Abstract:
|
Recently in neuroimaging, there has been a lot of attention drawn to the reproducibility crisis, where replications of studies fail to reproduce the same results. One of the primary contributions to this problem is the prevalence of low-powered neuroimaging studies. Small sample sizes have many negative impacts on an analysis including: true effects will go undetected, if a significant effect is found, the effect size tends to be inflated and significant results from underpowered studies are more likely to be false. Power analysis for neuroimaging is difficult due to the multiple comparison problem and temporal autocorrelation. In addition, there is often little intuition about reasonable effect sizes and collecting pilot data, for the sake of a power analysis, is quite expensive. Here is a discussion of the history of power analysis in neuroimaging, the positive impact it has had on the field and challenges that will need to be addressed in the future.
|