Andrew wrote:
> I'm sure that others will come in on this one. <warning of
> impending over-
> simplification - sensitive readers may wish to skip the rest of this
> paragraph> My first impressions as an outsider are that VGA, as a 2D
> analysis, is suited to modelling movement in public squares, where people
> are moving around the space in 2 dimensions. Axial maps, as a 1D analysis,
> are suited to modelling movement along streets, where people are
> predominantly moving through the space in 1 dimension.
I think that this is about right (at least so far as axial analysis is
concerned). We have found VGA to work quite well with building interiors,
with station concourses and some urban areas, but there are a couple of
complications that need to be pointed out. For instance, VGA is heavily
biased by large open spaces. If the graph is constructed on a regular grid
then an square or a park becomes a large clique in the graph and can skew
the distribution of integration in a surrounding street pattern. This leads
to relatively poorer (than axial) correlations between VGA and movement in
areas that have large open spaces along with street systems.
The second complication is of how best to attribute movement observations to
VGA analysis. For the axial map this s easy. A flow along a street is
attributed to the axial line passing along that segment of the street. But
in VGA there are any number of grid points that an observation of someone
moving along a street could be atrributed to. This is a classic problem of
how to associate linear movement data with point data in VGA. Pragmatically
it is easy to do anything you like in a GIS (set buffers, count people
within them, average the VGA values etc. However, currently no theoretically
substantiated methods exist to decide what should be done in any specific
case.
>
> I look forward to putting everything into a multiple regression analysis
> [MRA] and seeing what comes out.
>
> >2. Choice of variables to test
> > ... <snip> ...
> > * land use
>
> Hmm, tricky one to stick into an MRA. I guess we'll try various types of
> dummy variable, and see what comes out. All suggestions welcome - the
> first things that come to mind are a 0/1 indicator for presence or absence
> of retail, or a metric for closeness to some perfect mix of retail,
> employment, housing. This perfect mix would optimise the availability of
> the first two from any given dwelling.
I guess ideally one just wants for an MRA is a measure of metric area of
retail, other land uses, accesible from doorways opening onto a street. The
point is that what space syntax does is represent the configuration of space
through which we move. Some of that is internal to buildings other parts are
external, but so far as a user is concerned it amounts to much the same
thing. Land use dictates whetehr or not buildings are open to public access,
and the general density of occupancy of the buildiung interior. When we map
the city, we tend to map only the external space. The point of the MRA with
measures of density and land use would be effectively to look for the
effects of the interior space that we have not mapped. Land use and density
measures would effectively be used as proxies for a full configuration map.
My hunch is that the full map would be the best thing to construct.... but
an even greater effort than getting data out of the valiation office.
Alan
|