Factors associated with supermarket and convenience store closure: a discrete time spatial survival modelling approach

J R Stat Soc Ser A Stat Soc. 2018 Jun;181(3):783-802. doi: 10.1111/rssa.12330. Epub 2017 Oct 22.

Abstract

While there is a literature on the distribution of food stores across geographic and social space, much of this research uses cross-sectional data. Analyses attempting to understand whether the availability of stores across neighborhoods is associated with diet and/or health outcomes are limited by a lack of understanding of factors that shape the emergence of new stores and the closure of others. We used quarterly data on supermarket and convenience store locations spanning seven years (2006-2012) and tract-level census data in four US cities: Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; San Francisco, California. A spatial discrete-time survival model was used to identify factors associated with an earlier and/or later closure time of a store. Sales volume was typically the strongest indicator of store survival. We identified heterogeneity in the association between tract-level poverty and racial composition with respect to store survival. Stores in high poverty, non-White tracts were often at a disadvantage in terms of survival length. The observed patterns of store survival varied by some of the same neighborhood sociodemographic factors associated with lifestyle and health outcomes, which could lead to confusion in interpretation in studies of the estimated effects of introduction of food stores into neighborhoods on health.

Keywords: Bayesian modeling; integrated nested Laplace approximations (INLA); neighborhood sociodemographics; sales volume; spatially correlated random effects.