Research Concepts

Stratified Random Samples

Sometimes you can isolate characteristics of your population that might influence your study. For example, in studying eyestrain among workers, those subjects who wear contact lenses might be more susceptible to eyestrain than those who do not wear contacts. If your sample had a smaller proportion of contact lens wearers than in your population, your sample might result in biased results even if it were randomly selected.

Since approximately 11% of the US population wear contact lenses, we might want to construct our sample so that 11% of the subjects in our sample wear contact lenses. To construct a stratified random sample we would divide our population into two groups or strata: contact lens wearers and everyone else. We would then randomly select 11% of our sample from the contact lens wearers and 89% of our sample from everyone else. This strategy combines random sampling and stratified sampling.

This approach assures that our final sample -- the combination of the random samples from each strata -- has a similar composition to the population. This approach is often used to assure gender, ethnic, age, geographic and other kinds of balance for samples.

There are many technical challenges to random sampling. For example, to do all of the random selections we require lists from each category; in many cases these lists may not be readily available. Adding strata also increases the complexity of sample construction. FInally, to assure the confidentiality of your subjects, no strata should have fewer than three subjects; this can result in uneconomically large sample sizes.