Explaining socio-economic status differences in walking for transport: An ecological analysis of individual, social and environmental factors

Abstract

The identification of potential mechanisms of influence (mediators) of socio-economic status (SES) on walking for transport is important, because the likely opposing forces of influence may obscure pathways for intervention across different SES groups. This study examined individual, and perceived social and physical environmental mediators of the relations of individual- and area-level SES with walking for transport. Two mailed surveys, six months apart, collected data on transport-related walking and its hypothesized individual, social and environmental correlates. The sample consisted of 2194 English-speaking adults (aged 20–65) living in 154 Census Collection Districts (CCDs) of Adelaide, Australia. Individual-level SES was assessed using data on self-reported educational attainment, household income, and household size. Area-level SES was assessed using census data on median household income and household size for each selected CCD. Bootstrap generalized linear models examined associations between SES, potential mediators, and total weekly minutes and frequency of walking for transport. The product-of-coefficient test was used to assess mediating effects. Individual, social–environmental, and physical environmental factors significantly contributed to the explanation of the relations between SES and transport-related walking frequency. Educational attainment and area- and individual-level income played independent roles in explaining frequency of walking for transport, through opposing common and distinct pathways. While engagement in leisure-time physical activity was the most influential mediator of the association between educational attainment and frequency of walking for transport, the number of motorized vehicles and perceived levels of environmental aesthetics and greenery were the strongest mediators of the relations of frequency of transport-related walking with individual- and area-level income, respectively. Environmental interventions aimed at increasing residential density, reducing physical barriers to walking and traffic load, developing social-support networks, and creating greener and more aesthetically pleasing environments in more-disadvantaged areas may help to reduce SES inequalities in participation in physical activity, by facilitating walking for transport.

Introduction

Moderate-intensity physical activity incorporated into routine daily activities can help to reduce the risk of several chronic diseases (Leitzmann et al., 2007). In the last three decades, the physical activity levels of populations of industrialized countries have steadily declined (World Health Organization, 2004), although modest increases have been recently reported (Centers of Disease Control, 2007). The observed decline in level of overall physical activity has been attributed to technological and social changes affecting the behaviour of people in the workplace, at home, and in travel (Transportation Research Board, 2005). This paper focuses on travel-related physical activity and, specifically, transport-related walking, here defined as walking done to travel to and from work, to do errands, or to go from place to place (Craig et al., 2003).

Increased vehicle ownership and improvements in roadway infrastructure for the purposes of automobile use have resulted in significant reductions in the frequency of transport-related walking (Transportation Research Board, 2005). While the percentage of walking trips to work in the USA in 1960 was 10%, it was 3% in 2001; similar findings were reported in regard to travel for other purposes (Pucher & Renne, 2003). The identification of declining trends in transport-related physical activity, coupled with the negative impact of high volumes of motorized transportation on environmental air quality and urban congestion (Woodcock, Banister, Edwards, Prentice, & Roberts, 2007), and the substantial health benefits from active modes of transport (Hu, Pekkarinen, Hanninen, Tian, & Guo, 2001) have led to increased interest in understanding the determinants of transport-related walking (Lee and Moudon, 2006, Owen et al., 2007).

Designing effective physical activity promotional strategies and policy initiatives requires knowledge of socio-demographic differentials of engagement in specific forms of physical activity and of the mechanisms underlying such differentials (Taylor, Poston, Jones, & Kraft, 2006). Socio-economic status (SES) is a strong and consistent correlate of physical activity (e.g., Cole, Leslie, Bauman, Donald, & Owen, 2006) and is a major source of health inequalities (National Research Council, 1999). Low SES has been associated with higher odds of being overweight or obese and lower odds of engaging in obesity-protective behaviours, including physical activity (Harper & Lynch, 2007). Marked differences in engagement in leisure-time physical activity (LTPA or leisure activity) in favour of higher SES groups have been consistently reported (Berrigan et al., 2006, Cerin and Leslie, 2008, Hoehner et al., 2005). Several aspects of SES, including educational attainment, and individual- and neighbourhood-level household income independently contributed to leisure activity (Berrigan et al., 2006, Cerin and Leslie, 2008).

However, the associations of SES indicators with walking for transport are less clear. Some studies have identified negative relationships of individual-level (Badland and Schofield, 2006, Berrigan et al., 2006) and area-level SES with walking for transport (Giles-Corti and Donovan, 2002, Hoehner et al., 2005), but others reported non-significant (Frank et al., 2006, van Lenthe et al., 2005) or even positive associations (Ball et al., 2007, Cole et al., 2006). These contrasting results may be due to differences in measures, in attributes of the populations examined, or in adjustments for covariates. They may also result from the action of opposing, nearly-balanced mechanisms that may underlie the effects of SES on walking for transport.

Fig. 1 represents a hypothetical model of the mechanisms through which SES indicators might influence walking for transport. It is based on an analysis of the extant literature and relevant theoretical considerations (Sallis, Owen, & Fisher, 2008), which are summarized below.

Engagement in LTPA was positively associated with readiness to replace motorized with non-motorized trips (Badland & Schofield, 2006). As SES was positively related to leisure activity (Cerin & Leslie, 2008), a positive effect of SES on walking for transportation could be expected through adopting a more-active lifestyle (Fig. 1). On the other hand, higher SES, especially in the form of a higher household income, was associated with higher rates of automobile ownership and greater affordability of fuel (Giles-Corti & Donovan, 2002), which likely decreased walking for transport.

More-walkable neighbourhoods, with many diverse destinations and a developed street network, are busy places with increased levels of traffic and noise (Cerin et al., 2008, Zhu and Lee, 2008). Higher levels of traffic and noise can be attributes of lower SES neighbourhoods (Giles-Corti and Donovan, 2002, Taylor et al., 1982). Those with higher incomes have the material resources to allow them to avoid living in noisy neighbourhoods, while those with higher educational attainment may prefer to live in quieter environments due to their unwillingness to tolerate noise (Fyhri & Klæboe, 2006).

To complicate matters, there is evidence that those with higher incomes can ‘self-select’ to more desirable, more-walkable areas (Handy, Cao, & Mokhtarian, 2006) that are aesthetically pleasing and safer, which facilitate active modes of transport (Giles-Corti and Donovan, 2002, Pikora et al., 2003). Similarly, higher neighbourhood SES may encourage transport-related walking by having more appealing destinations, better walking infrastructure and lower traffic and crime (Pikora et al., 2003). These characteristics may result in higher levels of social capital (Du Toit et al., 2007, Phongsavan et al., 2006, Wood et al., 2008) defined by positive neighbourhood social norms, perceived safety, trust, social connections and reciprocity (Macintyre, Ellaway, & Cummins, 2002), which may further encourage walking. While transport-related walking may create social interaction opportunities that lead to better sense of community (Du Toit et al., 2007), higher levels of perceived social capital can facilitate walking in the neighbourhood of residence (Poortinga, 2006).

While the body of available evidence on recreational walking is substantial, no study has examined the independent contributions of individual- and area-level SES indicators to transport-related walking, together with the individual, social and environmental factors that may underlie these relationships. This is particularly important because the likely opposing forces of influence may obscure pathways for intervention across different SES groups. Using an ecological model of health behaviour applied to the study of physical activity (Sallis et al., 2008), we investigated the extent to which individual (perceived benefits of physical activity; engagement in leisure activity; number of motor vehicles in the household), social (sense of community and crime), and physical environmental factors (including residential density, traffic load, aesthetics, and walking infrastructure) mediated the relations of individual-level SES (educational attainment and household income) and area-level SES (median area-level income) with transport-related walking. As previous work showed differential associations of environmental attributes with weekly frequency and total minutes of walking for transport (Owen et al., 2007), these two outcome measures were analysed separately.

This study was conducted in Adelaide (Australia) in 2003–2004. A stratified two-stage cluster sampling design was used to recruit 2650 English-speaking adults (aged 20–60) who were residents of private dwellings and able to walk without assistance. Participant recruitment consisted of several steps. First, 425 Adelaide suburbs were ranked based on their objective transport-related walkability (residential density, street connectivity, land use mix, and net retail area; Leslie et al., 2007) and those falling in the lower and upper quartiles were identified. They were subsequently ranked according to their SES (census data on district median household income), and those falling into the upper and lower quartiles, within each walkability stratum, were identified. This led to choosing 32 suburbs, consisting of 154 census collection districts (CCD), eight by each walkability/SES stratum. The mean population (aged 20–60) and area of the selected districts were 290 persons (SD = 114) and 3.5 km² (SD = 5.2), respectively.

Simple random sampling, without replacement, was used to select households from residential addresses within the 32 suburbs, to which surveys were mailed out. The mail-out package included an introductory letter, a questionnaire, a token gratuity, a consent form and a reply-paid envelope. In households with more than one eligible participant, the person with the most recent birthday was asked to participate in the study.

Given the large amount of information requested from the participants (24 scales) and the fact that one study aim was to examine seasonal variations in physical activity, two surveys were mailed to participants with 6 months between the first (N = 2560) and second (N = 2194). Except for the items measuring physical activity, the surveys measured different constructs. This study used data from both surveys: data on socio-demographics, perceived neighbourhood environment, perceived benefits of physical activity, number of motorized vehicles in the household, weekly minutes of leisure activity and walking for transport from the first survey; and data on sense of community from the second survey. The study was approved by the Behavioural and Social Sciences Ethics Committee of the local university.

The overall response rate as a proportion of the households that received the survey invitation was 12%. Over 74% of those known to be contacted completed the first survey, and 83% of first-survey participants completed the second survey. Compared with the 2001 Census data, respondents were more likely to be female, older, and in paid work. However, no differences were found in income, educational attainment and household size.

The average participants' age was 46.3 years (SD = 12.0; range 20–60). The mean household size was 2.4 (SD = 1.3; range 1–18). Most study participants were female (60%), employed (63%), with tertiary education (46%), and living in a household with no under-18 children (69%). The percentage of participants with secondary and lower education was 29% and 23%, respectively. Approximately 34% respondents were from households with an annual income lower than AUS$ 31,200, while 42% respondents reported household incomes between AUS$ 31,200 and 77,999. The sample was roughly evenly distributed across categories of area-level median household income, with 36% respondents living in high-SES areas (household income > AUS$ 999/week) and 31% in medium (AUS$ 600–999/week) and 33% in low SES areas (<AUS$ 600/week).

The area unit in this study was a census district. Area-level SES was operationalised as the census-based median household income of a district adjusted for median household size. Area-level household income was categorized into <AUS$ 600/week (approximately the 1st and 2nd quintiles of gross household income in Australia); AUS$ 600–999/week (3rd quintile); and >AUS$ 999/week, corresponding to the 4th and 5th quintiles of income (Australian Bureau of Statistics, 2003).

Demographic and socio-economic measures obtained by self-report included age, sex, children under-18 in the household (none; one or more), employment status (employed; not employed), household size, educational attainment (year 10 or less; year 12, trade or equivalent; tertiary), and annual household income. Educational attainment and household income, adjusted for household size, were used as individual-level SES indicators.

Perceived benefits of regular physical activity were assessed using a 10-item scale developed by Hovell et al. (1989). Scores were computed by averaging the responses on the items. Participants were asked to report the number of motorized vehicles in the household. Leisure-time physical activity was measured using the long version of the International Physical Activity Questionnaire (IPAQ; Craig et al., 2003). Participants reported the frequency (number of days) and duration (hours and minutes per day) of leisure-time walking, and moderate- and vigorous-intensity physical activity undertaken in the last seven days. Total weekly minutes of leisure activity were computed and weighted by a multiple of the resting metabolic rate (MET; Craig et al., 2003).

Perceived physical neighbourhood attributes hypothesized to be related to walking for transport were measured using the Australian version of the Neighbourhood Environment Walkability Scale (NEWS-AU; Cerin et al., 2008). The NEWS-AU consists of the following scales: residential density, land use mix – diversity, access to services, street connectivity, infrastructure for walking, aesthetics and greenery, traffic load, traffic safety, crime, hilly streets, physical barriers to walking, lack of parking space, not many cul-de-sacs, and footpaths separated from traffic. Sense of community is defined as a sense of belonging to a residential neighbourhood. In this study, it was measured using three items rated on a 5-point Likert scale (Du Toit et al., 2007). Internal consistency for this scale was 0.65. Although other aspects of social capital, such as social cohesion, informal social control, and local social interaction were also measured, they were not included in the analyses due to their high correlations with sense of community (average r = 0.85) (Du Toit et al., 2007). Sense of community was selected for the purpose of this study because, unlike other measure of social capital, it indicates social connectedness and sense of belonging to a physical place.

Items from the IPAQ (Craig et al., 2003) were used to measure frequency and duration of transport-related walking during the past 7 days. Two outcome measures of walking for transport were computed: (1) frequency of walking, defined as the number of days during which participants walked for transport; (2) and total weekly minutes of walking.

The aim of this paper was to identify individual, social and environmental mediators of the relationships of SES with two measures of transport-related walking (weekly frequency and total minutes). In doing so, we employed the currently recommended product-of-coefficient test performed using bootstrapping re-sampling techniques (Cerin and Leslie, 2008, MacKinnon, 2008). This method tests the statistical significance of the product of two regression coefficients αβ, where α represents the effect of SES on the potential mediator (see Fig. 1), and β represents the SES-adjusted effect of the mediator on the measure of transport-related walking. Mediating effects (represented by the coefficients αβ) are considered statistically significant if their 95% bootstrap-based confidence intervals do not include zero. This study used cluster bootstrapping and bias-corrected confidence intervals (explained in detail in a previous paper; Cerin & Leslie, 2008).

The analysis was conducted in three steps. To identify potential mediators of the SES–walking relationships, we first examined associations of individual- and area-level SES indicators with individual, social and environmental factors. This step of the analyses provided estimates of the regression coefficients α (see Fig. 1). Analyses were performed using generalized linear models (GLM). All models were adjusted for socio-demographic confounders (age, gender, presence of children in the household, household size, and average weekly working hours). Only factors that showed a significant or nearly significant (p < .10) relationship with SES measures were entered in the subsequent mediating variable analyses models (steps 2 and 3).

In the second step of the analysis, the two outcome measures (frequency and total minutes of walking) were regressed onto the SES indicators, confounders and potential mediators identified in step 1. Step 2 of the analyses provided estimates of the regression coefficients β and direct effects of SES on walking (coefficients τ′; see Fig. 1). All regression analyses were performed using GLMs. While the model of frequency of walking used Poisson variance and logarithmic link functions, that of total minutes of walking used normal variance and logarithmic link functions.

In the third step of the analysis, the products of the coefficients αβ (representing the mediated effects) were computed across bootstrap samples and their means and 95% bias-adjusted confidence intervals obtained. The antilogarithms of these values were then calculated, representing the effects of SES indicators on walking for transport mediated by a specific factor. These values are interpreted as the proportional difference in walking for transport between levels of the SES indicator due to a specific mediator. All analyses were performed using Stata 10 (Stata Corp, 2007).