Incremental Spatial Autocorrelation guidance

Discussion created by tlg230 on Mar 6, 2012
I am using the new Incremental Spatial Autocorrelation script to establish valid distance bands for a point pattern consisting of prehistoric archaeological sites in the Kaibab National Forest in Arizona. I have successfully established the nearest neighbor tendencies and used the maximum neighbor distance to set the beginning distance and the average observed distance to set the distance increment. My results, however, demonstrate  Z-scores that consistently show at on the beginning distance threshold and drop from there. Out of curiosity, I set the beginning threshold to half of the maximum neighbor distance and saw the peaks fall earlier than in the previous iteration. While these results are generally considered to be statistically invalid, my question is why? Is it because the smaller distance is only capturing part of the pattern? If I have some extreme outliers, wouldn't omitting them from the Moran's I test still pick up relevant metrics on the remaining (majority) of the pattern? Is there ever a case where using a smaller beginning distance might be a good idea?

Tim Gibbs
Geospatial Analyst
Museum of Northern Arizona