Exploring dynamic process of regional shrinkage in Ohio: A Bayesian perspective on population shifts at small-area levels

Abstract

Recently shrinkage phenomenon in urban areas has gained broader attention to scholars, policymakers, and practitioners. There has been considerable effort devoted to unveiling various facets of shrinkage within the literature. From a different perspective, this study explicitly concentrates on the influence of the latent effects including space, time, and spatiotemporal interaction on dynamic shrinkage process based upon neighborhood-level (census tract) population shifts in Ohio for the period 1970–2010. To explore those random influences, we hypothesize that the shrinkage of population in census tracts over previous decades is in close association with growth/decline of neighboring tracts (spatial and temporal dependencies). The study adopts a full Bayesian hierarchical analytic framework that specifically utilizes the Conditional Autoregressive (CAR) and Penalized Complexity (PC) priors in order to account for the latent effects on dynamic process. For 635 census tracts in Cleveland-Elyria Metropolitan Statistical Area, Ohio in the U.S., we have found that when the size of initial population of a tract increases, it is more likely to observe population decline on average. Similarly, the relative share of African American population, unemployment rate, and poverty rate has shown a negative correlation with population growth. Our model reveals that there exists a significant influence of the random (latent) effects: a spatial correlation from adjacent tracts, a non-linear temporal dependence, and a spatiotemporal interaction complementing the random fluctuation. Moreover, we present areas of higher risks that are census tracts with higher probabilities (close to 1) of losing 10% or more of their relative population. We believe that the findings of our study allow us to support informed decision making for devising policies of shrinking regions.