[R-sig-Geo] relative risk (was: negative area of polygon in spatstat::owin)

Adrian Baddeley adrian.baddeley at uwa.edu.au
Sat Sep 21 03:48:54 CEST 2013


Jeff Marcus <jeff.n.marcus at gmail.com> hath writ:

  > Some context on problem: I am trying to use functions in spatstat for
 > spatial relative risk calculations, i.e, the spatial ratio of denstity of
 > cases vs. controls. For that I need an observation window and point plot
 > within that window. 
 > I could cheat and make the observation window a rectangle around massachusetts 
 > but that would presumably distort values near the coast.

In general, this is the correct approach, and this is why spatstat insists that every point pattern
must have a window.

However, for the special case of relative risk, cheating would be OK for some purposes.

The estimate of relative risk is the ratio of two density estimates, which are both subjected to the
same edge effects. So, for a given value of the smoothing bandwidth 'sigma'. the relative risk estimate
 will be the *same* if you use a larger window. This is one of the arguments made by Diggle and Rowlingson (1994).

However, automatic methods for selecting the smoothing bandwidth (such as 'bw.relrisk' or 'bw.diggle') 
will give a different answer in a larger window.


Prof Adrian Baddeley FAA
University of Western Australia  /and/ CSIRO Mathematics, Informatics & Statistics
Mail: <cet.uwa.edu.au>             Skype: adrian.baddeley

More information about the R-sig-Geo mailing list