Inputs for spatial autocorrelation

Question

I am trying to use the spatial autocorrelation function in the statistical analysis toolbox but I am unsure what I need to set as the inputs. I want to use it to test for autocorrelation in a set of 'samples' from the seafloor that have been used to test the accuracy of an automated classification of substrate type.

The Input Feature Class would presumably be the locations of the samples as X, Y points in the map, and then is the Input Field the 'value' for the substrate class assigned to that point?

Another issue is that the 'samples' we are using are along transects, how accurate can an autocorrelation test be when the points are located along a line?

Answer 1

You need to put some thought into your experimental design and talk with somebody well versed in statistical analysis. In what you are describing, your experimental units are the transects and not the observations along them. This severely limits the types of analysis that can be applied and power is likely an issue.

In terms of autocorrelation effects on a linear model, it is irrelevant if the transect-based observations are autocorrelated. Your degrees of freedom would be defined as [number of transects - 1] and residual error would be based on the aggregate value of each transect, which would define y.

If you used a mixed effect model you could use the transect-based observations and treat the transect as a random effect. In this case, if you had a high level of autocorrelation at the observation level then you could define a random effect term for this as well.

However, this would be tricky because the sample is constrained (conditional) to a linear feature and the assumption is that events occur probabilistically across a random field. This is akin to issues arising from quantifying spatial pattern along linearly constrained features, such as streams, and violates assumptions of stationarity (anisotropy). A standard statistic, such as Moran's-I, would be grossly biased.