Weekday Commuting Costs and Weekend Recreational Mobility Conditions: A U-Shaped Relationship in the Jobs–Housing–Recreation Spatial Structure

Abstract

Weekday commuting and weekend recreation are two mobility domains through which urban spatial structure shapes residents’ well-being and urban functioning, yet direct empirical evidence on how they are related remains limited. This study investigates how weekday commuting costs and weekend recreational mobility conditions are related within a jobs–housing–recreation spatial framework, using individual-level location-based services (LBS) data from the central urban area of Chongqing, China. Generalized additive models reveal a nonlinear and range-dependent commuting–recreation relationship. Distance-based and driving-time specifications provide the main evidence for a U-shaped relationship, whereas transit-time specifications do not clearly reproduce this pattern, reflecting short-distance cost overestimation and spatially shared public-transport constraints rather than realised mobility conditions. From a spatial-configuration perspective, this pattern suggests that work-related and recreational mobility conditions are unevenly combined across residential locations, rather than simply aligned or opposed. It also suggests that relatively favourable commuting and recreational mobility conditions can coexist within some residential contexts. Rather than establishing a universal rule, the Chongqing case provides a testable hypothesis that may be relevant to large cities with uneven and partially aligned employment, housing, transport, and recreational opportunities. The study provides an empirical entry point for integrated spatial-performance diagnosis and future evaluation of alternative jobs–housing–recreation configurations.

1. Introduction

Weekday commuting and weekend recreation represent two major domains of intra-urban mobility through which urban spatial structure shapes residents’ well-being and the functioning of cities. The former captures the most regular and unavoidable work-related travel, while the latter reflects one of the main ways in which residents access leisure, recovery, and social participation outside work routines.

Examining how jobs–housing–recreation spatial structure relates to these two mobility-performance dimensions is important for two reasons. First, both commuting and recreation are core components of urban spatial-structure performance. Commuting has long been central to urban mobility and jobs–housing research [1,2], whereas recreational mobility has received comparatively less attention. Yet weekend recreational mobility matters in its own right, as recreational activities support leisure participation, social interaction, recovery, quality of life, and the attractiveness of large cities [3,4,5]. In large Chinese cities, where congestion increasingly extends from weekday commuting peaks into weekend leisure periods, weekend recreational mobility has also become part of the broader challenge of urban mobility and quality of life. This makes it increasingly necessary to evaluate urban spatial-structure performance beyond a commuting-centred perspective.

Second, these two dimensions are interrelated rather than independent. Recreational opportunities may also matter for commuting performance indirectly through residential choice: if households value access to recreational facilities, open spaces, environmental quality, and other non-work amenities, the spatial distribution of these resources may shape where they live and, in turn, the commuting costs they bear [6,7]. In this sense, recreational conditions are not external to commuting-centred spatial performance; they may condition the effectiveness of jobs–housing balance itself. More broadly, weekday commuting and weekend recreation are jointly shaped by the spatial configuration of jobs, housing, and recreational resources. The key issue is therefore not only how jobs and housing are related, but how jobs, housing, and recreation are jointly configured to affect the combined performance of work-related and non-work mobility.

However, the value of an integrated jobs–housing–recreation perspective depends on whether the relationship between weekday commuting costs and weekend recreational mobility conditions leaves scope for coordinated improvement, which can only be assessed by clarifying its functional form empirically rather than assuming it in advance. In Alonso’s residential location framework [8], households trade off commuting costs against other locational advantages, and subsequent empirical research suggests that such trade-offs may also extend to recreational accessibility [2,3]. At the same time, unequal bidding power may sort some residents into locations with favourable conditions in both domains, while leaving others in locations with unfavourable conditions in both. In real cities, these mechanisms may coexist and interact, producing relationships more complex than either straightforward trade-off or straightforward alignment. What therefore remains unclear is whether this relationship is broadly positive, broadly negative, nonlinear, range-dependent, or shaped by more complex combinations of these tendencies. Clarifying this issue is therefore essential for assessing whether there is scope for jointly improving commuting and recreational mobility performance within a jobs–housing–recreation framework.

Despite its importance, the relationship between weekday commuting costs and weekend recreational mobility conditions has rarely been treated as an explicit empirical question [9], and direct research linking commuting to other activity or travel domains remains limited [10,11,12,13,14]. Relevant insights are instead dispersed across several related strands, each of which illuminates part of the problem without identifying the relationship itself.

A first relevant strand is commuting-centered spatial structure research. This literature is primarily organized around how urban spatial structure shapes commuting performance, with outcomes commonly measured in terms of commuting time, commuting distance, excess commuting, jobs–housing balance, and related indicators of spatial mismatch and accessibility [1,15,16,17,18,19,20]. Its evaluative scope therefore remains centered on spatial-structure performance from a commuting perspective, rather than on how commuting and recreational mobility performance are jointly configured.

A second relevant strand concerns recreation-related access and use, particularly in relation to parks and greenspaces. This literature focuses mainly on the distribution, accessibility, equity, and use of parks and greenspaces, examining how access and visitation vary with transport conditions, perceived quality, population demand, supply–demand balance, and spatial equity, increasingly with human mobility data [21,22,23,24,25,26,27]. Recent work on non-work travel further examines activity types and nonlinear built-environment associations, but remains centered on non-work activity demand rather than weekend recreational mobility conditions [28]. The analytical focus of this strand therefore remains on recreational opportunities, park access, greenspace use, and non-work activity patterns themselves, rather than on their relationship with weekday commuting costs.

A related strand concerns residential location and accessibility trade-off studies. By modelling how households choose among alternative residential locations, this literature inherently addresses trade-offs among housing attributes, commuting costs, neighbourhood conditions, and access to employment, services, open space, and other amenities [6,7]. The most closely related evidence shows that commuting burdens may be weighed against non-work dimensions of residential utility, such as open-space access, leisure-related facilities, or proximity to social contacts, and that residential location may shape both commuting and non-work travel outcomes [7,14,29,30,31]. These studies are closely related, but their analytical focus remains on residential choice and its travel implications, rather than on the relationship between weekday commuting costs and weekend recreational mobility conditions.

A fourth strand concerns built environment–travel behaviour studies and mixed-use planning. This literature examines how urban form and land-use characteristics—such as density, diversity, design, destination accessibility, and transit access—are associated with travel behaviour, and how mixed-use or compact development is expected to enhance proximity, accessibility, functional diversity, and transport sustainability [32,33,34]. Recent reviews further emphasize that mixed-use development is multidimensional and context-dependent, involving transport, social, economic, environmental, and regulatory dimensions [33,34]. Here, lower commuting burdens and better access to recreational or other non-work opportunities are often treated as parallel benefits of more integrated land-use arrangements, rather than as outcomes whose empirical relationship is itself examined directly.

Taken together, these strands provide important but partial foundations for understanding weekday commuting costs and weekend recreational mobility conditions. They show that commuting performance, recreational access and use, residential choice trade-offs, and built-environment effects on travel behaviour are conceptually connected. However, existing studies tend to examine these issues as separate or only loosely connected concerns, rather than treating the commuting–recreation relationship itself as the object of empirical analysis. As a result, limited evidence exists on how weekday commuting costs and weekend recreational mobility conditions are jointly structured within residential space. Whether the two dimensions are positively aligned, subject to trade-offs, or related through nonlinear and range-dependent forms therefore remains an open empirical question.

Against this background, this study examines the individual-level relationship between weekday commuting costs and weekend recreational mobility conditions using large-scale location-based services (LBS) data from the central urban area of Chongqing, China. The analysis is guided by two research questions: (1) What empirical relationship exists between weekday commuting costs and weekend recreational mobility conditions at the individual level, and what functional form does this relationship take across commuting-cost ranges? (2) How can the identified pattern be interpreted in relation to the jobs–housing–recreation structure of urban space, and what implications does it have for integrated spatial diagnosis and future spatial-performance research?

By addressing these questions, this study positions the commuting–recreation relationship as an empirical entry point for integrated spatial-performance research. The underlying idea is that weekday commuting and weekend recreation should be understood not as separate mobility outcomes, but as two related performance expressions of the same jobs–housing–recreation spatial structure. This relationship matters because residential locations simultaneously anchor residents’ work-related mobility costs and their access to weekend recreational opportunities, while recreational resources and other non-work amenities may also shape residential choice and, indirectly, commuting costs. Examining the commuting–recreation relationship therefore provides a way to assess how work-related and recreational mobility costs are jointly configured through the spatial relationships among jobs, housing, and recreation.

The study makes two specific advances. First, it examines whether there is empirical support for considering commuting and recreational mobility conditions jointly within a jobs–housing–recreation framework. It does so by analysing whether weekday commuting costs and weekend recreational mobility conditions are systematically related at the individual level, what functional form this relationship takes, and whether the spatial logic behind the relationship suggests scope for integrated spatial diagnosis and future evaluation of coordinated improvement possibilities. Second, it uses this empirical examination to outline how spatial-performance evaluation can move toward a jobs–housing–recreation perspective. In this perspective, residential locations are examined through the joint configuration of weekday commuting costs and weekend recreational mobility costs. Rather than claiming to identify causal mechanisms or optimization effects, the study points to the need for future research on how alternative jobs–housing–recreation configurations may support more favourable combined mobility conditions.

2. Data and Methods

2.1. Conceptual Definitions

Three concepts are central to the analysis: weekday commuting costs, weekend recreational activities, and weekend recreational mobility conditions.

First, weekday commuting costs refer to the time- and distance-based burdens of regular home–work travel on weekdays. In this study, commuting is treated as a recurrent and relatively fixed domain of daily mobility, structured by the spatial relationship between residence and workplace. As a necessary activity, its cost reflects the routine travel burden imposed by the existing jobs–housing arrangement.

Second, weekend recreational activities refer to non-work, discretionary activities undertaken during weekends for leisure, relaxation, entertainment, or social participation. This definition excludes work-related trips, weekday after-work activities, and non-work stops embedded in commuting chains. The focus is therefore on recreational behaviour as an independent activity domain rather than an extension of weekday work routines.

Third, weekend recreational mobility conditions refer to the realised travel conditions associated with weekend recreational trips. In this study, they are represented by travel distance and travel time between the individual’s residence and identified recreational destinations for home-based weekend recreational trips, rather than by subjective evaluation or potential accessibility alone. The concept is used to capture the realised mobility outcomes of discretionary recreational travel under the existing urban structure.

2.2. Data and Sample Construction

2.2.1. Study Area Selection

The central urban area of Chongqing (Figure 1) is selected as the study area because it provides a suitable metropolitan context for examining the relationship between weekday commuting costs and weekend recreational mobility conditions. This area contains a large population, intensive daily travel demand, and a complex spatial arrangement of residential neighbourhoods, employment centers, commercial facilities, parks, cultural venues, and other recreational destinations. It therefore offers substantial individual-level variation in both home–work commuting conditions and weekend recreational mobility.

The case is also relevant to the broader planning problem addressed in this study. Like many large Chinese metropolitan areas, Chongqing’s central urban area faces mobility pressures that are no longer confined to weekday commuting peaks; congestion increasingly extends into weekends as well. At the same time, weekend recreation is itself an important dimension through which urban space affects residents’ well-being and urban functioning. These features make Chongqing’s central urban area an appropriate case for examining the commuting–recreation relationship at the individual level.

2.2.2. Data Source

Location-based services (LBS) data were used to capture individual-level commuting and recreational mobility. Unlike survey-based data, LBS records capture observed travel patterns rather than self-reported activities. The distributions of residents and workers represented in the dataset are highly consistent with official population census and economic census distributions at the sub-district level in Chongqing, supporting the construction of individual-level travel indicators.

The dataset includes two components: anonymized and de-identified individual-level location records derived from multiple positioning sources, including GPS-based periodic signals and WiFi-based positioning, and anonymized socio-demographic labels, including age group, income level, occupation, and education level, provided by the data vendor in pre-processed form. No direct personal identifiers were available to the authors. These sociodemographic labels are used only as control variables, and their provenance and validation status are discussed in Section 2.4.4. Stable residence and workplace locations were provided by the data vendor as pre-identified baseline attributes inferred from a longer three-month observation window. The specific rules used to define residence and workplace are described in Section 2.2.3.

The records used to construct commuting and recreational indicators cover the period from 30 October 2023 to 3 December 2023. This observation window is concentrated mainly in November, when Chongqing generally has mild weather conditions and weekend out-of-home activities are less likely to be strongly constrained by extreme heat or cold. The period is therefore appropriate for identifying regular weekend recreational mobility during a non-extreme season, including visits related to shopping, cultural services, leisure and entertainment, and tourist attractions. Nevertheless, the observation window cannot fully capture seasonal variation. In particular, summer heat in Chongqing may reduce outdoor-oriented recreational activities and alter weekend destination choices; this limitation is discussed further in the Section 4.

2.2.3. Data Processing and Indicator Construction

To construct individual-level metrics of weekday commuting and weekend recreational mobility, several processing steps are applied.

First, residence and workplace locations are used as vendor-derived stable activity locations. According to the data vendor’s processing rules, residence refers to the dominant nighttime location during 22:00–07:00, while workplace refers to the dominant weekday daytime location during 10:00–17:00. These labels were inferred by the vendor from a longer three-month observation window and are used directly in this study rather than being re-identified by the authors.

Second, weekend recreational visits are identified using an AOI-based approach. Recreational activities are defined as non-work, discretionary activities undertaken on weekends for leisure, relaxation, entertainment, or social participation. The selected AOI categories are used as observable proxies for such activities and include shopping, tourist attractions, leisure and entertainment, and cultural services, as detailed in Appendix A 

Table A1. AOI categories used to identify weekend recreational visits.

Primary AOI Category Used in This Study	Secondary AOI Categories
Shopping	Shopping centre, department store, supermarket, market, home appliance and digital products, shopping service
Tourist attractions	Park, zoo, botanical garden, amusement park, museum, scenic area, tourist attraction, scenic spot, cultural relics/historic sites, church, temple
Leisure and entertainment	Resort, farmhouse recreation, cinema, KTV, theatre, bathing/massage, leisure square, leisure and entertainment venue, game venue
Cultural services/cultural media	Cultural service venue, cultural palace, exhibition hall

. A weekend visit is recorded when an individual’s location points fall within these AOIs during the weekend observation window. To further reduce misclassification, visits to AOIs located within 500 m of the individual’s workplace are excluded. This step helps remove workplace-adjacent observations that may reflect work-related presence, routine stops, or incidental activity rather than independent weekend recreation. In the subsequent construction of mobility indicators, these identified visits are treated as destinations of home-based weekend recreational trips originating from the individual’s residence.

Because a weekend day may involve visits to more than one recreational AOI, an additional person-day diagnostic was conducted to assess the prevalence of potential multi-AOI recreational activity patterns. For each individual on each weekend day, distinct recreational AOIs were counted after duplicate AOI records were removed. The results show that 76.64% of recreational person-days involved only one identified AOI, and 93.64% involved no more than two AOIs (

Table A2. Distribution of distinct recreational AOIs visited per person-day.

Number of Distinct Recreational AOIs per Person-Day	Person-Days	Share (%)	Cumulative Share (%)
1	1,406,190	76.64	76.64
2	311,840	17	93.64
3	78,149	4.26	97.9
≥4	38,577	2.1	100
Total	1,834,756	100	—

Note: The table counts the number of distinct recreational AOIs identified for each individual on each weekend day after duplicate AOI records were removed. The results show that 76.64% of recreational person-days involved only one identified recreational AOI, and 93.64% involved no more than two AOIs. Person-days with three or more AOIs accounted for only 6.36%. This diagnostic reduces, but does not eliminate, concerns that potential multi-AOI activity patterns dominate the construction of the individual-level recreational mobility indicators.

in Appendix A). Person-days involving three or more identified AOIs therefore accounted for only 6.36% of the total. This diagnostic reduces, but does not eliminate, concerns that multi-AOI activity patterns may affect the construction of the individual-level recreational mobility indicators.

Third, route-based metrics are constructed for the identified commuting OD pairs and recreational visit OD pairs. The LBS data allow the identification of activity locations and origin–destination relationships, but do not directly observe realised travel routes, travel time, or travel mode. Route distance and mode-specific travel times are therefore estimated using the Baidu Map navigation API, which provides route-based travel conditions under specified temporal settings. For commuting, route distance as well as driving and transit time are retrieved for each individual’s stable residence–workplace OD pair during weekday peak periods (07:30–09:30 and 17:30–19:30), so as to capture typical commuting conditions. For recreation, route distance as well as driving and transit time are retrieved for each OD pair between the residence and the identified recreational destination during weekend daytime periods (10:30–19:00), which correspond to the main temporal window of recreational activity. Because the actual travel mode of each observed visit cannot be reliably identified from the LBS data, these mode-specific travel times are treated as alternative route-based indicators of mobility conditions rather than direct observations of realised mode choice.

Finally, commuting measures are constructed from each individual’s stable residence–workplace OD pair, whereas recreational mobility measures are constructed from OD pairs between each individual’s residence and identified recreational destinations. When multiple weekend recreational visits are observed for the same individual, visit-level recreational measures are aggregated to the individual level using mean values.

2.2.4. Sample Selection

To ensure that weekday commuting patterns and weekend recreational mobility can be consistently identified at the individual level, several filtering criteria are applied. Individuals without regular commuting are first excluded, including those without stable workplace identification and those whose residence and dominant daytime activity locations coincide within campus-type areas. Such cases do not provide a reliable basis for identifying conventional weekday home–work commuting. In addition, individuals labelled as scenic-area staff in the occupation field provided by the data vendor are excluded, because visits to recreational destinations in such contexts cannot be reliably distinguished from work-related activities.

After these substantive filters are applied, the analysis is further restricted to a complete-case sample in which all variables required for model estimation are observed. To keep the main analysis within the principal empirical range of ordinary intra-urban mobility conditions, observations involving driving-time or transit-time variables greater than 90 min are excluded. In addition, observations with commuting or recreational distance greater than 30 km are excluded from the main analysis. This threshold lies in the extreme upper tail of the sample distribution (approximately the 99th percentile) and is used to limit the analysis to the main empirical range relevant to the present study. These very long-distance observations are few in number and are more likely to reflect exceptional cases than the dominant empirical pattern of interest. Taken together, these steps define the final analytical sample used in the main models.

2.3. Analytical Framework and Strategy

2.3.1. Research Logic

The empirical objective of this study is to identify the relationship between weekday commuting costs and weekend recreational mobility conditions at the individual level, especially its functional form. The analysis asks whether this relationship is broadly positive, broadly negative, or nonlinear and range-dependent. Because both mobility domains are embedded in the broader jobs–housing–recreation spatial structure of the city, the relationship cannot be assumed a priori to follow any simple directional or linear form.

The analysis is designed to estimate a citywide average commuting–recreation relationship among individuals residing in the Chongqing central urban area. It is not designed to estimate separate relationships for predefined intra-urban spatial contexts, such as urban centres, large residential communities, commercial–leisure districts, or urban fringe areas. This distinction is important because the purpose of the study is to examine whether weekday commuting costs and weekend recreational mobility conditions are empirically related within the overall jobs–housing–recreation structure of the city. Spatial heterogeneity across residential contexts is therefore not treated as a nuisance to be fully removed, but as part of the urban spatial structure within which the citywide average relationship emerges.

Accordingly, the methodological task is to estimate this relationship flexibly while preserving the interpretability of its overall shape. The analysis therefore emphasizes functional-form identification rather than threshold estimation, prediction-oriented optimization, or direct modelling of residential location choice. It is also not intended to estimate a de-spatialized individual-level effect by removing residential spatial context, because the residence-anchored spatial context is itself part of the mobility-cost configuration of interest.

2.3.2. GAM-Based Modelling Strategy

To address the task of functional-form identification, this study adopts a semi-parametric analytical strategy based on Generalized Additive Models (GAMs). GAMs allow the association between a continuous commuting variable and a recreational outcome to be estimated through a smooth function, without imposing a pre-specified linear, polynomial, or piecewise form.

For estimation, the recreational outcome variables were transformed using the natural logarithm because they are strictly positive and remain right-skewed after sample restriction. The GAMs were therefore estimated on the log-outcome scale. The main analysis was conducted on the filtered complete-case sample described in Section 2.2.4.

The log-scale GAMs were implemented using LinearGAM in pyGAM 0.10.1. Across all GAM specifications in Section 2.5.1, the commuting variable was entered as a penalized smooth term using pyGAM’s default cubic spline basis with 20 basis functions. This basis dimension was chosen to provide sufficient flexibility for identifying nonlinear functional forms while avoiding an overly restrictive smooth. In penalized GAMs, the basis dimension defines the maximum flexibility available to the smooth term rather than the final complexity of the fitted curve, because the final smoothness is controlled by the smoothing penalty. The resulting effective degrees of freedom (edf) were inspected to confirm that the fitted smooths did not simply exhaust the maximum flexibility allowed by the basis setting.

Income was entered as a linear term, while age group and education level were entered as factor terms, with an intercept included. Section 2.5.1 presents the formal model specification. Smoothing parameters were selected using grid search over 11 logarithmically spaced lambda values from 10⁻³ to 10³, with generalized cross-validation (GCV) as implemented in pyGAM’s gridsearch routine. A logarithmic grid was used because smoothing penalties can vary across orders of magnitude; the selected grid therefore covers a broad range from relatively weak to relatively strong penalization. This procedure was used to select an appropriate smoothing level for functional-form identification, rather than to conduct prediction-oriented hyperparameter optimization. All models were fitted using the full filtered analytical sample, because the objective was functional-form identification rather than predictive benchmarking.

Diagnostic checks were conducted on the model scale using residual-versus-fitted plots, residual histograms, normal Q–Q plots, and calibration plots comparing observed and fitted log outcomes across fitted-value bins. Term-wise effective degrees of freedom and model summaries were also inspected to assess the final estimates and the flexibility of the fitted smooth terms.

Consistent with the study’s analytical objective, the resulting curves are interpreted as estimated conditional functional forms for identifying the commuting–recreation relationship, rather than as individual-level predictions or causal effects of commuting costs on recreational mobility conditions.

2.4. Variables and Measurement

2.4.1. Rationale for Indicator Selection

The indicators used in this study are selected to operationalize residence-anchored mobility cost. In commuting, non-work travel, and urban green-space accessibility research, distance and travel time are commonly used to represent spatial and temporal impedance. Distance and travel time are directly observable and interpretable indicators for comparing origin–destination mobility conditions across individuals. Monetary cost, perceived effort, comfort, safety, and psychological distance may also affect travel behaviour, but these dimensions require additional fare, preference, or survey information that is not available in the LBS dataset used in this study. The present analysis therefore focuses on distance- and time-based impedance indicators.

For weekday commuting, commuting distance and commuting travel time capture the spatial and temporal burden associated with routine work-related mobility. These indicators describe how residential locations are connected to workplaces through the transport system and are commonly used to examine the spatial structure and burden of commuting [35].

For weekend recreation, recreational distance and recreational travel time capture the mobility cost of accessing realised weekend recreational destinations. They are not complete measures of recreational utility, which may also depend on destination quality, activity type, novelty, social arrangements, or household needs. Nevertheless, because weekend recreation is discretionary rather than compulsory, distance and travel time remain relevant constraints on participation frequency and destination selection [27,36].

Based on this logic, the empirical analysis uses both distance- and time-based indicators. Distance provides a relatively mode-neutral measure of spatial separation, while route-based travel time captures network impedance under specific transport conditions. Driving time is used as the primary time-based impedance indicator, and transit time is used in supplementary and transit-constrained specifications to examine how the estimated relationship changes under public-transport impedance.

2.4.2. Dependent Variables

The dependent variables measure home-based weekend recreational mobility conditions at the individual level. Three indicators are used: recreational travel distance, recreational driving time, and recreational transit time. Recreational travel distance captures the spatial extent of weekend recreational activity, while recreational driving time and recreational transit time capture route-based time conditions under driving- and transit-based route options estimated by the Baidu Map API. Because the LBS data do not allow the realised travel mode of each observed recreational visit to be identified directly, all three indicators are retained. Their respective analytical roles are discussed in Section 2.5.2.

2.4.3. Independent Variables

The independent variables measure weekday commuting costs and home–work spatial separation at the individual level. Three indicators are used: commuting driving time, commuting transit time, and commuting distance. Commuting driving time and commuting transit time capture weekday commuting costs, while commuting distance captures the spatial separation between residence and workplace. Their respective analytical roles are discussed in Section 2.5.2.

2.4.4. Control Variables

Control variables are included to account for key observable individual heterogeneity that may affect both weekday commuting costs and weekend recreational mobility conditions. The models include income, age, and education level, with income entered linearly and age and education level specified as categorical variables. This parsimonious control strategy is consistent with the study’s relationship-identification and residence-anchored design. The objective is not to maximize individual-level explanatory power, but to identify the commuting–recreation relationship as it is expressed within the existing urban spatial structure. Extensive spatial-context controls could absorb part of the spatial configuration that the study seeks to diagnose. Likewise, adding a large set of demographic, household, housing, or vehicle-related variables could partially absorb residential sorting and spatial differentiation, rather than merely adjust for individual heterogeneity. The selected controls therefore adjust for key individual differences while preserving the broader spatial configuration of interest.

The provenance and validation status of these control variables also require clarification. Income, age, and education were obtained from the data vendor as sociodemographic labels. Because the underlying classification procedures and source records were not accessible to the authors for privacy and contractual reasons, these labels should be interpreted as vendor-derived control labels rather than independently validated demographic measurements. To assess whether the main finding depends on these labels, the two primary GAM specifications in Section 2.5.1 were re-estimated without income, age, and education controls. The results are reported in Appendix A Figure A1.

2.5. Statistical Identification of Functional Form Using GAM

2.5.1. Model Specification

Because the recreational outcome variables are log-transformed prior to estimation, the general specification can be written as:

$$E\left[\log \left(Y_i\right)\right]=\mathsf{\alpha }+s\left(X_i\right)+{\mathsf{\beta }}_{1}Incom{e}_i+{\mathsf{\gamma }}^{\prime }Ag{e}_i+{\mathsf{\delta }}^{\prime }Educatio{n}_i$$

where $Y_i$ denotes the recreational mobility outcome for individual $i$, $X_i$ denotes the commuting measure of interest, $s\left(\cdot \right)$ is a smooth function estimated from the data, and the remaining terms represent control variables. $Ag{e}_i$ and $Educatio{n}_i$ denote vectors of dummy variables for age group and education, respectively.

In this specification, the commuting variable enters as a smooth term, income is included linearly, and age and education are specified as categorical controls. This allows the commuting variable to vary flexibly while keeping the control structure fixed and comparable across specifications. The specific variable combinations used in the analysis, however, do not carry identical analytical meanings and are therefore assigned different roles in the empirical strategy, as explained in the following subsection.

Because the analysis is residence-anchored, spatial clustering among individuals is not treated simply as a nuisance to be removed. Residents living in nearby areas share transport networks, employment access, housing-market conditions, and recreational opportunities, which are part of the spatial-structure conditions of interest in this study. At the same time, such clustering may reduce the effective independence of individual observations and may affect the standard errors of the smooth terms. The reported statistical significance should therefore be interpreted cautiously. Accordingly, the main interpretation of the results rests on the estimated functional form, its recurrence across the primary specifications, and its spatial interpretability, rather than on p-values alone.

2.5.2. Analytical Roles and Evidential Status of Alternative Specifications

The specifications are organized according to their evidential role rather than treated as interchangeable model variants. Distance provides a relatively mode-neutral measure of spatial separation or travel extent, but it is not a direct measure of travel cost because it does not capture network conditions, congestion, route choice, or travel speed. Time-based indicators are therefore more directly aligned with the cost dimension of mobility.

Among the time-based indicators, driving time is used as the main indicator because it reflects route-based impedance under road-network conditions and provides a consistent impedance measure across origin–destination pairs. It is not interpreted as evidence that all individuals travel by car, but as a route-based time-cost measure that is closely related to feasible motorized travel conditions and relevant to feasible motorized access to the observed destinations.

Transit time is interpreted more cautiously. Because the LBS data do not identify realised travel mode, transit time derived from the Baidu Map navigation API represents the impedance of a transit-based route option rather than observed travel time. For short-distance trips, public-transport travel time typically includes access, waiting, transfer, and route-indirectness components, so it may overstate realised mobility cost and shift otherwise short or low-cost trips out of the lowest transit-time range. For long-distance trips, residents facing high public-transport travel time may switch to other available modes, choose closer recreational destinations, or reduce discretionary recreational trips rather than experience the full transit-time impedance [37,38]. Transit time is therefore interpreted as a public-transport impedance measure, not as evidence of actual public-transport use or realised travel time. This measurement issue is more consequential for transit time than for driving time because transit routes more often involve access, waiting, transfer, and network-indirectness components, and because high transit impedance may induce mode substitution or destination adjustment.

The transit-based specifications are included because public transport constitutes an important mobility option in large Chinese cities. However, given the absence of observed travel-mode information in the LBS data, these specifications are interpreted as supplementary diagnostics based on public-transport impedance, rather than as direct representations of actual transit use or as equivalent substitutes for the primary distance- and driving-time specifications. The specification combining commuting transit time with recreational driving time examines how the curve changes when public-transport impedance is introduced on the commuting side. The specification using transit time on both sides is interpreted more narrowly as the association between two transit-time impedance indicators.

The role of distance differs between commuting and recreation. For commuting, distance mainly represents home–work spatial separation. Because commuting is a necessary, repeated, and destination-fixed activity, its cost is better represented by time-based impedance than by distance alone. For weekend recreation, however, distance is not only a cost-related measure but also an indicator of the realised spatial range of discretionary activity. Recreational distance is therefore treated as a primary outcome, whereas commuting distance is retained as a supplementary spatial counterpart.

Based on these distinctions, the primary specifications are:

recreational distance ~ commuting driving time

recreational driving time ~ commuting driving time

These specifications combine the principal commuting-cost indicator with two complementary recreational outcomes: one capturing the spatial extent of weekend recreation and the other capturing time-based recreational mobility conditions.

A key supplementary spatial specification is:

recreational distance ~ commuting distance

This specification assesses whether the main pattern is also visible when both commuting and recreation are represented in spatial terms.

An additional supplementary specification is:

recreational distance ~ commuting transit time

A separate transit-constrained specification uses transit time on both sides:

recreational transit time ~ commuting transit time

This specification is treated separately because transit-time measures on both sides are more likely to capture shared public-transport network and service constraints than the broader commuting–recreation relationship of interest.

2.5.3. Identification of Nonlinear Form

Nonlinear form is identified on the basis of the smooth-term significance, the effective degrees of freedom (edf), and the overall shape of the estimated smooth curve with its confidence interval. The analysis focuses on broad structural features of the relationship, such as monotonicity, curvature, directional reversals, and range-dependent variation, rather than on minor local fluctuations.

2.5.4. Estimation and Reporting

The GAM results are reported using smooth-term statistics and estimated smooth curves. Tables present smooth-term significance and edf, while figures display the estimated relationship with 95% confidence intervals on the log-transformed outcome scale. Interpretation is descriptive and focuses on the broad empirical form of the relationship. Basic diagnostic checks were conducted to confirm that the fitted models were suitable for functional-form interpretation.

3. Results

3.1. Descriptive Statistics of the Analytical Sample

The final analytical sample contains 654,037 individuals.

Table 1. Descriptive statistics of continuous variables in the analytical sample.

Variable	N	Mean	SD	P25	Median	P75	P90	Max
Recreational driving time (min)	654,037	19.2	9.8	11.7	18.1	25.5	32.7	75.5
Recreational distance (km)	654,037	9.3	6.4	4.1	7.9	13.1	18.8	30.0
Recreational transit time (min)	654,037	46.2	17.6	32.3	44.4	58.5	71.5	90.0
Commuting driving time (min)	654,037	22.0	14.1	10.4	19.4	31.0	42.1	89.8
Commuting distance (km)	654,037	8.39	6.6	3.08	6.6	12.17	18.3	30.0
Commuting transit time (min)	654,037	46.7	19.2	30.6	44.5	61.2	74.8	90.0
Income proxy	654,037	9919.5	6977.9	5146.0	8004.0	12,535.0	18,633.0	94,632.0

reports the distributions of the main continuous variables. Commuting driving time has a median of 19.4 min and a 90th percentile of 42.1 min, while mean recreational driving time is 19.2 min, mean recreational transit time is 46.2 min, and mean recreational distance is 9.3 km. The transit-time variables are much higher than the corresponding driving-time measures, and even their lower ranges remain relatively high, consistent with the interpretation of transit time as public-transport impedance rather than observed travel time. This feature may overstate the realised mobility cost of some short-distance trips and helps explain why the transit-based specifications require separate interpretation. The dispersion of the recreational outcomes is consistent with the use of log-transformed dependent variables in the GAM analysis.

Table 2. Selected characteristics of the analytical sample.

Characteristic	Category	N	%
Age group	20–24	101,415	15.5
	25–29	149,460	22.9
	30–34	158,281	24.2
	35–39	94,885	14.5
	40–44	53,068	8.1
	45–49	58,903	9.0
	50–54	13,075	2.0
	55–59	21,009	3.2
	60 and above	3941	0.6
Education	Below bachelor	177,978	27.2
	Bachelor	299,333	45.8
	Postgraduate and above	176,726	27.0

reports selected socio-demographic characteristics of the analytical sample. Age is distributed across multiple categories, with the largest shares in the 30–34 and 25–29 groups. Education is distributed across three categories, with bachelor’s degree holders accounting for the largest share. These distributions indicate meaningful variation in the main socio-demographic controls used in the analysis.

3.2. Overall Model Statistics and Explanatory Power

Table 3. GAM smooth-term statistics for primary, supplementary, and transit-constrained specifications.

Role	Specification	Edf of Smooth Term	Smooth-Term p-Value	N	Pseudo R2
Primary	Recreational distance ~ Commuting driving time	10.2	<0.001	654,037	0.0158
Primary	Recreational driving time ~ Commuting driving time	10.2	<0.001	654,037	0.0146
Supplementary	Recreational distance ~ Commuting distance	14.1	<0.001	654,037	0.0205
Supplementary	Recreational distance ~ Commuting transit time	11.9	<0.001	654,037	0.0158
Transit-constrained	Recreational transit time ~ Commuting transit time	11.9	<0.001	654,037	0.0274

summarizes the smooth-term statistics and pseudo-R² values for all GAM specifications. The smooth terms are statistically significant across all models, suggesting that the relationships between weekday commuting costs and weekend recreational mobility conditions are not well represented by a simple linear form. However, the pseudo-R² values are modest, ranging from 0.0146 to 0.0274. These values indicate that commuting costs explain only a limited share of individual variation in weekend recreational mobility conditions.

This limited explanatory power is consistent with both the behavioural complexity of weekend recreational mobility and the intentionally parsimonious design of the models. Weekend recreational mobility is shaped by many factors beyond commuting cost, including personal preferences, household responsibilities, weather, activity type, destination quality, vehicle availability, social arrangements, and other constraints. The models include key individual controls but do not attempt to absorb the full set of spatial-context, demographic, household, preference-related, destination-quality, and activity-specific factors that may shape individual recreational mobility. Accordingly, the GAM results should be interpreted as evidence of nonlinear average conditional relationships rather than as highly explanatory or predictive models of individual recreational behaviour. The subsequent interpretation therefore focuses on the estimated curve form, its recurrence across specifications with different evidential status, and the substantive magnitude of the fitted relationships.

3.3. Primary Specifications

3.3.1. Recreational Distance and Commuting Driving Time

Figure 2 presents the estimated smooth effect of commuting driving time on log-transformed recreational distance. The smooth term is highly significant (edf = 10.2, p < 0.001; Table 3). The curve declines sharply in the lower commuting range, reaches its lowest fitted value at approximately 13 min, and then rises steadily through the middle commuting range into the upper portion of the observed range. This shape indicates that the association between commuting driving time and recreational distance changes direction across commuting-cost ranges, rather than following a simple monotonic form. The confidence interval remains relatively narrow across the central portion of the distribution and widens toward the far right tail, where observations are sparser. The broad shape of the relationship is therefore well supported over the main analytical range, whereas the upper-tail behaviour should be interpreted more cautiously.

3.3.2. Recreational Driving Time and Commuting Driving Time

Figure 3 presents the estimated smooth effect of commuting driving time on log-transformed recreational driving time. The smooth term is highly significant (edf = 10.2, p < 0.001; Table 3). The curve closely resembles that in Figure 2: it declines in the lower commuting range, reaches a local minimum at approximately 15 min, and then rises through much of the middle commuting range before flattening toward the upper end. This similarity indicates that the primary pattern is not specific to recreational distance, but is also observed when weekend recreational mobility is measured by travel time. Confidence intervals remain relatively narrow across the central range and widen toward the far right tail, where observations are sparser.

3.3.3. Broad Features of the Primary Relationship

Taken together, the two primary specifications show the same broad pattern: an initial decline in the lower commuting range followed by a sustained increase through much of the middle and upper range. This recurrence indicates that the primary commuting–recreation relationship is nonlinear and range-dependent, rather than a simple monotonic trade-off in which longer commuting is uniformly associated with either shorter or longer recreational mobility. It provides the main empirical basis for identifying a U-shaped form in the primary specifications.

To assess substantive magnitude, fitted conditional contrasts were calculated for the two primary specifications. In the recreational-distance model, the fitted value at the left end of the commuting-driving-time curve is approximately 25.7% higher than the lowest fitted value around 13.3 min; the curve returns to a comparable level on the rising segment at approximately 42.9 min, close to the 45 min reference often used to identify long commuting in Chinese commuting studies and planning practice [2]. In the recreational-driving-time model, the corresponding left-end-to-minimum contrast is approximately 13.7%, and the curve returns to a comparable level at approximately 36.9 min. These contrasts are based on model-fitted outcomes rather than the centred smooth-effect scale. They indicate that, despite modest pseudo-R² values, the estimated average relationship shows a substantively meaningful valley-to-side contrast.

3.4. Supplementary Specifications

3.4.1. Recreational Distance and Commuting Distance

Figure 4 presents the estimated smooth effect of commuting distance on log-transformed recreational distance. The smooth term is highly significant (edf = 14.1, p < 0.001; Table 3).

The curve broadly resembles those of the primary specifications. It declines sharply in the lower commuting-distance range, reaches its lowest fitted value at approximately 3–4 km, and then rises steadily through the middle commuting-distance range. Toward the upper end, the curve continues to increase more gradually, with slight flattening and local upper-tail variation.

The confidence interval remains relatively narrow across most of the observed range and widens modestly near the far upper end, where observations become sparser. The broad early-decline-then-rise pattern is therefore well supported over the main analytical range, while the precise shape near the far upper end should be interpreted more cautiously.

This supplementary spatial specification is broadly consistent with the primary results, suggesting that the main empirical structure is also visible when commuting cost is represented in spatial rather than time-based terms.

3.4.2. Recreational Distance and Commuting Transit Time

Figure 5 presents the estimated smooth effect of commuting transit time on log-transformed recreational distance. The smooth term is highly significant (edf = 11.9, p < 0.001; Table 3).

The curve differs from the primary specifications mainly in the lower commuting range. Because commuting transit time starts at approximately 10 min rather than near zero, the lowest-cost range observed in the distance- and driving-time specifications is not represented in the same way. The fitted curve rises from the left tail, reaches a local high around 20 min, declines toward 30–35 min, and then increases steadily through the middle and upper transit-time ranges before flattening near the upper end.

The lower-range fluctuation should be interpreted cautiously because observations are relatively sparse and the confidence interval is wider near the left tail. By contrast, observations are more densely represented across the middle and upper transit-time ranges, where the increasing segment after approximately 30–35 min is better supported.

Taken together, this supplementary transit specification provides limited support for the primary empirical pattern. Introducing public-transport impedance on the commuting side changes the estimated form mainly in the lower range, while the middle-to-upper range remains closer to the increasing segment observed in the primary distance- and driving-time specifications.

3.5. Transit-Constrained Specification

Figure 6 presents the estimated smooth effect of commuting transit time on log-transformed recreational transit time. The smooth term is highly significant (edf = 11.9, p < 0.001; Table 3).

Compared with Figure 5, Figure 6 shows a further departure from the primary specifications. When both commuting and recreational mobility costs are measured by transit time, the lower-cost range is less clearly represented on both sides of the relationship, and the early-decline-then-rise structure observed in the primary models no longer appears. Instead, the curve rises through most of the observed range, with only slight flattening in the lower-middle range and near the upper end.

The rug plot and confidence intervals indicate that this broadly increasing form is better supported in the middle and upper transit-time ranges than near the lower left tail, where observations are relatively sparse. The lower-range behaviour should therefore be interpreted cautiously, but the positive association between the two transit-time indicators is well supported over the main analytical range.

This specification clarifies the evidential status of the transit-based results. When transit time is used on both sides, the estimated relationship is closer to a positive association between two public-transport impedance indicators than to a replication of the U-shaped pattern found in the primary specifications.

4. Discussion

4.1. A Nonlinear and Range-Dependent Commuting–Recreation Relationship

4.1.1. Empirical Findings Across Specifications

The direct empirical results show that the commuting–recreation relationship is nonlinear and range-dependent, but the U-shaped form is not equally supported across all specifications. The distance- and driving-time specifications provide the main evidence for an early-decline-then-rise relationship. In the two primary driving-time-based models, weekend recreational distance and weekend recreational driving time first decrease and then increase as commuting driving time rises. The supplementary distance-based specification broadly reproduces this form when commuting cost is measured by route distance, indicating that the pattern is not merely an artifact of the driving-time measure. Together, these results support a U-shaped relationship under distance- and driving-time-based operationalizations.

The transit-based specifications produce different curve forms. When commuting transit time is used as the commuting-cost variable while weekend recreational mobility is still measured by driving time, the estimated curve departs from the primary driving-time-based form, especially in the lower commuting range. When transit time is used on both the commuting and recreational sides, the estimated curve becomes closer to a broadly increasing relationship rather than an early-decline-then-rise form. The transit-based results therefore do not clearly reproduce the U-shaped pattern observed in the distance- and driving-time specifications.

4.1.2. Inferring the Realised Commuting–Recreation Relationship

As actual travel mode is not observed in the LBS data, the five specifications should not be treated as equivalent evidence of the same realised citywide commuting–recreation relationship. As discussed in Section 2.5.2, the distance- and driving-time-based specifications are more closely aligned with realised spatial separation and feasible road-network impedance, whereas the transit-based specifications represent public-transport impedance rather than observed transit use or realised travel time.

The transit-based specifications are therefore less reliable for inferring realised commuting–recreation behaviour. From a behavioural perspective, transit time may overstate realised mobility cost for short trips because residents may walk, cycle, drive, or use other modes instead of public transport; for long trips, high transit time may also induce mode substitution, destination adjustment, or trip suppression rather than represent realised travel time [37,38]. The model results reinforce this limitation. The transit-time variables have weak coverage of the very low-cost range, and the transit-based curves therefore do not provide a directly comparable test of the initial declining segment observed in the distance- and driving-time specifications.

Taken together, the transit-based results clarify the sensitivity of the estimated curve to public-transport impedance measures, but they should not be interpreted as overturning the distance- and driving-based evidence. The realised citywide commuting–recreation relationship is therefore more plausibly inferred from the specifications based on distance and driving time, which support a U-shaped, early-decline-then-rise pattern.

4.2. Spatial Interpretation from a Jobs–Housing–Recreation Perspective

4.2.1. Spatial Configuration Behind the U-Shaped Commuting–Recreation Relationship

The nonlinear relationship can be interpreted from the perspective of the jobs–housing–recreation configuration of urban space. This interpretation requires caution because commuting and recreational mobility behaviours reflect not only spatial constraints, but also multiple interacting mechanisms. The estimated curves should therefore be understood as average relationships formed within broader spatial constraints, rather than as precise representations of behavioural mechanisms.

The analysis also does not directly identify the spatial structure of jobs, housing, and recreational opportunities. The following interpretation is therefore a spatially grounded reading of the estimated curve form, not definitive evidence that a particular spatial configuration is solely responsible for the observed relationship. Other non-spatial mechanisms not captured here may also contribute. With this boundary in mind, different empirical relationships between weekday commuting costs and weekend recreational mobility conditions may correspond to different spatial-organization scenarios.

A broadly positive relationship would be consistent with a spatial structure in which employment and recreational advantages are concentrated in the same locations or strongly overlap across major centres. Residential locations close to these shared advantages would tend to experience both lower commuting costs and lower recreational mobility costs, while locations farther away would tend to face deterioration in both. By contrast, a negative relationship would be consistent with a more direct trade-off between employment and recreational advantages, especially where employment opportunities and recreational resources are spatially separated and residential locations are distributed between them.

The observed U-shaped pattern suggests uneven combinations of employment access and recreational access across residential locations: some locations may have strong employment access but weaker recreational accessibility, while others may combine acceptable commuting conditions with more favourable weekend recreational access. The latter combination indicates that commuting and recreational mobility costs are not structurally locked into opposition.

This spatial reading is important because it links the nonlinear curve to the possibility of integrated spatial diagnosis. If employment access and recreational access fully overlapped, commuting-centred evaluation would already capture much of the relevant spatial-performance variation. If they were strictly opposed, the two domains would form a simple trade-off. The observed nonlinear pattern instead suggests that employment and recreational advantages are unevenly but not oppositely distributed across residential space. More favourable combined commuting–recreation outcomes may therefore be possible under the current jobs–housing–recreation structure, depending on how employment distribution, residential development, recreational opportunity provision, and transport connectivity are configured together.

4.2.2. Self-Selection, Destination-Choice Heterogeneity, and Interpretive Boundaries

Residential self-selection is an important interpretive qualification, but it is not the spatial mechanism from which the U-shaped pattern is derived in this study. In built-environment and travel-behaviour studies, self-selection is commonly treated as a confounding mechanism: households may choose residential locations according to preferences, resources, or constraints that also shape travel behaviour [39]. This issue is relevant here because households may sort into different residential positions within the jobs–housing–recreation configuration, thereby affecting the observed composition of residents and the relative weight of different segments of the curve.

However, this study does not estimate the isolated causal effect of residential location on recreational travel. It examines the realised association between two residence-anchored mobility-cost dimensions within a given jobs–housing–recreation configuration. The conceptual simulation in Figure 7 supports this spatial-structural interpretation. When employment and recreational centres are unevenly arranged, an early-decline-then-rise relationship can emerge under minimum-commute, random-commute, and maximum-commute scenarios. Although these scenarios alter the depth, width, and local shape of the curve, they do not remove the overall U-shaped form. This suggests that the nonlinear relationship can be generated by the spatial configuration itself. Residential self-selection may shape how this relationship is observed in the empirical sample, but it does not replace or negate the spatial mechanism demonstrated by the simulation.

Weekend recreational destination choice introduces heterogeneity into the observed commuting–recreation relationship. Residents may choose more distant destinations because of destination quality, activity type, novelty, social arrangements, or household needs. These choices may affect the dispersion and local shape of the estimated curve, but they do not make mobility cost irrelevant. Because weekend recreation is discretionary, distance and travel time still constrain participation frequency and destination selection. Long-distance trips may provide high utility in particular cases, but they are less able to support frequent routine recreational participation than more accessible destinations. Thus, destination-choice heterogeneity may modify the observed relationship, while spatially structured recreational mobility costs remain central to interpreting the U-shaped pattern.

4.2.3. Chongqing’s Spatial Configuration and Broader Relevance

Chongqing provides a particularly relevant case because its mountainous terrain and river-valley constraints have shaped a polycentric, terrain-separated urban structure. Its central urban area consists of multiple clusters, with employment centres, recreational centres, and smaller dispersed recreational destinations distributed across the urban landscape. These conditions produce a partially overlapping yet spatially constrained jobs–housing–recreation configuration. Route-based commuting and recreational mobility costs therefore reflect the uneven joint arrangement of employment and recreational opportunities across residential locations. The same terrain-constrained context may also amplify distortions in transit-based travel time: short physical distances may involve stairs, steep slopes, or incomplete transit coverage, making transit time a less stable proxy for experienced accessibility than in flatter cities with more continuous transit networks.

The broader relevance of this finding does not lie in claiming that the same curve must appear in all cities. Chongqing’s terrain-constrained polycentricity may make the jobs–housing–recreation configuration especially pronounced, but many large Chinese cities also contain multiple employment and recreational centres, heterogeneous residential districts, and differentiated transport networks. The finding therefore suggests a testable spatial hypothesis: where employment and recreational opportunities are unevenly combined across residential locations, commuting and weekend recreational mobility costs may form nonlinear relationships rather than a simple trade-off or synergy. An early-decline-then-rise pattern may emerge when some locations combine relatively low costs in both dimensions, whereas others secure only one advantage or face high costs in both. In cities with more continuous metro and bus networks, transit-based models may reveal a similar nonlinear relationship more clearly.

4.2.4. From Trade-Off Mechanisms to a Spatially Conditioned U-Shaped Relationship

Existing studies most closely related to this topic have generally interpreted the relationship between commuting, residential location, and non-work activities through trade-off, compensation, or constraint mechanisms. These include studies on commuting–leisure compensation [9], commuting and health-related activities [12], and residential-location trade-offs involving open space, leisure facilities, and social contacts [7,30,31]. These interpretations are important because they show that commuting burden and non-work or recreational opportunities are not independent. However, they tend to frame the relationship as a monotonic or average effect. The present study partly supports this view: in the declining segment of the U-shaped curve, higher commuting costs are associated with lower recreational mobility costs, suggesting an apparent compensatory pattern between commuting burden and recreational accessibility. Its key contribution, however, is to show that this relationship is not a monotonic trade-off. Beyond a certain commuting-cost range, the curve turns upward, indicating that higher commuting costs can again coincide with higher recreational mobility costs. This nonlinear reversal—from apparent compensation to dual mobility disadvantage—is the key empirical pattern that previous trade-off-oriented interpretations have not captured.

This difference is partly related to data scope, measurement design, and functional-form assumptions. Many previous studies have relied on survey samples, specific non-work outcomes, or pre-specified functional forms. By contrast, this study uses large-scale individual-level LBS data, covers a broader set of weekend recreational activities, and applies GAM to estimate the relationship flexibly across the commuting-cost range. These features make it possible to identify a U-shaped empirical regularity that may otherwise be compressed into an average trade-off, compensation, or constraint effect. Mechanistically, this does not mean that individual trade-offs disappear. Rather, it suggests that they are spatially conditioned. Individual choices operate within a jobs–housing–recreation configuration in which employment access and recreational access are unevenly combined across residential locations. Different segments of the curve may therefore correspond to different opportunity combinations: some locations exhibit apparent compensation between commuting burden and recreational accessibility, whereas others face high costs in both dimensions. The commuting–recreation relationship is therefore not simply a one-dimensional trade-off, but structured by the spatial configuration in which mobility choices are made.

4.3. Planning Implications for Integrated Commuting–Recreation Performance Evaluation

4.3.1. Diagnostic Implications of the U-Shaped Relationship

The planning significance of the identified U-shaped relationship lies primarily in its diagnostic value. The pattern indicates that commuting and recreational mobility conditions are not distributed independently across residential space. Some residential locations may combine relatively low commuting costs with relatively low recreational mobility costs, whereas others may secure only one advantage or face high costs in both dimensions. The finding does not by itself identify which intervention would causally improve both outcomes. However, it shows that commuting performance and weekend recreational mobility conditions should be evaluated jointly when assessing urban spatial-structure performance.

This diagnostic perspective is especially relevant in large Chinese cities, where daily life is increasingly shaped by both weekday commuting and weekend recreational mobility. An evaluation focused only on commuting efficiency may overlook residential locations with poor recreational access, while an evaluation focused only on recreational accessibility may ignore the commuting burden associated with those same locations. A more useful evaluative question is therefore whether residential locations with relatively favourable conditions in both dimensions are sufficiently available and equitably distributed across the urban structure.

4.3.2. Combined Accessibility and Spatial Matching

From this perspective, the relevant planning concern is not proximity to any single type of opportunity, but the spatial matching among residential demand, employment supply, recreational supply, and transport connectivity. This perspective extends beyond the commuting-centred logic of jobs–housing balance [1,16,19,20] and the treatment of amenities as isolated residential-choice attributes [6,7,31]. For integrated spatial performance, the key issue is not only whether acceptable employment access and favourable recreational access coexist, but whether this coexistence is produced by effective spatial matching among residential distribution, employment locations, recreational opportunity hierarchies, and transport networks.

This matching perspective is important because employment, housing, and recreation are socially differentiated. Different population groups are associated with different workplace locations, residential preferences and constraints, mobility resources, and recreational demands. High-skilled workers, service workers, families with children, older adults, and lower-income households, for example, may face different feasible combinations of jobs, housing, and recreational destinations. Recreational resources should therefore not be evaluated only by their presence or proximity, but by how they are matched with the residential groups and employment opportunities that structure everyday mobility. The relevant issue is whether the jobs–housing–recreation configuration expands favourable combined accessibility conditions for different population groups. This should be understood as an evaluative and diagnostic principle, rather than a direct project-level planning prescription.

4.4. Toward a Framework for Integrated Commuting–Recreation Spatial Performance Analysis

A key requirement for integrated spatial performance analysis is to develop a rigorous framework for evaluating recreation-oriented spatial-structure performance. Such performance should not be equated with observed recreational travel outcomes alone. It should refer more specifically to the extent to which observed recreational mobility reflects the organizing effect of urban structure [1,40]. In commuting research, concepts such as excess commuting already move in this direction by evaluating structural inefficiency rather than travel distance or time alone [1]. By contrast, although existing studies have generated substantial knowledge on recreational accessibility, use, vitality, and equity [24,26], a comparable framework for evaluating the structural performance of recreation-oriented spatial organization remains underdeveloped.

Developing such a framework requires a clearer understanding of how recreational mobility is formed. Compared with commuting, recreation is less compulsory, less routine, and more heterogeneous. Realised recreational mobility may be shaped by recreational demand, facility hierarchy, destination quality, destination choice, mode choice, household constraints, and the spatial distribution of opportunities [28]. Without clarifying these mechanisms, it is difficult to construct performance measures that distinguish the contribution of spatial structure from observed travel outcomes.

Once a recreation-oriented framework is developed, a further task is to integrate it with commuting-oriented spatial-structure evaluation. Only then can cities be assessed in terms of how their spatial structure jointly shapes work-related and weekend recreational mobility. The longer-term agenda is therefore to move from identifying an empirical commuting–recreation relationship to evaluating alternative jobs–housing–recreation configurations, and eventually to informing planning interventions through better matching among residential demand, employment opportunities, recreational resources, and transport networks.

4.5. Limitations

Several limitations should be noted. First, the five model specifications do not have identical evidential status. The distance- and driving-time specifications provide the primary evidence for the commuting–recreation relationship, whereas the transit-time specifications represent public-transport impedance rather than observed public-transport use. The conclusions should therefore be interpreted mainly as evidence on distance- and driving-time-based mobility-cost relationships, not as a full account of actual multimodal travel behaviour.

Second, the spatial-structural interpretation of the curve requires further validation. Although the U-shaped relationship is consistent with the uneven joint configuration of employment access, residential locations, and recreational opportunities within the jobs–housing–recreation structure, the individual-level GAM results alone cannot rule out alternative explanations. Further spatial analyses are needed to examine how the curve relates to residential locations, employment access, recreational opportunity distributions, and transport networks.

Third, the temporal scope of the data is limited. The November observation window captures only one period of recreational behaviour in Chongqing and does not represent full seasonal variation. The estimated relationship should therefore not be interpreted as a year-round pattern without seasonal validation.

Fourth, the study is based on a single-city case. Chongqing provides a useful case for developing a testable spatial hypothesis, but the identified nonlinear relationship should not be directly generalized to other cities without multi-city empirical verification.

Fifth, recreational distance and travel time measure mobility cost and spatial accessibility, not the full utility of recreation. Longer trips may reflect higher destination quality, novelty, or household needs, while shorter trips do not always imply greater utility. Future research should incorporate destination quality, activity type, satisfaction, and visit frequency to assess recreational utility more fully.

Sixth, multi-AOI recreational activity and activity-chain reconstruction remain unresolved issues in interpreting recreational mobility costs and utility. This limitation is closely related to the preceding point, because recreational distance and travel time cannot be directly translated into recreational utility without considering how multiple destinations are combined within a weekend activity chain. For same-direction or sequential multi-AOI activities, AOI-based average distance or travel time may be close to chain-based travel burden. However, for multi-directional or reverse-direction activities, averaging cannot distinguish whether longer movement reflects additional recreational value, additional travel burden, or both. Future research should therefore use activity-sequence or trip-chain methods to examine multi-AOI recreational mobility more directly.

Taken together, these limitations define the directions in which the U-shaped relationship identified in this study should be further examined. Future research should therefore combine multi-city empirical analysis, more detailed spatial-configuration analysis, and conceptual simulations of alternative jobs–housing–recreation layouts to develop a more systematic understanding of when and how such nonlinear relationships emerge.

5. Conclusions

This study examined the relationship between weekday commuting costs and weekend recreational mobility costs as an empirical entry point for integrated spatial-performance research within a jobs–housing–recreation framework. By situating both mobility dimensions within the jobs–housing–recreation structure, the analysis moves beyond separate assessments of commuting efficiency or recreational accessibility and focuses on how the two are jointly configured across residential space.

Empirically, the central finding is a nonlinear and range-dependent relationship between weekday commuting costs and residence-anchored weekend recreational mobility costs in Chongqing. The distance- and driving-time specifications consistently reveal an early-decline-then-rise pattern: as weekday commuting costs increase, weekend recreational mobility costs first decrease and then increase. The transit-time specifications mainly indicate operationalization sensitivity under public-transport impedance, rather than realised public-transport use. Overall, the citywide commuting–recreation relationship is best characterized by the U-shaped pattern identified in the distance- and driving-time results.

Substantively, the U-shaped relationship moves beyond trade-off-oriented interpretations by linking commuting–recreation trade-offs to the uneven spatial combination of employment and recreational access. Individual trade-offs do not disappear, but they operate within a jobs–housing–recreation configuration in which residential locations are unevenly positioned in relation to employment access and recreational access. The declining segment is consistent with apparent compensation, whereas the upward segment reveals that higher commuting costs can again coincide with higher recreational mobility costs. The commuting–recreation relationship is therefore better understood as a spatially conditioned U-shaped relationship than as a one-dimensional trade-off.

The broader contribution of the study is to establish a diagnostic basis for integrated commuting–recreation spatial performance evaluation. The findings suggest that employment distribution, residential development, transport connectivity, and recreational opportunity provision should be assessed in terms of how they jointly shape combined mobility conditions. For planning, the jobs–housing–recreation framework should be used as a diagnostic lens for evaluating combined commuting and recreational mobility conditions before specific interventions are proposed.

These conclusions should be read within both evidential and contextual boundaries. The analysis identifies an empirical relationship and offers a spatial-structural interpretation, but it does not by itself establish a direct causal mechanism. The Chongqing finding should also be treated not as a universal rule, but as a testable spatial hypothesis. In cities where employment, housing, transport, and recreational opportunities are unevenly distributed and only partially overlapping, commuting and recreational mobility costs may also form nonlinear rather than simply positive or negative relationships. Future multi-city, cross-seasonal, and spatially explicit studies are needed to examine the external validity, seasonal stability, and spatial mechanisms of this relationship.