Skip to learning

Exploring One-Variable Data and Collecting DataTopic 5 of 6

Sampling Methods: Simple Random, Stratified, Cluster, and Systematic; Sources of Bias

A sample is useful only if the selection process supports the conclusions you want to draw. Sampling methods aim to obtain information about a population while controlling cost and avoiding systematic exclusion. Random selection addresses representativeness; it does not guarantee a perfectly balanced sample every time.

Name the population and sampling frame

The population is the full group you want to describe. The sampling frame is the list or mechanism from which individuals can actually be selected. If a city's residents are the population but the frame contains only registered library users, some residents have no chance of selection. That mismatch creates undercoverage.

A sampling unit may be an individual, household, school, or another defined unit. State it clearly. Selecting households and surveying every resident is not the same design as independently selecting individual residents.

Simple random sampling gives every sample a chance

A simple random sample of size n is selected so every possible group of n individuals from the population is equally likely. Number the population members, then use a random-number process to select n distinct labels without replacement.

Equal individual selection probabilities alone are not enough to guarantee a simple random sample. A design can give each person the same chance while allowing only certain combinations of people. The equal-likelihood condition concerns entire samples.

Example: stratify to represent important groups

A school has 600 lower-grade and 400 upper-grade students. To sample 100 students proportionally by grade group, form two strata and independently select 60 from the lower grades and 40 from the upper grades by random sampling.

Stratification samples from every stratum. It can improve precision when individuals within each stratum are similar in the response of interest and the strata differ. The grouping variable should be related to the study's goal, not chosen merely because it is available.

If equal numbers are sampled from unequal strata, an overall estimate may need weighting. A simple unweighted average would otherwise overrepresent the smaller stratum. Proportional allocation avoids that particular mismatch when response rates are comparable.

Cluster sampling selects groups, not individuals from every group

In cluster sampling, randomly select some clusters and survey all individuals in those selected clusters. For example, randomly choose 8 classrooms and survey everyone in those rooms. This can reduce travel and administrative cost.

Good clusters resemble small versions of the population. If students within a classroom are very similar, sampling whole classrooms can provide less information than the same number of independently selected students. Stratification and clustering have opposite selection patterns: every stratum contributes a sample, whereas only selected clusters contribute members.

Example: systematic sampling needs a random start

To sample about 100 entries from a list of 2,000, choose a random starting position from 1 through 20, then select every twentieth entry. This is systematic sampling.

A repeating pattern in the list can cause trouble. If every twentieth entry corresponds to the same shift or household role, the sample may miss important variation. A random start is essential, but it does not eliminate all risks from periodic ordering.

Distinguish major sources of bias

Convenience sampling chooses easy-to-reach individuals. Voluntary response lets participants choose themselves, often attracting people with unusually strong opinions. Nonresponse occurs when selected individuals do not participate; bias arises if their responses would differ systematically from those who answer.

Response bias concerns inaccurate answers caused by wording, interviewer influence, memory, social pressure, or measurement problems. A perfectly random selection cannot fix a leading question such as “Do you agree that the excellent new policy should continue?”

Increasing sample size reduces random sampling variability but does not remove systematic bias. A huge voluntary online poll can be less trustworthy than a much smaller well-designed random sample.

Try it yourself

A district randomly chooses 5 schools and surveys every student in them. Another district randomly surveys 20 students from every school. Identify the methods and explain the difference.

Answer: The first is cluster sampling: only selected schools contribute, and all their students are surveyed. The second is stratified sampling with schools as strata: every school contributes a random sample. The second design may require weighting if school sizes differ and the goal is a districtwide student proportion.

What to remember

Describe exactly how units are selected. Distinguish strata from clusters, identify frame and response limitations, and remember that a larger sample cannot cure a biased selection or measurement process.