Understanding false positives in control conditions: a simulation study of post-hoc testing in low-powered neuroimaging trials.

Abstract

Randomized controlled trials (RCTs) are essential for evaluating treatment efficacy, typically comparing active interventions to control conditions. In situations where blinding is impractical-such as in psychological therapies or physical rehabilitation-waitlist controls are often used to account for natural symptom progression and test-retest variability. This study examines the biases introduced by post-hoc analyses under conditions of low statistical power, particularly in neuroimaging research. Through large-scale simulations involving 100 million datasets with varying sample sizes, treatment effects, and test-retest variability, the study demonstrates that the common practice of conducting post-hoc tests only on brain regions showing significant interaction effects can substantially increase the false positive rate in the control condition. These findings underscore the relevance of Berkson's paradox in interpreting unexpected control group outcomes and caution against overinterpreting such results. A complementary neuroimaging simulation reinforces these conclusions, emphasizing the need for critical scrutiny when evaluating significant effects in control groups. Overall, this work challenges conventional post-hoc testing strategies and advocates for a more nuanced and statistically informed interpretation of results, especially in studies with limited power.