Jake Desyllas of Intelligentspace.com posted the following on 10/06/2003:
> For those interested in pedestrian modelling, this paper can be
> downloaded at http://www.casa.ucl.ac.uk/working_papers/paper62.pdf
> Summary details below:
> "Pedestrian Demand Modelling of Large Cities: An Applied Example
> from London"
> by Jake Desyllas, Elspeth Duxbury, John Ward, Andrew Smith
At the heart of the paper are what look like some good, if familiar, ideas,
which are unfortunately marred by what appears to be a misguided attempt
to cut out any trace of the origins of the research methods used, and some
rather suspect statistical analysis. In this note we will try and set the
record straight and go on to point out where these methods seem to us to
need to go from here.
A multivariate regression model to explain and predict pedestrian movement
using a set of independent variables including measures of the
configuration of space is an excellent idea. Many space syntax researchers
have been doing this for some years, and this presumably is why the authors
thought that these mail lists would be interested in their results.
The paper points out how important it is to avoid what Harvey (1973)
would call confounding spatial and social measures, and although it
misinterprets the ecological fallacy, the need to analyse space at a fine
resolution in order to avoid the problems posed by aggregation are well
The response to Lee's (1973) criticisms of large scale urban models through
the use of statistical and data based modelling are spot on.
All this is excellent, so it is a pity that the paper does not
properly review the background to the methods it has used nor the
extensive literature on related multivariate regression models of
pedestrian movement. A number of other extremely similar models have been
developed in recent years, most notably Penn et al's analyses of vehicular
and pedestrian movement in central London (1998). These regression models
used a wide range of variables, including land use, development density
measured by building height, street and pavement width, parking
availability and of course the full range of accessibility measures
provided by space syntax analysis using axial analysis. The observation
data are consistent, thorough and gathered by the same team working to
common observation methodologies and standards, for the sole purpose of
developing a high resolution data set of pedestrian movement. The studies
observed all street segments in seven areas of London (rather than sampling
a subset of segments) giving over 460 observation locations.
Along quite different lines Dongkuk Chang's 1998 PhD developed an
integrated regression based model for pedestrian movement in multi-level
circulation systems (see Chang & Penn, 1998). These models recognized that
spatial configuration alone would be hard put to it to explain movement
patterns in highly complex and unintelligible environments such as London's
Barbican and South Bank areas where invisible staircases and major
attractors or generators of pedestrian traffic play a powerful role.
Chang's methods quantified visibility, grade separation, and, along with
axial analysis, incorporated them in a multi-criteria model, which
effectively 'calibrated' the axial model for the effects of these other
Both sets of studies have been peer reviewed and published. Both are clear
precursors to the work on walkability in London that Andrew Smith initiated
at Transport for London along with Space Syntax Limited (Stonor, Campos &
Smith, 2002), however, all these are unmentioned, even though Alan Penn was
Jake Desyllas' PhD supervisor, Chang a contemporary doctoral student and
Smith a co-author. One wonders why? Perhaps it is merely an oversight in an
unreviewed working paper on the web.
But the oversights go further. One of the most serious omissions is
the origin of a key component of the model, Visibility Graph
Analysis (VGA). VGA is a technique developed at UCL by Turner and
Penn, through a joint project led by the Bartlett and CASA, on which
the first author was employed as a research assistant. It seems at least odd
at most obloquious that in the paper he should attribute the development of
the technique to Braaksma (1980). Braaksma does use visibility graphs of
service networks (although he uses the term 'adjaceny matrix'), but this is
more related to the networks of radio transmitters proposed De Floriani
(1994) and many other more traditional visibility graph applications. The
key to Visibility Graph Analysis of urban areas is the dense grid form of
the visibility graph, developed by Turner et al. (2001), and implemented in
Depthmap and subsequently in Intelligent Space's 'Fathom' program. The
term itself, 'Visibility Graph Analysis', and the application of these
techniques to analysis of pedestrian movement in building interiors (and
the possibility to analyse urban areas) originates in the Turner et al.
VGA was built on and resulted from research into space syntax. It extends
local and global measures from graphs of axial lines and convex spaces to a
complete visual domain. The CASA working paper also builds itself into this
following the introduction of VGA with a discussion of how it may be used
to examine features such as 'two step' relationships to an accessible
'neighbourhood'. Global analysis of spatial networks is described, but the
provenance of the ideas is misattributed to Wasserman and Faust (1994) who
applied two step measures in social network analysis. The idea of
examining global network relationships in spatial systems is introduced
in Hillier & Hanson (1984) and developed into radius measures during the
1990's. Indeed, one of the best correlates discovered for pedestrian
movement in space syntax research, radius-3 axial integration,
is the forrunner of radius based measures in VGA. That the
authors appear not to understand the methodological basis of these techiques
The proper referencing of the origins of the research is certainly our
major concern, but there is also a problem with data analysis using linear
regression that needs to be commented upon. Regression models are
notorious for giving false correlation coefficients with skewed data sets.
At the very least, some form of data normalization or test for normality
should be performed before claiming a linear regression result. As can be
seen from the published scatterplots in the paper, data on both axes are
highly skewed to the bottom end of the scale. This needs to be corrected
using a transformation of the data sets to produced normally distributed
data sets if any reliance is to be placed on linear regression.
After these negative paragraphs, it may seem odd for us to say that this
paper could be a good one. Much of the literature review is enlightening,
pointing towards the often overlooked origins of research into pedestrian
movement. Fruin's (1971) levels of service are nicely woven into the idea
of a complete pedestrian model. Using VGA to develop a regression model of
a large area of central London is also a huge improvement on the state of
the art in urban modelling in general. The parameters are well chosen: the
literature has shown that spatial variables of street width and land use
and transport interchanges are important factors in pedestrian movement.
The importance of incorporating tube stations and rail terminii is clear
given that these account for the major (and most concentrated) change of
mode sources in London. Shop frontages are also a sensible choice, at
least for the West End of London, where shopping streets see significantly
more usage than, as pointed out in the paper, nearby side streets.
This leads us to ask where this stream of research needs to go in the
Perhaps a crucial point lies in the choice of VGA as the analytic basis for
calculating accessibility. VGA is a really nice idea (although speaking as
its originators we would say that), however we have now been looking at it
for nearly 7 years and still in urban street systems axial mapping
significantly outperforms VGA as an explanatory variable of pedestrian
movement flows in urban systems (Turner, 2003). The reasons why this might
be the case are the subject of considerable research at the moment (see
2002 and Bill Hillier's paper in the Space Syntax Symposium next week (The
Architectures of Seeing and Going). However, from a pragmatic point of view,
where predictive power and empirical evidence
are the most important criteria, it is clear that axial mapping and high
resolution axial mapping could form a much more powerful basis for a
multicriteria regression model.
It might also be an idea to look more generally at the available land use
data since the previous studies found that office space also contributed
significantly to pedestrian movement, and that significant
differences at different times of day were best explained by looking in
detail at a full land use breakdown. Development density is yet another of
the significant factors and should ideally be incorporated in a model of
this sort. Perhaps most important however, was the finding that average
axial radius-3 integration was powerfully related to average movement flow
rates in large aggregated areas. We suspect that any model of this sort
will need to take account no only of the specific local spatial properties,
including those that are measured in terms of the global network, but will
also need to address neghbourhood properties that describe the degree of
grid intensification in different parts of the urban area.
In conclusion, whilst it is highly embarrassing that a paper in this state
published on a UCL website, there is significant data and a worthwhile
technique published within it, and we look forward to its republication in a
revised and more academically responsible form.
Alasdair Turner, Alan Penn & Bill Hillier
The Bartlett School of Graduate Studies
Faculty of the Built Environment
University College London
Braaksma, J P, Cook, W J, 1980, "Human orientation in transportation
terminals" Transportation Engineering Journal 106(TE2) 189 203
Chang, D., & Penn, A., 1998, "Integrated multilevel circulation in dense
urban areas: the effect of multiple interacting constraints on the use of
complex urban areas," Environment and Planning B: Planning and Design,
Vol. 25, 507-538, Pion.
De Floriani, L, Marzano, P, Puppo, E, 1994, "Line-of-sight communication on
terrain models" International Journal of Geographical Information
Systems 8 329 342
Harvey, D, 1973 Social Justice and the City (Edward Arnold, London, UK)
Hillier, B, 2003, "The Architectures of Seeing and Going", 4th Space Syntax
Symposium, London June 16-19th 2003.
Hillier, B, 1996 Space is the Machine (Cambridge University Press,
Hillier, B, Hanson, J, 1984, The Social Logic of Space (Cambridge
University Press, Cambridge, UK)
Lee, D B, 1973, "Requiem for large-scale models" Journal of the American
Institute of Planners 39 163 178
Penn, A., 2003, "Space syntax and spatial cognition: or why the axial
line?" Environment and Behaviour, (2003) 35 (1) 30-64, Sage Publications,
Penn, A., Hillier, B., Banister, D. & Xu, J. 1998, "Configurational
Modelling of Urban Movement Networks," in Travel Behaviour Research:
updating the state of play, Juan de Dios Ortuzar, David Henshar and Sergio
Jara-Diaz (eds), Pergamon press, Elsevier, Amsterdam, 1998, Ch19, 339-362.
Penn, A., Hillier, B., Banister, D., Xu, J., 1998, "Configurational
modelling of urban movement networks," Environment and Planning B: Planning
and Design, Vol 25, pp 59-84, Pion, 1998.
Stonor, T., Campos, M.B deA & Smith, A., "Towards a Walkability Index",
Proceedings of teh 3rd International Conference, Walk 21, Barcelona, 2002.
Turner, A, 2003, "Analysing the Visual Dynamics of Spatial Morphology",
Environment and Planning B: Planning and Design, 2003, vol 30, (in press).
Turner, A, Doxa, M, O Sullivan, D, Penn, A, 2001, "From isovists to
visibility graphs: a methodology for the analysis of architectural space"
Environment and Planning B: Planning and Design 28 103 121
Wasserman, S, Faust, K, 1994, "Social Network Analysis: Methods and
Applications" (Cambridge University Press, Cambridge, UK)