Andrew, If you plan to model the whole of London I can see at least one issue that you may want to consider further: the implications of very different densities in different areas of London. With regards to the absolute amount of movement in an area, there is an obvious link to the amount of development (density) in that area, as well as the type of land use. Basically a sq.km. of high density commercial development generates a lot more journeys than a sq.km. of low density residential development. A secondary effect of density is that it actually encourages a higher percentage of the journeys to be made on foot (since there are more potential destinations within sensible walking distance). In London, gross densities (i.e. sq.m. of building floorplate per sq.m. of land) vary greatly, from about 3.5 in the City to less than 0.5 for much of Outer London. However, the length of street network per sq.km. does not vary that much, being about 27 km per sq.km. for the City and around 24 for areas south of the Thames (Southwark, Lambeth), that have much lower densities. What this means is that the City of London may well present levels of pedestrian movement per km of network more than 10 times higher (probably a lot more) than suburban residential areas, only because of the differences in density and type of land use (regardless of spatial configuration, since they could conceivably have the same street layout). This would suggest that a model that only considers morphological properties cannot account for such diversity (unless density was also very closely correlated to integration, which Canary Wharf would seem to contradict). Space Syntax types of models can give some idea of how pedestrian flows are distributed throughout an area in a relative rather than absolute way. And my understanding is that they tend to give better correlations the more homogeneous an area is, and using local integration measures rather than global ones (although I may be wrong on this and someone from Space Syntax could comment). Therefore, I cannot see that a single model for the whole of London would give you a sensible approximation to overall levels of walking, without doing something about the density issue. One way round this problem would be splitting the analysis into a number of zones (probably very many), that have a certain degree of uniformity. Establish correlations for each zone (for which the correlation equation, slope and intercept, would be different) and then aggregate the various areas. The problem then is, how do you select the different zones and how many you go for. If you decided on a size in the region of 2 sq.km. you might end up with about 1,000 zones. But even if you decided to choose a fraction of that, say 200 zones, you would still need a good number of survey locations for each, covering the spectrum of integration values. I would guess that you might want at least 50 (possibly more), which would result in 10,000 survey locations. A different (and perhaps more logical) approach would be to incorporate land use and density within the modelling tool, which is something I have been working on. With this, you could produce indicators for absolute accessibility that could compare across the whole of London. The issue here is that you would need detailed land use data to incorporate into the model for the whole of London, but one would hope that the various local authorities would be able to provide much of it. I look forward to hearing further comments on this. Regards, Juan