The Comparison of Means When Samples Consist of Spatially Autocorrelated Observations

A basic assumption of many statistical methods, which is rarely satisfied by geographically-located observations, is that the data to be analysed represent (spatially) independent drawings from some population or populations. In this paper, we illustrate the disastrous consequences of the failure to meet this assumption upon applications of tests for means based upon Student's t distribution. Procedures are developed which enable such tests to be applied more appropriately in the presence of autocorrelated samples.