RE: Relationship between Getis Ord G* statistic and Kernel density function

[udaymanepalli]

Hi,

I was working on a points in a local region. I found that results analyzed by kernel density estimation and Getis Ord G* statistic yield same. I would like to know if there is any form of mathematical relationship among the two. Thanks much in advance

[LaurenRosenshein]

Hi Uday,

This is a great question.  The simple answer is that the Kernel density function and the Getis Ord Gi* statistic are completely different analyzes. The Kernel density function does a simple density calculation based on the user-specified radius and raster cell size.  For more information on the mathematics involved in Kernel Density calculations, you can check out the conceptual documentation here.  The Getis-Ord Gi* statistic, also known as Hot Spot Analysis, works by looking at the value of each feature in your dataset in the context of its neighbors' values.  The Hot Spot Analysis tool reports a z-score and a p-value, both of which represent statistical significance of clustering, which can help make the resulting maps less subjective (i.e. red represents statistically significant hot spots and blue represents statistically significant cold spots, as opposed to a density surface where the color choices are largely subjective cartographic decisions).  To learn more about how Hot Spot Analysis (Getis-Ord Gi*) works, you can take a look at the conceptual documentation here. 

There is also a video that looks at the differences between Hot Spot Analysis and Density surfaces, which you can see here: Performing Proper Density Analysis.

And for more information about the Spatial Statistics tools, you can check out our blog post which will point you to tons of videos, tutorials, seminars, and more.  The post can be found at http://bit.ly/spatialstats , and we update it whenever new materials are available!

Hope this helps!

Lauren Rosenshein
Geoprocessing Product Engineer
Spatial Statistics

[udaymanepalli]

Dear Lauren,

Thank you very much. The links you have provided are very useful and informative. Could you kindly go through the attached word file. I feel there is a connection internally. I may be incorrect. But I now have a better understanding than before. Thanks much in advance


Uday

[SalmanAhmed]

I think it is correct to assume that the hotspot tool would be a bit same as the density tool when no analysis field is specified and the point counts themseleves are taken as values to check clusturing. What makes it different from density tool is that it checks clustering of specific values of points for example if points represent cities and you input a schools field in the hotspot analsyis field then it will check if the cities with more schools or less schools cluster or not but it wont check if the cities themselves cluster which the density tool would do. So basically it checks clusturing of specific values (schools in this case) inside the points (cities in this case) as opposed to just number of points.