A power calculation guide for fMRI studies

Abstract

In the past, power analyses were not that common for fMRI studies, but recent advances in power calculation techniques and software development are making power analyses much more accessible. As a result, power analyses are more commonly expected in grant applications proposing fMRI studies. Even though the software is somewhat automated, there are important decisions to be made when setting up and carrying out a power analysis. This guide provides tips on carrying out power analyses, including obtaining pilot data, defining a region of interest and other choices to help create reliable power calculations.

Figures

Fig. 1

Examples of three different levels of power, assuming a normal distribution. In each case, the null distribution is centered at 0 and the alternative is centered at , both with a variance of 1. A statistic threshold of 1.64 controls the Type I error rate at 5% for a one-sided hypothesis. In all cases, power is the area under the alternative distribution for statistic values larger than the threshold of 1.64. The left panel shows a high power example. The middle panel illustrates that if the mean of the distribution is exactly the threshold, the power is 50%. Lastly, the right panel shows a case where power is very low.

Fig. 2

Two options for sample size specification. If an individual sample size is selected you will obtain power estimates for that sample size only (left), whereas a range of sample sizes supplies a power curve (right).