6

I am trying to locate areas with a high rate of accidents involving bicyclists or pedestrians per car accident of any kind. I have a point file of accidents and a point file of all those with bike/ped involvement. I want to use the Hot Spot Analysis tool, but all I can figure out is using it on each point file to tell me the overall crash hotspots and the bike/ped crash hotspots. Is there a way to normalize one by the other? Is there a good way to compare the two hotspot output files? is there a different tool or approach I should look into?

Improve this question
2
  • Are you aggregating points to some pre-defined areal unit? Or do you want to work with the point data directly? If you are using aggregate data, perhaps visualizing location quotients or Oden's IPOP statistic would be something your interested in.
    – Andy W
    Commented Jul 17, 2012 at 1:48
  • The points are aggregated using the "collect events" tool, but I want them to stay point data to get the finest grain possible in the analysis. Unfortunately I haven't the slightest idea how to go about using the statistic you suggested, or if the way it is standardized is compatible with my goal.
    – M.A.Primus
    Commented Jul 17, 2012 at 12:45

1 Answer 1

Reset to default
3

I can think of some more sophisticated methods outside of ArcGIS (the section on marked point patterns in the "Analysis of spatial point patterns in R" file is excellent ), but for a good start with ArcGIS:

  1. Make sure your "all accidents" file includes the ped/bike accidents. Combine into one if it does not.
  2. Run a Point Density (from the Spatial Analysis toolbox) measure on the "all accidents file" carefully defining the output raster to be reasonable given your study area.
  3. Run a Point Density measure on the ped/bike file making sure to define the output raster as identical to the all accidents density raster.
  4. Create a new raster using raster math equal to "ped/bike density" divided by "all density"

This is a map showing the relative share of accidents that are ped/bike.

Diagnostics can be more or less sophisticated depending on your skills and your audience. Map the outliers (the highest 5% and lowest 5% of cell values) Do they look clustered? Make a histogram of the distribution, what does it look like?

Again, if you want more sophistication you can do all sorts of things modeling the likelihood of getting values that look like the observed distribution, look at some of the text from the pdf above starting around page 165 and see where they take it.

Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.