Neuroscience› Cognitive Neuroscience› Functional Brain Connectivity Studies

Neuroimaging methods — Significance testing for multi-voxel pattern analysis of BOLD data

Submitted 2026-10-22

Field: Neuroscience › Cognitive NeuroscienceType: A method hit a wallArea: neuroimaging methodsTarget: Significance testing for multi-voxel pattern analysis of BOLD dataStage: full study

Hypothesis

Standard tests such as Student’s t and the binomial test validly assess whether multi-voxel classification performance exceeds chance.

What failed

Simulations showed those standard tests are invalid in typical MRI experiments, with only a carefully constructed permutation test assessing significance correctly; performance estimates also rose with both temporal and spatial signal correlation among voxels, so classification accuracy cannot be compared across areas, subjects or preprocessing pipelines without care.

Broken assumption

Classification accuracy in fMRI can be tested as though trials and voxels were independent samples.

Why

The authors trace the inflation to the slow BOLD response and to correlations among voxels, which break the independence the standard tests assume.

From the published paper: Schreiber et al. PLoS ONE, 2013. ↗CC BY

Research ideas from this failure

Unvetted lead — read the papers

New hypotheses this failure suggests, each with a crucial experiment, a rival explanation of the original result, and the papers it rests on. Written by AI from paper titles and abstracts found on OpenAlex (and Europe PMC, for life and health sciences).

Related failures (1)

Matched on shared genes, mechanisms and ways of failing as well as wording. Each shows why it matched — treat these as leads rather than verdicts.

neuroimaging methodsBrain-wide association studies at typical sample sizes
Associations between individual differences in brain structure or function and cognitive or mental-health phenotypes can be reliably detected at the sample sizes typical of neuroimaging studies (median about 25).
Across three large datasets totalling roughly 50,000 individuals, brain-wide association effects were smaller than previously thought, leaving typical studies underpowered, with inflated effect sizes and replication failures. Replication rates improved and inflation fell only as samples reached the thousands.
Why it matched: same research area

Links in published research

Where this experiment's genes are known to matter outside its own research area, and the papers behind those links. Sources: Open Targets Platform and Europe PMC.