A sample drawn from a population that is divided into strata from each of which a random sample is taken. The strata selected may relate to features of the population that are expected to alter the impact of the treatment being investigated. For instance, a vitamin may be anticipated to have smaller effects on families that grow many of their own vegetables compared to ones who do not. If the sample is not stratified to include an adequate proportion from those who do and do not grow their own vegetables, it may turn out, by chance, that all selected by random come almost exclusively from only one of these two groups. This may result in the effect of vitamin supplements being overestimated or underestimated for the population as a whole, underestimated if there were an unrepresentatively large group of vegetable growers in the sample, overestimated in the reverse case. The problem is reduced if the total sample size is large enough to allow for stratified sampling with respect to factors expected to alter the effectiveness of a treatment. In addition, if the stratified sample is large enough, the differential impact of the treatment on the different subgroups can be estimated.
Was this article helpful?