What Are Sampling Methods?
Sampling methods are the processes by which you draw a sample from a population. When performing research, you’re typically interested in the results for an entire population. Unfortunately, they are almost always too large to study fully. Consequently, researchers use samples to draw conclusions about a population—the process of making statistical inferences.
A population is the complete set of individuals that you’re studying. A sample is the subset of the population that you actually measure, test, or evaluate and base your results. Sampling methods are how you obtain your sample.
Before beginning your study, carefully define the population because your results apply to the target population. You can define your population as narrowly as necessary to meet the needs of your study—for example, adult Swedish women who are otherwise healthy but have osteoporosis. Then choose your sampling method.
Learn more about populations and samples, inferential vs. descriptive statistics and populations and parameters.
In research and inferential statistics, sampling methods are a vital issue. How you draw your sample affects how much you can trust the results! If your sample doesn’t reflect the population, your results might not be valid. It’s a crucial part of experimental design.
In this post, learn more about sampling methods, which ones produce representative samples, and the pros and cons of each procedure.
Probability vs Non-Probability Sampling Methods
Sampling methods have the following two broad categories:
- Probability sampling: Entails random selection and typically, but not always, requires a list of the entire population.
- Non-probability sampling: Does not use random selection but some other process, such as convenience. Usually does not sample from the whole population.
Probability sampling is typically more difficult and costly to implement, but, in exchange, these processes tend to increase validity by producing representative samples. In short, you can make valid conclusions about the population. A statistical inference is when you use a sample to learn about a population. Learn more about Making Statistical Inferences.
On the other hand, non-probability sampling methods are often easier and less expensive, but the trade-off is that the validity of your conclusions is questionable. You might not be able to trust the results. Sampling bias is more likely to occur.
Learn more about Validity in Research and Psychology: Types & Examples and Internal and External Validity.
Probability Sampling Methods
Given the benefits of using representative samples, you’ll typically want to use a probability sampling method whenever possible. Let’s go over the standard methods. They each have pros and cons. Click the links to learn more about each sampling method and see examples. Learn more about representative samples.
To use a probability method, you’ll first need to develop a sampling frame, which lists all members of your target population. Then you can use one of the following methods.
Learn more about Sampling Frames: Definition, Examples & Uses.
Simple Random Sampling (SRS)
In simple random sampling (SRS), researchers take a complete list of the population and randomly select participants from it. All population members have an equal likelihood of being selected. Out of all sampling methods, statisticians consider this one to be the gold standard for producing representative samples. It’s entirely random, leaving little room for accidentally biasing the results.
However, this sampling method has some drawbacks.
First and foremost, this method can be pretty unwieldy and require abundant resources. For one thing, it requires a list of all population members, which can be a tremendous hurdle by itself. Attempting to perform SRS with an incomplete population list causes undercoverage bias and a nonrepresentative sample.
Furthermore, while random selection is beneficial, it also ensures that the subjects are maximally dispersed, making them harder to contact.
SRS can exclude smaller but crucial subpopulations purely by chance. Additionally, this approach produces less precise estimates for subgroups and the differences between subgroups than some other probability sampling methods.
Learn more about Simple Random Sampling and Undercoverage Bias: Definition & Examples
Systematic Sampling
Systematic sampling is similar to SRS but attempts to ease some of the difficulties for researchers. There are several versions of this method.
One form uses a complete list of the population. The researchers randomly select the first subject and then move down the list choosing every Xth subject rather than using a randomized technique.
The other form does not use a complete list of the population. This sampling method is suitable for populations that are tough to document, such as the homeless, because a comprehensive list won’t exist. The essential requirement for this sampling method is knowing how to locate them. While it’s not perfect, it’s a feasible option when you can’t obtain the full list.
Suppose you want to survey theater patrons but lack a complete list. Instead, you can use systematic sampling and recruit every 20th person who exits the theater. This approach works because they leave randomly.
This sampling method has some disadvantages. The form that uses a complete list of the population can closely mirror the results of simple random sampling. However, the non-randomness increases the potential for manipulation, even if accidentally. Additionally, patterns in the list can unintentionally create a non-representative sample.
The form that doesn’t use a list has more potential problems. Namely, it increases the potential for missing subgroups and acquiring a non-representative sample. This sampling method increases the knowledge you must have about the population and their habits. Without that knowledge, you won’t be able to find subjects that reflect the whole population.
Learn more about Systematic Sampling.
Stratified Sampling
In stratified sampling, researchers divide a population into similar subpopulations (strata). Then they randomly sample from the strata.
This sampling method can guarantee the presence of small but vital subpopulations in the sample. Relative to SRS, this method can increase the precision of subgroup estimates and the differences between subgroups. In short, it helps researchers gain a better understanding of the subgroups. Dividing the whole population into smaller, more similar subsets can also reduce costs and simplify data collection.
The drawbacks are that this sampling method requires additional upfront knowledge and planning. The researchers must know enough about the subgroups to devise an effective strata scheme. Then they must have sufficient information about all population members to assign them to the correct strata.
Learn more about Stratified Sampling.
Cluster Sampling
Like stratified sampling, the cluster sampling method divides the whole population into smaller groups. However, unlike strata, each cluster mirrors the full diversity present in the population. Then the researchers randomly sample from some of these clusters.
The primary benefit of this sampling method is that it reduces the costs of studying large, geographically dispersed populations. Using this method, researchers don’t need to sample the entire geographic region but only certain areas because they know individual clusters are similar to the population. Additionally, they don’t need to develop a list of potential subjects for clusters from which they’re not sampling. These considerations can significantly reduce planning, administrative, and travel costs.
When researchers can’t create a list of the entire population, cluster sampling can be an excellent choice.
On the downside, cluster sampling increases the design complexity. Researchers must understand how well each cluster approximates the whole population. If the clusters don’t fully represent the population, results can be biased. In real-world studies, clusters tend to be naturally occurring groups that don’t mirror the population, which reduces the ability to draw valid conclusions.
Learn more about Cluster Sampling.
Non-Probability Sampling Methods
Non-probability sampling methods don’t use random selection, and they typically don’t use a complete population list. While these methods are simpler and less expensive, your results are more likely to be biased, reducing your ability to make sound conclusions.
Researchers often use non-probability sampling methods for exploratory research, pilot studies, and qualitative research. These sampling methods provide quick and rough assessments, help work kinks out of measurement instruments and procedures, and help refine the design for a more rigorous study in the future.
Below are several standard non-probability sampling methods:
- Convenience sampling: The main criteria for recruiting subjects are those who are easy to contact and willing to participate. There are no inclusion requirements. Online polls are a type of convenience sampling. Learn more about Convenience Sampling.
- Quota Sampling: Non-random selection of subjects from population subgroups that the researchers define. Learn more about Quota Sampling.
- Purposive sampling: Investigators use subject-area knowledge to handpick a sample they think will help their study. Learn more about Purposive Sampling.
- Snowball sampling: Researchers use subjects to find and recruit other subjects. This method is helpful when a population is hard to contact. When recruits help you find more recruits, and those help find even more, and so on, the total number snowballs. Learn more about Snowball Sampling.
As you can see, there are many sampling methods. Each one has benefits and disadvantages. When designing a study, evaluate the nature of your target population, your research goals, and the available time and resources to choose your sampling method. After deciding between the sampling methods, calculate your sample size using a power analysis.
Reference
Sampling in Developmental Science: Situations, Shortcomings, Solutions, and Standards (nih.gov)
Michael Green says
Hello Mr. Frost,
I would like to know whether people with mild Parkinson’s Disease symptoms are less likely to have kidney stones. Do PwP (People with Parkinson’s) have significantly less incidences of kidney stones than in the general population (~ 10%). So far, I have asked 12 people I know who has been diagnosed with Parkinson’s and 0% had kidney stones. I would like to increase my sampling size by randomly sampling members of a forum for PwP I belong to. Should I get a list of all forum subscribers and randomly select around 40 forum members to pose the question, “If you have been officially diagnosed with Parkinson’s, have also had a kidney stone?”. What would you suggest? I had posed the question in the forum before, but only PwP folks that had a Kidney stone responded.
Thanks,
Mike
Khursheed Ahmad says
I think stratified sampling will work __ mke two groups as stratas _ then use SRS to obtain a complete sample .
Marilen says
hi.what sampling technique will i use if my respondents are 1st yr college students awardees vs non awardees of different courses?