Role of multilevel analysis in exploring environmental health disparities

Multilevel modeling is a tool that can assist in exploring how the built, physical and social environment can perpetuate environmental injustices and create racial disparities in health. By combining individual level characteristics such as age, race/ethnicity, education, etc. with neighborhood level factors such as presence of liquor stores, absence of supermarkets, etc., the effects of levels of effects can be assessed and variance between levels estimated. Multilevel modeling makes it possible to study the effects of racial residential segregation, income inequality, disparate exposure and other factors that operate on a level greater than that of any individual's experience.

Multilevel modeling is becoming popular and is possible to perform using standard statistical programs including SAS, STATA and HLM. But multilevel analysis is more than just plugging in numbers and commands into the software.

A critical issue is the selection of what levels beyond the individual should be included in the analysis. Under best circumstances, level selection should be guided by an analysis of the factors being studied and the conceptual model driving the hypotheses being tested. Too often, level choice is opportunistically driven by the available data, which can result in inappropriate or misleading conclusions. A major related issue is how to define “neighborhood”. Researchers have used census block group, census tract, zip code and other metrics to parse out neighborhoods. Each has its advantages and disadvantages.

Used properly, multilevel modeling is a robust tool for furthering the science of environmental justice.