Data were collected in the Resilience for Eating and Activity Despite Inequality (READI) study during 2007–8 [14, 15]. The Deakin University Human Research Ethics Committee approved this study, with written informed consent obtained.
Participants
Postcodes in the bottom third of the Socio-Economic Index for Areas (SEIFA) [16] distribution in Victoria, Australia, were classified as ‘disadvantaged’. Neighbourhoods scoring in the bottom third of this aggregate index are characterized, for example, by higher proportions of residents with relatively low incomes, low education levels, and higher rates of unemployment. Forty urban and 40 rural areas were randomly selected. From the Australian electoral roll (compulsory from age 18), 150 women (18–45 years) were randomly identified in each area (n = 11,940; some areas had <150 eligible women). After excluding 861 ‘return to sender’ surveys, 4,938 (45%) women completed a postal questionnaire. Excluded were those moving from the sampled suburb prior to survey completion (n = 571), surveys completed by a non-intended person (n = 3), those withdrawing data (n = 2), or those <17- or >46-years old (n = 13). Of the remaining participants (n = 4,349), those pregnant (n = 210); with missing data for pregnancy (n = 40), physical activity (n = 165), or environment (n = 97); or with >1 missing psychosocial factor (n = 22) or demographic factor (n = 74) were excluded, leaving 3,765 participants.
Measures
Leisure time physical activity
Participants reported past week physical activity duration, frequency and intensity within leisure, transportation, occupational and domestic domains in the International Physical Activity Questionnaire (long version), which has demonstrated high reliability and acceptable validity across at least 12 countries [17]. Leisure time physical activity (LTPA) is discretionary and under volitional control (unlike for example, occupational activity) and therefore more likely to be amenable to intervention. In addition, LTPA conceptually is most likely to be influenced by local neighborhood environmental features (i.e., we would not expect to observe an association between perceived neighbourhood aesthetics and domestic activity, or between the local physical activity environment and occupational activity) [18]. For these reasons, LTPA was chosen as the focus for this paper, and included time spent walking for leisure, as well as moderate- and vigorous-intensity physical activity for leisure. Minutes/week of LTPA were calculated, then categorised because of high skewness resulting from the large number of zero values (a suitable transformation could not be determined). Domain-specific physical activity guidelines currently do not exist, so a categorical variable was calculated by classifying LTPA as 0 minutes/week (to account for the large number of zero values), then using tertile cutpoints to categorise the non-zero data as 1–119 minutes/week, 120–279 minutes/week, and ≥280 minutes/week. Tertiles were selected as they allow for an acceptable number of participants in each category while retaining sensitivity.
Psychosocial score
A ‘psychosocial score’ was created using four individual (self-efficacy, enjoyment, intentions, behavioural skills) and four social (child care, family support, friend support, dog ownership) variables previously found to be associated with LTPA in this [8] and other [2, 13, 19] samples. Where scales were used, internal reliability as assessed in this sample is presented as Cronbach’s alpha. Participants rated their confidence (self-efficacy) on a five-point Likert scale in being active in five difficult circumstances [20]; responses were summed (Cronbach’s alpha = 0.82). Responses to six items on a seven-point Likert scale about physical activity enjoyment [21] were summed (Cronbach’s alpha = 0.95). A seven-point Likert scale was used to assess activity intentions [22], and behavioural skills were assessed by summing two questions about past week frequency of goal setting and planning for physical activity [22] (Cronbach’s alpha = 0.83). Participants reported access to childcare if they wanted to be active (not applicable, yes, no), and dog ownership (yes, no). Family social support was assessed by asking how often in the past year family members engaged in physical activity with them or encouraged physical activity [23]; responses to the two questions were summed (Cronbach’s alpha = 0.76). The same questions were asked in reference to friends/work colleagues (Cronbach’s alpha = 0.69).
Childcare was scored 0 (not available) or 1 (available or not applicable), and dog ownership was scored 0 (no dog) or 1 (dog owner). To overcome differences in scales, each of the other four individual and two social variables were categorised into thirds and allocated a score of 0 (lowest third), 1 (middle third) or 2 (highest third). Scores for the eight variables were summed to create an overall ‘psychosocial score’, with a possible range of 0–14. This score was classified as low (0–6), medium (7–9) or high (10–13) based on tertile cutpoints, with higher scores indicating more favourable psychosocial characteristics. While some information on specific variables may be lost, this approach provides an opportunity to categorise discordant participant psychosocial ‘profiles’ in a parsimonious manner. That is, a range of psychosocial variables are incorporated into a single score, providing an aggregate indicator of psychosocial support, and reducing complexity and issues associated with multiple testing of numerous constructs in isolation. The internal consistency of the psychosocial score was reasonable (Cronbach’s alpha = 0.66).
Perceived environment score
A perceived neighbourhood ‘environment score’ was created from three self-reported variables described in our earlier work [8]. Briefly, personal safety (sum of three items; Cronbach’s alpha = 0.85), aesthetics (sum of five items; Cronbach’s alpha = 0.76), and the ‘physical activity environment’ (sum of seven items; Cronbach’s alpha = 0.80) were assessed on five-point Likert scales adapted from an existing measure [24]. To overcome differences in scales, each of the three environmental variables was categorised into thirds and allocated a score of 0 (lowest third), 1 (middle third) or 2 (highest third). Scores for the three variables were summed to create the score, with a possible and actual range of 0–6. This score was classified as low (0–1), medium (2–3) or high (4–6) based on tertile cutpoints, with higher scores indicating more favourable physical environments. The internal consistency of the environment score was reasonable (Cronbach’s alpha = 0.63).
Potential covariates
Self-reported demographic characteristics included age; marital status (married/living as married, previously married, or not married); level of education (low = less than Year 12; mid = Year 12, certificate/trade/diploma; high = tertiary); employment status (full-time work, part-time work, or not employed); children <18 years (none, 1, 2, 3+); birth country (Australia or outside Australia); height and weight, converted to body mass index (BMI) and categorised as healthy (including underweight), overweight or obese (≤25 kg/m2, 25–29.9 kg/m2, ≥30 kg/m2, respectively) [25]; presence of illness/injury/disability likely to affect physical activity (yes, no); current smoking (yes, no); and access to a motor vehicle (yes, no). Area of residence (urban/rural) was assessed as a potential covariate. Urban and rural are defined elsewhere in detail (Cleland, Hume, et al., 2010). Briefly, urban areas included metropolitan Melbourne, rural cities (defined by the Regional Infrastructure Development Act 1999 [Vic]) and all suburbs completely within a 10 km radius of the centroid of these cities, and all suburbs completely within a 10 km radius of the centroid of other Victorian cities with a population of ≥20,000. Rural areas were those falling outside of metropolitan Melbourne and outside of a 25 km radius of the rural cities.
Analyses
Demographic characteristics (potential covariates) are presented for the overall sample and across categories of LTPA. Differences across these categories were tested using one-way analysis-of-variance (equal variances) or Kruskal-Wallis equality-of-populations rank test (unequal variances) for continuous variables, and Pearson’s chi-square test for categorical variables. Mean (and standard deviation) and median (and inter-quartile range) LTPA is presented overall and across categories of environment score categories (low, medium, high) and psychosocial score categories (low, medium, high). Cuzick’s non-parametric test for trend was used to assess whether a significant trend was evident for LTPA across increasing environment score categories and across increasing psychosocial score categories.
Because of the ordered but not evenly spaced nature of the outcome data, generalised ordered logistic regression was employed by using the ‘gologit2’ command [26] in Stata, Version 10.2 (Statacorp, College Station, Texas) to calculate odds ratios (OR) and 95% confidence intervals (CI) for increasing categories of LTPA across categories of the environment score. These analyses were repeated stratified by psychosocial score categories to examine the association between the environment and LTPA across categories of psychosocial scores. In generalised ordered logistic regression, the actual values of the dependent variable (LTPA) are irrelevant except that it is assumed that larger values correspond to ‘higher’ outcomes. The results indicate the odds of participating in increasing LTPA associated with one-unit increases in the independent variables (e.g. the difference between a low and medium environment score). Unlike standard logistic regression where increasing categories are compared to a typical ‘referent’ group (e.g. the bottom/lowest category), ordered logistic regression compares each increasing outcome category to all data falling below the cutpoint for that category (e.g. <120 minutes/week is the referent group for calculating the odds of achieving ≥120 minutes/week LTPA, while <280 minutes/week is the referent group for calculating the odds of achieving ≥280 minutes/week LTPA). Demographic characteristics that were associated with LTPA and the environment score were included as confounders in regression models. Analyses were conducted in 2010 and included adjustment of the standard errors for the effects of clustering within the 80 areas (the unit of recruitment).