Drawing a stratified sample requires you to weight your overall estimate, or else it will be

biased. As an example, imagine that 15 percent of pediatric patients had an oral health condition,

and 50 percent of the rest of the patients had an oral health condition. In a stratified sample of 20

patients where you draw 10 from the pediatric population and 10 from the rest of the population,

because the pediatric population is oversampled (because they only make up 10 percent of the

background population but make up 50 percent of our sample), if weights are not applied, the

estimate of the percentage of the population with an oral health condition would be artificially

reduced. That is why it is necessary to apply weights to overall estimates derived from a

stratified sample.

If you are familiar with large epidemiologic surveillance studies such as the National Health

and Nutrition Examination Survey (NHANES) in the United States, you may be aware that

extremely complex stratified sampling is used in the design and execution of such studies.

Stratified sampling in these studies is unlike the simple example described earlier, where the

stratification involves only two age groups. In surveillance studies like NHANES, there may be

stratified sampling based on many characteristics, including age, gender, and location of

residence. If you need to select factors on which to stratify, trying looking at what factors were

used for stratification in historical studies of the same population. The kind of stratified sampling

used in large-scale surveillance studies is reviewed later in this chapter in the section “Sampling

in multiple stages.”

Engaging in systematic sampling

Earlier you considered a scenario where a clinic had a printed list of the entire population of patients

from which an SRS could be drawn. But what if you want to sample from the population of patients

who present to a particular emergency department tonight between 6 p.m. and midnight? There is no

convenient list from which to draw such a sample. In a scenario like this, even though you can’t draw

an SRS, you want to use a system for obtaining a sample such that it would be representative of the

underlying population. To do that, you could use systematic sampling.

Imagine you are surveying a sample of patients about their opinions of waiting times at a particular

emergency department, and you are doing this in the time window of between 6 p.m. and midnight

tonight. To take a systematic sample of this population, follow these steps:

1. Select a small number.

This is your starting number. If you select three, this means that — starting at 6 p.m. — the first

patient to whom you would offer your survey would be the third one presenting to the emergency

department.

2. Select another small number.

This is your sampling number. If you select five, then after the first patient to whom you offered the

survey, you would ask every fifth patient presenting to the emergency department to complete your