henhaobio
Sign in

bioRxivNeurosciencePreprint

Effect Size and Power in fMRI Group Analysis

Stephan GeuteriD, Guanghao QiiD, Robert C. WelshiD, Tor Dessart WageriD, Martin A. LindquistiD

PDFDOI
Structured summary available · Jump to summary ↓

Abstract

Multi-subject functional magnetic resonance imaging (fMRI) analysis is often concerned with determining whether there exists a significant population-wide activation in a comparison between two or more conditions. Typically this is assessed by testing the average value of a contrast of parameter estimates (COPE) against zero in a general linear model (GLM) analysis. In this work we investigate several aspects of this type of analysis. First, we study the effects of sample size on the sensitivity and reliability of the group analysis, allowing us to evaluate the ability of small sampled studies to effectively capture population-level effects of interest. Second, we assess the difference in sensitivity and reliability when using volumetric or surface based data. Third, we investigate potential biases in estimating effect sizes as a function of sample size. To perform this analysis we utilize the task-based fMRI data from the 500-subject release from the Human Connectome Project (HCP). We treat the complete collection of subjects (N = 491) as our population of interest, and perform a single-subject analysis on each subject in the population. We investigate the ability to recover population level effects using a subset of the population and standard analytical techniques. Our study shows that sample sizes of 40 are generally able to detect regions with high effect sizes (Cohens d > 0.8), while sample sizes closer to 80 are required to reliably recover regions with medium effect sizes (0.5 < d < 0.8). We find little difference in results when using volumetric or surface based data with respect to standard mass-univariate group analysis. Finally, we conclude that special care is needed when estimating effect sizes, particularly for small sample sizes.

Abstract summary

Claude Opus 5.5 · xhigh effort · 18 Oct 2026 · from the abstract · CC BY-NC 4.0

Subsampling 491 Human Connectome Project subjects puts numbers on fMRI group analysis: roughly 40 subjects catch large effects (d > 0.8), near 80 are needed for medium ones.

Group fMRI studies routinely test an average contrast of parameter estimates against zero without knowing how many subjects that test actually needs. Using a large task-fMRI collection as a stand-in population lets the authors measure, rather than assume, the sensitivity and reliability of typical sample sizes.

Key results

  1. ShownRequired sample size depends sharply on the effect size being chased: regions with large effects are generally detected with 40 subjects, while medium effects need samples closer to 80.

    Samples of 40 generally detected regions with Cohen's d > 0.8, while reliable recovery of regions with 0.5 < d < 0.8 took samples closer to 80, assessed by drawing subsets from a 491-subject population. · Abstract

  2. SuggestedChoosing volume-based or surface-based data makes little practical difference to a standard mass-univariate group analysis.

    Reported as little difference in sensitivity and reliability between the two representations; the abstract gives no numbers for the comparison. · Abstract

  3. ClaimedEffect size estimation needs special care, the authors conclude, particularly at small sample sizes, where they looked for bias that depends on the number of subjects.

    Framed as potential bias in estimating effect sizes as a function of sample size; the abstract states the conclusion without reporting the magnitude or direction of the bias. · Abstract

Methods, data and reproducibility

Each subject in a large task-based fMRI collection was analysed individually, the full set was treated as the population of interest, and subsets of varying size were analysed with standard group-level general linear model techniques to see how often population effects were recovered.

  • Task-based fMRI from the 500-subject release of the Human Connectome Project, used as the reference population · human · n = N = 491
Methods & tools
Human Connectome Project (HCP), general linear model (GLM), task-based fMRI
Reproducibility gaps
The data source is named but no accession, download link, analysis software or code repository appears in the abstract.
Limitations (3)
  • The authors caution that estimating effect sizes requires special care, particularly in small samples. · stated by the authors
  • All conclusions come from one task-fMRI collection analysed with standard mass-univariate methods, so the sample-size guidance may not transfer to other tasks, scanners or multivariate analyses. · noted in this summary
  • The 491 subjects are treated as the population, so the recovered effects are defined by that cohort rather than by an independent ground truth. · noted in this summary

Read this if you plan or review task-fMRI group studies and want empirical sample-size guidance tied to effect size, or are weighing surface against volume analysis.

Citations
83

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISGao '26Durnez '16Sadil '25Gerchen '21Turner '18Nee '19
75% semantic matchNeuroscience
R. S. Gao, Anna Aleksandrovna Ivanova·bioRxiv·Aug 18, 2026