If you wanted to study the average height of people living in your neighborhood, one strategy would be to take a random sample of the city's citizens, measure them and then calculate the mean and median heights. This would lead to misleading conclusions, however, since children would bring the average height down. Even if you excluded children from the sample, your final average would likely be too short for most men and too tall for most women. The solution is a technique called stratified sampling.
Stratified sampling is the practice of dividing a research population into distinct groups, called strata, and randomly selecting members of each group. In the height study, for example, researchers might create two separate strata, men and women, and randomly select population members into the sample. They might also stratify different age groups, like infants, toddlers, elementary school children, young teens and older teens. This technique would allow researchers to calculate the average height within each strata. They could also calculate an average height for the population as a whole by combining the data from each strata into a weighted average. If the town's population is 40 percent male and 60 percent female, for example, the average height for the whole population would be the average male height multiplied by 0.4 plus the average female height multiplied by 0.6.
When to Use Stratification
Stratification is most useful when researchers are dealing with highly heterogeneous populations, especially when measurements of different subgroups are expected to vary. For example, women are usually shorter than men, so those two subgroups are likely to yield disparate height data. Stratification allows researchers to compute more precise and meaningful statistics than simple random sampling because random samples lump everyone together. Researchers studying incidence of skin cancer, for example, might stratify the population according to whether they live in sunny or cloudy areas. A national average might over-represent the risk of skin cancer in Minnesota, but under-represent the risk in Arizona. Stratified sampling allows the researchers to measure risk in each state separately.
Sometimes, grouping the population into strata is easy. Classifications like gender are fairly easy to make, but sometimes researchers have to make more subjective choices. For example, a company wanting to compare the TV-viewing habits of young, middle-aged and elderly people would need to define what those terms mean. Since a 35-year-old might be young to some and middle-aged to others, stratification can introduce subjectivity into the research.
A Labor-Intensive Method
Stratified sampling allows researchers extra precision in their calculations, but those benefits can come at a steep cost. To execute a stratified sampling methodology, researchers need a complete list of all members of the population and the relevant stratifying variables for each member. In the height study, for instance, stratification would require identifying all members of the population plus their ages and sexes. This makes stratified sampling much more labor-intensive than simple random sampling.
- Photo Credit Dave & Les Jacobs/Blend Images/Getty Images
The Disadvantages of Stratified Random Sampling
Suppose you're designing a political poll. You want to get an accurate sample of the U.S. population, but you know that men...