Clustering around Zero-Weight Variables

[Tony]

Hey,

I'm trying to run an analysis similar to the one Kyle used in his
thesis (i.e. weight dependent variable at zero and then identify
negative clusters), but I think there's an issue. A zero-weight cell
should, under most circumstances, look like a hit histogram. But my
zero-weight variable looks like the map was constructed around it
indifferent to the weight setting. So my question is this: Is the
weighting system working properly?

Tony

[Kyle Thayer]

I would have to see the data and run the program on it myself to double check.

One possibility is that the more related your 0 weighted variable is to ones that are weighted, the more it will look like it was organized on it (since it was organized on things related to it).The less correlation it has the more random it should appear. I tried it on random data, not weighting on 2 of the channels and attached an image. When you set the min neighborhood (in advanced options) to 0, the randomness should be more apparent. Try it with your data again but with a min neighborhood of 0 and see if it still looks like it is being organized on those channels. If you think you are seeing problems beyond this, let me know.

Kyle

[Tony]

Hey Kyle,

You were right on both accounts. I had organized on closely
correlated values, and then I still had to set the min neighborhood to
zero. Good call, and thanks a lot!

Tony

>  rands_no_weight_on_1_and_3.JPG
> 75KViewDownload
>
>  rands_no_weight_min_neighborhood_0.JPG
> 79KViewDownload