Re: measuring change

j.Maxwell Legg (
Thu, 27 May 1999 11:49:30 +1200

bob green wrote:

... MDS analysis which uses as data distances
> between cities and then compares the spatial plot vs a map, is one thing.
> Explaining how ratings are transformed into distances and the assumption of
> similarity based on proximity, is another.


Yes there is a problem of assumptions based on the proximity within a 2D
plot. You will notice that for only two constructs (e.g., latitude and
longitude) the total variation of the grid vs. the number of elements
lies on a single plane. The MDS plot within WinGrid uses a stepping
bandwidth to average its order throughout this total variation while
spinning the grid across a range of dimensions.

Whereas to pinpoint a geographic location on a map requires a minimum of
latitude and longitude, Patrick Slater explained to me that these single
numbered grid ratings each described a knot of many coordinates in a
multi-dimensional space.

He chose PCA over Factor Analysis as an effective way of untangling that