DO NOT QUOTE WITHOUT AUTHOR ' S PERMISSIONUsing local census data to investigate scalee

In this paper we introduce a statistical model which may be used to explain the within area homogeneity that leads to aggregation eeects. The term`within area homogeneity' is used to describe the phenomenon where two individuals from the same small area tend to be a little more alike than two individuals, each from a diierent small area. Under the statistical model we explain how the available census data for a local authority may be used to estimate the within area homogeneity for a range of variables at two diierent scales: Enumeration District (ED) level and ward level. We discuss estimation approaches and focus on the moments approach, which makes full use of the available data. Under the statistical model, we can investigate the bias of using ED or ward level correlations to estimate their individual level population values. This gives us some insight into the extent of the ecological fallacy in census data.