Population data are often collected or presented for geographical areas which may have little or no connection to the processes generating the data. Such areal units are termed 'modifiable'. However analysis undertaken on such data is not independent of how these areal units are configured. Indeed, Openshaw (1984) and others have shown that the results of statistical analysis may differ wildly according to the scale and pattern of the areal units used. This phenomenon is called the modifiable areal unit problem (MAUP). It is clear that the MAUP exists, but far from clear about how often it occurs, how often it affects the conclusions from empirical data analysis, and in what contexts it makes most (or least) difference. British census data are well suited for investigating these issues, being available for different geographies which neatly nest within each other, and for a range of different variables of interest to central and local government and to many academic disciplines. This article is concerned with bivariate correlations (using Pearson's r) between pairs of variables. The aim is to see if any variables seem particularly liable to display MAUP effects, and if so, why. The conclusion is that MAUP in many cases makes little or no difference to the results, but there are some variable pairs where the effect is substantial.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below