Confidence interval for statistical power using a sample variance estimate

To improve estimation of statistical power by constructing its confidence interval.

Statistical power is often computed using a sample variance estimated from previous studies. The sample variance deviates from the population variance and its use makes the estimated statistical power less accurate. The estimated power thus computed may be higher than the actual statistical power in the planned study.

The standard deviation for body mass index is estimated by using the standard literature on weight-loss research.

This is a methodology paper.

Researchers avoid estimating the population variance in statistical power analysis by using the standardised effect size. The rule-of-thumb numbers for 'small', 'medium' and 'large' standardised effect sizes are .2,.5, and .8, but these numbers may not fit all the studies in different contexts. It is recommended that researchers start with a simple effect size that has some clinical importance and then estimate the population variance in running statistical power analysis. The use of an estimated variance in power analysis introduces uncertainty to the computed statistical power. To account for the uncertainty in estimating the variance, researchers can use the confidence interval to accurately assess the actual statistical power in planned hypothesis testing.

The confidence interval provides a more realistic view of power than a single-value estimate of the nominal power does, which tends to be higher than the actual power in a clinical study.

The authors have introduced a useful technique to construct a confidence interval on statistical power using a sample variance from a previous study. The method can be easily implemented to plan statistical power and sample size in a clinical study.