New Approaches to Simulating Immigration and Integration Trajectories

• The project is an exploratory, modularly structured methodology project. The aim is to develop and test innovative simulation approaches for analyzing immigration, emigration, integration, and settlement processes. The focus is on combining different methods, including agent-based simulations, probabilistic models, and AI-supported methods such as large language models, in order to better explain migration dynamics and investigate counterfactual scenarios.  
• Module 1 focuses on modeling individual migration decisions using bottom-up approaches that link micro and macro data and can simulate policy changes or exogenous shocks.  
• Module 2 examines the use of artificial intelligence to collect and evaluate open-ended response formats in quantitative surveys, in particular for analyzing narrative integration processes.

The scope of application of AI-based machine learning tools for hypothesis testing instead of descriptive exploratory work has not yet been sufficiently explored in social and behavioral science migration research, which is the starting point of this project.

Overall, the project aims to establish new methodological tools for migration-related research at DeZIM to gain theoretically sound, empirically reliable findings.

• Module 1 examines the feasibility of a bottom-up simulation model of individual migration decisions. Instead of modeling migration directly, the underlying decision-making processes of migrants are formalized, empirically informed using survey data, and calibrated with macro data to ensure external validity. Methodologically, the module combines agent-based simulations with stochastic and probabilistic methods. The aim is not to predict migration, but to conduct a causal analysis of key migration mechanisms and simulate counterfactual scenarios, such as political changes or exogenous shocks.  
• Module 2 examines the use of language-based AI models for collecting and evaluating open-ended response formats in quantitative surveys. Based on existing narrative interview data, language models are trained to convert complex qualitative statements into standardized, comparable survey data. In addition, the module examines whether AI-supported, dialogical survey formats can be used to systematically collect relevant information on migration and integration-related research questions.

Module 1 laid the theoretical and methodological foundations for a bottom-up model of individual migration decisions. A probability-theoretical framework for analyzing path dependencies in migration trajectories was developed, and a standardized ODD+D protocol for model description was created. 

Liam Haller 2025 Are Forced Migrant Trajectories Path-Dependent? A Markov Analysis International Migration Review https://doi.org/10.1177/01979183251319015 

Liam Haller 2025 A Call to Embrace Uncertainty: Rethinking Statistical Inference in Migration Research