For nonparametric reference limits with n > 120, which combination defines the reference limits and associated confidence interval?

For nonparametric reference limits with n > 120, which combination defines the reference limits and associated confidence interval?

• A. 2.5th and 97.5th fractiles; 90% CI
• B. 5th and 95th fractiles; 90% CI
• C. 2.5th and 97.5th fractiles; 95% CI
• D. 1st and 99th fractiles; 90% CI

Answer: (A)

For nonparametric reference limits with a large sample, you base the limits on the empirical percentiles and express precision with a confidence interval around those limits. Since you want a central reference interval that covers 95% of the population without assuming any distribution, the endpoints are the 2.5th and the 97.5th percentiles of the data. These two points define the central 95% reference interval. The precision of those endpoints is shown with a confidence interval around each limit, and for n greater than about 120, the recommended confidence level for those limits is 90%. In other words, report the lower limit at the 2.5th percentile and the upper limit at the 97.5th percentile, with a 90% confidence interval around those limits to reflect sampling variability. Using 5th and 95th percentiles would yield a central 90% interval rather than the standard 95% reference interval, and 1st/99th percentiles would produce much wider limits with a different confidence-interval profile.

For nonparametric reference limits with a large sample, you base the limits on the empirical percentiles and express precision with a confidence interval around those limits. Since you want a central reference interval that covers 95% of the population without assuming any distribution, the endpoints are the 2.5th and the 97.5th percentiles of the data. These two points define the central 95% reference interval.

The precision of those endpoints is shown with a confidence interval around each limit, and for n greater than about 120, the recommended confidence level for those limits is 90%. In other words, report the lower limit at the 2.5th percentile and the upper limit at the 97.5th percentile, with a 90% confidence interval around those limits to reflect sampling variability.

Using 5th and 95th percentiles would yield a central 90% interval rather than the standard 95% reference interval, and 1st/99th percentiles would produce much wider limits with a different confidence-interval profile.