In stratified sampling, the population is first divided into groups, also called strata, according to some variable (for example, school district). Then, elements from each stratum are selected at random according to one of the two ways:

1. the number of elements drawn from each stratum depends on the stratum´s size in relation to the entire population (“proportionate” sampling),
2. the number of elements sampled from each stratum is not proportionate to the size of the stratum (“disproportionate” sampling).  Often an equal number of elements is typically drawn from each stratum and the overall results are weighted according to the stratum´s size in relation to the entire population.

Stratified sampling is used to assure representation of the entire population – for example, assuring that, in a survey of student attitudes, 3rd graders are equally represented with 5th graders.

Disproportionate sampling (usually equal numbers from each stratum) is used when you want to draw conclusions about subgroups (say, opinions of blacks and whites) and you want the confidence intervals for those subgroup conclusions to be based on similar sample sizes.