A systematic review of the reporting of sample size calculations and corresponding data components in observational...

Anecdotal evidence suggests that functional magnetic resonance imaging (fMRI) studies rarely consider statistical power when setting a sample size. This raises concerns since undersized studies may fail to detect effects of interest and encourage data dredging. Although sample size methodology in this field exists, implementation requires specifications of estimated effect size and variance components. We therefore systematically evaluated how often estimates of effect size and variance components were reported in observational fMRI studies involving clinical human participants published in six leading journals between January 2010 and December 2011. A random sample of 100 eligible articles was included in data extraction and analyses. Two independent reviewers assessed the reporting of sample size calculations and the data components required to perform the calculations in the fMRI literature. One article (1%) reported sample size calculations. The reporting of parameter estimates for effect size (8%), between-subject variance (4%), within-subject variance (1%) and temporal autocorrelation matrix (0%) was uncommon. Three articles (3%) reported Cohen's d or F effect sizes. The majority (83%) reported peak or average t, z or F statistics. The inter-rater agreement was very good, with a prevalence-adjusted bias-adjusted kappa (PABAK) value greater than 0.88. We concluded that sample size calculations were seldom reported in fMRI studies. Moreover, omission of parameter estimates for effect size, between- and within-subject variances, and temporal autocorrelation matrix could limit investigators' ability to perform power analyses for new studies. We suggest routine reporting of these quantities, and recommend strategies for reducing bias in their reported values.