ABSTRACT
Environments tend not to change in a fashion that is really random with movement across a local landscape. Such patterns of persistence and coherent change cause measurements made close together to have more similarity than those made farther apart, or so-called spatial autocorrelation (Bivand et al. 2008; Isaaks and Srivastava 1989; Webster and Oliver 2001). This is a major reason that maps help bring order to operations in landscape ecological analysis and regional planning. There is thus practical purpose in systematically capturing such coherence so that it is more easily exploited (Cressie 1993; Waller and Gotway 2004). However, these proximal propensities are also problematic for conventional statistical inference based on assumptions of information independence among observational instances (Schabenberger and Gotway 2008; Webster and Oliver 1990). In effect, a newly observed instance is only partially new if others occur in the same regional regime. Thus, there are effectively fewer degrees of freedom for assessing signi¢cance, which leads to stronger indications of signi¢cance than is warranted. Therefore, it may become necessary to appeal to pseudosigni¢cance through Monte Carlo methods rather than running tests in the regular way. We explore this sort of spatial structure in the current chapter to support subsequent scenarios.
