I'm a psychology researcher planning a study on the impact of background noise on cognitive task performance, and I'm stuck on the experimental design. I want to compare three noise conditions (silence, white noise, and cafe chatter) but I'm concerned about order effects and participant fatigue if I use a within-subjects design. However, a between-subjects design would require a much larger sample size, which my lab's budget may not support. I'm also unsure about the best way to randomize the task order and control for individual differences in noise sensitivity. Any advice on balancing these methodological constraints would be greatly appreciated.
Good plan to stay within-subjects but guard against order and fatigue. Use a Latin square to rotate the three conditions so each appears in every position roughly the same number of times, and keep sessions short with built-in breaks. A day or two between sessions helps mitigate carryover.
From an analysis standpoint, go with a linear mixed-effects model. Treat participant as a random intercept, include fixed effects for condition and order, and covariates like baseline attention or self-reported noise sensitivity. Interactions can show if sensitivity moderates the noise effect.
Pilot the design and pre-register your analysis. Build in a washout period or rest breaks and include a brief fatigue check after each block to detect carryover. If possible, split testing across two or three shorter sessions rather than one long session to reduce fatigue.
Power-wise, within-subjects designs are efficient, but you still need a reasonable effect size. If the budget is tight, consider a Bayesian or sequential design to stop early when evidence is strong, rather than sticking with a fixed N.
Randomization detail: with three conditions there are six possible orders. Use complete counterbalancing across participants so each order occurs roughly equally; if that’s not feasible, a randomized block approach can achieve similar balance.
Your measures should include both objective and subjective data: accuracy and reaction time, plus a brief noise annoyance rating and a mood/state anxiety check. Pre-register hypotheses and the data-collection plan to minimize flexible analyses.