For each Adult Health Behavior indicator, we used multivariate analysis (logistic regression) to explore relationships between a set of predictor variables and the variable of interest. We tested age, race, Hispanic ethnicity, sex, HSA, and income. We found age, race, and sex to be the most consistent statistically significant predictor variables. Income was also significant in a number of cases.

We then turned our sights to developing synthetic estimates. The basic methodology was to develop age, sex, and race specific estimates for each indicator from the 5,000+ cases in the Virginia database for 2003-2004. We then derived estimates of the total population by age, sex, and race from the 2000 Census and 2005 population estimates from SRC Corp. We then developed BRFSS estimates for 2005 by applying the age, sex, and race specific BRFSS estimates for 2003 to the age, sex, and race specific population estimates for 2005.

We left income out of the equation for multiple reasons. First, income was a factor in some, but not all BRFSS variables. Second, we were only able to synthetically estimate income as a percentage of poverty from the 2003 BRFSS data (no actual estimates were provided). Third, there were not enough BRFSS cases to support reliable estimates of BRFSS indicators by age AND race AND sex AND income. Fourth, we would have had to synthetically estimate actual poverty levels for 2005, as no actual data are available.