Small-scale health-related indicator acquisition using secondary data spatial interpolation
1 School of Planning, University of Waterloo, Waterloo, Ontario, N2L3G1, Canada
2 Department of Health Studies and Gerontology, University of Waterloo, Waterloo, Ontario, N2L3G1, Canada
3 Department of Statistics and Actuarial Science, University of Waterloo, Waterloo, Ontario, N2L3G1, Canada
International Journal of Health Geographics 2010, 9:50 doi:10.1186/1476-072X-9-50Published: 13 October 2010
Due to the lack of small-scale neighbourhood-level health related indicators, the analysis of social and spatial determinants of health often encounter difficulties in assessing the interrelations of neighbourhood and health. Although secondary data sources are now becoming increasingly available, they usually cannot be directly utilized for analysis in other than the designed study due to sampling issues. This paper aims to develop data handling and spatial interpolation procedures to obtain small area level variables using the Canadian Community Health Surveys (CCHS) data so that meaningful small-scale neighbourhood level health-related indicators can be obtained for community health research and health geographical analysis.
Through the analysis of spatial autocorrelation, cross validation comparison, and modeled effect comparison with census data, kriging is identified as the most appropriate spatial interpolation method for obtaining predicted values of CCHS variables at unknown locations. Based on the spatial structures of CCHS data, kriging parameters are suggested and potential small-area-level health-related indicators are derived. An empirical study is conducted to demonstrate the effective use of derived neighbourhood variables in spatial statistical modeling. Suggestions are also given on the accuracy, reliability and usage of the obtained small area level indicators, as well as further improvements of the interpolation procedures.
CCHS variables are moderately spatially autocorrelated, making kriging a valid method for predicting values at unsampled locations. The derived variables are reliable but somewhat smoother, with smaller variations than the real values. As potential neighbourhood exposures in spatial statistical modeling, these variables are more suitable to be used for exploring potential associations than for testing the significance of these associations, especially for associations that are barely significant. Given the spatial dependency of current health-related risks, the developed procedures are expected to be useful for other similar health surveys to obtain small area level indicators.