A Latin square design is a statistical method used to control for two sources of variability in experimental research, particularly when dealing with three factors. It arranges treatments in a square grid so that each treatment appears exactly once in each row and each column, minimizing the potential for confounding effects.