What is stratified sampling? Definition and examples
Stratified sampling can improve the accuracy and representation of research, statistical analysis and decision-making. Learn how and why to use stratified sampling in your research.
Stratified sampling is a type of sampling design that randomly collects samples from distinct subgroups based on a shared characteristic. These samples represent a population in a study or a survey.
Let’s explore the basics of stratified sampling, how and when to collect a stratified sample and how this sampling method compares to others.
What is stratified sampling?
Stratified sampling is a type of probability sampling. Researchers and analysts use stratified sampling to minimise bias and ensure that they can make valid inferences about their target population from the sample data. With this sampling method, every individual in the given population has the same chance of being selected.
Stratified sampling divides its population into subgroups called strata. Samples are then drawn from each singular subgroup (or stratum) using another probability sampling method.
For example, imagine you wanted to assess student performance at a university where the student body comprises 60% males and 40% females. You could use stratified sampling to represent each subgroup in your study proportionally. For accurate representation, you would start by ensuring that 60% of your sample population is male and the remaining 40% is female. You could then use another sampling method to select samples from each subgroup.
What is stratified random sampling?
Although the terms ‘stratified sampling’ and ‘stratified random sampling’ are often used interchangeably, there’s a subtle difference between the two. The main distinction lies in how samples are collected from each subgroup.
Stratified sampling: To collect a stratified sample, divide your population into strata and then use a separate sampling method to select participants from each stratum.
Stratified random sampling ensures that each individual within your strata has an equal chance of being selected. To collect a stratified random sample, you would randomly select individuals from each stratum.
For example, a nurse splits a group of patients into strata based on their injuries: arm, leg or head.
To pull a stratified sample, she might select the first three names in each subgroup or the last three names on her list. For a truly random stratified sample, she could pull participants from each stratum out of a hat or by rolling dice.
What is the difference between stratified and cluster sampling?
Cluster sampling is a type of sampling design where samples are selected from random clusters within a larger group. This method simplifies the sampling process while maintaining accuracy.
For example, a company may want to conduct a survey to gain a better understanding of its employees’ preferences and needs. To collect a cluster sample, the company would divide its workforce into clusters based on specific characteristics (age, gender, location, etc.) and randomly select individuals from each cluster until it obtained its desired sample size.
For a stratified sample, the company would organise its workforce into strata and collect samples from each stratum using a secondary sampling method of their choice. The company may select participants based on their department, length of service or location.
How to do stratified sampling
Sample each group in your population fairly and rationally by using our step-by-step approach:
Step 1: Determine your overall sample size
First, look at your overall population and determine the sample size that you will need based on your desired margin of error. Your margin of error helps you understand the extent to which your survey results may differ from those of your overall population.
Your confidence level can also influence your sample size. The confidence interval represents a statistical range where it’s likely that the true result lies. For example, a 95% confidence interval indicates that if you sampled the same population numerous times, your true result would lie within the interval in approximately 95% of the samples.
In stratified sampling, you need to work out how many samples to take from each stratum. A sample size calculator can help you determine how many samples to take from each stratum in order to properly represent each group.
Once you’ve calculated the overall sample size for the study, divide the sample among your subgroups. Select samples from each stratum (subgroup) until you obtain your desired sample.
Step 2: Divide the population into strata (subgroups)
Divide your overall sample into smaller subgroups based on common characteristics. Common characteristics could include:
- Age
- Gender
- Region
- Education
- Religion
- Race
- Socioeconomic class
Your single characteristic should differentiate participants and yield accurate results.
Step 3: Decide between proportional and disproportionate representation
Each stratum may not yield an equal population size, because each stratum represents a particular demographic or shares a specific characteristic. For example, if you categorise your population according to gender identity, 60% may be male, 30% may be female and 10% may be non-binary.
You’ll need to determine whether you should use proportionate or disproportionate stratified sampling.
- Proportionate sampling collects a sample size from each subgroup that reflects its population share. For a proportionate sample, you would need to ensure that your final sample consists of 60% men, 30% women and 10% non-binary participants.
- Disproportionate sampling doesn’t require the precise proportions of each subgroup within the sample; instead, it focuses on greater representation of underrepresented groups. Using this method, you might evenly divide your sample size among each gender identity, selecting the same number from each group. Alternatively, if it makes sense for your research, you might use a different proportion for various subgroups.
Step 4: Randomly pick from each group
To collect your survey sample, you’ll pick individuals randomly from each of your subgroups. How you select participants will be up to you: you might choose every fifth name from a list, only select participants born after a certain year or use a completely random method.
Step 5: Combine your strata
After selecting a sample of each of your subgroups, combine them to form your representative sample. You’ll use this sample in your research, statistical analysis, forecasting, market research or other work.
Before you start, review your survey methodology to double-check that you have what you need to gather the most useful, precise data.
When to use stratified sampling
Use a stratified sampling method to represent specific population subgroups adequately. Benefits include:
- Improved accuracy and precision: This method reduces variability within each group for more precise and reliable subgroup estimates.
- Balanced representation: Dividing the population into key subgroups helps to ensure adequate representation of each stratum and prevents over-representation of dominant groups or under-representation of minority groups.
- Minimised bias: Guaranteeing proportional representation reduces the risk of certain segments being ignored or over-represented.
- Enhanced comparisons: Stratified sampling facilitates detailed analysis of different subgroups by ensuring that you have data from each group.
- Greater efficiency: This method enables you to achieve the same level of accuracy as you might with simple random sampling but with a smaller sample size.
However, when using stratified sampling, consider these common roadblocks:
- Greater complexity: This method requires detailed advanced knowledge of the population’s characteristics and can be more time-consuming than other approaches (e.g. simple random sampling).
- Selection bias risk: If you have prior knowledge of your population’s characteristics, you may be more prone to selection bias when defining your strata.
- Classification difficulties: It can be challenging to distinctly classify every member of the population sample. For example, ethnicity may be complicated, and a subject might fall into multiple subgroups, which could cause overlap issues.
- Strata inaccuracies: When strata are improperly defined or there is overlap, you face a higher risk of errors in your results.
- Higher costs: It’s likely that a more complex and time-consuming method will require more resources than other sampling methods that you might use to obtain a representative sample (e.g. systematic sampling).
Stratified sampling use case examples
Researchers use stratified sampling when representation is uniquely important to the accuracy and reliability of their results. Use cases include:
- Marketing: Marketers who need an in-depth understanding of their target markets can use stratified sampling to compare target audience strata or focus on a specific stratum.
- Politics: Political pollsters who must accurately represent various demographics (e.g. age, race, region, religion) can use stratified sampling to gauge voter preferences.
- Health: Medical researchers conducting a study can use stratified sampling to understand the impact that a treatment has on different age groups, genders or those with different health conditions.
- Education: Educational institutions seeking to understand how socioeconomic background or previous educational experience affects retention can use stratified sampling.
Interpreting your stratified sampling survey results
Once you have conducted your study or survey, bear the following best practices for reliable and meaningful analytics in mind:
Ensure that your interpretation is accurate by doing the following:
- Run a t-test (especially between different strata) to check whether two numbers are significantly different from one another.
- If you sampled disproportionately, use a weighted analysis to match population distribution and avoid biased results.
- Check whether your results are statistically significant and consider the effect size to determine whether an observed effect or difference has real-world meaning.
- Interpret both within and across strata. Explore how different strata compare to one another. You might reveal unique insights for a particular subgroup by comparing results within a single stratum.
- Avoid generalisations about the whole population based only on one stratum.
- Apply multiple comparison corrections if testing across several strata to maintain the integrity of your conclusions.
Better research insights with SurveyMonkey
Collecting representative samples using the stratified sample method can lead to more useful, accurate surveys and research data.
Learn more about how SurveyMonkey Audience can help you connect with your target audience and conduct market research that yields the insights you need.