Modelling context-specific relationships between neighbourhood socioeconomic disadvantage and private car use

Abstract

Car use generates negative externalities, which are responsible for many health, environmental and economic problems. To tackle this issue, more work is needed to identify better the correlates of car use, especially at the contextual level. In this study, a mobility-focused questionnaire involving 1722 working French adults living in the Paris area (France) was used to explore gender-stratified relationships between residential socioeconomic deprivation and car use as the main transport mode against public transport, after controlling for potential confounders. While the vast majority of similar studies have assumed linear and global statistical relationships, the present work involved a random slope hierarchical generalised additive modelling framework, which revealed both non-linear and territorially-varying relationships. Among women, living in a more deprived neighbourhood was associated with an increase in the odds of reporting car use up to a certain threshold, after which the relationship plateaued, while among men, this relationship is linearly negative. In the most deprived department of the Paris region (Seine-Saint-Denis), living in a more deprived neighbourhood was associated with a lower odds of car use among men while a more complex nonlinear bell-shaped relationship was observed among women. The opposite was found in the wealthiest department (Yvelines), with a negative relationship among women and a U-shaped one among men. In Paris inner city, again a strong opposite trend was distinguished according to sex, with a negative relationship among women. These findings suggest that spatial contexts, characterised by complex interactions between socioeconomic factors, the built environment and the distance to Paris, play the role of moderators in the relationship between residential deprivation and car use. In conclusion, this study reinforces the idea that environment-transport relationships should be understood through local analyses (e.g. random slope multilevel or spatially-varying coefficients models) rather than global ones only, in order to guide specific public policies more effectively.

Introduction

Private car use is associated with significant environmental and health costs. As a completely inactive mode, car use has been associated with health-related issues, such as cardiovascular diseases (Warren et al., 2010) and obesity (Wen et al., 2010). It was also responsible for 49 road deaths per mln. Inhabitants in the European Union in 2018 (European Transport Safety Council, https://etsc.eu/euroadsafetydata/) and health issues due to air pollution (Krzyżanowski et al., 2005). Moreover, cars generated more than half of the French transport sector CO₂ emissions in 2017 (Citepa, 2019), thus contributing to global warming.

These effects have prompted the scientific community to explore the determinants of car use in order to identify potential levers for a shift toward alternative and healthier modes, including active mobility (walking and cycling) and public transport. Many studies have focused on the relationship between car use and built environment characteristics, highlighting negative relationships with public transport accessibility, residential density, activity space walkability and land use mix, and positive relationships with the presence of parking places at work/home (den Braver et al., 2020; Ewing and Cervero, 2010). Other studies have also revealed strong relationships between socioeconomic status and car use at the individual level, indicating that disadvantaged communities are less likely to have access to a car (Licaj et al., 2012; Rachele et al., 2015).

However, very little work has been done on the relationships between car use and residential neighbourhood deprivation, independently of individual and built environment factors. Moreover, these rare studies have reported inconsistent results, revealing the complexity of the process mechanism involved. In the European context, a first hypothesis assumes that the most deprived areas are those pushed further away from city centres and thus remote from the associated opportunities and services – while having fewer public transport facilities and other amenities favouring alternative modes to the car – generating more frequent private car use on average (Goodman, 2013; Xiao et al., 2018). In the economic literature, this hypothesis is consistent with a monocentric model in which deprived households have a lower valuation of accessibility and amenities of the Central Business District (CBD) than of dwelling size (Alonso, 1964; Brueckner et al., 1999; Mills, 1967; Muth, 1969; Wheaton, 1977) or with a Tiebout-like (1956) mechanism which leads the richest households to concentrate in the CBD and make housing unaffordable for poor households there. Car ownership is also shown to be related to household composition (number and age of children) and the bargaining power of spouse in couples (Picard et al., 2018) which also depend on household income.

A second hypothesis assumes an inverse relationship, based on the fact that the ownership and number of cars per household is lower on average in the most deprived neighbourhoods, implying a lower car use (Licaj et al., 2012). It is very likely that the remoteness of neighbourhoods from urban amenities interacts with this relationship between the residential socioeconomic context and car use through mediation or moderation effects (Lucas et al., 2018). Therefore, this relationship cannot be correctly considered without integrating a territorial or spatial dimension, and thus an appropriate modelling framework.

Based on a mobility-focused questionnaire involving 2002 working French adults living in the Paris area, we explored this research question through successive statistical models. This relationship between residential neighbourhood deprivation and car use was first modelled with an adjusted parametric logistic model, then with a generalised additive model to capture the non-linearity of the relationship. Because neighbourhood deprivation is spatially structured, this non-linearity amounts to spatial heterogeneity of relationships, which was emphasised through a random slope hierarchical (multilevel) generalised additive model. To the best of our knowledge, this is the first application of this method in the field of transport geography.