Statistical modelling and analysis of big data on pedestrian movement

Paper in proceeding, 2019

This work follows a long line of studies and empirical investigations in Space Syntax research, that, in general, try to conceptualise, describe and quantify the relation between physical space and human agency. How many people share public space is known to affect many socio-economic processes in cities, such as segregation, vitality and local commercial markets. Observing and measuring pedestrian movement through surveys, as well as statistically analysing it have been at the core of diverse investigations not least in the field of Space Syntax, not only a means to validate and measure the dependence of pedestrian movement on spatial configuration, but also as a means to forecast and predict pedestrian flows. However, these studies do not necessarily provide us with comparable, let alone generalisable findings that can lead to generalisable propositions. They remain in most cases specific investigations of particular cities, neighbourhoods or types of areas (e.g. city centres). Another issue, as will be elaborated in this paper, is that the typical statistical methods used, such as multivariate regression models, are not always the optimal or even suitable for modelling pedestrian movement, typically measured in pedestrian counts.
 
This paper aims therefore, to directly address three methodological challenges: first, construction of comparable GIS-models; second, gathering large scale pedestrian data; third, applying advanced statistical modelling suitable for pedestrian data.The ultimate goal is to estimate the impact of spatial form on urban life in a way that is methodologically sound and can provide robust results that can be generalisable, and allows us to speak of the relation between spatial form and pedestrian movement in a way that is not specific to a certain area, or types of areas or streets, or even to a specific city.
 
The results show, first, high and consistent correlations between spatial form and pedestrian movement in a study of unprecedented size that comprises three cities, including a large range of neighbourhoods of varying morphological types, from villa areas to urban cores, and offer convincing proof that the tested morphological variables have a strong impact on the spatial distribution of pedestrian flows in cities. Second, the study shows that the model with all explanatory variables has the highest explanatory power and the best model fit where Angular integration and Accessible FSI are the explanatory variables with the largest effect on pedestrian movement, but others are significantly contributing to the predictive power of the model. Third, the study contributes to the advancement of the statistical modelling that is suitable for the specificities of the data used, proposing the use of a negative Binomial model instead of regression models, most common in the field.