Empirical calibration of full scale agent-based models of school choice: computational challenges and model validation
Talk, ODISSEI Conference for Social Science in the Netherlands 2022, Utrecht, The Netherlands
School segregation is widely associated with existing inequalities and their reproduction. Although it has been studied for decades and using various methods/techniques, it still is a persistent problem in society. Currently employed methodologies often treat households on the micro-level as utility maximising individuals that decide in isolation or analyse macroscopic trends and correlations. However, these methodologies might miss important interactions within and between these levels. For example, parents rely on their social networks, observe current school compositions, school profiles, live in segregated neighbourhoods and are subject to institutional rules, hence they interact with each other and their environment. Simulation-based techniques, such as Agent-Based Models (ABM) provide a way to explicitly model these features and have shown promising results in other fields of science. Existing ABM of school choice are mostly based on theoretical rules and smaller scales, hence do not take the full scale of the city into account. On the other hand, currently used fully data driven methods, such as discrete choice analysis, do not consider the potentially complex interactions. Therefore, we present results of one of the first empirically calibrated ABM of school choice on the Amsterdam scale. However, key challenges of ABM are empirical calibration and validation of their simulated (household) behaviour. These are important for confidence in the model and could inform potential policy, but requires a lot of data and computation. Multiple runs are needed to grasp how sensitive the model is to its input parameters, to quantify uncertainty and analyse the impact of specific scenarios. We show some of the benefits of our modelling approach as well as some of the computational difficulties and existing challenges we encountered while modelling school segregation at the Amsterdam scale.