My terse review of the literature suggests there are still many questions about how spatial autocorrelation (SA) should be included within ecological models. For example, what is the impact of not including SA on either the parameter estimates or standard error estimates? What are the conditions under which either may become biased, such as the small sample sizes that are typical of ecological research? Asked differently, how much data is needed in the presence of SA to accurately detect it?
Also, for the various models that are typically used, what are the differences in terms of structure as well as what they are able to detect? Which model should thus be used under any given scenario?
Last updated: 9/27/07.