# What are best practices to compare densities and locations between two sets of points?

I have two sets of point shapefiles. They represent two different customer groups for a retail store. Shapefile A represents online customers, Shapefile B represents in-store customers.

What I would like to do is create some sort of heat map that displays where the concentrations of these customers are. That way I can hopefully show the differences in where these two different sets of customers are located in comparison to our actual store locations. I even have how much money each individual customer has spent attributed in both shapefiles if I wanted to organize this by amount spent. What do you think would be the best methodology to show this? I have the Spatial Analyst and Business Analyst extensions installed.

-

The methodology you are looking for is called 'spatial statistics', and specifically 'spatial point-pattern analysis'.

You could do a kernel-smoothing of the two point patterns and then compute the ratio of the kernel-smoothings. However, this is sensitive to choice of kernel shape and bandwidth, so if you wanted to draw robust statistical conclusions (something like "there are 2x the online customers than retail at this point") then you have to properly formulate the model and choose you bandwidth carefully. If you just want a pretty map to wave at the PHBs, just choose something that smooths things "enough" so it doesn't look dotty and doesn't smudge out all the detail.

If you have a natural set of areas, such as electoral regions, or census districts then you could count the points of each type (or the sum of spend for each type) in each district and plot a map of the ratio. Again, these will be subject to statistical variation as well as genuine variation so if you want to state robust statistical conclusions you need some fancy schmancy statistical model that computes the ratios and the errors on the estimates of the ratios.

And if you want to do that, you probably want to get a book on spatial statistics (Cressie, perhaps) and a copy of R (Open source stats system) which has some add-on mapping functionality in a bunch of packages.

-

I watched this ESRI video on hotspot analysis a while back. The speakers give many good tips and pointers. The example is local in scope, but the methods should work at most scales. They demonstrate how to collect events into locations that can then be analyzed. They also walk through many of the supporting processes.

-