Back to Annual Meeting

Back to Annual Meeting

APHA Scientific Session and Event Listing

Kelly L. Knott, MS¹, Barbara C. Tilley, PhD¹, Rickey E. Carter, PhD¹, Paul J. Nietert, PhD¹, Kit N. Simpson, DrPh², and Brent M. Egan, MD³. (1) Department of Biostatistics, Bioinformatics, and Epidemiology, Medical University of South Carolina, 135 Cannon Street, Suite 303, Charleston, SC 29415, 843 876 1065, knottk@musc.edu, (2) Center for Health Economics, Medical University of South Carolina, 151B Rutledge Ave, Charleston, SC 29425, (3) Department of General Internal Medicine and Hypertension, Medical University of South Carolina, 135 Rutledge Ave, Rm 1104, Charleston, SC 29425

The variability of ranked data is usually not considered when rankings are used to make data comparisons. In this work, methods already developed for single indicators were extended to compute confidence intervals (CI) for rankings based on multiple indicators. A Monte Carlo (MC) simulation was designed to estimate an empirical distribution of ranks. All data were simulated at the entity level. First, an empirical distribution for each indicator was resampled to derive a score for each indicator. Once these values were scored, the values were pooled across the indicators to derive an entity-specific composite score, which was then ranked by entity. This process was repeated 10,000 times to obtain stable estimates of the rankings. Across the 10,000 MC replications of constructed ranks, the 5th and 95th percentiles of the ranks distribution were used to create 90% CIs for the ranks. Alternative methods were developed to simulate the data to account for the correlation among indicators. Using these methods, 90% CIs for the 2003 America's Health: State Health Rankings were constructed. The CIs were widest for the middle ranking states and the CIs were the narrowest for the lower and upper ranking states. The influence of the truncation of the ranks on the CI width motivated an alternative classification of states. The same MC simulations were used to estimate the probability that a state's rank was above, at, or below average. Taken together, these two methods provided a better understanding of the rank when the ranks are comprised of multiple indicators.

Learning Objectives:

• Those attending will learn about the variability associated with ranked data, specifically with the State Health Rankings, the importance of including confidence intervals for ranks, and the implications of using ranked data without confidence intervals.

Keywords: Policy/Policy Development, Statistics

Presenting author's disclosure statement:

Any relevant financial relationships? No