The County Health Rankings show us that where we live matters to our health. The health of a community depends on many different factors - ranging from health behaviors, education and jobs, to quality of health care, to the environment.

Home > Rankings > Ranking Methods
Print

Data Quality

The County Health Rankings team draws upon the most reliable and valid measures available to compile the Rankings.

Data Reliability

Reliability depends on the specific measures. Mortality data, for example, are extremely reliable. Deaths are reported 100% of the time (or very nearly so), and so the death rates are based on a "census", (i.e., all deaths that occurred rather than a sample of deaths), and thus have high reliability. Other measures, for example, air quality and binge drinking, are based on sampling methods, and so we provide more detailed background information to help users understand the quality of those measures. For example, although survey data are very useful, they have some well-known limitations (e.g., when they rely on the availability of land line telephones or when they involve self-reports with no way of determining if people respond and answer truthfully). However, when all of the measures are combined, we are confident that the Rankings provide a solid picture of overall health in a community, and we have found in Wisconsin that the results are not surprising: the counties at the bottom of the list for overall health are counties that have had challenges for decades with respect to employment, income and education but also unhealthy behaviors and health care systems that are not in great shape. Looking measure by measure, there may be concerns about reliability but, overall, the summary message of overall health in a community comes through despite some of the data limitations.

Statistical Uncertainty

Where possible, we provide the margin of errors (95% confidence intervals) for our measure values. In many cases, the values of specific measures in different counties are not statistically different from one another; however, when combined using our model, those various measures produce the different rankings.

Insufficient Data for Ranking

One hundred seven counties or county equivalents (e.g., in Alaska) were not ranked at all.  Counties were not ranked if they had a missing value for Years of Potential Life Lost, or if they had an unreliable value for Years of Potential Life Lost and no other measures of morbidity were available. They were also not ranked if the only measures of Health Outcomes available were an unreliable YPLL value and an unreliable Low Birthweight value. Unreliable measures were identified when the standard error of the estimate was more than 20% of the estimate value.

Treatment of Missing Values

In many ranked counties, some individual measures do not have a large enough sample size to report data for that measure. In these counties, we assigned the state average for any missing measure in order to calculate a rank for that category.  On the map for each measure, counties with missing values will be identified with N/A (not available).

Age-Adjustment of Measures

Age-adjustment is a useful strategy to increase the comparability of health measures. Unfortunately, because of data limitations and sample size issues we can only age-adjust some of the measures used in the Rankings. The following Health Outcome measures are age-adjusted: premature death (YPLL), self-reported health, physically unhealthy days, and mentally unhealthy days.