seminar: Crime, Data and Networks
speaker: Paul Ormerod (Volterra Consulting)abstract: In general in the physical sciences, theories can be confronted with experimental data. In the social sciences, experiments cannot usually be conducted, and model validation has to be carried out with respect to a single series of data.
Further, data in the social sciences is often measured with error. This is particularly so with crime, especially with data over time.
This raises issues of model validation, which I discuss.
There is also the important methodological issue of the type of behavioural rule which is assigned to agents, either explicitly in an agent based model or implicitly in a set of differential equation.
We can posit two extremes in terms of agent cognition. At one extreme, we have the fully rational model of economics in which agents gather full information and use optimisation as their behavioural rule. At the other, we can have ‘zero cognition’, in which agents act purely at random.
I suggest that the best modelling strategy is to adopt as the ‘null model’ one which is close to zero cognition i.e. minimal content of the behavioural rule. This links with criminology views of agents in this content. We then add to the content of the rule, to the extent that the initial choice can be improved.
Finally, we know that network topology can be of decisive importance in the spread or containment of behaviour, and I suggest that discovering an approximation to the relevant network in specific contexts should be an important focus of mathematical modelling of crime.
timetable:
Fri 18 Apr, 9:00 - 10:00, Aula Dini
<< Go back