Why are bias-adjusted modeled estimates different?
At the state level, for selected statistics, you can choose between bias-adjusted modeled estimates and BRFSS direct estimates. You will find that the numbers are different. For example, the direct estimate for current smokers in Kentucky is 24.8%, while the bias-adjusted modeled estimate is 30.0%.
One big difference is that the direct estimate is from data collected in 2010, while the bias-adjusted estimate relies on pooled data from 2000 through 2003.
Newer bias-adjusted estimates are being modeled now and should be available by mid-2012.
Another difference is that the modeled estimates use data from BRFSS as well as another survey (NHIS) that has a smaller but more representative sample. For similar time periods, the modeled estimates may be closer to the true value.
See the Data Type Explanation.
The model uses multiple years to provide reasonable sample sizes (mainly for estimates at the county level), and needs at least two years of data from each survey. For the screening questions, which are not asked every year, that will mean grouping 2004 to 2009. For smoking, the period will be 2007 to 2009.