Spatial Autocorrelation and Experimental Design
I felt that I learned a good lesson the other day from Dr. Prasit about doing experimental work. I normally study variation in natural populations and in this situation, it is best to minimize the spatial autocorrelation between samples. In this way, you can be sure that each sample is independent. The easiest way to do this is to simply distribute the plots as far apart as possible across the sample area.
But, this is a bad idea for setting up experiments in which you want to compare the effect of different treatments. Because you want the treatment to be the ONLY difference between your different treatment plots. Let’s say, if you have one control and two treatments, these need to be set up in groups of three, in close proximity. In this situation, you actually want to maximize spatial autocorrelation between treatments. Now, of course, you should then separate the replicates of the experiment across the range of natural environmental factors. So, depending on your objectives, you may want to maximize spatial autocorrelation within each replicate of your experiment but then minimize spatial autocorrelation between replicates.
I know I talked about this in the field but I thought this was an interesting topic and I wanted to make it clear. Is it clear to you?