Properties of Simple Models of Spatial Interaction

Talk given at the MacArthur Workshop on Urban Modelling & Complexity Science, UCL, 20/10/2019.

Abstract:

With so much data available we can often fine tune complicated models to produce useful results.

However, working with simple models can still have advantages. They can highlight the role of simple

principles in the way that the structure of a large neural network may not (interpretability). In some examples,

such as the historical contexts I have studied, good data is not available. Finally, simple models

can provide a useful benchmark (null models) in many different contexts.

In this talk I will look at simple models, both old models like the Gravity model

and new versions of old models such as the Radiation model. These are both large families of models

and the choice of exactly which one to use can be overwhelming. My aim is to understand the key features

of these models which in turn can help us choose the most appropriate model for the task at hand.

This especially useful when used by experts in other areas such as archaeologists.

I will look at theoretical, numerical and data driven approaches I have used to study the properties

of these models.