ABSTRACT
The term regression modeling is used loosely to denote a situation where a disease outcome variable is geo-referenced and is to be related to predictors and/or covariates. These predictors can be observed at different levels of spatial aggregation and can be spatially-referenced or not. Ecological analysis is special case of regression modeling where a relation is estimated at an aggregation level and inference is to be made at a lower aggregation level. An initial example will serve to motivate the discussion of modeling issues. Some of these issues are purely general in that they relate to Bayesian modeling. Whereas others are specifically related to either spatial or spatial epidemiological contexts. Low income and poverty both could lead to a higher low birth weight (LBW) rate in counties due to poor prenatal care access or cultural factors. In addition, ethnicity may also confound due to behavioral and cultural modalities in different ethnic groups.
