Skip to main content

Bayesian Agent-based Population Studies: Transforming Simulation Models of Human Migration

Periodic Reporting for period 3 - BAPS (Bayesian Agent-based Population Studies: Transforming Simulation Models of Human Migration)

Reporting period: 2020-06-01 to 2022-05-31

The aim of the Bayesian Agent-based Population Studies (BAPS) project was to develop a ground-breaking simulation model of international migration, based on a population of intelligent, cognitive agents, their social networks and institutions, all interacting with one another. The ambition of the project was to transform the study of migration – one of the most uncertain population processes and a top-priority EU policy area – by offering a step change in the way it can be understood, predicted and managed. In this way, the project sought to effectively integrate behavioural and social theory with modelling. To develop micro-foundations for migration studies, the model design followed the recent developments in demography, statistics and uncertainty quantification, cognitive psychology and computer science, and was subsequently corroborated by insights from ethnographic research. At a high level, the interdisciplinary setup proved necessary for designing and analysing more realistic simulation experiments of migration processes, and for modelling the simulated individuals and the way they make decisions. Following the inductive principles of the scientific enquiry, the model construction process was iterative, with increasingly more realistic versions built based on the new empirical data collected in the areas with identified information shortages. By following this approach, we proposed a way for filling an important epistemological gap in demography, with respect to building formal theories based on empirically grounded models. The project also offered a pioneering environment for building and analysing simulation models with a range of languages and formalisms, and for documenting their provenance in a parsimonious and logical way.
In the project, we focused on the recent Syrian asylum migration to Europe. The empirical evidence base for the model was collated and assessed in an inventory of secondary data, available on the project website, followed up with primary data collection in a dedicated ethnographic study. In parallel, we have designed and executed the five rounds of psychological experiments. The first three focused on the role of risk and loss aversion in migration decisions, on the importance of information sources, showing high levels of trust in international organisations, and on the relative weightings of different migration drivers. The final two experiments, currently being analysed, focused on the role of immersion in experimental studies, aiming to increase their ecological validity in the migration context.

In terms of the modelling, five versions of the model of migrant journeys were developed, analysed, and tested, focusing at the role of information spread over social networks in shaping the emergence and changes, and persistence of migration routes. Statistical analysis confirmed the pivotal role of information in shaping migrant journeys. One model version was implemented in parallel in two languages, general-purpose and domain-specific, enabling the analysis of trade-offs in terms speed, descriptive detail, flexibility and generalisability. The computational side of the project also involved developing the semantics for the modelling language, implementing the model by using an efficient simulator, and unifying the modelling and simulation environment. In addition, through provenance modelling, we demonstrated the usefulness of a formal framework for documenting the modelling process and its different constituting elements.

Dissemination highlights include publishing the book monograph Towards Model-Based Demography (Springer 2021), with a dedicated virtual launch event held in December 2021. We delivered an online course on Agent-Based Modelling for Social Research in November 2020. We also successfully organised a two-day scientific and policy workshop on Uncertainty and Complexity of Migration in London in November 2018 and a technical workshop in Southampton in January 2020, gave 26 project-related talks at different fora, and have so far published seven articles and four refereed conference papers, with ten more papers in the pipeline. Details are available on the project website:
In all five project areas: simulations, data analysis, cognitive experiments, computations, and statistical analysis, the project made progress beyond the state of the art. Simulation modelling and the analysis of model results confirmed that while migration processes are complex and uncertain, information flows mattered a lot for model outcomes, with trade-offs between theoretical insights from models and their empirical alignment.

The analysis of secondary sources enabled separating the bias and variance in available data. The ethnographic study corroborated other findings, and identified different forms of capital, information and chance as key factors shaping the refugees’ journeys. One surprising finding was related to gender, with women travelling on their own, contrary to popular perceptions.

The results of cognitive experiments indicated that the migration decision-making patterns are consistent with the prospect theory, with loss and risk aversion, while information from official organisations and those with previous experience were most influential on migration decisions. The subsequent research on immersive experiments additionally pointed out to the importance of simple design to make the results interpretable.

For computations, we found trade-offs between speed, productivity, expressiveness and ease of development. We defined formal semantics that meets the modelling requirements, based on precisely defined stochastic processes. The language development offered solutions with adapted rule-based syntax and separation of different concerns (e.g. model logic or implementation) in an internal domain-specific language. Formal description of the model provenance allowed documenting, querying and visualising the model-building process and its different elements.

The project results confirm key limitations of existing social simulation approaches, which do not make full use of the advances in different disciplines. We have demonstrated the usefulness of a unified, formal and iterative model-building process, following the inductive principles. Adding detail increases the model complexity and uncertainty, but also our understanding of social processes, for which uncertainty and complexity are key features. Overreliance on data, while reducing uncertainty, risks losing the richness of the description.

With the caveats about high uncertainty, the models also enabled answering substantive questions on migration processes and the impact of policies and interventions, at least in qualitative terms. In conclusion, the complexity and uncertainty of migration should not be underestimated or, worse, ignored. Such results call for humility in the face of complex processes that result from the agency of thousands of actors exercising their free will based on the best information they possess and exchange through their social networks.

The results confirmed large unused potential of social simulation studies. We hope that the project will open up exciting research possibilities beyond population studies, both methodological and substantive, and will generate academic and practical impact, offering methodological advice for policy-relevant simulations (see book review:
Example screenshot from the route formation model