How the relationship between socio-demographics, residential environments and travel influence commuter choices

ABSTRACT Individual socio-demographic characteristics influence the composition of residential environments, employment considerations and transport-dedicated resources, all of which influence individual travel behaviours. To analyse these interrelationships, we employ generalised structural equation modelling using individual-level data from the 2016 Irish Census on workers across the Republic of Ireland alongside highly spatially disaggregated residential built and social environment data. This allows us to consider the non-linear relationships between multiple variables known to influence travel behaviours and provide direction for future policymaking. We find that regardless of socio-demographic compositions, increased developmental compactness and infrastructure quality are associated with increased public and active travel.


INTRODUCTION
Tiebout sorting is a phenomenon whereby people determine their residential locations as a function of heterogenous tastes in amenities and public goods (Tiebout, 1956).Sorting of this nature, although stemming from individual tastes and preferences, can also stem from socio-economic considerations, such as household mobility and employment considerations (Nechyba & Walsh, 2004).Subsequently, this form of sorting can spawn social exclusion through socio-economic segregation and potentially create spatial mismatches in the accessibility of economic opportunity (Lenaerts et al., 2022;Nechyba & Walsh, 2004).Historically, the sorting of relatively deprived persons occurred in central areas, but decades of urban sprawl have reduced the relative costs of peripheral living, potentially altering existing sorting dynamics and the spatial distribution of employment opportunities (Travisi et al., 2010).Here we analyse how individual socio-demographics, residential built and social environments, and travel considerations are related while also investigating how these interrelationships influence commute mode choices across the Republic of Ireland.
Sprawling developments catalyse increases in car use at the expense of alternative transport modes through landuse segregation and inefficient geographical expansion (Nechyba & Walsh, 2004).Environmental consequences aside, this can exacerbate social exclusion by inducing car dependency if relatively deprived persons are unable to access economic opportunities locally, which can worsen the travel-related burdens of already strained households (Bastiaanssen et al., 2022;Guerra et al., 2022).Consequently, when analysing the implications of regional development patterns on transport accessibility, researchers should consider the relationship between transport policies and land use.Whilst circular in their relationship, coordinated efforts can alleviate the risk of social exclusion by encouraging developments that facilitate the integration of multi-modal transport networks, the mixing of residential and employment centres, and the minimising of regional time-space geographies (Allen & Farber, 2019;Button, 2010;Curl et al., 2018;Thomas et al., 2018).
Often quipped as spawning a 'flight from blight', urban sprawl has been linked with sorting mechanisms that reflect middle-and upper-class desires to leave cities in response to rising social stress (Boustan, 2010;Nechyba & Walsh, 2004).In this context, spatial mismatch hypotheses suggest that combined residential and commercial decentralisation is what disconnects populations from economic opportunities by reducing relative accessibility (Lenaerts et al., 2022;Picard & Zenou, 2018).This can happen because commercial decentralisation generally concentrates around relative prosperity and connectivity, creating accessibility inequities (Guirao et al., 2018;Nechyba & Walsh, 2004;Picard & Zenou, 2018).In a sprawling context, these inequities can be exacerbated because peripheries are generally characterised by segregated land and limited alternative transport accessibility.We adopt a spatial mismatch framework to analyse commute mode choices and how these choices relate to socio-demographics, the social and physical characteristics of residential environments, and travel considerations.
We make three contributions to existing literature.First, we incorporate data that are typically segregated across studies which allows us to examine the relative influence different features of residential areas have on travel behaviours.For instance, when analysing the built environment's influence on commute mode choice, research typically focuses on either land-use metrics (Ton et al., 2019) or population measures (Guerra et al., 2022), and seldom both (see Guerra et al., 2018, for an exception).Similarly, comprehensive multi-modal transport network data are usually lacking, with studies generally only using road networks (Guerra et al., 2022).Few studies explicitly measure the underlying attributes of residential environments, such as the quality of existing infrastructure and the underlying social circumstances of areas (Eldeeb et al., 2021;Lee et al., 2018).Our first contribution stems from our synergy of different built environment indicators into one analysis.By doing this we answer calls by Guerra et al. (2018) who explicitly note how data availability on built environments restricted their analysis of travel behaviours, while numerous other authors call for greater quantification of the built environment determinants of travel behaviours (Bastiaanssen et al., 2022;Eldeeb et al., 2021;Gan et al., 2021;Guerra et al., 2022;Ton et al., 2019).
Second, we extend the scope of analysis typically employed in transportation studies.Ordinarily, studies analysing travel behaviours are confined to metropolitan areas due to a lack of data available at other spatial scales (Cheng et al., 2020;Guerra et al., 2018Guerra et al., , 2022)).Our ability to use a nationally representative sample comprised of 235,936 individuals residing across 3409 localities allows us to investigate interregional disparities in travel behaviours and their social and physical environmental drivers, something only before attempted by Bastiaanssen et al. (2022).This allows us to answer calls by Ahrens and Lyons (2019), who determine that 'link [ing] census data on commuting flows between small areas to administrative data … ' offers exciting avenues for future research (Ahrens & Lyons, 2019, p. 12) and Wu et al. (2021) who call for the incorporation of more geographically diverse datasets when analysing travel behaviours.
Typically studies employ a single-equation framework, such as multinomial regression, when analysing travel behaviours (Eldeeb et al., 2021;Gan et al., 2021;Guerra et al., 2018;Ton et al., 2019).However, this framework risks collapsing a multi-stage decision into a single equation (De Vos et al., 2021;Ewing et al., 2016).We analyse the interrelationships between socio-demographic characteristics, residential environments and travel considerations, alongside how these interrelationships influence commute mode choice.This requires a methodological framework capable of testing multiple interrelated hypotheses simultaneously (De Vos et al., 2021;Ewing et al., 2016;Gim, 2018).Accordingly, our final contribution stems from our use of a novel generalised structural equation modelling (GSEM) framework which facilitates the use of categorical and binary response variables, a capability which has allowed GSEM to outperform traditional structural equation modelling structures when analysing travel behaviours (Yin et al., 2020).
Empirically, we use a stratified random sample from the 2016 Place of Work, School, or College Census of Anonymised Records (POWSCAR) dataset that is derived directly from the 2016 Irish Census which covers all persons working in the Republic of Ireland.This allows us to observe individual socio-demographic and commute characteristics, while also accounting for residential built and social environment characteristics within electoral divisions (ED), of which there are 3409 in Ireland.We link individual travel behaviours to socio-demographic characteristics, local built and social environments, and travel considerations by modelling the determinants of commute mode choice for 235,936 individuals, across 3409 localities, using five mode choice optionswalking, cycling, buses, trains/trams and cars/motorbikes/vans.
The remainder of the paper is structured as follows.Section 2 documents a literature review surrounding the determinants of travel behaviours.Section 3 describes the data and methodology used.Section 4 presents our results.Section 5 discusses their implications.We conclude the analysis in section 6 by investigating the policy implications of this work while also highlighting this study's limitations and future research avenues.

Travel behaviour
Travel is a good/service typically characterised by derived demand, whereby journeys serve as a means to an end (Button, 2010;Cheng et al., 2019).When commuting, this relates to how individuals travel between their home and workplace.In this context, private cars generally exhibit characteristics of normal goods due to their flexibility and reliability (hence explaining positive relationships between car use and income), whereas public transport is typically perceived as an inferior good due to negative perceptions of service quality (Friedrich et al., 2011;Giuliano, 2012).Therefore, unlike typical goods, the monetary costs associated with travel may not influence mode use as strongly as factors pertaining to reliability, convenience and quality (Button, 2010).This implies that commute mode choices may be highly sensitive to land-use and transport policy instruments (Button, 2010;Vandenbulcke et al., 2009).
A substantial body of evidence suggests that individual socio-demographic considerations distinctly influence travel behaviours (Nechyba & Walsh, 2004;Vandenbulcke et al., 2009).However, evidence suggests that these behaviours are sensitive to other considerations, such as the physical characteristics of residential environments.Evidence suggests that reducing required travel distances is associated with reduced car use due to the heightened competitiveness, convenience and efficiency of alternative modes (Button, 2010;Millward et al., 2013;O'Riordan et al., 2022).From a land-use perspective, this involves increasing developmental compactness, while from a transport perspective, this involves allocating more resources to the provision of multimodal transport infrastructure (Gim, 2012).The efficacy of these policy instruments is determined by the spatial distribution of infrastructural investment, something historically linked to local socio-demographic compositions and, subsequently, phenomena such as Tiebout sorting (Boustan, 2010;Nechyba & Walsh, 2004).
Although these considerations often hold high explanatory power over individual travel behaviours, individuals do not operate in a vacuum.Evidence suggests that larger family units typically travel greater distances, subject to residential environment characteristics (Commins & Nolan, 2011;Ermagun & Levinson, 2016).For larger family units (i.e., households with children), this suggests that the ability/need to trip-chain (i.e., combining the trips of multiple individuals) significantly influences the travel behaviours of parents and children, thereby intertwining behaviours across individuals (Commins & Nolan, 2011;Ermagun & Levinson, 2016).Trip-chaining is a travel option typically reserved for private cars due to their flexibility, convenience and reliability. 2Subsequently, depending on the characteristics of residential environments and required travel distances, public and active transport may be inconvenient, unreliable and too costly, thereby increasing car use (Curl et al., 2018).
Individuals typically have ordinal preferences regarding their consumption of goods and services (Samuelson, 1954).While in a travel behaviour context, these preferences manifest through mode choices, in a broader regional development context, these preferences (when pertaining to the consumption of public goods, such as land-use configurations and transport infrastructure) can be deduced through residential locations (Tiebout, 1956).Here, the decision to move (or not to move) becomes somewhat analogous to willingness to pay in economies characterised by high levels of human mobility (Samuelson, 1954;Tiebout, 1956).Although high levels of mobility is symptomatic of society in the developed world, the ability of individuals to realise these preferences is highly sensitive to the accessibility of travel-related resources and employment opportunities, something obviously linked to socio-demographic considerations, but also to regional land-use configurations and transport infrastructure provision (Bastiaanssen et al., 2022;Curl et al., 2018;Guerra et al., 2022).

Built environments and travel behaviours
Regional development is a spatiotemporal process characterised by the expansion of urban areas whereby rural/ semi-rural areas are integrated into urban infrastructures (Travisi et al., 2010).This links the topography of regional developments to economic, social and environmental outcomes by defining the severity of the disadvantages associated with separation from regional cores (Button, 2010;Trannoy et al., 2011;Vega & Reynolds-Feighan, 2016).Historically, this topography has been defined by monocentric expansion and urban sprawl across much of Europe, whereby land-use developments are characterised by low residential densities, high degrees of land-use segregation and widespread car use (Reynolds-Feighan, 2003;Vega & Reynolds-Feighan, 2008, 2009, 2016;Weilenmann et al., 2017).
Evidence increasingly points to aspects pertaining to the built environment, such as land-use mix, transport infrastructure provision and infrastructural quality, as important features influencing commuting decisions (Ewing & Cervero, 2010;Gan et al., 2021;Guerra et al., 2018;Ton et al., 2019).But urban sprawl has also resulted in population dispersions characterised by relatively affluent populations residing in peripheral areas and commuting to employment hubs, whilst relatively deprived populations tend to live and work in the same localities (Button, 2010;Nechyba & Walsh, 2004). 3 Over time, regional infrastructure investment has become increasingly concentrated in areas characterised by relative affluence.This creates spatial imbalances in the quality of local infrastructure and may warp the spatial distribution of commercial decentralisation, something particular evident in Ireland through uneven regional growth (Vega & Reynolds-Feighan, 2016).
Subsequently, because land-use and population distributions determine the provision and accessibility of regional transport infrastructure, the consequences associated with spatially segregated, sprawling developments may be compounded by spatial mismatches in the provision of infrastructure and the accessibility of jobs (Button, 2010;Nechyba & Walsh, 2004).Spatial mismatch hypotheses conclude that regional investment is inequitably distributed across space and that residential/commercial decentralisation coupled with inadequate public/ active transport networks disconnects populations from employment opportunities (Nechyba & Walsh, 2004;Vega & Reynolds-Feighan, 2009, 2016).In a sprawling landscape, the quality of local infrastructure and the accessibility of employment will be lower in relatively deprived areas, heightening sensitivities to the interconnectedness of localities and the relationship between land-use configurations and transport infrastructure/service provision (Curl et al., 2018).In other words, if services are not available locally, access to distant services may be prevented by inadequate transport infrastructure, a scenario whose severity hinges on the transport infrastructure accompanying development patterns (Curl et al., 2018;Guerra et al., 2018).
Spatially segregated, low-density settlements increase regional time-space geographies and reduce the efficiency and effectiveness of public and active transport (Vega & Reynolds-Feighan, 2016).Accordingly, evidence suggests that land-use-induced challenges to employment accessibility can be mitigated by increasing the accessibility and efficiency of public and active travel (Millward et al., 2013;Tyndall, 2017).Improved transport accessibility is positively associated with regional economic performance and reduces the risk of social exclusion regardless of local socio-demographic compositions, thereby abating landuse-induced accessibility inequities (Bastiaanssen et al., 2022;Heuermann & Schmieder, 2019).

Travel considerations, social environments and travel behaviours
The physical configurations of residential environments clearly exert a distinct influence on required travel distances.Mixed-use landscapes exhibit higher levels of accessibility to local employment opportunities than areas characterised by land-use segregation and low residential densities (Friedrich et al., 2011;Giuliano, 2012; Organisation for Economic Co-operation and Development (OECD), 2018).Urban sprawl has resulted in relatively affluent populations developing an inwardcommuting culture over longer distances, while relatively deprived populations usually reside near their workplace, meaning relatively affluent populations typically exhibit larger commute distances (Curl et al., 2018;Guerra et al., 2022;OECD, 2018;O'Driscoll et al., 2022;Thomas et al., 2018).
Evidence suggests that variables concerning travel considerations, such as trip durations, can influence (and potentially restrict) travel behaviours and that these variables are explicitly linked to Tiebout sorting mechanisms (Chen & Vickerman, 2017;Vega & Reynolds-Feighan, 2009).If infrastructure and regional investment is disproportionately concentrated in suburban peripheries, it implies that although travel distances to urban centres and employment opportunities may be larger, the existing infrastructure will be of a higher quality (Vega & Reynolds-Feighan, 2016).But this also suggests that while relatively deprived populations may face shorter travel distances, local infrastructure may be inadequate (Boustan, 2010;Giuliano, 2012).Therefore, if the employment opportunities relatively deprived populations cluster around relocate to suburban areas, these populations may need to adapt their travel behaviours or relocate their residence to avail of new opportunities (Curl et al., 2018;Guerra et al., 2022).
Because peripheries are inherently inaccessible by public and active transport in a sprawling landscape, this can force trade-offs between affordable housing and transport needs, heightening the risks of transport poverty, 4 social exclusion and unemployment (Curl et al., 2018;Hu & Schneider, 2017).Therefore, socio-economic exclusion can be conceptualised as a problem of accessibility, whereby economic and social outcomes are directly influenced by the usability of regional transport infrastructure (Picard & Zenou, 2018;Tyndall, 2017).This implies that socio-demographics directly influence travel behaviours through individual preferences, convenience and reliability, and indirectly influence these behaviours through the consequences of Tiebout sorting.This connects socio-demographic considerations to the underlying qualities of residential environments and the relative accessibility of employment opportunities, mechanisms directly linked to travel behaviours (Bastiaanssen et al., 2022;Guerra et al., 2022;Lenaerts et al., 2022).

Theoretical model
Figure 1 builds upon the literature reviewed in previous subsections by synergising our conceptual framework.We posit that socio-demographics (i.e., age, sex, socioeconomic group and household composition) distinctly influence physical (i.e., relative urbanisation, land-use characteristics and transport infrastructure provision/quality) and social (i.e., area-level deprivation) characteristics of residential environments, travel considerations (i.e., journey durations and whether people live and work in the same locality) and commute mode choices through Tiebout sorting mechanisms.These relationships are presented as arrows that stem from socio-demographics and directly point toward the other elements of our model.
We contend that socio-demographics are the principal exogenous variables within this research context, hence the lack of arrows leading to socio-demographics in Figure 1.These variables share a direct relationship with the types of built environments people reside in and the underlying social characteristics of these environments through Tiebout sorting mechanisms, mechanisms illustrated through arrows stemming from socio-demographics in Figure 1 (Boustan, 2010;Nechyba & Walsh, 2004).This builds upon Cheng et al. (2019), who investigate the interrelationships between individual socio-demographics, household considerations, trip purpose and the frequency of mode use.
Our Tiebout sorting-spatial mismatch theoretical framework suggests that relatively deprived populations typically live and work in the same locality while relatively affluent populations usually do not (Boustan, 2010;Guerra et al., 2022;Nechyba & Walsh, 2004;Tyndall, 2017).This implies that relatively deprived populations may have shorter travel distances.Because travel distances define the mode choice-sets available to individuals, shorter travel distances imply larger choice sets.However, this conceptual framework also suggests that infrastructural investment may disproportionately concentrate in areas characterised by relative affluence (Nechyba & Walsh, 2004;Vega & Reynolds-Feighan, 2016).This may leave relatively deprived areas with lower quality transport infrastructure which reduces the efficiency and feasibility of public and active transport (Allen & Farber, 2019, 2020;Curl et al., 2018).Subsequently, we argue that the inherent quality of infrastructure and the wider social environment will influence travel behaviours by defining the efficiency of travel and the mode choice set available to individuals.In other words, because relatively deprived populations will typically live and work in the same locality, they may use public and active transport more regularly than relatively affluent populations, but low-quality transport infrastructure may counteract this.This provides the rationale for our arrows stemming from social environment and ending in travel considerations and commute mode choice in Figure 1.
Advancing the work of Idris et al. (2015), who analyse the interrelationships between psychological considerations, the appraisal of transport services and travel mode choices, we posit that the physical characteristics of built environments, such as developmental

Data
We use individual-level data from the Place of Work, School, or College Census of Anonymised Records (POWSCAR) from the 2016 Irish Census.This provides detailed information on individuals' socio-demographic and commuting characteristics.We also incorporate regional infrastructural data inclusive of transport networks, land-use metrics and area quality indicators derived from OpenStreetMap, The 2018 CORINE Land Cover index, The 2016 Deprivation Index (Teljeur et al., 2019) and The Property Price Register (2016).These data measure the individual determinants of travel behaviours based on socio-demographic, built and social environment characteristics, and travel considerations.Our other variables are divided across four groups: socio-demographics, travel considerations, built environments and social environments.Our variables representing socio-demographics capture each person's age, sex, socioeconomic group and household composition.Household composition and socio-economic group categorise the household structure of individuals' and the socio-economic group to which individuals belong.We argue that these variables capture individual-level tastes/preferences and facilitate investigations into Tiebout sorting mechanisms.We conceptualise travel considerations as journey durations (min) and whether people live and work in the same locality at the ED and small area (SA) level. 6 Our built environment variables comprehensively capture the physical characteristics of residential environments by accounting for land use, population and transport infrastructure while also proxying for their underlying quality using property market values.Specifically, CORINE classifies each residential ED according to its majority land-use classification.This was calculated in ArcGIS whereby EDs are characterised according to the 2018 CORINE land-use class which covers the largest share of ED land.Residence area type measures the degree of urbanisation per residential county expressed as population counts. 7To capture transport infrastructure provision and the relative accessibility of said infrastructure, we incorporate network access nodes.Road and Active represent the logged length of roads and active travel infrastructure per ED, 8 allowing investigations into the relationship between mode-specific (i.e., road) infrastructure provision and mode-specific (i.e., car) use.Supplementing these are measures which account for transit node provision, such as Bus stops and Rail stops, which count the number of bus and rail stops within each ED. 9 Finally, we measure local infrastructural quality through the lens of property market values whereby the logged average house price per ED is assumed to be positively correlated with the provision of high quality infrastructure, and subsequently subject to spatial mismatches (Allen & Farber, 2020;Guerra et al., 2022).
To capture the social environments of residential areas, we use Deprivation, which is an area-based index capturing the underlying social circumstances of localities through employment levels, socio-demographic characteristics, and car ownership levels aggregated across Electoral Divisions (Teljeur et al., 2019).Deprivation is specified in quintiles whereby a score of 1 represents the least deprived 20% of EDs and a score of 5 represents the most deprived 20% of EDs.This proxies for the relative socio-economic prosperity of localities and is therefore related to spatial mismatches and Tiebout sorting mechanisms.Table 1 defines and summarises these variables.

Econometric modelling
We employ a series of generalised structural equations to estimate the relationships between socio-demographics, built and social environment configurations, and travel considerations, while also investigating how these variables (and interrelationships therein) influence commute mode choices.We do this because the factors influencing travel behaviours often interact, illustrating the need to adopt a methodology capable of investigating multiple interrelated hypotheses simultaneously (De Vos et al., 2021;Ewing et al., 2016;Gim, 2018).Generalised structural equation modelling (GSEM) facilitates this by allowing us to assume and evaluate relationships between different mediating variables.It does this by evaluating theoretical models with data through the use of path diagrams (Ewing et al., 2016;Gim, 2018;Idris et al., 2015).
In our path diagram (Figure 1), we postulate that socio-demographics are the main exogenous variable, meaning these factors influence residential built and social environments, travel considerations, and commute mode choices.This follows our Tiebout sorting theoretical framework by suggesting that although built and social environment configurations (alongside travel considerations) directly influence travel mode choices, sociodemographics are what influence the infrastructural make-up and area-level deprivation of these environments (Acker et al., 2007;Ding et al., 2017).By assuming relationships between these mediating variables, we can partly account for residential self-selection by capturing the indirect effect of personal characteristics on commute mode choice via residential built and social environments (Ding et al., 2017(Ding et al., , 2018;;Mokhtarian & Cao, 2008).Therefore, this theoretical model tracks how individual, environmental, and trip-specific considerations influence travel behaviours while also investigating how these considerations interact.
Our modelling framework simultaneously accounts for endogenous and exogenous categorical and continuous variables which are interrelated (Acker et al., 2007;Cheng et al., 2019;Ding et al., 2017;Golob, 2003;Idris et al., 2015;Yin et al., 2020).We follow existing literature by employing the maximum likelihood method to estimate our GSEM (Cheng et al., 2019;De Vos et al., 2021;Ding et al., 2018;Golob, 2003).Specifically, our specification builds upon that detailed by Cheng et al. (2019), who posit that a GSEM which includes no latent variables takes the generic form of: where y is column vector of n dependent variables, x is a column vector of m independent variables, B is a matrix (n × n) of direct relationships, Г is a matrix (n × m) of regression coefficients, and ζ is a column vector of error terms (Cheng et al., 2019).Goodness-of-fit tests for traditional structural equation models, such as the root mean square error of approximation (RMSEA), are not applicable in this context due to the generalised structure of response variables (Yin et al., 2020).This renders comparisons between the Akaike information criterion (AIC) and Schwarz Bayesian Information Criterion (SBIC) scores of alternative modelling structures as the principal robustness test for our modelling structure (Cheng et al., 2019;De Vos et al., 2021;Ding et al., 2017Ding et al., , 2018;;Gim, 2018;Hooper et al., 2008;Yin et al., 2020).In this regard, we evaluate four alternative modelling specifications which can describe the relationships in our data, whereby we determine the best fitting model according to AIC and SBIC scores (De Vos et al., 2021;Gim, 2018;Golob, 2003).This approach illustrates that our proposed modelling specification (Figure 1) best describes the relationships within our data, something corroborated by results which are produced using alternative econometric techniques (i.e., multinomial logit regression) that are consistent with the findings presented in this article (De Vos et al., 2021;Gim, 2018;Golob, 2003).The supplemental data online demonstrates consistent coefficients when the key elements of this model are estimated using a standard spatially disaggregated multinomial logit framework and highlights how our modelling structure produces the lowest (i.e., best) AIC and SBIC scores relative to alternative specifications.
Our primary dependent variable is commute mode choice: (1) Walk; (2) Bike; (3) Bus; (4) Train/Tram; and (5) Private Motorised.This generalised response variable takes a multinomial logit structure, meaning it must satisfy the assumption of the Independence of irrelevant alternatives (IIA).To satisfy IIA, each option within the choiceset must be independent.Following established literature, AIC and BIC scores were used to analyse the efficacy of our dependent variable specification (Guerra et al., 2018), while Hausman tests analysed the validity of assuming IIA.The optimal specification of our dependent variable emerged as previously outlined in section 3.1.Our other variables are specified as a mix of continuous and binary variables, with many serving as independent and dependent variables. 10

RESULTS
Table 2 presents the results of the marginal effects derived from our GSEM estimation on the relationships between socio-demographic considerations, built and social environments, and travel considerations.How the relationship between socio-demographics, residential environments and travel influence commuter choices

REGIONAL STUDIES
illustrates the results of the marginal effects derived from the same GSEM estimation on the influence of sociodemographics, built and social environments, and travel considerations on commute mode choice.Within our GSEM models, commute mode choice is estimated using a multinomial logit structure, but other categorical dependent variables (Residence Area Type, Land-Use (CORINE) and Deprivation) are treated as continuous variables.When response variables are polytomous (commute mode choice), coefficients are presented as the marginal effects at means derived from odds ratios.All other response variables are treated as continuous variables, meaning they reflect the marginal effects at means derived from linear regression.

Tiebout sorting and travel behaviours
Our model implies that populations are roughly distributed according to Tiebout sorting mechanisms.That is, relatively deprived populations will live and work within the same locality so that local time-space geographies can be minimised, while relatively affluent populations will disperse in residence to peripheral areas and commute inward (Nechyba & Walsh, 2004;Vega & Reynolds-Feighan, 2016).Accordingly, the key relationship is whether the relative deprivation of localities shares a direct relationship with living and working in the same locality.Columns II and III in Table 2 show that 1 unit increases in deprivation are associated with increased instances of workers living and working in the same locality at the ED and SA levels.
When analysing the spatial distribution of different socio-economic cohorts, the relationships between most socio-demographic variables and residential built and social environment variables are significant at the 1% level.We find that male workers typically live in more urbanised areas and areas which are associated with greater levels of socio-economic affluence relative to female workers (Table 2, columns IV-V and X-XI). 11Gender shares a somewhat ambiguous relationship with commute durations in that male workers typically have longer commutes, but inconsistently relate to living and working in the same localitya characteristic obviously linked to commute durations (Table 2, columns I-III).We interpret this to be the consequence of a trip duration variable which is insensitive to the mode used when commuting.We find that older workers usually reside in urban areas (Table 2, column V), and usually live and work in the same locality (Table 2, columns II and III), a locality which is generally relatively affluent (Table 2, column XI).The built and social characteristics of residential environments become more significantly associated with household structures as household structures become more complicated, and, by extension, individual travel behaviours become more intertwined (Table 2, columns IV-XI).We observe no significant relationship between individual socio-economic groups and the physical characteristics of built environments (Table 2, columns IV-IX).However, advances in individual socio-economic groups (i.e., moving from non-manual to employers/managers) are associated with increased affluence in residential areas (Table 2, columns X and XI), reduced instances of workers living and working in the same locality (Table 2, columns II and III), and longer commutes (Table 2, column I).
Theoretically, Tiebout sorting mechanisms should exert different environmental influences on different socio-demographic cohorts because socio-demographics proxy for group tastes and preferences surrounding the built and social characteristics of residential environments (Nechyba & Walsh, 2004;Tiebout, 1956).Subsequently, these mechanisms are explicitly linked to local infrastructure provision, land-use configurations, and the accessibility of employment, characteristics which directly influence travel behaviours (Curl et al., 2018;Guerra et al., 2022;Tyndall, 2017).This inference is strengthened upon analysing the relationships between the physical and social characteristics of residential environments and travel considerations.Here, we find that although the relative deprivation of localities insignificantly affects trip durations, an outcome which is not sensitive to mode choice, it positively influences the number of people living and working in the same locality (Table 2, columns II and III), a consideration clearly linked with reduced travel burdens (Table 2, column I).
Complementing these insights are results highlighting how increases in socio-economic status are associated with increased likelihoods of using rail-based transit and private cars relative to alternatives (Table 3, columns IV and V), while increases in the relative deprivation of localities is associated with increased likelihoods of public and active transport use relative to cars (Table 3, columns I-V).This suggests that because more relatively deprived workers tend to live and work in the same localities that these populations are more inclined to avail of public and active travel (Table 3, columns I-IV).

Built environments, trip durations and commute mode choices
Considering the relationship between socio-demographics and the built and social characteristics of residential environments, it is important to analyse the effect these environments have on travel considerations and commute mode choice.In this context, we find no significant relationship between the physical characteristics of residential environments and journey durations (Table 2, column I).We interpret this to be the consequence of a trip duration variable which is insensitive to the mode used when commuting.Similarly, our results do not extract meaningful relationships between the number of workers living and working in the same locality and urbanisation or transport infrastructure provision (Table 2, columns II and III).However, we do find that residential property market values negatively influence the number of workers living and working in the same locality, supporting previous results that the relative deprivation of residential areas is positively associated with workers living and working in the same localities (Table 2, columns II and III).As   How the relationship between socio-demographics, residential environments and travel influence commuter choices 647 REGIONAL STUDIES  highlighted within our theoretical model (Figure 1), residential built environments do not influence residential social environments, meaning the model reports 'n.a.' values.
Table 3 shows how decreases in the continuity of urban fabrics (i.e., increases in land-use segregation), are associated with lower likelihoods of active and public travel but higher likelihoods of car use.Similarly, increases in the populations within, and the urbanisation of, residential counties are associated with increased public and active travel and lower car use.Additionally, we find that living and working in the same locality increases the likelihood of walking and reduces the likelihood of using cars; effects which vary by spatial scale (Table 3, columns I and V).
While these variables are explicitly related to the landuse configurations of residential environments, intrinsic to these compositions is the accompanying transport infrastructure.In this context, we find that active transport infrastructure provision produces results contrary to our expectations in that increases in active transport infrastructure insignificantly influence the likelihood of using any mode (Table 3, columns I-V). 12However, we observe that the provision of road networks significantly affects all modes at the 1% level, illustrating how increased road network provision reduces the likelihood of walking, cycling, and bus use, whilst increasing the likelihood of car use (Table 3, columns I-III and V).We also find that transit node provision positively influences the likelihood of using public transport (Table 3, columns III and IV).Increases in bus stop provision is associated with lower likelihoods of car use and insignificantly impacts alternative mode-use, whilst increases in rail stop provision increase the likelihoods of walking, reduce the likelihood of bus use, and insignificantly affects alternative mode use.These results detail how increased transport network provision is associated with increased likelihoods of people using these modes, a relationship that is particularly strong between road network provision and car use.
Our model also tests whether infrastructure provision alone is enough to stimulate changes in travel behaviours.To test this, we analyse infrastructural quality through the lens of property markets.We find that increased property market values within localities are associated with increased likelihoods of public and active transport use relative to cars (Table 3, columns I-V).While this may imply that the quality of transport infrastructure may distinctly influence travel behaviours, part of this effect could be attributable to confounding variables, such as population densities.Nonetheless, when comparing these results with our infrastructure provision variables, we argue that the quality of infrastructure may share a stronger association with travel behaviours than infrastructure provision itself.For modes other than cars, building infrastructure on its own may not be sufficient to attract greater alternative transport users, but the infrastructure that is built must be of a high quality.Future research may wish to test this claim with more comprehensive data capturing transport infrastructure quality.

Robustness checks
Several robustness checks were conducted to ensure the validity of our results.Principal of those is our evaluation of the theoretical model presented in Figure 1.We propose four alternative modelling specifications using the same key variable groups (socio-demographics, built/social environments and travel considerations).We do not consider omitting any individual group entirely on the grounds that each hold significant empirical and theoretical justification for inclusion.Consequently, we primarily test whether model goodness of fit improves (according to AIC and SBIC scores) when relationships (i.e., arrows in the path diagram) are removed.More information on these alternative specifications can be found in the supplemental data online.We also use a multinomial logit specification which provides an alternative modelling framework to test these relationships.We find support for our results as the multinomial logit produces similar marginal effects to those of the GSEM (see Table B1 in Appendix B in the supplemental data online).We test for multicollinearity in our variables using a correlation matrix which failed to identify issues relating to multicollinearity (see Table A1 in Appendix A online).We test the efficacy of our use of specific independent variables by omitting each from our modelling framework and comparing AIC and SBIC scores.Additionally, we test whether Deprivation and House Price variables appropriately proxy the underlying social conditions of localities and the quality of existing infrastructure by substituting household incomes and the proportion of local populations reliant on social welfare.In all cases, AIC and SBIC scores indicate that the variables employed in this analysis produce a model with superior fit.

DISCUSSION AND IMPLICATIONS
The first stage of our model examines the relationship between socio-demographics, the characteristics of residential environments and travel considerations.From a demographic perspective, our key finding is that more complicated household structures generally reside in environments characterised by lower levels of active and public transport infrastructure, high degrees of land-use segregation, and that fewer commuters in these households live and work in the same locality.From a socio-economic perspective, we find that greater numbers of relatively deprived populations tend to live and work in the same locality and that people residing in these localities typically face shorter commuting times.This suggests that relatively affluent cohorts reside peripherally and commute inward for work, whilst relatively deprived populations live and work locally, possibly due to their reliance on the accessibility of local employment opportunities.This inference is supported through the relationships found between commute mode choice, individual socio-economic status and the relative deprivation of localities, whereby people residing in relatively deprived areas are more likely to travel via public and active transport than cars, and that these people generally live and work in the same locality.
The second component of our model investigates how residential environments influence travel considerations and commute mode choices.We observe that increases in the segregation of urban fabrics and lower degrees of urbanisation deter public and active transport use at the expense of cars.Complementing these patterns are our findings that living and working in the same locality significantly increases active travel and public transport use.These results suggest that regardless of the socio-demographic composition of localities, areas which minimise required travel distances encourage greater transit use and active travel.We interpret this to be the result of increases in the efficiency of public and active transport infrastructure/services.
We also find that larger road networks adversely affect active and public transport use, while increased transit node provision increases the likelihood of using transit.These results echo existing literature by showing that greater mode-specific transport infrastructure provision increases mode-specific use (Millward et al., 2013).However, this is only really the case for cars and roads, as building infrastructure on its own negligibly, but positively, impacts the likelihood of travelling via transit and does not significantly influence the likelihood of travelling actively.We observe a stronger relationship between infrastructural quality and the likelihood of using public and active transport as we find that increases in local infrastructure quality significantly increase the likelihood of transit use and active travel.This implies that increased public and active transport use is only significantly related to the provision of high-quality infrastructure, and that providing infrastructure alone may be insufficient.
These results align with global policymaking objectives.To meet the targets laid out by the United Nations Sustainable Development Goals relating to regional development, land-use developments must be coordinated around compactness and mixed use.These developments reduce regional spatial demands and use existing infrastructure more efficiently (Eldeeb et al., 2021;O'Driscoll et al., 2023).This increases multi-modal accessibility (and subsequently regional connectivity) by reducing the spatial expanse of developments (Gim, 2012;O'Riordan et al., 2022;Zhang & Zhao, 2017).These initiatives have been shown to reduce the environmental degradation attributable to travel behaviours and also facilitate economic growth by increasing the accessibility of economic opportunity (Chen & Vickerman, 2017;Heuermann & Schmieder, 2019).Our research adds to the evidence base supporting these frameworks by showing how the coordination of land-use and transport policies around mixed use and compactness offers robust mechanisms to reduce the environmental degradation, economic inefficiencies and social exclusion attributable to regional development.

CONCLUSIONS
This research investigates the distinct influences sociodemographics, built and social environments, and travel considerations have in determining commute mode choices, and the interrelationships therein.To do this, we employ GSEM to assess these interrelationships across the Republic of Ireland.Considering these findings, we discuss planning mechanisms which can work harmoniously to increase regional connectivity, reduce social exclusion, and facilitate shifts towards greater sustainable transport use.
Our principal finding is that the associations between residential environments, travel considerations and commute mode choice typically reported in existing literature are possibly inflated by Tiebout sorting mechanisms.We find that these mechanisms directly influence the types of environments people reside in and subsequently their travel behaviours.Overall, we find that relatively deprived populations tend to live and work in the same locality and are subsequently more likely to use public and active transport.However, we also find that there is no significant relationship between the physical characteristics of built environments and socio-demographic characteristics, indicating that regardless of socio-demographic composition, areas characterised by higher levels of mixed use and compactness are generally associated with increased likelihoods of using public and active transport.Additionally, we observe that transport infrastructure provision alone only positively affects the likelihood of car use, whereas increases in infrastructural quality are associated with increased likelihoods of using public and active transport.This suggests that public and active transport use may be particularly sensitive to infrastructural quality, and that infrastructure provision alone may not be sufficient to encourage greater use, a finding which suggests that the implicit costs associated with transport (i.e., service quality) are a stronger predictor of mode choice than the monetary costs (Button, 2010).
These results illustrate that those residing in extremely deprived areas are highly sensitive to the accessibility of economic opportunities.Should employment centres continue to decentralise and become inaccessible by modes other than cars, people residing in relatively deprived areas may be forced into car use to avail of economic opportunities (Allen & Farber, 2019;Bastiaanssen et al., 2022;Tyndall, 2017).This exacerbates the risk of creating transport poverty and social/economic exclusion because of land-use and transport policy configurations.Considering this, we support calls for the development of planning frameworks centred around accessibility and mixed use.Evidence suggests that these increase regional connectivity, reduce the risks of socio-economic exclusion, and alleviate the environmental degradation attributable to land-use and transport policies (Eldeeb et al., 2021;Guerra et al., 2018;Millward et al., 2013).We concur that this joint focus can limit the creation of spatial mismatches in economic accessibility by keeping places of interest accessible via multiple modes (Bastiaanssen et al., 2022;Tyndall, 2017).
This analysis calls for improvements in the quality of land-use data/metrics when analysing travel behaviours.Particularly pertinent is our call for studies which improve 650 Conor O'Driscoll et al.

REGIONAL STUDIES
upon the limitations of this study by incorporating: (1) more thorough land-use classification data and land-use configuration metrics; and 2) more comprehensive measures of public transport service quality, more precise measures of infrastructural quality and a more comprehensive mode choice set which accounts for trip-chaining.
Empirically, this will likely involve the incorporation of opensource datasets, the innovative use of land-use classification metrics, and the mapping of transit schedules/services (alongside congestion effects).Finally, future research could incorporate temporal population dynamics, such as shrinkage, into travel behaviour literature.

DATA AVAILABILITY
The

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.

NOTES
1.In Ireland, evidence suggests that females are less likely to commute actively than males, favouring cars instead, an imbalance seemingly caused by demographic and location effects (Carroll et al., 2020).2. While this has historically been the case, the commercialisation of cargo-bikes may change this landscape.
3. Appendix D in the supplemental data online presents the spatial distribution of socio-demographic deprivation and mode use.4. Transport poverty is the accumulation of socio-economic and transport accessibility disadvantage whereby the social mobility of relatively deprived persons is hindered due to the inaccessibility of amenity/employment destinations (Allen & Farber, 2019). 5.The DART is a commuter train network; the LUAS is a tram line in Dublin.A van is a private commercial vehicle that carries goods and industrial equipment.6.Small areas are the smallest spatial unit in the Irish Census.There are 18,641 small areas in the Republic of Ireland.7. Ewing and Cervero (2010) highlight how variables such as population density typically display weak associations with travel behaviours once land-use characteristics are controlled for and should be treated as a mediating variable, hence its exclusion here.8. For more details on this type of transport network data, see O'Driscoll et al. ( 2022).
9. The Irish train network is not nationally comprehensive and principally serves to connect major cities with limited access outside these areas.There is only one tram network across Ireland, in Dublin.Subsequently, bus is the only nationally comprehensive Irish public transport network.10.A correlation matrix of the variables are presented in Table A1 in Appendix A in the supplemental data online.The correlation coefficients are sufficiently low so as not to raise multicollinearity concerns.11.In 2016, it was reported that 62% of households with two partners were dual earners.However, in single-earner households, 72% of the workers were male (Central Statistics Office (CSO), 2016).Considering our sample consists of workers, this may explain this somewhat counterintuitive finding.
12. We interpret this to be the result of imprecise data as road networks with low-speed limits are used to proxy for footpaths to get a comprehensive network.
compactness, directly influence travel behaviours by defining local time-space geographies, influences which are represented by arrows stemming from built environment and leading to travel considerations and commute mode choice in Figure1.From a travel behaviour perspective, this suggests that areas characterised by compactness and mixed use will facilitate greater levels of public and active travel than areas characterised by land-use segregation and low residential densities(O'Riordan et al., 2022).
POWSCAR comprises 3,058,607 individuals.Given our focus on commuting behaviour within the Republic of Ireland, our sample includes only those who(se) (1) live and work in the Republic of Ireland; (2) are gainfully employed; (3) do not report null values for key variables; (4) are of working age (20-65 years); or (5) employment does not entail primarily mobile/tele work.From this potential sample of 1,299,821 individuals, we use a 20% sample comprised of 235,936 individuals, which constitutes a stratified randomly selected sample with proportionate weighting of individuals within each socio-

Figure 1 .
Figure 1.Theoretical model visualised through a path diagram describing a generalised structural equation modelling (GSEM) framework.Source: Authors.
errors are clustered around residential electoral divisions (ED) in the generalised structural equation modelling (GSEM) prior and are shown in parentheses.***Significance at 1%.

Table 1 .
Key variables used.

Table 2 .
Marginal effects of socio-demographic changes on residential environment characteristics and travel considerations.

Table 3 .
Marginal effects socio-demographic, residential built and social environment, and travel consideration changes on commute mode choices.
data used stem from the Central Statistics Office (CSO).Individual-level data are not available publicly and access can only be granted through Research Microdata File (RMF) approval.Built environment data are available via OpenStreetMap (https://www.openstreetmap.org)as well as The Copernicus Land Monitoring Service (https://land.copernicus.eu/pan-european/corine-landcover/clc2018).