Unexpected LISA results for polygons contained by another polygon

[anitagraser]

As mentioned on https://gitter.im/pysal/pysal, LISA results for polygons that are completely contained within another polygon (therefore only have one neighbor) are often "not significant" even though visual inspection of the input values would suggest that they should be part of the hotspots/coldspots:

Data and notebook to reproduce:

https://github.com/anitagraser/pysal-temp/blob/main/notebooks/moran-local-holes.ipynb

[jGaboardi]

Possibly related to pysal/spaghetti#527.

[sjsrey]

I think what is going on is that the hole polygons have neighbor cardinality=1. As was discussed on gitter, this impacts the variance of the local statistics:

Updated notebook.

[anitagraser]

Thank you, @sjsrey! I'll look into distance-based weights instead to try avoid this effect.