Chapter 9 · Designing the study
Whom do you pick, and how?
Probability and non-probability sampling, from simple random to snowball, and the WHO cluster survey.
20 TNMGRMU questions answered hereset 22 times, 2003–2025
The problem first
You are a Community Medicine resident measuring anaemia among adolescent girls in your district. To save time, you examine every girl attending the hospital OPD for a month. Most of them are anaemic, and you report that figure as the district's prevalence.
The examiner asks one question.
“Would a girl who never visits a hospital be in your sample?”
Start here
Why not study everyone?
Studying everyone costs too much and takes too long, and some tests destroy what they measure. Sampling is the method of choosing which part of the population to study.
A sample describes the population only if it is chosen in a way that represents that population. This chapter covers which people to choose, and how. The sample size chapter covers how many.
| Feature | Census (complete enumeration) | Sample |
|---|---|---|
| Covers | Every unit in the population | A selected part of the population |
| Sampling error | None | Present, and calculable for a probability sample |
| Cost and time | High | Lower |
| Quality of each measurement | Many field staff, harder to train and supervise | Fewer staff, who can be better trained and supervised |
| Tests that destroy the item | Not possible | Possible, as in testing tablets from a drug batch |
| Small subgroups | Counted exactly | May contain too few people for a precise estimate |
| Used for | The national census, disease registers, a hospital's record of all admissions | Most surveys and research studies |
Before you pick
Whom are you trying to describe?
You want an answer about a big group, but you can only pick from a list you actually have.
Before you pick
What do you pick, what do you measure, and what do you estimate?
An anaemia survey may pick households, examine the girls living in them, and report one prevalence for the district. Each of these steps has its own term.
| Term | Meaning | Anaemia survey (teaching example) |
|---|---|---|
| Sampling unit | The unit that the selection procedure picks | A household |
| Unit of study (unit of inquiry) | The unit on which data are recorded | Each adolescent girl in a selected household. A household may have none, one or several |
| Parameter | A value for the whole population, usually unknown | Anaemia prevalence among all 60,000 girls in the district |
| Statistic (estimate) | The same value calculated from the sample | Prevalence among the 600 girls examined |
| Sampling fraction | n ÷ N | 600 ÷ 60,000 = 1% |
| Sampling weight | N ÷ n: the number of population units that each sampled unit stands for | 60,000 ÷ 600 = 100 girls for each girl examined |
An estimate is unbiased when its average over all possible samples equals the parameter. Any one sample still differs from the parameter by chance, which is the sampling error.
Precision depends mainly on the number sampled, not on the sampling fraction. For a true prevalence of 50%, a random sample of 400 people from a town of 10,000 (4%) has a standard error of 2.45 percentage points. A sample of 400 from a state of 1,000,000 (0.04%) has 2.50. A 10% sample of a population of 1,000, which is 100 people, has 4.75.
Two types of sampling
Did everyone have a known chance of being picked?
Yes: it is a probability sample, and you can calculate how far your answer might be from the truth. No: it is a non-probability sample, quicker, but you cannot.
Probability sampling
Four ways to give everyone a known chance
Same crowd of 40, four different ways of choosing. Red people are the sample.
The exact definition of simple random sampling: every possible sample of the chosen size has the same chance of being drawn. Units are normally drawn without replacement, so a unit once picked cannot be picked again.
Systematic sampling when N ÷ n is not a whole number. A register of 250 antenatal bookings and a sample of 40 gives 250 ÷ 40 = 6.25. With k = 6 and a random start from 1 to 6, the sample has 41 or 42 women instead of 40. In circular systematic sampling the random start is any number from 1 to 250, and every 6th entry is taken, continuing from the top of the list after entry 250, until 40 women are chosen. A start of 233 gives 233, 239, 245, then 1, 7, 13 and so on.
Systematic sampling without a list. Patients arriving at a clinic can be sampled as they come: estimate the number of eligible patients expected in the period from last month's register, calculate k, choose a random start, and enrol every k-th eligible patient.
Comparison
Stratified or cluster: both split people into groups. What is the difference?
What the groups are like, and whether you sample from every group or only some.
| Feature | Stratified | Cluster |
|---|---|---|
| Groups are | Strata: people alike inside, different between strata | Clusters: mixed inside, each a mini-population |
| You sample from | Every stratum | A random few clusters |
| Inside each group | A random sample of individuals | Everyone (or a sample, in multistage designs) |
| List needed | Individuals within each stratum | Only the clusters |
| Main gain | Better precision, and every subgroup is represented | Lower cost and travel |
| Main cost | Needs stratum information in advance | Less precise, and needs a design effect |
| Example | Urban and rural blocks sampled separately | Villages picked at random, all children surveyed |
Interactive
Draw repeated samples with each method and watch how far the estimates spread
Try this: press "Draw 20 more samples" a few times. Watch the cluster row spread much wider than the other three, then compare the standard errors in the table.
Teaching example (made-up numbers): 400 households in 20 localities of 20 households. In the 10 urban wards, 180 of 200 households have piped water. In the 10 rural villages, 40 of 200 do. The true value is 220 of 400, or 55%. Every sample takes 40 households.
20 samples drawn with each method
| Method | How the 40 households are chosen | Estimates so far | Standard error, all possible samples |
|---|---|---|---|
| Simple random | 40 of the 400 households by random numbers | 40% to 70% | 7.5 points |
| Systematic | Random start from 1 to 10 on the list, then every 10th household | 47.5% to 62.5% | 5.5 points |
| Stratified | 20 urban and 20 rural households, each at random | 50% to 62.5% | 5.3 points |
| Cluster | 2 of the 20 localities at random, all 40 households in them | 15% to 92.5% | 24.6 points |
The standard error is the standard deviation of the estimates from every possible sample drawn by that method. A smaller standard error means estimates closer to the true value.
- All four methods give estimates centred on the true 55%. None of them is biased.
- Stratified sampling takes 20 urban and 20 rural households every time, so chance cannot change the mix of urban and rural. Its standard error (5.3 points) is smaller than that of simple random sampling (7.5 points).
- The list for systematic sampling runs locality by locality, so every 10th household gives 2 households from each of the 20 localities. The sample is spread evenly across the list.
- Cluster sampling studies 2 whole localities. Households in the same locality resemble each other, so 40 households from 2 localities give fewer independent answers than 40 households chosen separately. A cluster sample here can give anything from 7.5% to 97.5%, and its standard error is 24.6 points. This is why a cluster survey needs a design effect in its sample size.
Stratified sampling
How many people from each stratum, and how are the strata combined?
The way the total sample is shared between strata is the allocation. When a stratum receives more than its share of the population, each person in it must count for less in the combined estimate.
Teaching example: a hypertension survey in a district with 30,000 urban and 70,000 rural adults, and a total sample of 600.
| Allocation | Rule | Urban, rural | SE of the district prevalence |
|---|---|---|---|
| Proportionate | n in each stratum in proportion to its size, so the sampling fraction is the same in every stratum | 180, 420 | 1.68 points |
| Equal | The same n in every stratum | 300, 300 | 1.75 points |
| Optimum (Neyman) | n in proportion to stratum size × stratum SD | 205, 395 | 1.67 points |
The survey used equal allocation. It found hypertension in 96 of 300 urban adults (32%) and 54 of 300 rural adults (18%).
Unweighted: 150 ÷ 600 = 25.0% (urban adults are over-represented)
Weighted prevalence = Σ Wh × ph = 0.30 × 32% + 0.70 × 18% = 22.2%
- W h
- share of the population in stratum h: 30,000 ÷ 100,000 = 0.30 urban, 0.70 rural
- p h
- prevalence found in stratum h
- Sampling weight
- N h ÷ n h: 30,000 ÷ 300 = 100 for each urban adult, 70,000 ÷ 300 = 233.3 for each rural adult
A proportionate sample is self-weighting: the ordinary sample percentage already equals the weighted estimate. Equal allocation suits a survey that needs a separate estimate of similar precision for each stratum. Optimum allocation gives more people to large and variable strata. Here the SDs are √(0.32 × 0.68) = 0.466 urban and √(0.18 × 0.82) = 0.384 rural, so the urban share is 30,000 × 0.466 ÷ (30,000 × 0.466 + 70,000 × 0.384) = 34.2% of 600, or 205.
Weighting each person by the inverse of their chance of selection is also the idea behind inverse probability weighting in observational studies, covered in Stratification and propensity scores.
Probability sampling
What if the area is huge?
Pick in stages. Each stage is a random pick inside the one before.
Probability sampling
How do you pick villages that differ in size?
Give each village a chance of selection in proportion to its size. This is probability proportional to size (PPS) sampling, used at the first stage of cluster surveys.
Teaching example: 8 villages with 10,000 households in all. Pick 3 villages, then 20 households in each.
- List the villages with the number of households in each, and add the numbers up cumulatively: A 1 to 1,200, B 1,201 to 1,650, and so on to H, which ends at 10,000.
- Sampling interval = total ÷ villages to pick = 10,000 ÷ 3 = 3,333.3.
- Choose a random start between 1 and 3,333, say 1,480. Add the interval twice: 1,480, 4,813 and 8,147.
- Each number picks the village whose cumulative range contains it: villages B, D and G.
- In each chosen village, select 20 households at random from the household list.
Chance for one household = (3 × H ÷ 10,000) × (20 ÷ H) = 60 ÷ 10,000 = 0.6%
- H
- number of households in that household's village
H cancels out, so every household in the district has the same chance and the sample is self-weighting. If the 3 villages were picked with equal chances instead, a household in a village of 300 would have a 2.5% chance and one in a village of 2,800 a 0.27% chance. A village with more households than the sampling interval is certain to be picked and can be picked twice.
The equal chance holds only if the household counts used for selection are correct. The 2018 WHO coverage survey manual warns that assuming a self-weighting sample is rarely valid, because sampling frames are often out of date, so weights are usually calculated from the actual counts found in the field.
Interactive
Pick villages with PPS or with equal chances, and watch each household's chance of selection
Try this: draw a few samples with PPS, then switch to “equal chance for every village” and compare the bottom chart. Watch the chance of a household in village H.
Teaching example (made-up): 8 villages with 10,000 households. Stage 1 picks 3 villages. Stage 2 takes 20 households in each chosen village.
The chances below do not change between draws
Random start 1,480, then add 3,333 twice: 1,480, 4,813, 8,147. These numbers fall in villages B, D, G.
Every household has the same chance, 3 × households ÷ 10,000 × 20 ÷ households = 0.60%. The sample is self-weighting.
Community Medicine
How do you check vaccination coverage in a district without a list of children?
There is no register of every child. But there is a list of villages and their populations. So select villages first, then find children by visiting households.
WHO has since changed its recommendation. Its 2018 Vaccination Coverage Cluster Surveys: Reference Manual (WHO/IVB/18.09) recommends probability-based sampling at every stage. Its reasons: with the next-nearest-house rule, households have unequal and unknown chances of selection, so 30 × 7 was not a true probability sample. Field workers could choose easy-to-reach houses. One fixed design was used whatever the survey's goals.
Non-probability sampling
When there is no list, or no time
Useful for pilot studies, qualitative research and hidden groups. The drawback: you cannot calculate sampling error, and the sample may not represent anyone else.
Convenience samples in hospital research take several forms. Volunteers are often healthier and more motivated than other people. Easily reached groups such as medical students or hospital staff differ from the general population in age, health and education. Referred patients at a tertiary centre are often more severe or more complicated than patients treated elsewhere. Online and telephone surveys reach only people who have access and who choose to reply.
Consecutive sampling comes closest to a random sample of clinic attenders when every eligible patient in the period is enrolled, and the period is not chosen for the kind of patients it brings. Enrolling only on the days one consultant runs the clinic, or only during an outbreak, selects a particular group of patients.
Interactive
Read each study and name its sampling method
Try this: read the study, pick the sampling method, and read the explanation. If your first answer is wrong, try another.
Study 1 of 11 (made-up examples)
From a list of 150 anganwadi centres in the block, 20 centres are selected at random, and every child registered at those 20 centres is examined.
What can go wrong
Is a bigger sample always more accurate?
It reduces only one kind of error.
Non-response is a non-sampling error. People who refuse or cannot be found often differ from those who take part, and enlarging the sample does not remove that difference. Ways of analysing data with missing answers are covered in Missing data.
Interactive
Change the sample size and the source of the sample, and watch where 200 estimates fall
Try this: move the sample size from 10 to 500 with a random sample, then switch to girls attending the OPD and do it again. Watch where the estimates fall and how far apart they are.
Teaching example (made-up numbers): 40% of adolescent girls in the district are anaemic. Among girls who attend the hospital OPD, 70% are anaemic. Each bar counts how many of 200 samples gave an estimate in that 2-point range.
Larger sample → estimates closer together
average 39.6% · standard error 6.9 points · true district value 40%
A random sample from the district gives estimates centred on the true 40%. With 50 girls per sample, the standard error is 6.9 percentage points. A larger sample makes the sampling error smaller.
Clinical trials
Do clinical trials use random sampling?
Usually not. They enrol eligible patients as they present, then use chance to decide who gets which treatment.
Decide
Which sampling method fits your study?
Answer a few questions. Each answer links back to its picture.
Do you need a number (prevalence, average) that applies to a bigger group?
Which method: all possible answers
- Stratified random sampling: Split the list into strata, then draw a random sample inside each, usually in proportion to its size.
- Systematic sampling: Interval k = N ÷ n, random start between 1 and k, then every k-th unit.
- Simple random sampling: Every unit has an equal chance, by lottery, random-number table or computer.
- Multistage sampling: A random pick at every level. Allow for a design effect in the sample size.
- Cluster sampling: Pick whole groups at random and study everyone in them. Multiply the sample size by the design effect.
- Consecutive sampling: Enrol every eligible patient in the period. Say clearly that results describe people who reach your clinic.
- Purposive (judgement) sampling: Choose information-rich participants deliberately. Common in qualitative research.
- Quota sampling: Fill fixed numbers per category. Quick, but who fills each quota is not random.
- Snowball sampling: Each participant refers the next. It reaches hidden groups but cannot give a prevalence.
By specialty
Sampling problems by specialty
Immunisation coverage or anaemia prevalence across a district, with no list of every child or adolescent.
Cluster or multistage sampling, with the design effect built into the sample size.
A study of patients attending the diabetes OPD.
Consecutive sampling is fine, but the results describe clinic attenders, not the community.
Outcomes of a new technique, with the surgeon choosing which patients get it.
Picking "suitable" cases is selection bias. Enrol all eligible patients consecutively.
Screening schoolchildren for refractive error across a city.
Select schools as clusters, then children within them. Get parental consent.
An antenatal clinic register with every booking listed in date order.
Systematic sampling works well, provided the register has no repeating pattern.
Understanding people who use drugs and avoid services.
Snowball sampling reaches them, but it cannot tell you how common drug use is.
Auditing reports from thousands of specimens in the lab register.
Simple random or systematic sampling from the specimen log.
Interviews on why patients default from treatment.
Purposive sampling of people with the experience, continued until interviews stop adding new themes.
Common myths
Common mistaken beliefs
Myth: "Random" means picking whoever, casually.
Random needs a chance mechanism: lottery, random-number table or computer.
Myth: A big enough sample fixes a bad selection.
Bias does not decrease as n increases. In 1936 a magazine poll with about 2.4 million replies predicted the wrong winner of the US election. Gallup predicted the right winner with about 50,000.
Myth: My OPD patients represent the community.
They represent people who reach your OPD.
Myth: Randomised trials use random sampling.
They use random allocation. Recruitment is usually consecutive.
Myth: Stratified and cluster sampling are the same: both split people into groups.
Stratified samples from every group. Cluster picks a few groups and takes them whole.
Myth: Systematic sampling is always as good as random.
Not if the list repeats in a pattern that lines up with the interval.
Why this matters to you
Five consequences of choosing the sample badly
A wrong prevalence
Your estimate is wrong, and you cannot say by how much.
Selection bias noted by the examiner
Results from a convenience sample cannot be generalised, and the examiner will say so.
A common viva error
Calling consecutive or convenience sampling "random" in your protocol.
Misreading large polls
A large self-selected online survey represents only the people who chose to answer.
An underpowered survey
A cluster design without a design effect ends up with too few independent answers.
History
History of sampling methods
- 1895
Anders Kiær, Norway
Presented the "representative method" at the International Statistical Institute session in Berne: study a part to describe the whole.
- 1931
P.C. Mahalanobis, India
Founded the Indian Statistical Institute in Calcutta. He developed interpenetrating samples for large surveys, and India's National Sample Survey was established in 1950.
- 1934
Jerzy Neyman
Compared purposive selection with random sampling and concluded that stratified random sampling is the method to advise for general use. He also showed the best way to divide a sample across strata.
- 1936
Literary Digest vs Gallup
About 2.4 million mailed-back ballots predicted Landon would win the US election. Roosevelt won. Gallup predicted it correctly from about 50,000 responses. The larger sample did not correct the bias.
- 1948
Dewey vs Truman
Polls predicted Dewey, but Truman won. The failure was partly blamed on quota sampling, and pollsters moved to probability sampling.
- 1961
Leo Goodman
Published the paper "Snowball sampling" in the Annals of Mathematical Statistics.
- 1982
Henderson and Sundaresan, WHO
Described the 30 × 7 cluster survey for immunisation coverage.
- 2018
World Health Organization
Its vaccination coverage survey manual recommended probability-based sampling at every stage, replacing 30 × 7.
Guidelines
Sampling in reporting guidelines
Observational studies must “give the eligibility criteria, and the sources and methods of selection of participants”.
Trials must report the settings and locations where the trial was conducted (item 11) and the eligibility criteria for participants (item 12a). These were items 4b and 4a in CONSORT 2010.
Vaccination coverage cluster surveys should use probability-based sampling at each stage, with households selected centrally rather than by interviewers in the field.
Reading a paper
Five checks on any study's sample
- Who was the target population, and who could actually be reached?
- What was the sampling frame?
- Which method was used, and was it truly random?
- What was the response or participation rate?
- How did the people who did not take part differ from those who did?
Why is it called that?
Word surgery
“Random”
Haphazard, careless, "anything goes".
A deliberate procedure in which chance, not the researcher, decides who is picked, and each unit has a known chance.
“Population”
The people living in a place.
The whole set of units you want to describe: patients, hospitals, blood samples or prescriptions.
“Error”
A mistake.
Sampling error is the chance difference between a sample estimate and the population value. It is present even when the sampling is done correctly.
Stratified: sample from every group. Cluster: choose some groups and study everyone in them.
Sample
Latin exemplum: "that which is taken out"
The part taken out to stand for the whole.
Population
Latin populus: "a people"
In statistics, the whole group you want to describe, even if it is hospitals or blood samples, not people.
Stratum
Latin stratum: "a thing spread out", a layer
Plural strata: subgroups of the population that differ from each other and are sampled separately.
Cluster
Old English clyster: "things growing naturally together"
A naturally occurring group: a village, a school, a ward.
Systematic
Greek systema: "an organised whole", from syn- (together) + "to stand"
Picking by a fixed rule, every k-th unit, instead of a fresh random draw each time.
Quota
Latin quota pars: "how large a part"
A fixed number for each category, filled until that number is reached.
Exam answers
Model answer outlines
Probability sampling techniques (list and describe)
Ph.D. RM 2014, 2015, 2016, 2022, 2025 · MD Community Medicine 2004, 2009, 2013, 2022Definition: every unit in the population has a known, non-zero chance of selection. It needs a sampling frame, and it allows sampling error to be calculated and results generalised.
Basic terms: population, sampling frame, sampling unit and unit of study, parameter (population value) and statistic (sample value), sampling fraction n/N, sampling weight N/n.
- Simple random: equal chance for each unit, using a lottery, random-number tables or computer-generated numbers. Strictly, every possible sample of size n has the same chance.
- Systematic: sampling interval k = N/n, a random start between 1 and k, then every k-th unit. If N/n is not a whole number, use circular systematic sampling.
- Stratified random: population divided into homogeneous strata (age, sex, urban/rural), then a random sample from each, with proportional or disproportional allocation. The allocation can be proportionate, equal or optimum (Neyman). Weight the strata when it is not proportionate.
- Cluster: natural groups selected at random, and all units in the chosen clusters studied. Cheaper, but less precise (design effect).
- Multistage: random selection in successive stages (district → village → household).
- Multiphase: part of the information from the whole sample, details from a sub-sample.
- Probability proportional to size (PPS): larger clusters get a larger chance of selection. With a fixed number of households per chosen cluster, every household has the same overall chance (self-weighting).
Non-probability sampling techniques, including snowball and judgement sampling
Ph.D. RM 2014, 2023, 2024 · MD Community Medicine 2018Definition: selection does not give every unit a known chance. It relies on availability or the researcher's choice.
- Convenience: readily available subjects. Quick and cheap, but high selection bias.
- Consecutive: all eligible subjects over a set period. Common in hospital studies.
- Purposive / judgement: researcher selects information-rich or typical cases. Used in qualitative and expert studies.
- Quota: fixed numbers per subgroup, filled non-randomly.
- Snowball: initial participants recruit others from their networks. Used for hidden or hard-to-reach groups.
Uses: pilot studies, qualitative research, no sampling frame. Limits: selection bias, sampling error cannot be estimated, poor generalisability.
Sampling methods in public health: merits and demerits
MD Community Medicine 2003, 2009, and undated| Method | Merits | Demerits | Example use |
|---|---|---|---|
| Simple random | Unbiased, simple analysis | Needs a full list, costly if spread out, may miss small groups | Households from a village register |
| Systematic | Quick, spreads the sample evenly | Biased if the list has a repeating pattern | Every 10th OPD card |
| Stratified | Represents subgroups, more precise | Needs stratum information in advance | Urban and rural prevalence survey |
| Cluster | Cheap, no list of individuals needed | Less precise, needs a larger n (design effect) | Immunisation coverage survey |
| Multistage | Practical over large areas | Complex analysis, sampling error from each stage adds to the total | State-wide household survey |
| PPS (first stage of a cluster survey) | Self-weighting when a fixed number is taken per cluster | Needs a recent count of the size of every cluster | Villages for a coverage survey |
| Convenience / purposive | Fast, cheap, feasible | Selection bias, not generalisable | Pilot study, qualitative interviews |
WHO 30 × 7 EPI cluster sampling technique
MD Community Medicine 2011, 2020, 2021, and undated- Origin: WHO Expanded Programme on Immunization, described by Henderson RH and Sundaresan T (Bull World Health Organ 1982;60:253–60).
- Purpose: estimate immunisation coverage in an area without a list of children.
- Sample size logic: n = z²pq/d² with p = 0.5 and d = ±10 percentage points gives 96; allowing for clustering almost doubled this, and 210 was adopted.
- Stage 1: from the list of sites (villages or wards), select 30 clusters at random.
- Stage 2: in each cluster, pick a random starting household, then go to the next nearest household until 7 children of the target age group are found.
- Total: 30 × 7 = 210 children. Record vaccination status for each.
- Merits: quick and cheap, and needs a list of sites, not a register of every child.
- Demerits (WHO 2018 reference manual): households have unequal, unknown chances of selection, so it is not a true probability sample. Field workers may prefer easy-to-reach houses. One fixed design is used regardless of survey goals.
- Current WHO advice: probability-based sampling at each stage (WHO/IVB/18.09, 2018).
Sampling methods in clinical trials
DM Medical Oncology 2024- Define target population and eligibility criteria (inclusion, exclusion).
- Recruitment is usually consecutive enrolment of eligible, consenting patients: a non-probability method.
- Random allocation (simple, block or stratified randomisation) then assigns treatment, which maintains internal validity.
- External validity is improved with broad eligibility, multiple centres and a recruitment log of screened and excluded patients.
- Report the settings and locations (CONSORT 2025 item 11) and the eligibility criteria (item 12a).
- Sample size is fixed in advance from a power calculation.
The one thing to remember
How you pick matters more than how many you pick.
Sources: Kiær AN, ISI session, Berne 1895. Neyman J. J R Stat Soc 1934;97:558–625. Goodman LA. Ann Math Stat 1961;32:148–70 (doi:10.1214/aoms/1177705148). Henderson RH, Sundaresan T. Bull World Health Organ 1982;60:253–60. WHO Vaccination Coverage Cluster Surveys: Reference Manual, WHO/IVB/18.09, 2018. von Elm E et al. STROBE statement. Lancet 2007;370:1453–7. Hopewell S et al. CONSORT 2025 statement. BMJ 2025;389:e081123.
Past papers
Every TNMGRMU question on sampling
Each one is answered in this chapter. Tap a sitting to open the university's own question paper.
- ×2Sampling methods
- ×2Non-probability sampling techniques
- ×1Explain the classification of data with suitable examples. Explain the basic concept of sampling. How do you determine the sample size of a research study?
- ×1Types of Sampling
- ×1Describe the various sampling methods in clinical trials
- ×1Snowball sampling
- ×1Judgment Sampling
- ×1Snowball method of sampling
- ×1Write briefly on sampling techniques. Describe the commonly used sampling technique to evaluate the immunization status of children below five years of age in a district?
- ×130 * 7 EPI cluster sampling technique
- ×1List the sampling methods and discuss in detail the random sampling method
- ×1Discuss the types of sampling using suitable examples
- ×1Probability sampling techniques
- ×1Types of sampling methods
- ×1W.H.O. cluster sampling technique
- ×1Discuss in detail the procedures adopted for different sampling methods
- ×1What is sampling? What are the different types of sampling technique? Describe any two methods with suitable examples
- ×1Describe the sampling methods, which are used in public health practice giving their relative merits and demerits. (25)
- ×1What is Sampling? Discuss the sampling methods and their applications in Community Medicine
- ×1Cluster sampling technique – Role in sampling
Further reading