California Exodus? A network model of population redistribution in the United States (Q6577559)

The authors propose a new approach to modelling and analysing the drivers of migration flows in a multi-regional setting. Their motivation stems from the debates on California's net migration loss, which has triggered a considerable debate in the US on the causes of California's out-migration, such as de-urbanisation, political and racial factors, etc. To address the issues raised in the debate, the authors model migration flows between US counties. The proposed modelling approach is based on graph theory and uses the valued exponential family random graph models with graph vertices representing individual counties. To examine the contribution of different factors to migration flows, the authors use counterfactual `knockout' experiments in which different factors are switched on or off. The authors suggest that racial dynamics contribute to the California exodus, urbanisation ameliorates it, and political climate and housing costs have little impact. Moreover, the severity of the California exodus depends on how it is measured, and California is not the state with the largest population loss.\N\NAs such, the paper introduces a methodological approach that is new to the field and therefore deserves further development, popularisation among demographers and practical testing. Multi-regional migration modelling and projection is one of the most methodologically challenging demographic problems, and a particular methodological problem is the unsatisfactory representation of complex phenomena, such as joint origin-destination drivers of migration, correlations between migration flows, etc., which the authors explicitly aim to address in their model. This undoubtedly contributes to the relevance and need for further elaboration of their model. In particular, it is worth noting that the approach presented is a rare attempt to use the concept of the migration system, which is as old and relevant to real migration as it is, unfortunately, rarely used in actual modelling and projection of migrations. Nevertheless, the model needs to be better popularised among demographers, and it seems to share the problem of computational complexity with other multi-regional migration models.\N\NIn terms of content, there are a number of potential shortcomings that may need to be addressed when interpreting the work and developing the model further. Although the authors have done a good job in addressing a wide range of possible migration drivers, this does not seem sufficient, especially when taking the migration system approach. For example, one would need to address the issue of illegal border crossings. How and where do they end up in migration statistics? If they do not result in `immigration' statistics for California, but ultimately contribute to out-migration from the state, this would create a fictional exodus effect that should be examined, discussed and perhaps modelled. Given the racial and income distribution dimension, and also within the ideology of the migration system, one would need to address the heterogeneity of the population, which I believe is evident in the results but not explicitly addressed in the model. For example, high rental costs, which indicate opportunities for higher-skilled workers, may be associated with out-migration of lower-paid workers and in-migration of higher-paid workers. Similarly, with regard to the racial distribution dimension, it would be useful to examine migration flows by race separately. Rural and urban places may also be associated with different types of migration flows.\N\NFinally, the model needs to be properly tested and validated. Given the complexity and non-linearity of the factors involved, it would be desirable to use the same approach as in machine learning, with separate training and test data sets.\N\NAll in all, this is a very interesting work with a new and promising model that certainly deserves further elaboration and popularization in the field.