Spatial Path Models with Multiple Indicators and Causes: Population Psychiatric Outcomes in US Counties
*Peter Congdon, Queen Mary University of London 

Keywords: Spatial, Structural equation, Social Capital, Latent construct, Path model, Bayesian.

This paper considers a structural model for the impact on area health outcomes of spatially structured latent constructs, namely unobserved ecological risk factors. Psychiatric health outcomes (e.g. depression, suicide) are considered in relation to four constructs: deprivation, social capital, social fragmentation and rurality. These are measured by multiple observed social indicators, and here allowed to be correlated both between and within areas. However, particular latent constructs may also be influenced by known variables or by other constructs; that is, a spatial structural equation scheme may also involve multiple causes as well as multiple indicators,and possibly path sequences between constructs. For example, area social capital may be measured by indicators such as community organization memberships or charitable activity, but influenced (as causes) by other constructs (e.g. fragmentation) and by observed features of the demo-ethnic structure of areas. The model is applied to depression prevalence and suicide mortality in 3141 US counties, which are related to the latent spatial constructs and to observed variables (e.g. county ethnic mix)