"How many households do we need to survey?" is one of the first questions in any study — and one of the most consequential. Too few and your results are unreliable; too many and you burn budget. Here is how to think about household sample size.
What drives sample size
For estimating a characteristic in a population, four factors set the number:
- Confidence level — how sure you want to be, usually 95%
- Margin of error — how precise, e.g. ±5%
- Expected proportion — the prevalence you are estimating (50% is the conservative worst case)
- Population size — matters mainly when the population is small
A worked intuition
To estimate a proportion at 95% confidence with a ±5% margin of error, you need roughly 385 households for a large population. Want a tighter ±3% margin? The number jumps to around 1,067. Precision is expensive — each step tighter costs many more interviews.
Adjustments household surveys must make
The textbook number is only the starting point. Real household surveys adjust for:
Design effect (clustering)
Household surveys usually use cluster sampling (surveying groups of households in selected villages/blocks). Because households in a cluster resemble each other, you get less information per interview. You multiply the base sample by a design effect (often 1.5–2.0) to compensate.
Non-response
Some selected households will refuse or be absent. If you expect an 85% response rate, inflate the sample so you still reach your target number of completes.
Subgroup analysis
If you need reliable estimates for subgroups (e.g. by district or gender), each subgroup needs an adequate sample of its own — which can drive the total up substantially.
Putting it together
A realistic household sample size calculation looks like:
Base sample × design effect ÷ expected response rate, then check subgroup needs.
So a base of 385 with a design effect of 1.6 and an 85% response rate becomes roughly 725 households to plan for.
Tools to use
- Online sample size calculators for the base figure
- Statistical software (R, Stata) for complex, stratified, or clustered designs
- A statistician's input when subgroups and clustering interact
The trade-off nobody puts in the brief
Sample size is usually presented as a statistical calculation. In practice it is a negotiation between three things that cannot all be maximised: precision, the number of subgroups you want to report on, and the budget.
A useful way to run that negotiation is to price precision explicitly before choosing.
| Target precision | Households needed (DEFF 1.6, 85% response) | Indicative field cost (INR) |
|---|---|---|
| ±10% | 181 | 1,80,000 |
| ±7% | 369 | 3,50,000 |
| ±5% | 723 | 6,50,000 |
| ±4% | 1,129 | 10,00,000 |
| ±3% | 2,008 | 17,50,000 |
Presented this way, the conversation changes. A funder asking for ±3% is asking to spend nearly three times what ±5% costs, to narrow the interval by two percentage points. Sometimes that is exactly right — a national estimate feeding policy needs precision. Often, once the cost is visible, ±5% or ±7% turns out to be entirely adequate for the decision at hand.
Subgroups are where budgets actually break
The overlooked multiplier is not precision — it is disaggregation. Each subgroup you intend to report on separately needs its own adequate sample, not a share of the total.
| Reporting requirement | Effective sample requirement |
|---|---|
| One district-level estimate | 1 × base sample |
| Split by gender of household head | Base sample, if the split is roughly even |
| Split by three income terciles | Roughly 3 × base for equal precision on each |
| Split by four blocks | Roughly 4 × base |
| Blocks × gender, reported separately | Roughly 8 × base |
| A group that is 12% of the population | Base ÷ 0.12 to get that group alone |
That last row is the one that catches everyone. If female-headed households are 12% of the population and you need a ±7% estimate for them specifically, you need roughly 370 female-headed households — implying about 3,100 households overall under proportional sampling.
The efficient answer is almost never to inflate the whole sample. It is to stratify and over-sample the small group, then apply weights at analysis. Taking 350 female-headed households deliberately and weighting them back gets you both the subgroup estimate and the correct district figure, at a fraction of the cost.
Allocating sample across clusters
For a fixed total, how you distribute households across villages changes your precision substantially.
| Design | Villages | Households each | Relative precision |
|---|---|---|---|
| A | 20 | 50 | Lowest |
| B | 33 | 30 | Better |
| C | 50 | 20 | Best |
| D | 100 | 10 | Marginally better, much costlier |
More villages with fewer households each always improves precision, because you are capturing more independent variation. It also costs more in travel, supervision and time. The practical sweet spot for most Indian district surveys sits at 20 to 30 households per village across 30 to 50 villages — beyond which the precision gains flatten while travel costs keep climbing.
The exception is when within-village variation is high for your key outcome. For individual attitudes, which cluster weakly, larger clusters cost you little. For infrastructure access, which is close to determined at village level, taking 50 households in one village is close to taking one.
Two worked examples
A CSR livelihoods programme. Target: detect a 6 percentage point rise in the share of households with a secondary income source, from a baseline of about 30%. Two rounds, treatment and comparison villages.
| Step | Value |
|---|---|
| Base per arm, 80% power, 5% alpha | 861 |
| Design effect, 1.7 | 1,464 |
| Non-response, 88% | 1,664 |
| Attrition to endline, 15% | 1,957 per arm |
| Total across both arms | 3,914 households |
That is a large and expensive study, and the number is worth putting in front of the funder early. The honest options are to accept the cost, to raise the detectable effect to 10 points and survey roughly 1,400 per arm, or to accept lower power. What should not happen is fielding 600 per arm and hoping.
An EIA socio-economic baseline. Target: describe the affected population within 10 km, no change detection required, one composite estimate at ±6%.
| Step | Value |
|---|---|
| Base at ±6%, proportion near 50% | 267 |
| Design effect, 1.4 | 374 |
| Non-response, 90% | 416 |
| Total | 420 households |
An order of magnitude smaller, because nothing has to be detected — only described. Confusing these two purposes is the most common reason a survey is either wildly over-specified or unable to answer its own question. See socio-economic surveys in environmental impact assessment.
What to write in the sampling note
Reviewers and funders reject sample size sections not because the number is wrong but because the assumptions are absent, so nobody can check it. Six lines is enough:
- The primary indicator the calculation is based on.
- Its expected baseline value, and the source of that expectation.
- The smallest change or margin of error that matters, and why.
- Alpha, power, and the design effect assumed.
- Expected non-response and, for panel studies, attrition.
- The final number, with the subgroups it can support and those it cannot.
Point six protects everyone. Stating up front that the sample supports district-level and gender-disaggregated estimates but not block-level ones prevents the later request for a breakdown the data cannot honestly provide.
For the underlying formulas and reference tables see how to calculate sample size for your survey or study; for the design decisions, sampling methods for household surveys; for the budget context, how much does a socio-economic survey cost. Our socio-economic and baseline survey service produces this note as part of every design. OpenEpi handles the base calculation free in a browser.
The takeaway
Household sample size is not a single formula but a chain of adjustments — precision, clustering, non-response, and subgroups. Work through each deliberately, and you will field a survey that is large enough to be credible and no larger than it needs to be.
Frequently Asked Questions
How do I calculate sample size for a household survey?
Start from the base requirement for your target precision, then multiply by the design effect from clustering, divide by the expected response rate, and for panel studies divide again by one minus expected attrition. Each adjustment is real, and skipping them typically leaves a survey half the size it needed to be.
How many households should I survey in a village?
Twenty to thirty in most cluster designs. Fewer villages with more households each raises the design effect and reduces precision, while more villages with fewer households each improves precision but increases travel and supervision cost.
Why does my sample size need to be bigger for subgroups?
Because each subgroup you report on separately needs its own adequate sample, not a share of the total. Rather than inflating the whole survey, stratify and deliberately over-sample the small group, then weight the results back for population-level estimates.
What sample size do I need to detect programme impact?
Considerably more than one needed merely to describe the population. It depends on the smallest change worth detecting, the baseline level, clustering, non-response and attrition between rounds. Detecting a 6 percentage point change commonly requires 1,500 to 2,000 households per arm.
Is 384 households enough for a district survey?
Only for a single district-wide estimate at ±5% under simple random sampling with perfect response. Add a design effect of 1.6 and 85% response and the same precision requires about 723. Any subgroup reporting pushes it higher again.
What response rate should I plan for?
Eighty-five to ninety percent is achievable in rural Indian household surveys with proper revisit protocols and advance community notification. Urban and peri-urban response rates run lower, often 70 to 80%. Build the assumption into the sample size and report the rate you actually achieved.
How DSTechnoverse Runs Your Survey
DSTechnoverse is a full-service socio-economic and baseline survey agency based in Indore. We do not just advise — we run the whole study end to end and hand you evidence you can defend:
- Design — objectives, indicators, and sampling (random, stratified, cluster, or multi-stage), with a piloted questionnaire translated into local languages.
- Field data collection — trained enumerators using CAPI (tablet) tools such as KoBoToolbox, ODK, and SurveyCTO-style apps, with GPS stamping, audio audits, and built-in skip-logic validation.
- Quality control — back-checks, spot-checks, and daily high-frequency data monitoring so problems are caught in the field, not after.
- Analysis — SPSS, R, Python, and Excel for indicators, cross-tabulations, and impact estimates, including baseline–endline difference-in-differences.
- A detailed report — cleaned datasets, indicator tables, charts, GIS maps, and clear interpretation, with interactive Power BI / Tableau dashboards on request.
Our survey experience in Indore includes Smart City citizen-feedback surveys, Startup India & MSME surveys, Swachh Survekshan (Swachh Bharat) cleanliness surveys, and skill-development impact evaluations — delivered for government, municipal, corporate, and NGO clients.
Planning a socio-economic, baseline, endline, or willingness-to-pay survey? Talk to our survey team for a scope and quote.