Examining the correlation between individual metrics and overall score in a health index

Using America’s Health Rankings as an example, this presentation will show how tightly the individual metrics used to create composite indices of health are correlated.

In 2014, 29 individual metrics were used to construct the rankings, including the prevalence of smoking, obesity and binge drinking, immunization rates, cancer, cardiovascular disease and infant mortality, violent crime, air pollution and the density of primary care physicians and dentists. The correlation between  individual metrics and the overall score  were examined. The overall score is the sum of the z-scores of all 29 metrics multiplied by the metric’s assigned weight. The states are ranked based on the overall score. The correlations range from more than r=0.8 for smoking, obesity and diabetes to r < 0.4 for public health funding, teen immunization, and disparity in health status by educational attainment.

Over the last ten years, the correlation of many individual metrics with the overall score has remained relatively constant. For example, the prevalence of smoking has remained correlated at 0.76 <r < 0.79  and high school graduation has remained between 0.55 < r < 0.67.

Modeling the components within the rankings shows that changes in the prevalence of smoking, infectious disease, children in poverty and obesity have the largest impact on a state improving in its comparative health. However, even large changes in any single metric will have minimal impact on the rank because of the high inter-correlation of all of the components and the small weight of each individual metric.