This isn't a statistical methodological answer to Peter's question - more a report of long term practice. My approach is to avoid
uni-dimensional index construction and instead classify EDs using sets of relevant indices as input to cluster analysis procedures.
The really significant alogarithm choice is of the set of variables used to construct the clusters. I then map by colouring in a
physical map with coloured pencils with a different colour for each cluster. I am sure there is a much more high tech method of
mapping but I actually rather enjoy doing this. My experience, and I have done this a lot over the years, is that you don't get
'chaotic' clusters. I am assuming Peter is using Census data so there is no real (10% sample is very big and most indices I use come
from the 100% set) sampling issue. If you find something odd it is really there but in general what you get is adjacent areas of EDs
constituting neighbourhoods which triangulate very well with administrative data and qualitative reporting.
An alternative would be to use Census Tracts - a very handy size between Wards - far too big - and EDs which are small.
David Byrne
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|