Sampling Methods in Dental Research
On this page
Direct answer
Sampling is the disciplined act of studying a part to estimate the whole, and every dental survey stands on one of two platform families. Probability sampling — simple random, systematic, stratified, cluster, multistage and multiphase — gives every unit a known, non-zero chance of selection so that results can be generalised with calculable error. Non-probability sampling — convenience, judgement or purposive, quota, consecutive and snowball — has no such guarantee and serves exploratory, qualitative or hypothesis-generating work. The craft lies in matching the method to the geography and the question: a national oral health survey cannot list every Indian child, so it uses multistage stratified cluster sampling, exactly the logic of the WHO pathfinder design. Two terms keep the reasoning honest: sampling error, the chance wobble that shrinks with bigger samples, and sampling bias, the systematic skew that no sample size can repair.
What you must remember
- Simple random sampling: every unit has an equal chance, selection by lottery or random number tables; it needs a complete sampling frame, which is its practical weakness at population scale.
- Systematic sampling: every k-th unit from a randomly chosen start, where k = population size divided by sample size; elegant and quick, but vulnerable to hidden periodicity in the list.
- Stratified sampling: the population is divided into internally homogeneous strata — urban and rural, boys and girls, fluoride belt and non-belt — and samples are drawn from each, guaranteeing representation of small but important subgroups.
- Cluster sampling: naturally occurring groups (schools, villages, wards) are sampled wholesale; economical for fieldwork but observations within a cluster are alike, so the effective sample carries less information than its headcount suggests.
- Multistage sampling: sampling in steps — state, then district, then village, then school, then children — the backbone of national surveys including India's National Oral Health Survey.
- Non-probability set: convenience (whoever is available), judgement or purposive (hand-picked for the question), quota (fixed numbers per category), consecutive (all eligible arrivals in a period), snowball (participants recruit peers — the method for hidden populations such as smokeless tobacco users in slums).
- Sample size drivers: expected prevalence or effect size, desired precision (margin of error), confidence level, and for cluster designs the design effect — halving the margin of error roughly quadruples the required sample.
Designing a district survey's sample
Suppose a district dental officer needs caries prevalence among 12-year-olds across 800 schools. Listing every child is impossible, so build a multistage design. Stage one, stratify the district by block and by urban or rural status. Stage two, randomly select, say, 20 schools with probability proportional to enrolment. Stage three, within each school randomly pick one section per grade and invite all its children — a cluster, because classmates share water source, tuck-shop diet and teachers. The stratification protects representation of the remote tribal block that pure chance might skip; the clustering keeps travel affordable. Contrast the shortcut a student might propose: examining the nearest convent school — convenience sampling, fine for piloting the questionnaire, indefensible for the district estimate, since private schools under-represent the very children the programme exists for. The final planning touch is inflation: multiply the calculated sample by the anticipated non-response and by a design effect of about 2 for cluster sampling, or the study will be underpowered in the field.
Where students slip
The recurring confusion is stratified versus cluster sampling: strata are created by the researcher to be different from each other but internally alike, and units are sampled from within every stratum; clusters are pre-existing groups sampled wholesale, with the sampling unit (school) different from the observational unit (child) — saying "we stratified by school" when schools were randomly selected whole is the classic error. The second slip is treating snowball sampling as if it could yield prevalence — it reaches hidden people but reaches them through networks, so estimates from it carry unknowable bias. Third, students conflate sampling error with bias: the first is reduced by number, the second only by design. Finally, in quota sampling, remember that the interviewer chooses who fills the quota, which imports exactly the selection discretion that probability methods exist to remove.
Frequently asked questions
What separates probability from non-probability sampling?
Probability methods give every unit a known, non-zero selection chance and allow generalisation with measurable error; non-probability methods do not, and serve exploratory purposes.
How does systematic sampling work?
From a random starting point, select every k-th unit on the list, where k equals the population divided by the desired sample size.
Why do national oral health surveys use multistage sampling?
Because no complete frame of individuals exists, sampling proceeds in stages — for example state, district, school, class — combining economy with representative structure.
When is snowball sampling appropriate?
For hidden or hard-to-reach populations, where existing participants recruit others like them, since no sampling frame exists.
What is the design effect in cluster sampling?
The factor by which the required sample size must be inflated because observations within a cluster are correlated and carry less independent information — commonly around 2.