| Rankings are a common approach to summarizing and communicating public health data to foster community health improvement. However, data limitations, including small samples size, high variability, and censored/suppressed estimates can reduce the utility of such rankings. Not only are some entities unable to be ranked, our ability to assess similarities or differences among counties is limited because rankings do not generally report confidence intervals or other measures of precision regarding the ranks.;To address these problems, we developed multi-level models using aggregate, reported data to improve the estimation of health-related measures (premature mortality, percent reporting fair or poor health, poor physical and poor mental health days, and low birth weight births) across U.S. counties. These models capture components of variation at the state and county levels as well as sampling variation, and better approximate the true rates and ranks of county-level health measures. By leveraging existing data from all United States counties, these models naturally impute values for counties with missing or censored values. We are also able to calculate ranks adjusted for community demographics or "case mix." Most importantly, our approach allows for the calculation of confidence intervals for ranks. |
