Drawing the lines: the same city, different areas

Vancouver, 2015 and 2016. Does who lives in an area predict where police record an incident? Police use maps to decide where patrols go. Cut the city up a different way and see how the picture changes.

household size contribution
++
income contribution
++
the model’s constant term
++
spatial error we account for
++
residual error
what is left over
==
reported

Circle area is the size of the number. In each term, red adds to the prediction and blue takes away.

How much of the pattern the model explains
Which terms (don’t) hold up
Are the residual errors clustered in space? (Moran's I)
The area under the pointer
Point at the map, or tap an area.
Cut the city into
Reports of
The model
Neighbours are

This page does what it is describing. It turns a city into boxes with numbers in them, and everything it cannot count is missing from it too. The 996 smallest areas stay drawn under every grouping, so what the grouping throws away is never quite off the screen.