Examining neighborhood effects on mental health utilizing a novel two-stage modeling approach

Neighborhood conditions may affect health, but health may also determine a preference for where to live. This study estimates the effect of neighborhood characteristics on mental health while aiming to adjust for this residential self-selection.

A two-step method was implemented using register data from Statistics Netherlands from all residents of the city of Rotterdam relocating within the city in 2013 (N = 12,456). First, using a conditional logit model, we estimated for each individual the probability of relocating to a neighborhood over all other neighborhoods in Rotterdam, based on personal and neighborhood characteristics in 2013. Second, we corrected this selection process in a model investigating the effects of neighborhood characteristics in 2014 on reimbursed anti-depressant or anti-psychotic medication in 2016.

Personal and neighborhood characteristics predicted neighborhood choice, indicating strong patterns of selection into neighborhoods. Unadjusted for selection log neighborhood income was associated with reimbursed medication (β = -0.040, 95% CI = -0.060, -0.020), but the association strongly attenuated after controlling for self-selection into neighborhoods (β = -0.010, 95% CI = -0.030, 0.011). The opposite was observed for contact with neighbors; unadjusted for self-selection there was no association (β = -0.020, 95% CI = -0.073, 0.033), but after adjustment increased neighborhood contact was associated with an 8.5% relative reduction in reimbursed medication (β = -0.075, 95% CI = -0.126, -0.025).

The method illustrated in this study offers new opportunities to disentangle selection from causation in neighborhood health research.