Analysing Modelling Errors in Migration Models Using a Simulation Approach
2026-07-24
A major challenge of migration modelling is that the actual migration process is unknown to researchers. Previous migration modelling has been done in a black‐box manner. Only the overall performance of a migration model is evaluated, while the contributions of specific errors remain unclear. There are at least four modelling errors: errors caused by missing any significant factors of migration, incorrect model form, the wrong assumption of a random process and the random process. These four errors are mixed, and it is hard to identify each error clearly in empirical models. They affect accurate analyses of the migration process with biased parameter estimates and large modelling errors, reducing the applicability of migration models. This research uses a known migration model to generate simulated migration data to assess the first, third and fourth modelling errors. Hundred sets of simulations are conducted. If a migration model misses an explanatory variable (model LMb) or a model with a wrong random process (model PM) is used, it is found that the model performance is significantly different from that of the correct model. The impact on estimated parameters is more complex. The results confirm concerns about modelling errors in previous studies, such as missing an important variable and using the wrong random process. But the findings indicate that the impacts of modelling errors are not the same for all origin and destination variables and random processes. Migration researchers need to pay attention to such differential impacts of modelling errors.