SOME EFFECTS OF SPATIAL AGGREGATION ON MULTIVARIATE REGRESSION PARAMETERS

The Modifiable Area Unit Problem (MAUP), a term introduced in Openshaw and Taylor’s (1979) classic paper, has long been recognized as a potentially troublesome feature of spatially aggregated data, such as census data. Aggregation of unit-level (or other high-resolution) data to lower resolution areas is an almost unavoidable feature of large spatial datasets due to the requirements of privacy and/or data manageability. When the original data are aggregated, the values for the various univariate, bivariate, and multivariate parameters will change because of the loss of information. This phenomenon is called the scale effect. The N spatial units to which the higher-resolution data are aggregated, such as census enumeration areas or tracts, postal code districts, or political divisions of various levels, are arbitrarily created by some decision-making process and represent only one of an almost infinite number of ways to partition a region into N zones. Each partitioning will result in different values for the aggregated statistics; this variation in values is known as the zoning effect. The two effects are not independent, because the lower-resolution spatial structure may be built from contiguous higher-resolution units, such as census tracts from enumeration areas, and the resulting aggregate statistics will be different for each aggregation.