Abstract
Applications In Statistics: What's Hot / What's Next VII, Statistics & Public Health
School of Public Health, University at Albany East Campus Rensselaer, NY
March 23, 2002
Indirect Small Area Estimation of Behavioral Risk Factors by Logistic Regression:
Application to Estimating ZIP Code-Level Smoking Rates in New York State
Glen D. Johnson
New York State Department of Health, Bureau of Environmental and Occupational Epidemiology, Geographic Research and Analysis Section
and
University at Albany, School of Public Health,
Department of Environmental Health and Toxicology
New York State Department of Health, Bureau of Chronic Disease Epidemiology and Surveillance, Diabetes Epidemiology and Surveillance Section
and
University at Albany, School of Public Health,
Department of Epidemiology
Recent demand for mapping behavioral health risk factors at the sub-county level has created a statistical challenge when there is no data available that relates individuals to a geographic unit smaller than their county of residence. In order to obtain sub-county estimates for mapping, we must develop a method of indirect estimation that takes advantage of surrogate variables. To this end, a two-stage method was developed. First, a logistic regression function is fit to statewide survey data in order to model the relationship between a risk factor and socio-demographic variables with respect to a particular study area, such as a state. Second, community-level prevalence of the risk factor is predicted for individual mapping units by applying the model to the same predictor variables that are, in turn, aggregated from census data within the desired mapping units.
A mathematical argument is presented that justifies applying the model, whose parameters are fit using individuals in a statewide survey, to community-level aggregated population proportions. The method is then illustrated by fitting the model using the CDC-sponsored Behavioral Risk Factor Surveillance System (BRFSS) for New York State, then applying it to predicting smoking prevalence at the ZIP code level.