Results of impacts( ) for partitioned impact measures

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

Results of impacts( ) for partitioned impact measures

Ted P.
Hi all -

I am creating partitioned impact measure estimates using impacts( ) with the argument `Q' set to the desired number of orders of neighbourhood, say, Q = 5. The model is a Spatial Durbin model and the code is fairly standard:

trMC <-- trW(as(W, "CsparseMatrix"), type="MC")
imp <- impacts(test.sdm, tr=trW, R=1000, Q=5)
summary(imp, zstats=T, reportQ=T, short=T)

I have a general question about the results.

I have read that the results for direct impacts of order 1 (immediate neighbours) should be 0 and, similarly, that for indirect impacts of order 0 (local area) should also be 0. I have certainly seen some examples that reflect this, and as I understand things, the logic makes sense.

However, not only do my results not have zeros in these slots, but as I search for confirmation online I also see other examples that don't.  Mine look more like this:


Impact components
Q1 0.065757531
Q2 0.003234086     <== Should be 0?
Q3 0.006264031
Q4 0.002231203
Q5 0.001677652

Q1 0.02869280     <== Should be 0?
Q2 0.06544605
Q3 0.04367716
Q4 0.03408384
Q5 0.02472906


I am assuming that the expectation of zeros as indicated is correct and that my results are flawed. However, I have yet to locate a problem and the fact that I see other examples with non-zero entries in these slots has left me confused.

My question is: Is there ever a situation in which it is reasonable to expect non-zero entries in these order-of-neighborhood slots as indicated above? If there is, how might these non-zero entries be explained given that for indirect impacts it seems reasonable to expect no effects from the local area, for example?

 I'm sensing something obvious here (apologies in advance) - perhaps lock-down is getting me frazzled. Any help appreciated. In the meantime, I'll maintain my assumption that the results are wrong and keep re-checking everything that has led me here.

Keep well, all.


Ted Pope

        [[alternative HTML version deleted]]

R-sig-Geo mailing list
[hidden email]