< County Data
Sampling Method for State Survey
Respondents to the state BRFSS survey are selected at random using a “Disproportionate Stratified Random Sampling” design. The selection process occurs in two stages. First, a random telephone number is selected from listed and unlisted telephone numbers. When a residence is selected, a random member of the household (aged 18 or older) is selected for the interview according to BRFSS protocol.
Generating Local Area Estimates
The telephone number and zip code are used initially to determine number of responses to the state survey by county. Since the sample size for each county is based on the telephone numbers, each county will be represented roughly in proportion to the number of residential telephone lines in the county. This results in wide variations in the number of interviews from each county. For example, Pulaski County--the largest in the state--had 660 complete interviews in 2005. On the other end, Lee County, which is much smaller, had only 18 interviews completed--too small for a reliable estimate. There are two ways to address this issue: 1) combine data for multiple years or 2) combine data for adjacent counties. The Center for Health Statistics and epidemiologists at ADH decided to use data for single years and combine the data from adjacent counties to get enough interviews to develop a reliable estimate. This method is used for all counties.
For example, Chicot County had 22 complete responses in 2005. When this number is added to the complete responses for the adjacent counties of Desha (26), Drew (33) and Ashley (45), the total is 126 complete responses.
These 126 responses are then adjusted so that the age, race, and gender characteristics match those of the population of Chicot County. This is done with a special weighting program developed by Dr. John Senner that considers the age, race, and gender composition of both the sample and actual population to appropriately project the sample to the population. For example, if males are 60 percent of the sample but only 45 percent of the actual population for Chicot County, the male population is over-represented, which leads to a biased estimate. The weighting program removes this effect by giving less weight to the male respondents in the dataset. Individual weights are developed for each county.
Finally, a SAS program is used to calculate the estimates for health risk factors for each of the counties.