On Thu, Oct 13, 2011 at 1:40 PM, <[email protected]> wrote:

> I’m still not entirely sure what you’re getting at:  spatial correlation of
> origin/destination locations to each other?  Or some characteristic of the
> origin/destination pairs?  It sounds like the latter, but then the question
> is “correlated with what”?****
>
>
>

It sounds like you've got a bivariate point pattern - a set of origin points
and a set of destination points - and you want to ask questions about the
relationship of the underlying processes...

 Read:

http://www.amazon.co.uk/Statistical-Analysis-Spatial-Point-Patterns/dp/0340740701

 which should give you an outline.

 You probably want to use some kind of bivariate K-function. This estimates
the number of events that occur within a distance of an event, based on the
type. If type 2 and type 1 events are being generated by the same process,
then the number of type 1's and type 2's within a distance from a type 1 or
a type 2 event should be the same (within estimation errors, in practice).
For example, if thefts mostly happen in the east side of town, and
recoveries on the other side of the tracks, then you'll see more thefts near
thefts than recoveries near to thefts. If the number of recoveries near
thefts is the same as the number of other thefts, then you can possibly
conclude the reverse, that its all happening everywhere. An interesting
complication could be that you can link each theft event with its recovery
event, so maybe what you really have is a spatial line process. Possibly you
might want to look at the length, direction, orientation, and
endpoint-locations of the lines...

 The spatstat package can do all the point-pattern stuff, and splancs has
some K-function code too...

Barry

        [[alternative HTML version deleted]]

_______________________________________________
R-sig-Geo mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo

Reply via email to