Body Mass Index and Neighborhood Characteristics: Assessing Selection and Causation Mechanisms Using Mover-Stayer Models

Ken R. Smith, University of Utah
Ikuho Yamada, University of Utah
Barbara Brown, University of Utah
Heidi Hanson, Huntsman Cancer Institute
Cathleen Zick, University of Utah
Jessie Fan, University of Utah

Studies linking the physical environment to the risk of being overweight are limited by the fact that residents are not randomly distributed by neighborhood. If associations are found between neighborhood characteristics and body mass indices (BMI) in observational studies, one cannot confidently draw conclusions about causality. We assess the influence of causal effects in the presence of non-random residential selection using longitudinal mover-stayer models. There is no definitive statistical test of the relative contributions of causation and selection when dealing with non-experimental data. We consider mover-stayer models to assess these two effects. We use Utah birth certificates from 1989-2007 for women with 2+ children (N=135,300) because they contain longitudinal data on self-reported pre-pregnancy weight and residential information at the time of each birth. We find evidence of non-random selection into neighborhoods based on individual BMI and neighborhood characteristics. Results are compared to models that ignore the potential influences of selection.