Sanjay,

Can I suggest that you take a look at the papers first - they are all
about trying to explain what on the face of it is an unexpectedly strong
correlation... One answer is that it is obvious. The configuration of
the route network affects people's choice of routes with the result that
when one looks at the results of ALL choices - ie. the average flows in
the network - they turn out to be strongly correlated with
configurational measures. Since origins and destinations of trips are
pretty liberally distributed everywhere perhaps one would not expect
these to have a particularly strong effect on aggregate flows.
Alternatively - and this is in fact what I think - the processes of land
use allocation to urban parcels in cities that have evolved, leads to
those land uses that value accessibility by people locating in greater
densities in more accessible locations in the configuration. These land
uses are those that 'attract' the largest volume of trips. There is
therefore a process by which land uses and development densities
(origins and destinations for trips) evolve over time to correlate with
spatial configuration.

One key question is why people have not found effects of configuration
of the network to be important before. This I believe is the result of
our selection of the axial line as the basic element of the network.
Axial lines give rise to non-planar graphs, whilst the selection of
intersections as nodes and road segments as links in conventional
traffic models give rise to planar graphs. Planarity coupled to metric
distance based 'costs' lead to measures of configuration in the
conventional traffic modellers network representation being almost
entirely dependent on the selection of the boundary to the area mapped -
the accessible location is in the middle of whatever patch of the
surface of the world one has mapped. This means that configurational
measures in these representations are essentially arbitrary. This is not
the case with measures of configuration based on axial graphs. Here the
centre of the map can be (and often is) segregated and the edges
integrated. More importantly though, if you select a patch of a
continuous urban system mapped axially and analyse it, and then move the
boundary of the patch a bit and analyse again, and so forth, the pattern
of integration remains remarkably stable regardless of boundary. For a
node map the integrated centre moves with the bondary. The stability of
the axial map regardless of choice of boundary results from its being
non-planar.

There is a another element to the 'explanation' of why the axial line
should be a sensible representational element. One answer to this is
that it represents something close to what people moving through a
system of space experience as the locally stable elements of space. The
argument is that it is linear elements that are subject to least change
(in visual flow terms) as the observer moves. Changes of direction
conversely involve a high degree of change in local information from the
visual field. I have discussed this 'cognitive' argument and the
empirical evidence for it in:

        Space syntax and spatial cognition: or why the axial line?
(Penn, A.) Environment and Behaviour, (2003) 35 (1) 30-64,  Sage
Publications,  ISSN:0013-9165.


Alan

>
> >I understand your suggestion, but perhaps the key might be the other
way
> >round. How about using space syntax measures - which already explain
> >over 80% of the variance in traffic flows in the London network - in
the
> >matrix estimation phase of the construction of traffic models? This
> >would allow those models to use the other factors you mention to help
> >explain the 20% of variance remaining unexplained by the
configuration
> >of the network.
>
> thanks for the paper references.
> 80% confidence in variance prediction is an excellent result. could
you
> please briefly discuss the reasons/characteristics that lead to such
> excellent outputs?
>
> thank,
> sanjay.