Dear all,
I have been trying to test the statistical significance
of spatial clustering of a continuous variable X.
As an analogy to Moran's I,
I calculated the following statistics:
A1 = Sigma of Distance(i,j) * (Xi - Xj)**2 for all the
combinations of locations, i and j.
A2 = Sigma of ( 1/Distance(i,j) ) * (Xi - Xj)**2 for all the
combinations of locations, i and j.
I thought A1 would be unusually large and A2 would be unusually
small if X is spatially clustered.
Locations of all observations were randomly permutated many times
(e.g. 100000), thereby a unimodal distribution was obtained for
each of A1 and A2. Then, P value was calculated for
actual A1 and A2.
As a result, P value for A1 and P value for A2 were very different.
The above assumption of mine is correct?
Your suggestion of any sort would be appreciated.
Yoshiro Nagao
---------------------------------
Let's Celebrate Together!
Yahoo! JAPAN
