![[ Visit Client Website ]](images/banner.jpg)
The Behavioral Risk Factor Surveillance System (BRFSS) is a state-based telephone health survey that collects data on health conditions and risk behaviors among adults aged18 years or older. For more than two decades, BRFSS data were collected through landline telephone interviews and weighted by post-stratification. Two changes were recently made in BRFSS protocol: the addition of cell phone interviews and introduction of iterative proportional fitting or raking as the new weighting method. Raking allows BRFSS to incorporate cell phone survey data and calculate estimates with smaller samples. BRFSS data released for 2011 reflects these changes. The objective of this study is to examine the effect of addition of cell phone respondents and the change in BRFSS weighting methodology on prevalence estimates reported for Kentucky.
METHODS:
We used data from 2010 and 2011 BRFSS for Kentucky. From the 2010 survey, three groups of prevalence estimates were created for comparison: estimates calculated with post-stratification weights (landline-only), those calculated with raking weights (landline-only), and estimates calculated using raking weights (combined landline and cell phone). Raking was the only weighting method used in 2011. Hence, we compared 2011estimates calculated for landline-only respondents to 2011 estimates calculated for combined landline and cell phone respondents. The magnitude and direction of change was compared. The impact on prevalence estimates were examined for several health risk behaviors and chronic conditions.
RESULTS:
For 2010 landline data, prevalence estimates calculated with raking weights were higher for cigarette smoking (+5.0), no leisure-time physical activity (+3.2), asthma (+1.4), diabetes (+1.2), and obesity (+1.2) compared to estimates calculated with post-stratification weights. When raking was used as the sole weighting method, estimates calculated with combined landline and cell phone data were higher for binge drinking (+2.4), cigarette smoking (+1.1) and lower for obesity (-1.5) compared to landline-only estimates. From the 2011 survey (weighted by raking), prevalence estimates calculated with combined landline and cell phone data were higher for binge drinking (+2.5) and lower for obesity (-1.3) compared to landline-only estimates.
CONCLUSIONS:
Prevalence estimates calculated with raking weights were generally higher when compared to estimates calculated with post-stratification weights. With raking as the sole weighting method, most of the differences in prevalence were less than two percentage points when combined landline and cellphone estimates were compared to landline-only estimates. It is important that public health officials are aware of these changes in BRFSS protocol and how the changes affect BRFSS prevalence estimates.