Completely Randomized, Blocked, and Matched Pairs Designs

A completely randomized design assigns all subjects to treatments by chance, with no grouping first. That works, but if subjects differ a lot on some trait that affects the outcome, like age affecting running speed, that variation can drown out the effect you're trying to measure. Blocking fixes this: you sort subjects into groups (blocks) that are similar on that trait, then randomize treatment separately within each block. The random assignment still happens, so you keep the cause-and-effect logic of an experiment, but now you've controlled a known source of variability instead of leaving it to chance.

A matched pairs design is blocking taken to its extreme: each block has only two observations, often the same subject measured twice, or two very similar subjects paired up. Because each subject is compared only to themselves (or their matched partner), any individual differences in baseline ability cancel out. Within each pair, which treatment comes first, or which side of the body gets which treatment, is still randomly assigned, so chance still decides the order or placement.

Both designs sit between full randomization and no randomization at all: you deliberately reduce noise by grouping on a variable you suspect matters, but you preserve randomness where it counts, within each group. The payoff shows up in the final analysis: less unexplained variability means a real treatment effect is easier to detect. This connects directly to later work on experimental design, where you'll judge whether a study's structure lets you draw a valid cause-and-effect conclusion.