In this dissertation, I study the persistence of household residential choice using data from Boston during the Great Migration, where the in-migration of African Americans was a large demographic shock to existing neighborhood composition. Using novel panel data that tracks individual residential choices, I document stark differences in out-migration responses to this shock among two religious groups. To understand differences in neighborhood attachment, I estimate a dynamic panel data discrete choice model that disentangles state dependence in neighborhood choice from unobserved heterogeneity in preferences for neighborhood characteristics. I find that the differences in the two groups’ out-migration behavior are mostly explained by their differential state dependence. I show that this state dependence is strongly correlated with observable measures of investments in durable local infrastructure, such as the size of religious gathering places and the number of religious schools. My counterfactual policy analysis suggests that maintaining a mixed racial composition of urban neighborhoods in Boston during the relevant period would have required a voucher equivalent to 25 percent of the rent.