Hot Spot Results seem dubious

[PaulaMcNamara]

Hi,
I have been experimenting with Hot Spot analysis a bit since my last post and managed to get it working for some point data we required (at least i think i did).
I've now been tasked (after they liked it so much) to extend on it and I'm now working with volume data based on area boundaries.  They would like hot spots done for these.
I went through the same sort of procedure as for points (except not need to integrate and collect events) and I have experimented with analysis based on state or national scales, various distances (based on doing the autocorrelation, and classification types.
Attached are the results of my analysis using Zone of Indifference at a state based distance of 45,000m.  My polygons are very uneven and unevenly dispersed.  some also have no known volume.

When you look at the results (and it gets more confusing when looking at all the different results from different options) you can see where the highest volumes are on the bottom image.  how come after hotspot analysis not all these high volumes are included in a hot spot?  I would have though all these high volumes together would make one hotspot, but do the polygons have to be touching for that to occur?

Other results were putting hotspots where there was low volume counts.  even more confusing.

Am i missing something here?

[PhilipGibbons]

Hi Paula,

I have experimented with Hot Spot Analysis over the years and have encountered similar problems/questions (of which I never have been able to answer).  Keep in mind that your hot spots and cold spots are based off of statistically significant Z Scores.  So, you may have "high" values for certain polygons, but they may not be statistically significant as compared to other polygons with high values.  I find when I am doing cluster analysis that it is useful to not only map out the statistically significant Z Scores, but to also map out ALL of the Z-Scores to simply explore the data visually.  Also, you can run the Local Moran's I statistics and map out the Moran's I results as well as statistically significant Local Moran's I Z Scores to further explore the data.  Why do this, you ask?  Well, it just helps to really understand your data a bit more.  ESRI has great resources for understanding how all of these statistics work, so read up on them.  Also, Andy Mitchel's book on Spatial Statistics (Vol. 2) is a great resource for understanding cluster analysis.  And lastly, if you are working with polygons rather than point data, perhaps change your conceptualization of spatial relationship to something else: polygon contiguity? 

-Phil