Caterina Gratton, Steven M. Nelson, Evan M. Gordon
5 min
For years, human neuroscience has relied on cross-sectional studies to link brain measures—such as those from resting-state fMRI—to behavioral traits like cognitive performance or mental health. However, a recent large-scale analysis by Marek et al. (2022) reveals that these associations are often vanishingly small. By examining data from thousands of participants across major consortia like the ABCD study and the Human Connectome Project, the authors demonstrate that typical effect sizes are far too small to be reliably detected in the sample sizes (N=25–100) common in the field. This statistical reality means that many published findings are likely artifacts of sampling variability rather than true biological signals.
The authors argue that the field must move away from the current standard of underpowered cross-sectional research. They propose two distinct, valid paths forward:
Consortium-Level Research: By pooling data across thousands of participants, researchers can identify small but robust population-level associations. This approach is analogous to genomics and is well-suited for informing public health policy, though it is limited by high costs and a tendency to favor established, low-risk measures.
High-SNR, Individual-Focused Research: Researchers can instead maximize statistical power by increasing the signal-to-noise ratio (SNR) in smaller samples. This involves using within-subject designs, repeated measures, and experimental manipulations that induce large, observable changes in brain and behavior. This approach is particularly valuable for clinical translation, as it focuses on understanding the dynamics of individual patients.
This shift is essential for the future of neuroscience. Continuing to conduct small, underpowered cross-sectional studies is a dead end that wastes resources and generates non-replicable literature. By choosing between large-scale consortia for population-level questions and high-precision, within-subject designs for mechanistic or clinical questions, the field can move toward a more reliable and cumulative science. Funding agencies and journals must adapt by prioritizing these high-powered approaches over the traditional, yet unreliable, small-sample cross-sectional model.
Sam: [direct] That is the primary tension the authors identify. There is a fork in the road. You can join a massive consortium — which is well-suited to population-level questions and policy-relevant findings, but tends to be too risk-averse for genuinely novel, exploratory science. Or you change your design entirely.
Alex: [curious] What does that second path actually look like? [[RP_SECTION:within-subject-research-designs|Within-Subject Research Designs]]
Sam: [deliberate, teaching mode] You shift from cross-sectional to within-subject designs. Instead of looking for tiny differences across a heterogeneous population, you measure the same person repeatedly — under different conditions, across time, or in response to a controlled intervention. That isolates the effect of interest and dramatically improves your signal-to-noise ratio.
Alex: [the penny drops] So rather than trying to hear a whisper across a crowded room, you bring the speaker into a soundproof booth.
Sam: [calm] That is exactly it. By inducing large, controlled changes within an individual, you can recover effect sizes an order of magnitude larger than the cross-sectional correlations the field has been chasing. The biology hasn't changed — you've just stopped fighting the noise.
Alex: [deliberate] So the recommendation is to abandon the middle ground — those medium-sized cross-sectional designs that are too small for population trends but too heterogeneous for precision mapping.
Sam: [measured, honest] Precisely. The intermediate design is a dead end. If you want to characterize population-level traits, you need the scale of a consortium. If you want to understand the mechanisms of brain function, you need high-precision, within-subject work. A few hundred participants in a one-shot cross-sectional study gives you neither.
Alex: [quieter, reflective] That has real implications for how grants get scored and how manuscripts get reviewed. [[RP_SECTION:future-research-directions|Future Research Directions]]
Sam: [slower, deliberate] It does. Funding agencies and journals need to become skeptical of cross-sectional studies with small samples — not because the researchers are doing anything wrong, but because the design is structurally incapable of producing reliable answers to the questions being asked. The field needs to stop rewarding the discovery of statistical ghosts and start valuing either the scale of consortium data or the precision of individual-level mapping.
Alex: [measured] It's a significant constraint — but it sounds like the only route toward findings that will actually hold up in clinical translation.
Sam: [quiet conviction] It is. And while it feels restrictive in the short term, it is really a correction — moving neuroimaging toward the kind of rigor that makes findings worth building on. The science doesn't get smaller; the questions just get sharper. Thanks for listening to ResearchPod.