PRESENTING THE UNCERTAINTY OF SAMPLE SIZE CALCULATIONS

Home Chirostat News Multiple Imputation course About Chirostat

Eddie Channon, Chirostat Statistical Consulting

PSI conference May 2007

ABSTRACT

Classical sample size calculations assume that the parameters are known. When the parameters have been estimated from a small reference study, there may be considerable uncertainty. To illustrate this, a statistician could present 10 equally likely possibilities for the power of a new study with the proposed sample size.

As an example, when proportions on treatment and placebo were estimated from a reference study in 80 patients, the proposed sample size for the new study had power ranging from 50% to 100% (average 90%). After seeing these calculations, the clinician identified additional reference studies which led to much more precise sample size calculations. A second example looks at the uncertainty when standard deviation has been estimated from a small reference study.