Sampling: universe, sample and sampling techniques

1. Universe, census and sample

The universe (or population) is the whole group about which the researcher wants to draw conclusions: all undertrial prisoners in Karnataka, all motor-accident claims decided by a tribunal in 2025, all advocates enrolled with a State Bar Council. A study of every member of the universe is a census. A sample is a part of the universe selected so that the conclusions drawn from it can be extended to the whole.

Sampling is used because a census is usually too costly, too slow or impossible, and because a well-drawn sample, studied carefully, can be more accurate than a census done hurriedly. Sampling is unsuitable where the universe is small, or where every unit matters (a doctrinal study of all Constitution Bench decisions on a point should examine them all).

Key terms. The sampling unit is the element selected (a person, a case file, a court). The sampling frame is the list from which the sample is drawn: a cause list, a jail register, a Bar Council roll. A defective frame — one that leaves out part of the universe — biases the result however carefully the sample is drawn from it.

2. Characteristics of a good sample

• representative of the universe in the respects that matter;
• adequate in size for the analysis planned;
• free of bias in selection;
• practical within the resources of the study.

3. Probability sampling

In probability sampling every unit has a known, non-zero chance of selection. Only probability samples allow the researcher to estimate sampling error and to generalise statistically.

• Simple random sampling. Every unit has an equal chance; selection by lottery, random-number tables or software. Example: drawing 200 case numbers at random from the register of a family court.
• Systematic sampling. Every kth unit from a list after a random start. Simple and quick, but biased if the list has a periodic pattern (every tenth entry in a register being a particular kind of case).
• Stratified sampling. The universe is divided into strata that differ in an important respect (urban and rural courts; male and female prisoners), and a random sample is drawn from each, proportionately or otherwise. It ensures that small groups are represented and improves precision.
• Cluster sampling. The universe is divided into natural groups (districts, villages, jails); some clusters are chosen at random and all or many units in them are studied. Cheaper for scattered populations, but less precise if clusters differ.
• Multi-stage sampling. Sampling in stages: select districts, then courts in them, then case files. Large national surveys use it.

4. Non-probability sampling

Here the chance of selection is unknown. Results cannot be generalised statistically, but the methods are cheap, quick and often the only practical option in legal fieldwork. They suit exploratory and qualitative research.

• Convenience (accidental) sampling. Whoever is available: litigants waiting outside a court. Easy, but the most open to bias.
• Purposive (judgment) sampling. The researcher selects units thought to be typical or especially informative: senior judges of commercial courts, NGOs working on child labour. Common in qualitative legal research; its value depends on the researcher’s judgment and its justification must be stated.
• Quota sampling. The researcher fixes quotas for categories (so many women, so many men; so many advocates of each experience band) and fills them by convenience. It looks like stratified sampling but is not random within the strata.
• Snowball sampling. Initial respondents refer others. Used for hidden or hard-to-reach groups: victims of trafficking, bonded labourers, transgender persons facing discrimination. It tends to over-represent people in the same networks.

Worked choice. A study of how motor-accident claims tribunals in Karnataka assess compensation could stratify tribunals by region, select some tribunals at random within each region (cluster), and then draw every tenth disposed claim from each tribunal’s register (systematic) — a multi-stage design. Interviews with claimants’ advocates, added to explain the patterns, would use a purposive sample of those with the most claims.

5. Sample size

No fixed rule decides size. It depends on the variability of the universe (more variation needs a larger sample), the precision required, the number of sub-groups to be analysed, the method of analysis, and resources. Larger samples reduce sampling error, but with diminishing returns. In qualitative research size is set by saturation: interviewing continues until new interviews produce no new themes.

6. Errors

7. Sampling in legal research: practical points

• Court records are often the best frame, but access may need permission from the court administration. The National Judicial Data Grid publishes aggregate data on pending and disposed cases, which can help in building a frame and choosing strata.
• Reported judgments are not a random sample of decisions: they are chosen for reporting because they are important or unusual. A study of “how courts decide” that relies only on reported cases overstates the appellate and the novel.
• Vulnerable groups (prisoners, children, victims) raise consent and confidentiality issues; see the note on research ethics.
• State the sampling method and its limits in the methodology chapter. An examiner reads that paragraph closely.

How this is examined

For a short note: definition of universe and sample, the reasons for sampling, the two families of techniques with one example each, and a line on sampling error. For an essay on empirical research, use the legal points in section 7 to show that you understand the limits of reported cases as data.