What is statistical power?
robability of correctly rejecting the null hypothesis when a true effect exists (i.e., detecting real effects).
Low power equals to what kind of error
high Type II error (false negatives)
What determines power?
Sample size (n)
Effect size (true brain–behavior associations)
Alpha level (α) (Type I error threshold)
What additional structural challenges reduce power in fMRI?
Mass-univariate testing (hundreds of thousands of voxels → huge multiple-comparisons penalty).
Small typical sample sizes (n ≈ 20–30).
→ Even with perfect design, this severely reduces power.
Consequences of low power
- high rate of missed effects: Weak diffuse effects—common in real fMRI datasets—are almost never detected with typical sample sizes
- inflated effect sizes: Significant voxels in small samples dramatically overestimate the true population effect size. This happens because only the “lucky” overestimated voxels pass the significance threshold. The weaker and more diffuse the true effect, the worse the inflation.
- misleading inferences about brain organization: Small samples produce results that look highly localized and strong, even when the true underlying pattern is weak and distributed.
- poor reproducibility due to low power
Why is power especially bad in fMRI
- huge number of dependent variables requiring strict multiple-comparison correction
- typical sample sazes are too smole
- between subjects analyses
- effects in fMRI are often weak and diffused, not strong and localized
What did the simulations show?
Weak diffuse effects require huge samples; n=30 massively overestimates effect sizes and shows almost no overlap across replications. Strong localized effects fare better.
what did HCP data show?
Real fMRI effects (esp. individual differences) resemble weak diffuse scenario, confirming low power in typical studies.
How to improve statistical power in fMRI research
- conduct proper power analuses and report uncertainty
- increase sample size (most direct and reliable way to increase power - more subjects rather than more tasks per subject)
- less stringent statistical thresholds, for example uncorrected p < .001
- reduce number of tests
- meta analysis: combine many underpowered studies to detect consistent patterns
What is the main criticism of mental categories (attention, memory, emotion) in neuroscience?
They are epistemically sterile: they do not map cleanly onto neural architecture and provide poor explanatory value.
Why are mental categories circular?
They define the phenomenon and its mechanism using the same term (e.g., “attention is caused by attentional mechanisms”).
Why can’t brain regions be assigned to single mental domains?
Regions like the amygdala participate in many circuits (cognition, emotion, action) due to extensive interconnectivity
What is the alternative to mental categories?
Describe behavior in terms of dynamic, interacting brain circuits rather than labels like “attention” or “emotion.”
Why is the “old vs new brain” idea scientifically inaccurate?
Evolution is mosaic; subcortical regions are not primitive remnants, and cortex is not the exclusive seat of cognition.
What evidence undermines the idea that cortex is necessary for cognition?
Birds show high cognition with no layered neocortex; pallial regions perform similar functions via different architectures.
What actually changes in evolution: new regions or connections?
Primarily reorganization of connectional systems, not addition of new “modules.”
What should researchers study instead of mental categories?
Naturalistic, ethologically relevant behaviors (e.g., threat assessment, foraging)
Why is naturalistic behavior better?
Because real behaviors involve integrated cognitive, emotional, perceptual, and motor components—mirroring the brain’s organization
What is “dynamic brain–behavior coupling”?
Mapping how patterns of neural activity evolve together with ongoing behavior over time.
What methodological shift do the authors recommend in terms of new brain vs old brain?
Move from searching for region-specific functions toward circuit- and network-level explanations.
How can comparative neuroanatomy guide study design?
By showing which circuits and building blocks are conserved across species, helping identify meaningful neural units.
Why don’t mental categories accurately reflect brain function?
Because the brain is heterarchical and massively integrated; categories artificially separate processes that the brain blends
What is wrong with modular explanations of mental functions?
They assume stable, circumscribed neural systems that do not exist; neural circuits dynamically reconfigure depending on context.
