4

I have two raster datasets and I'm trying to determine the extent of their spatial relationship. I suspect they are closely related but I would like to produce a map output to illustrate this relationship.

The datasets are:

1 - a kernel density of prevalence of gambling locations (from their point locations)

2 - a deprivation index by Census Output Area converted to raster.

I don't know if it's best doing this as polygons or as rasters and which technique will provide me with what I want. Does anyone have any ideas of how I can investigate and represent this relationship in Arc?

  • To make sure I understand this: you want to calculate the correlation between gambling points and deprivation index? Or density of gambling points and deprivation index? – Erica Apr 4 '14 at 13:21
  • 1
    Yes, density of gambling points and deprivation index. Basically to illustrate the connection between deprivation and access to gambling facilities. – MyFamily Apr 4 '14 at 13:26
4

You might want to consider using some freeware such as GeoDa or Crimstat for running spatial correlation analysis. I found them way more useful than ArcGIS for that type of thing if you're willing to give them a go!

http://geodacenter.asu.edu/software

http://www.icpsr.umich.edu/CrimeStat/

  • I guess to use Geoda I would have to do a spatial join of the gambling facility kernel density values (after they've been converted to point data) to the deprivation Output Area polygons? Would this get them in the same attribute table for Geoda? – MyFamily Apr 4 '14 at 14:08
  • I'm not 100% sure on what format your data table will need to be in (sorry, it's been a long time since I've used GeoDa), so you'd probably be best checking forums for that. I remember there being some excellent step-by-step pdf guides for GeoDa available somewhere. – pvdev Apr 4 '14 at 15:27
  • @MyFamily You're welcome, glad it worked out well! – pvdev Apr 7 '14 at 8:59
2

I would personally do this as rasters, each normalized on a 0-1 scale. Multiply the two together, and you have a quick-and-dirty visualization: the closer to one, the higher the correlation.

A more statistical (and perhaps more robust) approach would be to use Band Collection Statistics, which gives you text output of the correlation between the two rasters. (This tool would also work with >2 rasters if desired, so you could also throw in, say, average age of residents if you felt like it.)

I believe both approaches require the spatial analyst extension.

  • Many thanks for this, I had already done the Band Collection Statistics and the result came out how I hoped. I just needed a more visual representation. I'll give your a suggestion a try now :) – MyFamily Apr 4 '14 at 13:44
  • Also, how would that account for locations with a low deprivation score and low gambling facility density - they have a high correlation but would get a low score in the model you suggest? I may be wrong! – MyFamily Apr 4 '14 at 13:50
  • This is why I say it's quick-and-dirty. It isn't a true "correlation" surface. A better method may be dividing deprivation by gambling to get a ratio. – Erica Apr 4 '14 at 13:57
  • Also, if you're mapping "correlation," your map will just be "strong correlation" everywhere (whether it's high gambling high deprivation or low gambling low deprivation). You may just need a different way of visualizing this. – Erica Apr 4 '14 at 14:00
  • Hi Erica, I find your first approach is misleading - where have you seen this used? – mwil Apr 4 '14 at 14:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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