In an experiment, we’re going to present randomly generated stimuli to multiple subjects one trial at a time and measure neural responses to them. The stimuli are randomly generated to satisfy the conditions we want to compare, say for example half of the trials will use stimuli randomly drawn from an infinite pool of condition A stimuli and half will be drawn from an infinite pool of condition B stimuli.
It’s the same amount of effort for us to use the same random generation of the stimuli to every subject, versus randomly generating entirely new stimuli for every subject while always having half of the stimuli from condition A and half of the stimuli from condition B. Our interest is to make inferences on the neural responses to condition A versus condition B. Which of the two would be preferred?