Sampling methods in research: types and examples

Most studies cannot ask everyone. Mohit could not visit every farming household in Hisar district, so he would ask some and use their answers to speak for all. Choosing those few is called sampling. A short review by Elfil and Negida in Archives of Academic Emergency Medicine (2017) calls the sampling method "the process of selecting a sample population from the target population".

A sample is useful only because it stands in for the population. If it does not look like the population, no analysis later can repair it. Elfil and Negida put it plainly: "The main methodological issue that influences the generalizability of clinical research findings is the sampling method." The same holds outside medicine.

Every sampling method belongs to one of two families. What separates them is who decides who gets picked: chance, or the researcher.

Elfil and Negida note that samples "selected using probability sampling methods are more representatives of the target population". Etikan, Musa and Alkassim, in the American Journal of Theoretical and Applied Statistics, explain why non-probability sampling is still common: it helps when randomisation is impossible or when time, money and workforce are short. The cost is that such a sample may not represent the population well.

Neither family is wrong. A study that wants to say what share of farmers in the district uses drip needs probability sampling. A study that wants to understand why needs people chosen for what they can tell you.

There are four main types. Each needs some kind of list, and each uses chance at the moment of choosing.

Each has a catch. Systematic sampling goes wrong if the list has a pattern that matches your interval: if every tenth house on a list is a corner house with a bigger plot, an interval of ten keeps picking the same kind of house. Stratified sampling needs each person's group known in advance. Cluster sampling saves travel, but often needs a larger sample for the same accuracy, because people in one village tend to be alike.

Large surveys usually combine these in stages, called multistage sampling: villages or wards first, then households inside the chosen ones.

Mohit's plan became a multistage one. He listed every village in the district and drew 10 by lottery. He split each chosen village's household list into marginal, small and larger farmers, and drew households at random from each group. "Now every farming household in the district had a chance," Dr. Malik said, "including the ones who never come to a fair."

Here the researcher, or the participants themselves, decide who ends up in the sample. That is fine as long as you say so openly.

Elfil and Negida call convenience sampling "the most applicable and widely used method in clinical research". For snowball sampling they describe a survey of street children, in which one child leads the researcher to others. That is its strength: it reaches groups no list would ever show you.

In qualitative research, purposive sampling is usually the right tool, not a compromise. For the second part of his study, interviews on why farmers give up drip, Mohit did not want any farmer at all. He wanted farmers who had installed drip and then gone back to flooding their fields. He found the first two through his survey, and each of them named others. That is purposive sampling with a little snowball, and he wrote it exactly that way.

Qualitative studies also decide their size differently. Instead of a formula, many keep interviewing until saturation, which Guest, Bunce and Johnson, in a 2006 paper in Field Methods, describe as the point at which no new information or themes are observed in the data.

Start from your research question, not from a senior's thesis. If your question is still loose, sharpen it first with our guide on how to write a good research question. Then ask these, in order.

1. Do you need to generalise?If your result must speak for a whole population, such as the share of farmers in a district who use drip, you need probability sampling. If you want depth and reasons, purposive sampling suits better.
2. Is there a list?Simple random and systematic sampling need a list of the whole population. If none exists, cluster sampling lets you start from a list of places, such as villages, wards or schools, and make lists of people only inside the chosen ones.
3. Are there groups you must not miss?If small groups matter, such as women farmers or first-generation students, stratify, so that each group is sampled on its own.
4. What can you afford?A smaller probability sample done well beats a large convenience sample.
5. Can you defend it?Your examiner will ask why. Write one sentence that ties the method to your question.

Mohit's answers were clear. He needed a percentage for the district, so the survey had to be a probability sample. He needed reasons, so the interviews had to be purposive. One study, two methods, each matched to its half of the question.

Two different things can make a sample's answer differ from the truth. Sampling error is chance: even a perfectly random sample will not match the population exactly, and the gap tends to shrink as the sample grows. Sampling bias is a fault in how the sample was chosen, and it pushes the answer in one direction. A bigger sample does not cure bias. It only measures the wrong group more precisely.

The famous example comes from 1936. The American magazine The Literary Digest polled about ten million people before the US presidential election, and 2.38 million replied. It predicted that Alf Landon would beat Franklin Roosevelt with about 57 percent of the vote. Roosevelt won with 60.8 percent and carried every state except Maine and Vermont. Its names came from its own readers and from lists of car owners and telephone users, groups richer than average. Later research found the bigger problem was who chose to reply: people who strongly disliked Roosevelt were more likely to answer.

A big sample from the wrong list is still the wrong list.

So how many people do you need? It depends on how precise your answer must be, how much people vary, and the analysis you plan. Surveys that estimate a percentage use a standard formula; studies that compare groups use a power calculation. Name and cite the method you used, and for clinical work, ask a statistician.

Then plan for people who will not take part. Mohit drew a few extra names in each group and recorded how many said no. A low response rate is a warning sign, because those who refuse may differ from those who agree.

The sampling paragraph of your methodology should let a reader repeat exactly what you did. Our guide on how to write a research methodology shows where it sits. Include:

• the population and the sampling frame you drew from
• the sampling method, named correctly, and why it suits your question
• the sample size and how you decided it
• how the selection was done in practice: the lottery, the random number tool or the interval
• your inclusion and exclusion criteria
• how many people were approached, how many took part, and why the others did not

Name the method honestly. Stopping people "at random" in a crowd is convenience sampling, not random sampling, and reviewers notice. If you are still at the proposal stage, write the same paragraph in the future tense, as our guide on how to write a research proposal explains.

Mohit's final sampling paragraph was six sentences long. Dr. Malik read it and asked only one thing: "And the farmers' fair?" He smiled. He would still go next year, but to share his results, not to collect them.