> Andrew Smith:
> So a model that explains 80% of the traffic flows in Central London
> (before congestion charging), based on an axial map, will predict no
> changes after such road pricing is introduced, as the axial map does
> not change. Similarly with bus lanes, traffic light phasing, traffic
> calming: all the standard tools of urban traffic management. But
> flows have changed significantly ... and a useful transport model
> would predict those changes reliably.
I'd love to get at the data on the changes pre/post congestion charging,
Andrew (hint hint ;-) Can you tell me whether there have been substantial
changes in relative flows between different streets within the zone? I can
imagine several reasons why there could be and I believe that some of these
might be predictable in principle from syntax analysis. For example, I would
anticipate a reduction in the proportion of through trips across the zone,
and so a reduction in radius of integration of the explanatory variables in
the regression. I would also hope that there would be a reduction in
congestion (that was the idea wasn't it), and so a reduction in the
proportion explained by road width.
But just to come back on the matter in hand. The policy of congestion
charging was presumably subject of a substantial traffic modeling exercise.
Now, following its implementation, how good were the predictions? Can we
establish a best practice 'standard error of estimate' for the pre/post
changes at the individual link level? Also of course at the overall area
average level - how much did the price point of the charge affect demand,
and how good were the traffic models at predicting this?
> As traffic (or more usefully, general transport) models use
> origin/destination matrices, I can't see how one would implement Alan
> Penn's suggestion of: "using space syntax measures ... in the matrix
> estimation phase of the construction of traffic models", but am
> willing to be enlightened. I'm often agnostic as to the usefulness of
> models based on such O/D matrices, but they do have lots of
> advantages in terms of being able to inform policy decisions on land
> use and transport.
Well, I have to come clean on this - I am no expert on matrix estimation
either - the idea was suggested to me some years ago by Pilo Willumsen who
is. I believe that the process would be that you would use our regression
equation as a proxy for flows on the network, insert that in the model and
use it to estimate what the O-D matrix must be to give rise to those flows.
The point here is that although our regression is based on a sample of
flows, the configurational analysis allows a good estimate for ALL links.
This might offer a substantial improvement on current methods of estimating
O-D matrices. But as I say I have never done this and am no expert....