Sample survey

Also called: Sample survey method · Topic: Economic Data: Census, NSS, Surveys and Statistical Tools · NCERT: Class 11, Ch 2 "Collection of Data"; Class 11, Ch 8 "Use of Statistical Tools"

Meaning

A sample survey collects data from only a part of the population (called the sample). The result is then used to estimate the value for the whole population. A census is different: it collects data from every single unit.

Most official data in India comes from sample surveys, such as those of the NSS and the PLFS. How good that data is depends on how well the sample is chosen.

• Sampling error = Sample estimate − Population parameter
• Chance of selection (sampling fraction) = n / N, where n = sample size and N = population size.

Explanation

Key building blocks

• Population (also called the universe): every unit that has the feature we want to study. A unit can be a person, a household, a farm or a firm.

• The purpose of the study decides the population. To study farm wages, the population is all agricultural labourers, not all people.

• Sample: the part of the population from which data is actually collected.

• Representative sample: a sample that looks like a small copy of the population.
• Example: if 60% of households in the population are rural, about 60% of the sample should be rural too.

• Such a sample gives fairly accurate results at lower cost and in less time.

• Sampling frame: the complete list of units from which the sample is drawn. Without a good list, a good sample is not possible.

• Population parameter: the true value for the whole population, for example the true average income of all households.
• Sample estimate (also called a statistic): the value worked out from the sample. It stands in for the parameter.

Two kinds of error

• Sampling error: the gap between the sample estimate and the population parameter. It happens only because we looked at a part and not the whole.
• Worked example: 5 farmers earn ₹500, ₹550, ₹600, ₹650 and ₹700.
  • True mean (the parameter) = ₹600.
  • A sample of 2 farmers (₹500 and ₹600) gives a mean of ₹550. This is the estimate.
  • Sampling error = 550 − 600 = −₹50.

• A larger, well-drawn sample usually makes this error smaller.

• Non-sampling errors can happen in a census as well as a sample. Examples:

• wrong answers;
• non-response (people who do not reply);
• recording mistakes;
• a biased questionnaire.

How the sample is chosen

• Random sampling: every unit has an equal chance of being picked. The investigator's choice plays no part.
• "Random" does not mean "haphazard". It follows a strict, planned procedure.
• Lottery method. NCERT example: 300 households in a locality are studied to see how a rise in petrol prices affects them. Write all 300 names on slips, mix them well and draw 30.
• Random number table. Number the units 001–300, then read numbers from a table such as Tippett's. Today, computers generate random numbers instead.
• Worked example: each household's chance of selection = n/N = 30/300 = 0.1 (10%). This is also the sampling fraction.

• Counting possible samples (NCERT exercise): choosing 3 of 10 students gives ¹⁰C₃ = (10×9×8)/(3×2×1) = 120 possible samples.

  • Each particular set of 3 has a 1/120 chance of being drawn.
  • Each student has a 3/10 chance of being chosen.

• Types of random sampling:

• Stratified sampling: first split the population into groups called strata (for example rural/urban, or districts). Then sample randomly inside each group. This makes sure every group is represented.
• Systematic sampling: take every k-th unit from the list, where k = N/n.
  • Example: N = 300 and n = 30, so k = 10.
  • Pick a random start between 1 and 10, say 4. Then take units 4, 14, 24 … 294.

• Multi-stage sampling: sample in steps.

  • First choose villages or blocks. These are the first-stage units (FSUs).
  • Then choose households inside them.
  • This saves travel, and you do not need a list of every household in the country.

• Non-random sampling: units do not have equal chances. The investigator picks them by judgement, purpose, convenience or quota (a fixed number from each group).

• Result: bias creeps in, and the sampling error cannot be measured properly.
• NCERT example: picking 10 of 100 households because they are nearby or known to you. Neighbours tend to have similar incomes, so the sample misses the variety in the population.

Why samples are used, and where they fall short

• Why most surveys are samples:
• Cheaper and quicker: fewer units means fewer interviews and less data processing.
• More detail: with fewer respondents, each interview can go deeper. This is called an intensive enquiry.

• Smaller team: a small team is easier to train and supervise, so there are fewer non-sampling errors.

• Where a sample cannot replace a census. Some tasks need data on every unit:

• electoral rolls;
• delimitation (redrawing constituency boundaries by population);
• small-area counts for a single village or ward;
• caste counts for every locality.

In India

• National Sample Survey (NSS):
• Set up in 1950 on the advice of Prof. P. C. Mahalanobis, then Statistical Adviser to the Cabinet [4].
• It was set up because the National Income Committee had found big gaps in data on the unorganised/household sector [4].
• NSS is now part of the National Statistics Office (NSO) under MoSPI, headed by a Director General [4].

• It completed 75 years in 2025. This was the theme of the 19th Statistics Day, held on Mahalanobis's birth anniversary [5].

• NSS sample design: stratified multi-stage sampling.

• The FSUs are census villages in rural areas and Urban Frame Survey (UFS) blocks in urban areas.

• Households are selected at the final stage [3].

• Periodic Labour Force Survey (PLFS):

• Launched in 2017. It gave quarterly data for urban areas only, and annual rural+urban data for each July–June year [3].
• Revamped from January 2025 [3]:
  • 22,692 FSUs a year (12,504 rural and 10,188 urban), up from 12,800.
  • 12 households per FSU, up from 8. That makes about 2,72,304 households, which is 2.65 times the earlier ~1,02,400.
  • The district is now the basic stratum for selecting FSUs.
  • Rotational panel: each household is visited 4 times in 4 consecutive months.
  • Monthly all-India estimates, with the first bulletin for April 2025.
• Reference periods [3]:
  • Current Weekly Status (CWS) looks at a person's activity in the last 7 days.
  • Usual Status looks at the last 365 days.

• Results from 2025 onward should not be compared directly with results up to December 2024, because the sample design changed [3].

• Scale: census vs sample:

• Census 2027 uses about 31 lakh enumerators and supervisors [2].

• The revamped PLFS covers about 2.72 lakh households a year (from 2025) [3].

• The sampling frame comes from the census:

• PLFS picks its FSUs from the list of Census 2011 villages and UFS blocks [3].

• The number of census villages was 6,40,932 in 2011 and is 6,39,902 for Census 2027 [2].

• NCERT example (Churachandpur, Manipur): the population is all agricultural labourers in the district. The sample is 10% of them.

• Exit polls are sample surveys of voters.
• Section 126A of the Representation of the People Act, 1951 bans conducting or publishing them from the start of polling in the first phase until half an hour after polling closes in the last phase.

Don't confuse with

• Census (complete enumeration): covers every unit of the population. A sample survey covers only a part and estimates for the rest. Only a sample survey has sampling error.
• Sampling error vs non-sampling error: sampling error happens only in samples (estimate − parameter). Non-sampling errors, such as non-response or wrong answers, happen in both a census and a sample.
• Parameter vs statistic (estimate): a parameter is the true value for the whole population. A statistic is the value worked out from the sample.
• Random vs haphazard/non-random sampling: random sampling gives every unit an equal chance through a planned procedure. Judgement, convenience or quota selection is non-random and brings in bias.

Prelims Hooks

• A sample survey studies part of the population and estimates for the whole. Sampling error = estimate − parameter.
• In random sampling, every unit has an equal chance of selection. Tippett's table is a random number table. Random ≠ haphazard.
• Systematic sampling interval k = N/n. ¹⁰C₃ = 120 possible samples.
• NSS was set up in 1950 on Mahalanobis's advice. It now works under NSO, MoSPI [4].
• PLFS (since 2017) uses a stratified multi-stage design. FSUs are Census 2011 villages / UFS blocks, and final-stage units are households [3].
• Trap: PLFS CWS = last 7 days. Usual Status = last 365 days [3].

Mains Points

• A census and sample surveys work together; one does not replace the other.
• Surveys (NSS/PLFS) are cheaper, faster and more detailed.
• But a survey is only as good as its sampling frame. PLFS in 2025 still draws from a 2011 list [3].

• So new settlements and fast-growing towns may be missed. The delayed Census 2021 also weakens survey data.

• The PLFS revamp trades cost for timeliness and detail.

• The sample grew 2.65 times, the district became the basic stratum, and monthly estimates were added [3].
• Policy can now respond faster to changes in jobs.

• The costs: the new series cannot be compared directly with the old one, and running it costs more.

• Sampling method decides credibility.

• Exit polls go wrong because of non-response, poorly chosen booths and false answers.
• This shows how weak sampling misleads the public. Section 126A of the RP Act, 1951 balances free speech against fair elections.

Related concepts

• Census
• Population
• Sample
• Representative sample
• Population parameter
• Random sampling
• Non-random sampling
• Sampling frame
• Lottery method
• Random number table

Read more

• Census or sample: populations, samples and sampling methods