For example say you wanted to study people's preference on pillow firmness. To help ensure you do not obtain a biased sample you could first split the population into male and females, then random sample each of these "strata"