Better power by design: A quasirandomization strategy to boost power in repeated-measures experiments

During an experimental session, participants adapt and change due to learning, fatigue, fluctuations in attention, or other physiological or environmental changes. This temporal variation affects measurement, potentially reducing statistical power. We introduce a restricted randomization algorithm, Permuted-Subblock Randomization (PSR), that boosts power by balancing experimental conditions over the course of an experimental session. We used Monte Carlo simulations to explore the performance of PSR across four scenarios of time-dependent error: exponential decay (learning effect), Gaussian random walk, pink noise, and a mixture of the previous three. PSR boosted power by about 13% on average, with a range from 4% to 45% across a representative set of study designs, while simultaneously controlling the false positive rate when time-dependent variation was absent. An R package, explan, provides functions to implement PSR during experiment planning.