This paper is concerned explicitly with spatial aspects of primary school performance results in northern England. In particular, the effect of the data framework (e.g. the area of each individual observation and the spatial arrangement of observations as an ensemble) on observed relationships between school performance and catchment characteristics is investigated. This work is particularly relevant to the interpretation of more substantive studies involving observed relationships that necessarily depend upon the data framework. Five geometric approaches to catchment area definition are evaluated in conjunction with a simple regression model. These approaches to the spatial linkage problem are compared to an autoregressive modelling approach, in which the structure of the spatial data is treated as part of the modelling process rather than a separate spatial allocation problem.