Using a Large Control Group for Statistical Comparison: Evaluation of a Between-Groups Median Test

Abstract

Abstract Research situations occur in which data are obtained from a small group of rare experimental participants. An alternative strategy to comparing the experimental group with a control group of comparable sample size is to use a large control group for statistical comparison. That strategy increases the probability that the control group will provide an accurate benchmark for statistical comparison and is recommended as a method to increase statistical precision. However, traditional statistical procedures cannot ensure accurate error rates when data with highly unequal cell frequencies are being analyzed. The author presents an alternative statistical test (BOOTmed) for the 2-group situation when a small experimental group is compared with a large control group. BOOTmed is a between-groups median test derived via bootstrapping techniques. Empirical evaluation indicated that BOOTmed maintains relatively robust error rates during a variety of test conditions.