Theorising and testing environmental pathways to behaviour change: Natural experimental study of the perception and use of new infrastructure to promote walking and cycling in local communities

Panter, Ogilvie, 2015, in BMJ Open

doi:10.1136/bmjopen-2015-007593
Location Southampton, Cardiff, and Kenilworth, UK
Population General
Sample size 1465
Factor analysis type principal components, varimax rotation
Stepwise regression no
Removal of insignificant variables no
Reviewed by LCM

Abstract

Objective: Some studies have assessed the effectiveness of environmental interventions to promote physical activity, but few have examined how such interventions work. We investigated the environmental mechanisms linking an infrastructural intervention with behaviour change. Design: Natural experimental study. Setting: Three UK municipalities (Southampton, Cardiff and Kenilworth). Participants: Adults living within 5 km of new walking and cycling infrastructure. Intervention: Construction or improvement of walking and cycling routes. Exposure to the intervention was defined in terms of residential proximity. Outcome measures: Questionnaires at baseline and 2-year follow-up assessed perceptions of the supportiveness of the environment, use of the new infrastructure, and walking and cycling behaviours. Analysis proceeded via factor analysis of perceptions of the physical environment (step 1) and regression analysis to identify plausible pathways involving physical and social environmental mediators and refine the intervention theory (step 2) to a final path analysis to test the model (step 3). Results: Participants who lived near and used the new routes reported improvements in their perceptions of provision and safety. However, path analysis (step 3, n=967) showed that the effects of the intervention on changes in time spent walking and cycling were largely (90%) explained by a simple causal pathway involving use of the new routes, and other pathways involving changes in environmental cognitions explained only a small proportion of the effect. Conclusions: Physical improvement of the environment itself was the key to the effectiveness of the intervention, and seeking to change people's perceptions may be of limited value. Studies of how interventions lead to population behaviour change should complement those concerned with estimating their effects in supporting valid causal inference.

Factors

Variable Loading
There are pavements suitable for walking (Other) 0.732
There are special lanes, routes or paths for cycling (Safety) 0.688
The routes are pleasant for walking or cycling (Comfort) 0.706
The routes for walking and cycling are generally well lit at night (Safety) 0.695
Variable Loading
Walking is unsafe because of the traffic (Safety) 0.809
Cycling is unsafe because of the traffic (Safety) 0.804
The level of crime or antisocial behavior means walking or cycling is unsafe (Safety) 0.678
Variable Loading
I see people in my neighborhood cycling for travel (Social norms) nan
I see people in my neighborhood walking for travel (Social norms) nan
I see people in my neighborhood cycling for recreation (Social norms) nan
I see people in my neighborhood walking for recreation (Social norms) nan

Models

Dependent variable Use of intervention (yes/no)
Model type Logistic regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Odds_ratio p-value
Residential proximity to intervention (km) 1.85 0.001
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in time spent walking and cycling (min/week)
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Use of intervention (yes/no) 31.16 0.063
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in infrastructure
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Residential proximity to intervention (km) 0.05 0.03
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in safety
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Residential proximity to intervention (km) 0.03 0.182
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in visibility
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Residential proximity to intervention (km) 0.05 0.013
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Use of intervention (yes/no)
Model type Logistic regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Odds_ratio p-value
Change in infrastructure 1.23 0.008
Change in safety 1.31 0.001
Change in visibility 1.33 0.001
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in time spent walking and cycling (min/week)
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Change in infrastructure -2.51 0.736
Change in safety 9.19 0.215
Change in visibility -6.21 0.398
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in visibility
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Change in infrastructure 0.06 0.039
Change in safety 0.03 0.328
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan
Dependent variable Change in safety
Model type Linear regression
Sample size 969-1139
R2 nan
Adjusted R2
Pseudo R2 (nan) nan
AIC nan
BIC nan
Log-likelihood at zero nan
Log-likelihood at constants nan
Log-likelihood at convergence nan
Variable Coefficient p-value
Change in infrastructure -0.03 0.215
Time spent walking and cycling at baseline CTRL nan
Sex CTRL nan
Age CTRL nan
Ethnicity CTRL nan
Presence of children under 16 CTRL nan
Height CTRL nan
Weight CTRL nan
Education CTRL nan
Income CTRL nan
Employment status CTRL nan
General health CTRL nan
Presence of long-term illness or disability limiting daily activities CTRL nan

The Attitudes and Travel Database is produced with support from the Center for Teaching Old Models New Tricks at Arizona State University, a University Transportation Center sponsored by the US Department of Transportation through Grant No. 69A3551747116.

sha256:a08d9e369743bf7e6d1c40d27347318209b40a7fb1543813fdcf31b898918815