Statistical Analysis of Experiments Conducted at Multiple Sites

Ecological phenomena are often spatially variable, and hence it is desirable to simultaneously conduct experiments at multiple sites. However, the number of sites is often constrained to be few, resulting in many more replicates within sites than there are sites. Such experiments are commonly analyzed by mixed model ANOVAs (where site is the random effect), which are very powerful at detecting differences among sites and siteby-treatment interactions. But, since few sites are sampled, there is little statistical power to make inferences concerning the treatment effects themselves. This lack of power is unfortunate whenever the primary reason for conducting the experiment is to test for differences among treatments, as is true for most ecological field experiments. In this note we suggest that, when there are few sites, the treatment effect should be examined only within sites using a nested ANOVA.