>There is another aspect of modelling that could be queried when comparing
>model - if we have competing models, one with few variables and one with
>many more variable with similar explanatory power, which one should you
I recogmend doing Prof Buxton's Pattern matching course in computer
science. One of the problems he points out that if you provide to many
variables and to many model parameters then you will effectivly 'end up
modeling the noise in the system not the underlying features'. This is a
problem people doing neural networks and pattern matching have all the
time. If you make the network to big and give it lots of data then the net
will try to predict the noise in the system. This explains why neural nets
have been so poor at predicting stockmarket prices. Try doing the courses
exersises if you don't belive me.
Using Ocams razer, the simplest explination for the most phenonoma is best.
Prof Batty also said
>All this is why we need new theories of urban morphology based on streets
>which link to what we know about geometry as well as toplogy and all else
>in urban geography and transportation besides
Well if you read my paper on Fractional Angular Intergration and the new
paper by Bill you will see how we do merge geometry back into the Space
Syntax representaion and get improved results against observed movement.
The angular approach appears to be the bridge between Space Syntax and the
Axial line and the tradional Junction centric models of Transportation