Exploring Different Types of Sampling Methods

Imagine trying to study the eating habits of every adult in India. With over a billion people, surveying each one is impossible. Researchers solve this through sampling – selecting a manageable subset that mirrors the larger population. But not all samples are created equal. The method used to choose participants shapes whether the findings can be generalized to millions or remain limited to a specific group. Understanding the types of sampling methods is therefore central to designing credible research in population and family health studies.

The two broad families: Probability and non-probability sampling 🔗

All sampling methods fall into one of two categories. The defining feature that separates them is whether or not random selection is used. In probability sampling, every member of the population has a known, non-zero chance of being selected, which allows researchers to make strong statistical inferences about the whole group. In non-probability sampling, participants are chosen based on convenience, judgment, or other non-random criteria, meaning some individuals have no chance at all of being included.

The trade-off between the two is essentially one of rigor versus practicality. Probability sampling supports generalizability and is the gold standard when researchers want to estimate disease prevalence, fertility rates, or vaccination coverage across an entire state or country. Non-probability sampling, on the other hand, is faster and cheaper, and is often the only feasible option when the population is hidden, hard to reach, or when the study is exploratory. According to a review on sampling in epidemiological research, the cost of this convenience is that the effect of sampling error cannot be estimated, and the results are usually not generalizable.

Probability sampling methods 🔗

Probability methods rely on a complete or near-complete list of the population, called a sampling frame. From this frame, individuals are picked using a random process. There are five main techniques used in large-scale health research.

Simple random sampling 🔗

This is the most basic form. Every individual in the sampling frame has an equal and independent chance of being selected, usually through a lottery draw or a random number generator. For example, if a district health officer wants to study the immunization status of 1,000 children out of 50,000 registered births, they could assign each child a number and let a computer pick 1,000 at random. The method is straightforward and unbiased, but it requires a complete list of the population, which is rarely available for large, scattered groups.

Systematic sampling 🔗

Here, the researcher picks every k^th unit from the sampling frame after a random starting point. If a hospital wants to audit patient files and has 10,000 records, it might select every 100^th file starting from a randomly chosen number between 1 and 100. Systematic sampling is easier to execute than simple random sampling, but it can introduce bias if there is a hidden periodic pattern in the list – for instance, if every 10^th ward in a hospital register happens to be a maternity ward.

Stratified sampling 🔗

Populations are rarely uniform. Stratified sampling divides the population into non-overlapping subgroups, or strata, based on a characteristic that matters to the study – such as age, sex, caste, or rural-urban residence – and then draws a random sample from each stratum. This ensures that even small subgroups, like tribal communities or women above 60, are adequately represented. A national nutrition survey, for example, would stratify by state and by urban-rural residence so that no region is underrepresented in the final estimates.

Cluster sampling 🔗

When a population is spread across a huge geographic area, listing every individual is impractical. Cluster sampling involves dividing the population into groups, or clusters, and then randomly selecting entire clusters to study. Villages, schools, urban blocks, or hospitals are common clusters. If a researcher wants to study malnutrition in children under five across a state, instead of listing every child, they might randomly select 30 villages and survey every eligible child in those villages. This dramatically reduces travel time and cost, though it usually requires a larger sample size to achieve the same precision as simple random sampling.

Probability proportional to size (PPS) sampling 🔗

PPS is a refinement used when clusters vary widely in size. Instead of giving every cluster an equal chance of selection, larger clusters have a higher probability of being chosen, proportional to their population. This ensures that each individual in the population still has roughly the same chance of being included in the final sample. The National Family Health Survey (NFHS-5) used a two-stage design where villages were selected as Primary Sampling Units with probability proportional to size, followed by a random selection of 22 households in each PSU. Census Enumeration Blocks served the same role in urban areas. This is also the method recommended by the Government of India’s sampling guidelines, which note that PPS ensures a constant selection probability at the final stage since each cluster has a fixed number of units sampled.

Non-probability sampling methods 🔗

When random selection is impossible, too expensive, or simply unnecessary, researchers turn to non-probability methods. These are widely used in market surveys, qualitative studies, pilot research, and studies of hidden populations.

Convenience sampling 🔗

Also called accidental or opportunistic sampling, this method selects participants who are easiest to reach. A student conducting a quick survey on smartphone usage among classmates, or a clinic interviewing whoever walks in on a given day, is using convenience sampling. It is cheap and fast, which is why it dominates undergraduate projects and pilot studies. The downside is severe: the sample may have nothing in common with the broader population, so the results carry limited weight.

Judgment sampling 🔗

Also known as purposive sampling, this approach lets the researcher use their expertise to pick participants who are believed to be most representative or informative. A public health researcher studying tuberculosis treatment adherence might deliberately select districts known for high TB burden and good record-keeping. According to the review on epidemiological sampling, judgment sampling is essentially an extension of convenience sampling, where the researcher picks one town and one rural area thought to be typical of the country. The method’s value depends entirely on the researcher’s skill – a poor judgment leads to a poor sample.

Quota sampling 🔗

Quota sampling is the non-probability equivalent of stratified sampling. The population is divided into subgroups, and the researcher decides in advance how many participants are needed from each. A market researcher studying soap preferences in a city might be told to interview 50 women aged 18-30, 50 women aged 31-50, and 50 women above 50. Once a quota is full, the interviewer stops recruiting from that group. Quota sampling is often preferred over other non-probability methods because it forces the inclusion of members from different subpopulations, though the selection within each group is typically left to the interviewer’s discretion. This makes it popular in consumer research, opinion polls, and political surveys.

Snowball sampling 🔗

Some populations are nearly invisible – sex workers, undocumented migrants, people living with HIV, drug users. There is no list, no easy way to find them, and they are often reluctant to come forward. Snowball sampling solves this by starting with one or two known members of the group, who then refer the researcher to others in their network, who in turn refer more. The sample grows like a snowball rolling down a hill. The method has been particularly useful in surveys of risk behaviours among intravenous drug users, where participants nominate others to be interviewed. However, because referrals come from social networks, the sample tends to over-represent people with shared characteristics, introducing selection bias.

Choosing the right method 🔗

No single sampling method is best for every study. The choice depends on the research question, the population, the available resources, and the level of generalizability needed. Large-scale demographic surveys like the NFHS use a combination of stratification, clustering, and PPS because they must produce reliable estimates for hundreds of districts. A start-up testing a new mobile health app on a few hundred users might rely on convenience or quota sampling and accept the limits on generalizability. A qualitative researcher exploring the lived experience of caregivers of dementia patients in Kolkata might begin with purposive sampling and expand through snowballing.

The Government’s guidelines also point out that sample size depends on factors such as margin of error, variability, confidence level, and population proportion – not just on the size of the population itself. A good sampling plan balances scientific rigor with practical constraints, and it is always disclosed transparently so that readers can judge the strength of the findings.

What do you think? If you were tasked with measuring contraceptive use among newly married women in a remote tribal district, which combination of sampling methods would you choose, and why? And in an era of digital surveys and online panels, do you think traditional probability sampling is becoming harder to defend on cost grounds, or more important than ever?

References

1. https://www.scribbr.com/methodology/sampling-methods/
2. https://pmc.ncbi.nlm.nih.gov/articles/PMC4817645/
3. https://www.healthknowledge.org.uk/public-health-textbook/research-methods/1a-epidemiology/methods-of-sampling-population
4. https://www.nfhsiips.in/nfhsuser/nfhs5.php
5. https://dmeo.gov.in/sites/default/files/2022-06/Sampling_Guidelines_21062022.pdf
6. https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch13/nonprob/5214898-eng.htm