Stratified Sampling: What Is Stratified Sampling?

The stratified sampling technique, also known as stratified random sampling, is a data collection method that breaks a larger population into different strata (subgroups). The number of strata and the sample size of each stratum depends on the total number of respondents in a study. One way to use this probability sampling method is to break the entire population of a study into specific demographics. For instance, researchers can stratify a population by breaking it up by gender or age into non-overlapping groups.

Stratified sampling has multiple uses.

• When the total sample size is unwieldy: Stratified sampling breaks large populations into manageable subsets for more efficient data analysis.
• To offset disproportionate sampling: Sometimes certain demographic groups are overrepresented or underrepresented in simple random sampling. By breaking groups into smaller strata, researchers can get representative sample units for each target population. Alternatively, some stratified samples intentionally use disproportionate sampling to capture information about groups that might be too small to show up in a random sample.
• For targeted sampling: Unlike simple random sampling, which assesses a broad population with minimal selection criteria, stratified sampling begins by identifying target populations. Once it stratifies respondents, it then brings in random sampling methods to get more information about each sampling fraction.

Both stratified sampling and cluster sampling break large sample groups into smaller ones. Consider the differences between these two types of sampling methods.

• Homogeneity vs. heterogeneity: The stratified sampling method creates homogeneous strata where all members share a common trait. The cluster sampling method creates clusters of equal size, but these groups are heterogeneous and not selected for a common trait.
• Accuracy vs. cost: Stratified sampling creates precision by limiting the size of the stratum and by ensuring the demographic similarity of people within that stratum. Cluster sampling is popular for market research, particularly when researchers must cover multiple geographical areas. Some cluster sampling goes into great depth. Single-stage cluster sampling only takes one sample of a population, but two-stage cluster sampling and multi-stage cluster sampling go even further. Yet all of these methods can be more cost-effective than stratified sampling or random probability sampling of large populations.

Data scientists and survey professionals see several advantages in stratified sampling. The largest of these advantages is that it can create survey samples that have the same demographic proportions as those of the general population. Stratification leads to a second advantage in that it is less likely to produce over-sampling or under-sampling errors that can be found in truly random sampling.

The primary drawback to stratified sampling is that it can be hard to have all the right conditions in place to take it on. The sampled population must be divided into non-overlapping strata, and this can sometimes be difficult. For instance, if one stratum is English speakers and another is French speakers, someone who speaks both languages couldn't be in either stratum. It can also be difficult to get the right size stratum to do scientifically sound statistical analysis.

Get the MasterClass Annual Membership for exclusive access to video lessons taught by science luminaries, including Terence Tao, Bill Nye, Neil deGrasse Tyson, Chris Hadfield, Jane Goodall, and more.