Abstract
Background
Migrants experience substantial changes in their neighborhood physical and social environments along their migration journeys, but little is known about how perceived changes in their neighborhood environment pre- and post-migration correlate with their mental health. Our aim was to examine the associations between recalled changes in the perceived neighborhood physical and social environments and migrants’ mental health in the host city.
Methods
We used cross-sectional data on 591 migrants in Shenzhen, China. We assessed their risk of mental illness using the General Health Questionnaire (GHQ). Neighborhood perceptions were collected retrospectively pre- and post-migration. We used random forests to analyze possibly non-linear associations between GHQ scores and changes in the neighborhood environment, variable importance, and for exploratory analysis of variable interactions.
Results
Perceived changes in neighborhood aesthetics, safety, and green space were non-linearly associated with migrants’ mental health: A decline in these characteristics was associated with poor mental health, while improvements in them were unrelated to mental health benefits. Variable importance showed that change in safety was the most influential neighborhood characteristic, although individual-level characteristics—such as self-reported physical health, personal income, and hukou (i.e., the Chinese household registration system)—appeared to be more important to explain GHQ scores and also strongly interacted with other variables. For physical health, we found different associations between changes in the neighborhood provoked by migration and mental health.
Conclusion
Our findings suggest that perceived degradations in the physical environment are related to poorer post-migration mental health. In addition, it seems that perceived changes in the neighborhood environment play a minor role compared to individual-level characteristics, in particular migrants’ physical health condition. Replication of our findings in longitudinal settings is needed to exclude reverse causality.
Learning variable importance to guide recombination
