Accurate estimation of standard errors when comparing simulated power functions

Abstract

We propose a simple technique for accurately estimating standard errors when comparing simulated powers of two tests designed to detect the same alternative. The procedure incorporates the correlation between the competing test statistics to yield reduced estimates of standard error. As a result, desired precision can be achieved with fewer replications when comparing, correlated test statistics. The technique is applied to compare tests based on two popular rank correlation coefficients.

Keywords:

• Simulation
• power function
• standard error