top of page

Assessment of the presence and magnitude of Modifiable Areal Unit Problem atvarious levels of geographic data aggregation in Ontario Health Central

This report was prepared for the client, the Ontario Community Health Profiles Partnership (OCHPP) team at the MAP Centre for Urban Health Solutions at St. Michael’s Hospital. The aim of this project was to assess the presence and magnitude of the Modifiable Areal Unit Problem (MAUP), the problem of different spatial units yielding different results, at various levels of geographic data aggregation in Ontario Health Central. Recently, new “neighbourhood” boundary regions, the smallest spatial units available, were created in Toronto, and this will be expanded for other Ontario regions. In addition to neighbourhoods, there are multiple administrative regional boundaries in Ontario used in

health research, such as various types of Census regions or provincial boundaries. It follows that health research in Ontario would also be impacted by the MAUP, but the extent of these effects tend to be overlooked and are largely unstudied in Ontario.

This study used disease data provided by the OCHPP and census data from Statistics Canada to conduct Multivariate Regression Analysis and Predictive Analysis on 366 neighbourhoods in southern Ontario. The results indicate that significant predictors at different geographic levels are distinct, and that a model from a high aggregation level will not be able to accurately predict the dependent variable at a low aggregation level, but vice versa. Furthermore, as the levels of aggregation increase, there is only an increase in correlation between a few variables and others being weakened.

Thus, this study provides support that the MAUP influences diabetes research when using different levels of geographic data aggregation. In this project’s findings, some variables do not remain significant at the CSD level. This was contradictory to typical findings in the literature seeing stronger correlations with higher data aggregation, but this finding is likely attributed to a small sample size and consequent lack of power in the CSD analyses. However, and fortunately for the sake of confidence in the diabetes literature, the overall correlations between the predictors and diabetes prevalence remained generally consistent. We further discuss limitations and recommendations to handle the MAUP.

16 views0 comments


bottom of page