Suppression rules for unreliable estimates of health indicators

Health statistics for small groups defined by such characteristics as age, race/ethnicity, socioeconomic position, and geography are of interest to policymakers. However, some estimates for small groups and rare health conditions are unreliable, whether based on surveys or complete counts such as death certificates. Federal reports and online tables use different criteria for flagging or suppressing estimates failing to meet specific reliability standards. Suppression criteria vary across data systems and may be expressed in terms of a relative standard error (for example >30%), number of events (for example < 20), width of the confidence interval, and/or sample size. These criteria can lead to different decisions. For example, suppressing rates with < 20 events always leads to rates with a low relative standard error (< 25%) but may not meet usual confidence interval rules. This presentation compares suppression rules used for tabular data dissemination at the National Center for Health Statistics. Using the minimum sample size to express the different criteria in a common unit, the comparison shows how rules can be relevant for some scenarios but not for others. With increasing demand for dissemination of small group estimates, the need to clearly convey data limitations also increases.