior college
15.2	125	Bachelor’s degree
4.8	40	Master’s degree or above
18.9	156	Below RMB 3000	Monthly income
54.9	453	RMB 3000–5000
17.0	140	RMB 5000–8000
9.2	76	RMB 8000 and above
68.0	561	Owner-occupied housing	Housing tenure
24.0	198	Rental housing
8.0	66	Other

reports the demographic and household profile of the sample. The gender distribution is relatively balanced (52.0% male and 48.0% female). The sample is comparatively young, with 40.4% aged 18–24 and 34.5% aged 25–34. In educational terms, 42.2% completed high school or technical secondary school, 24.2% completed junior college, and 20.0% held a bachelor’s degree or above. Enterprise employees account for the largest occupational group (59.5%), followed by self-employed or freelance respondents (25.7%). Monthly income is concentrated in the RMB 3000–5000 range (54.9%), and 68.0% of respondents live in owner-occupied housing. Overall, the sample shows variation across the key socioeconomic characteristics relevant to low-carbon consumption behavior.

Computation and preprocessing of objective indicators. Objective indicators were first computed on a 500 m grid covering the study area. The grid was then overlaid with subdistrict/township boundaries, and the median of grid values within each boundary was used as the unit-level indicator. Using medians helps reduce sensitivity to local extremes and measurement noise. All 5D indicators were standardized (z-scores) prior to modeling. For distance-based measures (transit distance and shopping accessibility), smaller values indicate better accessibility; for density, mix, and connectivity, larger values indicate stronger structural advantages.

3.3. Variable Measurement and Indicator Selection

The five objective indicators were selected according to the classic 5D framework because each captures a distinct way in which neighborhood conditions can enable or constrain low-carbon daily behavior. Density reflects the concentration of population and activities; diversity captures the mix of destinations and functions; design represents street-network connectivity; distance to transit measures the effort required to reach public transport; and destination accessibility reflects the convenience of reaching everyday shopping opportunities. Together, these indicators distinguish structural advantages from accessibility disadvantages and are therefore suitable for testing whether different dimensions operate through different direct and mediated pathways.

The 5D built-environment indicators were operationalized as follows: density (population density), diversity (POI/land-use mix measured by an entropy-style index), design (road-network or intersection density as a proxy for street connectivity), distance to transit (distance or network time from unit centroids to the nearest bus stop), and destination accessibility (time-based accessibility to daily shopping facilities). The spatial patterns of these indicators and residential clusters are summarized in Figure 3.

Table 3. Operationalization of 5D built environment indicators.

Dimension	Operational Definition	Unit	Direction (Higher = …)	Spatial Unit	Data Year/Source	Notes
Density	Population per built-up area	persons/km2	better	Subdistrict/township unit	Local stats + GIS grid	Z-standardized prior to merge
Diversity	POI entropy index (0–1)	index	more mixed (better)	Subdistrict/township unit	POI providers/OSM + own calc	Higher = more land-use mix
Design	Intersection density/network density	junctions/km2 or km/km2	more connected (better)	Subdistrict/township unit	OSM road network	Proxy for street connectivity
Distance to transit	Centroid-to-nearest bus stop (network/Euclidean)	m or min	worse	Subdistrict/township unit	Transit stops (GTFS/OSM)	Higher = farther/longer; direction unified before z
Destination accessibility (shopping)	Time to nearest daily shopping facility	min	worse	Subdistrict/township unit	POI + network time	Higher = longer; direction unified before z

provides definitions, units, directions, spatial units, and data sources for each indicator.

4. Methodology

As shown in Figure 4, the analysis proceeded in four linked steps. First, GIS data were used to construct objective 5D indicators for the study area using ArcGIS 10.8 (Esri, Redlands, CA, USA). Second, questionnaire data were collected to measure perceived built environment, psychological factors, low-carbon consumption behavior, and control variables. Third, the GIS indicators and survey responses were linked at the subdistrict/township level through the GIS–survey fusion procedure. Fourth, SEM and bootstrap mediation tests were used to estimate direct effects, indirect effects, and the serial pathway from objective spatial conditions to behavior through perception and psychological factors.

4.1. Workflow and SEM-Based Analytical Strategy

Data preparation and descriptive statistics were conducted using SPSS 29.0 (IBM Corp., Armonk, NY, USA). The analysis employed structural equation modeling (SEM) via AMOS 29.0 (IBM Corp., Armonk, NY, USA) to estimate (i) the measurement model for latent constructs and (ii) the structural relationships among objective 5D indicators, perceived built environment, psychological factors, and low-carbon consumption behavior. SEM is appropriate here because the mediators and outcome are latent variables measured by multiple Likert-scale items, and because SEM allows direct and multiple indirect effects to be estimated within a single coherent system. To make inference on mediated effects without relying on normality assumptions, nonparametric bootstrapping with 5000 resamples was used, and percentile 95% confidence intervals are reported.

4.2. Model Specification

In the structural model, perceived built environment, psychological factors, and low-carbon consumption behavior were specified as endogenous latent variables. The five objective 5D indicators entered the model as exogenous predictors. Demographic characteristics were included as controls to account for systematic differences in behavior and attitudes across individuals. All reported path coefficients are standardized to facilitate comparison across indicators and constructs.

4.3. Bootstrap Inference and Effect Decomposition

Mediation was evaluated using bootstrap percentile 95% confidence intervals: an indirect effect is considered significant if its confidence interval excludes zero. For each 5D indicator, the total effect was decomposed into four components: (a) a direct effect on behavior, (b) an indirect effect via perceived built environment, (c) an indirect effect via psychological factors, and (d) a chained indirect effect via perceived built environment → psychological factors. Mediation is classified as “full” when the direct effect is not statistically distinguishable from zero while at least one indirect component is significant; otherwise, the pattern is classified as “partial”.

5. Results

5.1. Descriptive Statistics and Correlations

Descriptive statistics and bivariate correlations were first examined, followed by evaluation of the measurement model using confirmatory factor analysis. Internal consistency and convergent validity were satisfactory across constructs, with high Cronbach’s α and composite reliability, and AVE values exceeding common thresholds. Discriminant validity was also supported by HTMT values below strict cutoffs. Correlation patterns are consistent with the hypothesized chain: perceived built environment is positively associated with psychological factors and with low-carbon consumption behavior, and psychological factors exhibit the strongest association with behavior (Figure 5).

Descriptive statistics indicate that the three latent constructs all fall within a moderate range on the five-point scale, with mean values of 3.68 for perceived built environment, 3.66 for psychological factors, and 3.64 for low-carbon consumption behavior. This pattern suggests that residents in the study area already report some low-carbon practices, but the overall behavioral level remains moderate rather than consistently strong. The current baseline therefore reflects partial but incomplete alignment between neighborhood conditions, behavioral readiness, and low-carbon daily routines.

Table 4. Item-level descriptive statistics for the eight low-carbon consumption behavior indicators (N = 825).

Agree/Strongly Agree (%)	SD	Mean	Behavior Indicator	Item
58.2	1.05	3.64	Walk/cycle/public transport for daily trips	Q9
61.1	0.97	3.69	Buy energy-efficient home appliances	Q10
57.3	1.01	3.62	Turn off lights/AC and set temperatures reasonably	Q11
59.4	1.05	3.64	Buy locally produced goods	Q12
58.7	1.02	3.64	Reduce disposable products	Q13
57.8	1.02	3.63	Participate in waste sorting	Q14
58.4	0.98	3.65	Choose eco-friendly or reusable products	Q15
58.4	1.03	3.62	Use household energy-saving devices	Q27

provides item-level descriptive statistics for the eight low-carbon consumption behavior indicators, which helps establish the current behavioral baseline more explicitly.

Among the eight behaviors, prioritizing energy-efficient home appliances shows the highest average score (Q10: mean = 3.69, 61.1% agreeing or strongly agreeing), whereas routinely turning off lights or air conditioning and setting temperatures reasonably shows the lowest average score (Q11: mean = 3.62, 57.3% agreeing or strongly agreeing), followed closely by household use of energy-saving devices (Q27: mean = 3.62). The narrow spread of item means (3.62–3.69) indicates that low-carbon consumption in the study area is present across multiple domains, but remains moderate rather than uniformly strong. In substantive terms, the baseline reflects partial adoption of low-carbon routines in mobility, purchasing, reuse, waste sorting, and household energy-saving practices, leaving clear room for improvement.

5.2. Decomposition of Effects and Hypothesis Testing

Table 5. SEM-based bootstrap (5000 resamples; percentile 95% CI) decomposition of total, direct, and indirect effects (including chained mediation).

Objective BE Indicator	Total Effect β	95% CI	Direct Effect β	95% CI	Indirect via Perception β	95% CI	Indirect via Psychology β	95% CI	Chain Indirect (Perception → Psychology) β	95% CI	Total Indirect β	Mediation Type
Distance to transit (bus distance)	−0.112	[−0.277, −0.005]	−0.041	[−0.148, 0.045]	−0.055	[−0.101, −0.025]	0.020	[−0.066, 0.095]	−0.037	[−0.068, −0.017]	−0.072	Full
Design (street connectivity/road network density)	0.391	[0.308, 0.463]	0.106	[0.029, 0.176]	0.051	[0.027, 0.075]	0.200	[0.134, 0.276]	0.034	[0.016, 0.054]	0.285	Partial
Destination accessibility (shopping accessibility)	−0.167	[−0.303, −0.006]	−0.058	[−0.160, 0.056]	−0.027	[−0.054, 0.000]	−0.065	[−0.151, 0.025]	−0.018	[−0.038, −0.0002]	−0.110	Full
Density (population density)	0.256	[0.199, 0.317]	0.127	[0.074, 0.185]	0.033	[0.017, 0.051]	0.074	[0.028, 0.118]	0.022	[0.011, 0.036]	0.129	Partial
Diversity (land-use mix/POI mix)	0.481	[0.408, 0.564]	0.162	[0.087, 0.240]	0.080	[0.046, 0.118]	0.185	[0.126, 0.249]	0.054	[0.028, 0.083]	0.319	Partial

presents the decomposition of the total, direct, and indirect effects based on the SEM bootstrap analysis.

H1 is supported (structural advantages = partial mediation): as illustrated in the analysis of total effects (Figure 6), density (population density), diversity (POI mix), and design (road network density) show significant positive total effects on low-carbon consumption behavior. Furthermore, the pathway decomposition (Figure 7) reveals that these total effects consist of both direct and indirect components.

H2 is supported (accessibility disadvantages = full mediation): the direct effects of distance to transit and shopping accessibility are non-significant (the 95% CI of the direct effect crosses zero), but their indirect effects through perceived built environment and psychological factors are significant, indicating that negative impacts are mainly transmitted via mediated pathways.

H3 is supported (psychological pathway dominance): in the structural model, the standardized coefficient from psychological factors to low-carbon consumption behavior is the largest (β = 0.545, p < 0.001), exceeding the direct effect of perceived built environment on behavior (β = 0.170, p < 0.001). Perceived built environment also affects behavior indirectly via psychological factors (β = 0.209, p < 0.001), confirming the perception-to-psychology pathway.

Overall, the decomposition results (Figure 7) indicate that built-environment effects in Ordos are dimension-specific rather than uniform. Density, land-use mix, and street connectivity show partial mediation, which means that they influence low-carbon consumption behavior both by creating direct spatial opportunities and by improving residents’ experience of neighborhood convenience and behavioral readiness. By contrast, distance to transit and shopping accessibility show full mediation, which indicates that their effects are transmitted mainly through perceived feasibility and psychological support rather than through a residual direct pathway. These findings help explain why built-environment interventions in energy-based cities cannot be treated as one-size-fits-all: some dimensions work as structural enablers, whereas others function primarily as barriers whose behavioral consequences are filtered through perception and motivation.

In Figure 8, Solid arrows indicate statistically significant paths, and dashed arrows indicate non-significant paths. The model includes direct effects of the five built-environment dimensions (5D) on low-carbon consumption behavior, indirect paths via perceived built environment and psychological factors, and the serial path from perceived built environment to psychological factors. Coefficients shown are standardized estimates from the final SEM specification (sample size: N = 825). Significance levels: *** p < 0.001, ** p < 0.01, * p < 0.05, ns p ≥ 0.05.

Figure 8 provides a compact visualization of the estimated structural paths. To ensure robustness of the chained-mediation decomposition, additional multicollinearity and common-method bias diagnostics were conducted. VIF diagnostics and measurement-quality checks indicate that the core conclusions are not driven by multicollinearity or measurement artifacts (

Table 6. Robustness checks and diagnostics checklist.

Diagnostic/Robustness Check	Rule/Threshold	Statistic	Conclusion
Internal consistency (Cronbach’s α)	α ≥ 0.70	Perceived = 0.919; Psychology = 0.895; Behavior = 0.937	Pass
Convergent validity (CR, AVE)	CR ≥ 0.70; AVE ≥ 0.50	Perceived: CR = 0.919, AVE = 0.587; Psychology: CR = 0.895, AVE = 0.631; Behavior: CR = 0.937, AVE = 0.651	Pass
Discriminant validity (HTMT)	HTMT < 0.85 (strict)/0.90 (lenient)	HTMT: Perc–Psych = 0.501; Perc–Beh = 0.601; Psych–Beh = 0.722	Pass (<0.85)
Multicollinearity (VIF) among 5D predictors	VIF < 10 (common)/<5 (conservative)	VIFs (bus distance, road density, shopping accessibility, population density, POI mix) = 5.29, 2.40, 5.01, 1.81, 1.73 (max = 5.29)	Acceptable (<10); borderline under strict < 5
Measurement model fit (CFA)	χ2/df < 5; CFI/TLI > 0.90; RMSEA < 0.08; (GFI/AGFI > 0.90)	χ2/df = 1.129; GFI = 0.977; AGFI = 0.971; RMSEA = 0.013; TLI = 0.998; CFI = 0.998; NFI = 0.982; IFI = 0.998	Good fit
Bootstrap mediation inference	percentile 95% CI; indirect significant if CI excludes 0	5000 resamples; percentile 95% CI; N = 825 (see Table 5 for effects)	Implemented

). Results are summarized in Table 6.

6. Discussion

6.1. Translating the SEM Results into Neighborhood Design Conditions

The results suggest that the effects of the 5D built environment on low-carbon consumption behavior should be interpreted in neighborhood design terms rather than as a single undifferentiated built-environment effect. The three structural-advantage dimensions (density, diversity, and design/street connectivity) show partial mediation, indicating that they influence behavior through both direct and indirect pathways. In practice, this means that neighborhood form affects low-carbon behavior not only by shaping objective opportunities, but also by improving how residents perceive and use everyday environments.

A design-oriented reading of the results is especially important. As shown in Table 5, diversity and design (street connectivity) play particularly strong roles, alongside density. This pattern suggests that promoting low-carbon consumption in an energy-based city is not simply a matter of increasing density. Rather, service clustering, mixed-use spatial layout, and connected street networks appear to be key neighborhood conditions that make low-carbon choices more feasible and more attractive in daily life. In this sense, the usability of neighborhood structure is at least as important as compactness alone.

The partial mediation pattern for density, diversity, and design also implies two complementary channels. First, these dimensions create direct spatial opportunities for walking, short-distance trips, and trip chaining by improving proximity and route options. Second, they improve perceived convenience, comfort, and controllability, which then enhance psychological readiness and support low-carbon behavior through the perception-to-psychology pathway. This interpretation helps explain why structural advantages retain direct effects while also showing meaningful indirect effects.

By contrast, distance to transit and shopping accessibility show full mediation. Their direct effects become non-significant after the mediators are introduced, while indirect pathways remain meaningful. In design terms, this indicates that distance-based disadvantages do not operate only as objective metric penalties. Instead, they are translated into behavior primarily through perceived difficulty and reduced psychological readiness to choose low-carbon options. Therefore, in peripheral or weakly serviced neighborhoods, the key issue is not only objective distance itself, but also whether neighborhood layout, route clarity, transfer conditions, and service-node configuration make low-carbon options feel practical and manageable in everyday routines.

6.2. Dialogue with Existing Studies: Design Implications for Energy-Based Cities

Compared with evidence from high-density and transit-oriented metropolitan contexts, the present results suggest a stronger role for mediated pathways and a weaker residual direct role for distance-based accessibility variables in this energy-based city setting. This pattern supports a context-sensitive interpretation: in resource-dependent and spatially uneven urban environments, neighborhood-scale design quality may determine whether objective infrastructure is translated into actual behavior.

For a buildings and urban design audience, the key implication is that interventions should not be framed only as “more infrastructure” or “more density.” Greater attention should be paid to (i) neighborhood structure, (ii) the spatial layout of daily destinations, (iii) street-network connectivity and route continuity, (iv) clustering of basic services around everyday anchors, and (v) urban design quality that improves perceived comfort, safety, and legibility. These neighborhood conditions help explain why structural-advantage dimensions (especially diversity and connectivity) show both direct and indirect effects, while accessibility-disadvantage dimensions operate mainly through perception and psychology.

This perspective also refines the interpretation of core-peripheral differences in energy-based cities. Core-area advantages are not simply the result of macro compactness; they are more plausibly related to finer-grained neighborhood form, including better connected blocks and streets, denser service clustering, and higher-quality everyday walking environments. Conversely, peripheral disadvantages are not reducible to longer travel distances alone; they also reflect weaker route integration, more dispersed service layouts, and lower perceived usability of low-carbon options in daily life.

The main contribution of this study is not simply to confirm that the built environment matters, but to show that different 5D dimensions operate through different behavioral mechanisms in an energy-based city. Previous studies in compact, transit-oriented metropolitan settings often emphasize aggregate built-environment advantages, whereas the present findings show that, in a resource-dependent urban context, some dimensions retain direct behavioral leverage while others act mainly through perception and psychological readiness. This distinction adds conceptual clarity to the literature by showing that built-environment effects are conditioned not only by urban form itself but also by the way residents interpret and respond to everyday accessibility conditions in a carbon-intensive urban setting.

6.3. Methodological Implications

Methodologically, the findings demonstrate the value of integrating GIS-derived 5D indicators, perceived built environment, and psychological factors within a single SEM framework. This approach does more than identify whether the built environment matters; it also clarifies how objective neighborhood conditions are translated into behavior through perception and psychological mechanisms.

In particular, decomposing effects into direct, indirect, and chained pathways helps distinguish design levers that operate mainly through objective opportunity structures (e.g., connectivity and service-rich layouts) from those that depend more strongly on perception and motivation (e.g., distance-sensitive accessibility conditions). This GIS-SEM approach therefore provides a useful bridge between urban form indicators and actionable neighborhood-scale intervention priorities.

The present findings should be interpreted as evidence about behavioral tendency rather than as a direct measurement of household carbon emissions. The study evaluates self-reported low-carbon consumption behavior and therefore does not estimate a quantified reduction in carbon footprint per respondent. Its practical significance lies in identifying which neighborhood conditions are more likely to support low-carbon daily practices and in clarifying where design and planning interventions may produce the strongest behavioral leverage in energy-based cities.

6.4. Limitations

This study has several limitations. First, objective 5D indicators were aggregated at the subdistrict/township level, whereas perceptions and behaviors were measured at the individual level. Although grid-based computation and median aggregation were used to reduce noise, future studies could apply multilevel SEM or cluster-robust approaches when larger and more balanced within-unit samples are available.

Second, the analysis focuses on a single energy-based city. Validation in other resource-dependent cities with different climatic and urban-form conditions would strengthen external validity.

Third, behavioral outcomes are self-reported and may involve recall bias or social-desirability bias. Future studies could combine surveys with behavioral traces (e.g., mobility logs or consumption records) to improve behavioral measurement and causal inference.

Fourth, the current “design” dimension is operationalized mainly through road-network/intersection density as a proxy for street connectivity. While appropriate within a 5D framework, this proxy does not fully capture micro-scale urban design quality, such as sidewalk continuity and width, crossing safety, frontage activity, shading/shelter, winter wind comfort, and streetscape legibility. Incorporating these finer-grained design-quality indicators would improve explanation of how neighborhood environments are perceived and used in everyday low-carbon decision-making.

7. Design and Planning Implications for Neighborhood-Scale Low-Carbon Consumption

Building on the differentiated mechanisms identified above, design and planning interventions in energy-based cities should be matched to the dominant pathway for each type of built-environment condition. Specifically, accessibility-disadvantage dimensions (full mediation) require interventions that reduce both objective access burdens and perceived difficulty, whereas structural-advantage dimensions (partial mediation) require neighborhood design upgrades that convert existing spatial strengths into consistently high-quality everyday experiences. This framing emphasizes actionable neighborhood-scale design levers rather than macro-level governance rhetoric.

7.1. Peripheral Neighborhood Accessibility and Service-Node Design (For Full-Mediation Barriers)

Because distance to transit and shopping accessibility operate mainly through mediated pathways, interventions in peripheral and low-density areas should focus on making low-carbon options easier to understand, access, and trust in daily routines, rather than only increasing nominal service supply.

(1)

Demand-responsive transit (DRT) embedded in neighborhood structure: In low-density peripheral towns, app-based on-demand minibuses can complement fixed-route services, but their effectiveness depends on spatial integration. Priority should be given to stop placement near neighborhood anchors (e.g., community centers, schools, clinics, and local retail clusters), short and legible walking approaches, sheltered waiting areas, and clear transfer points. These design features can reduce both actual and perceived last-mile burden.

(2)

Hybrid service nodes and service clustering: Co-locating basic retail, public services, and pickup/drop-off functions around key facilities can reduce multi-purpose trip costs and support trip chaining. From a spatial-layout perspective, this implies designing neighborhood-scale nodes with sufficient service mix and walkable catchments, rather than distributing single-purpose facilities in a fragmented pattern. In contexts with weak shopping accessibility, service clustering is likely to improve perceived convenience more effectively than isolated facility additions.

(3)

Pilot-based corridor and node retrofits: Under fiscal constraints, energy-based cities can prioritize small-scale upgrades around selected corridors and neighborhood nodes (e.g., access routes to transit stops, community-service clusters, and local centers). Pilot-first implementation allows planners to test how changes in route continuity, node usability, and layout quality affect perceived accessibility and behavior before wider rollout.

7.2. Core-Area Connectivity and Urban Design Quality Upgrades (For Partial-Mediation Advantages)

For structural-advantage dimensions (especially diversity and street connectivity), the findings indicate that objective spatial strengths do not automatically translate into full behavioral benefits. Their effects are stronger when residents experience these environments as convenient, safe, comfortable, and easy to use. Core-area interventions should therefore focus on urban design quality and the performance of everyday neighborhood layouts.

(1)

Fifteen-minute living circles through layout optimization: In core and semi-core districts, the priority is not only to maintain relatively high density, but also to optimize the spatial arrangement of daily destinations so that residents can complete routine errands through short, connected trips. This includes improving the placement and complementarity of grocery, public service, leisure, and community facilities within walkable catchments.

(2)

Connectivity quality, not connectivity quantity alone: Since street connectivity shows a strong total effect, network improvements should focus on route continuity and intersection permeability while also addressing usability. In practice, this means reducing detours, improving crossing quality, strengthening sidewalk continuity, and minimizing barriers that make connected networks difficult to navigate on foot.

(3)

Urban design quality to support perceived walkability: The mediation results indicate that perception remains important even in structurally advantaged areas. Core-area upgrades should therefore include comfort- and legibility-oriented design elements, such as pavement continuity, lighting, wayfinding cues, shaded/sheltered walking and waiting spaces, and climate-responsive design (including wind and winter comfort where relevant). These measures help convert objective connectivity and service density into stable low-carbon routines.

7.3. Perception-Supportive Soft Interventions Embedded in Everyday Spaces

The significant role of psychological factors and the perception-to-psychology pathway indicates that soft interventions remain important, but they are likely to be most effective when embedded in well-designed everyday spaces rather than implemented as stand-alone campaigns. In other words, communication and behavioral prompts should complement neighborhood design, not substitute for it.

Low-carbon cues (e.g., green-default options, simple social-norm prompts, and neighborhood-level visual feedback on participation) can be placed at transit nodes, community facilities, and service clusters where residents repeatedly make daily decisions. Time-limited events (e.g., car-free days or community challenges) may also create low-friction trials of low-carbon behavior; however, persistence is more likely when these trials occur in neighborhoods with improved accessibility, connected routes, and visible service convenience. This reinforces the central implication of the SEM results: neighborhood design quality and behavioral support mechanisms should be coordinated at the same spatial touchpoints.

8. Conclusions

This study examined how the built environment shapes residents’ low-carbon consumption behavior in an energy-based city by integrating GIS-derived 5D indicators with perceived built environment, psychological factors, and SEM-based mediation analysis. The empirical results provide support for all three hypotheses. H1 is supported: density, land-use mix, and street connectivity show positive total effects on low-carbon consumption behavior and retain direct effects after the mediators are included, indicating partial mediation. H2 is also supported: distance to transit and shopping accessibility operate mainly through perceived built environment and psychological factors, and their residual direct effects become non-significant after mediation is introduced, indicating full mediation. H3 is supported as well: psychological factors are the most proximate determinant of behavior and show the strongest direct path to low-carbon consumption behavior.

These findings indicate that built-environment effects in energy-based cities are dimension-specific rather than uniform. Structural advantages such as mixed-use layout, connected street networks, and relatively higher density appear to support low-carbon behavior both by creating direct spatial opportunities and by improving residents’ experience of everyday accessibility and convenience. By contrast, distance-based disadvantages act mainly by reducing perceived feasibility and behavioral readiness, which means that physical accessibility barriers are translated into behavior through subjective and psychological pathways.

The results should be interpreted as evidence about behavioral tendency rather than as a direct estimate of household carbon emissions. The study measures self-reported low-carbon consumption behavior rather than kilograms of CO2 reduction. Its practical value therefore lies in identifying which neighborhood conditions are more likely to support low-carbon daily practice and where design and planning interventions may have the greatest behavioral leverage.

For planning and design, the results suggest that low-carbon transition in energy-based cities should not rely on density increase alone. Greater attention should be paid to service clustering, mixed-use spatial arrangement, route continuity, and perception-supportive neighborhood design that makes low-carbon options easier to access, understand, and trust in daily life.

Methodologically, the GIS–SEM framework provides a useful bridge between objective urban form, subjective neighborhood experience, and behavioral response. By separating direct, indirect, and chained pathways, the study helps translate statistical evidence into more targeted neighborhood-scale intervention priorities.

Overall, low-carbon consumption in energy-based cities appears more likely to improve when spatial upgrading and behavior-supportive neighborhood design are coordinated rather than treated as separate policy domains.