Study aim, design and setting
This study was a subcomponent of a cross sectional study conducted in the district of Colombo with a sample of 1320 to assess the prevalence and correlates of PA reporting that 79.4% walked for transport and only a mere 14.5% walked for leisure . The GIS analysis was carried out only in the CMC area, with the district which is the island's economic capital, by far the country’s biggest city, and the most urbanised. The aim was to report on the objectively measured physical environmental factors associated with walking using GIS. The main reason for selecting only the CMC area for GIS analysis was the availability of digital data such as the road structure, land use and other services making it possible to carry out environmental data analysis using GIS. Data collection was carried out from September 2010- February 2011.
Selection of participants
The sample was selected using a cluster sampling method with the primary sampling unit being a Grama Niladhari Division (GND), the smallest administrative entity for which information on socio demographics are produced by Department of Census and Statistics in Sri Lanka. The GNDs selected are shown in Figure 1. From each GND 40 households were selected. The detailed sampling mechanism has been described previously (7).
Primary data collection
Survey measures: The study instrument was a reliable and pre-tested interviewer administered questionnaire. Socio-demographics measures that were assessed were the age, sex, ethnicity, civil status, educational level, average monthly household income and self-rated health status. Walking was assessed using the validated International Physical Activity Questionnaire, Long Form (IPAQ) . It is a widely used, standardized instrument to measure the habitual practice of PA of populations over the last 7 days [23, 24,25]. Participants reported the frequency and usual duration of each type of activity including walking, undertaken during the previous week in the different domains of job, transport, domestic/garden chores and leisure. However, for this analysis only the transport and leisure related walking was selected as the neighbourhood environment features are mostly related to them [17,21].
Location measures: The data collectors with a medical background were trained on measuring the coordinates of the housing location. After completing the questionnaire, the coordinates of the locations were recorded using Megallon eXplorist 510 GPS units.
Secondary data collection
Population estimates from the most recent census data in Sri Lanka was used to determine population density . Neighborhood environment characteristics were obtained from the Colombo land use maps (1:50,000) which were the recent most data which were, geocoded and verified by the Survey Department, Sri Lanka. The details of the digital data layers including the scale and format are shown in Figure 2. No parcel data were available.
Using the above secondary data sources the following density and distance measures were calculated.
Density measures: Density measures were carried out within the 200 m, 400 m and 600 m buffer limit of the participant’s residence. All buffers were straight-line buffers as the network structure was incomplete for Colombo. The neighbourhoods were defined by creating a 600 m radius “straight line” around each geo-coded participant’s address. Small radii of 200 m and 400 m were also evaluated as it was hypothesised that a smaller area around one’s home might be more influential in individual’s choice to walk . Population and housing density was calculated using data from the national census (26). The built area, access and connectivity density measures were calculated from the digital data from the survey department. The land use measure is a measure of ground covered with buildings, which cannot be used for walking. The number of buildings and the building foot area were then calculated within the buffer zones. Road lengths (both main and other roads) per unit area, intersections per unit area, and number of bus stops per unit area were used to assess access and connectivity.
Distance measures: Distance to the nearest major roads, nearest beach and the nearest park was calculated using the data from the survey department (Figure 3) which were all were straight-line distances as network work structures were not available.
GIS spatial analysis
Quality of data was verified before GIS spatial analysis. The secondary data that were collected were the most recent data. However, these were not assessed for count error (incomplete data), attribute error (inaccurate classification of facilitates or characteristics) and the positional error (inaccurate geo coding) due to logistic constraints.
The primary data collected, were rechecked for its accuracy in 10 housing locations by the experts in GPS. The coordinates of the household were downloaded from the GPS units using the vantage point software and was further visualised using the OZI explorer software. Thereafter, it was converted to shape file format as a point layer by the OZI explorer software. These were converted from their original spatial reference parameters of WGS84 to Kandawala system to be appropriate for Sri Lanka.
Initially, all the spatial data were converted in to ArcGIS geodatabase data model. Thereafter, a feature dataset was created. ArcGIS 9.3 software and its extensions such as 3D Analyst, Network Analyst, and Data interoperability were used in the study. Further, free extensions available such as Hawth’s tools, X tools etc were also used for the various spatial analysis in the study. The physical environmental variables were created and were based on the literature available, especially the GIS protocol developed for the Twin City study . The Concept and formulae in conducting GIS spatial analysis, for the indices are outline in Figure 3.
Analysis of walking
Total number of minutes of walking carried out during transportation and leisure was calculated as per the IPAQ protocol (23, 24)
Statistical analysis was performed using SPSS version 17. The physical environmental variables created through the GIS systems for each individual was correlated with his/her minutes of walking a week. Spearman correlation coefficient was calculated as the data were not normally distributed. A threshold of 0.05 for statistical significance was used.