Randomized block design

Blocks are formed before any randomizing, using a variable you expect to affect the response, so units inside a block are similar on that variable. For example, to test two fertilizers on 40 plots that run from sunny to shaded, form 4 blocks of 10 plots by sunlight level, then randomly assign each fertilizer to 5 plots in every block. Because all treatments occur within every block, you can compare the treatments without variation from the blocking variable clouding the comparison. A matched pairs design is the two-treatment case: units are paired on an extraneous variable and one treatment is randomly assigned within each pair, or a single unit receives both treatments in a randomized order.