Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Epidemiology / Doing epidemiology: surveys, evidence and ethics

Planning and running a population survey

14 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Plan a household survey from objective to analysis, with a sample size and design effect

  • Describe multistage cluster sampling and survey weights

  • Apply data-quality and field-management practices

A cross-sectional survey is the most common field study in public health and the basis of an MPH thesis. A well-planned small survey is worth more than a sloppy large one.

The planning sequence

  1. 1

    1. Question and objectives

    State the primary objective in one sentence with the outcome, the population and the place. 'To estimate the prevalence of hypertension in adults aged 18 years and above in Kyotera District in 2026.'

  2. 2

    2. Define the target population

    Who is included and excluded, the age range, residence rules (lived there at least 6 months), and the sampling unit (household, individual).

  3. 3

    3. Sampling frame and design

    A list from which to sample: villages from the census, households from a fresh listing. Choose stratified, cluster or multistage sampling.

  4. 4

    4. Sample size

    From the primary outcome, expected prevalence, precision, design effect and non-response.

  5. 5

    5. Tools and measurement

    Validated questionnaires (translate and back-translate; pre-test), calibrated equipment, standard operating procedures.

  6. 6

    6. Ethics and permissions

    Ethics committee, national registration, local government and community permissions, informed consent.

  7. 7

    7. Training and pilot

    Train for several days; pilot in a non-study area; refine tools and timing.

  8. 8

    8. Data collection and supervision

    Electronic data capture (KoboToolbox, ODK, REDCap) with built-in checks, daily supervision, repeat measures on a subsample.

  9. 9

    9. Cleaning, analysis and reporting

    Apply weights and account for clustering; report with STROBE.

Worked example: how many to sample

Estimate hypertension prevalence expected at about 30%, with precision ±5 percentage points, 95% confidence.

  1. Simple random sample: n = 1.96^2 × 0.30 × 0.70 / 0.05^2 = 322.7, so 323.
  2. Households are clustered in villages; assume a design effect of 2.0 (intracluster correlation about 0.04 with 25 people per village). 323 × 2 = 646.
  3. Allow for 10% non-response: 646 / 0.9 = 717.8, so 718 adults.
  4. With 30 interviews per village: 718/30 = 23.9, so 24 villages.

If the analysis will also compare men and women, size the study for the smaller group and each subgroup, or the precision will be inadequate.

Multistage cluster sampling

The usual approach in low-resource settings, because no list of all individuals exists.

  1. Stage 1: select villages (enumeration areas) with probability proportional to size (PPS) from the census list.
  2. Stage 2: in each village, list households (or use a map and a random start) and select a fixed number, systematically or randomly.
  3. Stage 3: within the household, select one eligible adult at random (Kish grid or electronic randomisation).

Because people have unequal chances of selection (one adult in a large household is less likely to be chosen than one in a small), the analysis uses design weights (inverse of selection probability) and adjusts for non-response. The standard errors must reflect clustering (survey procedures such as svy in Stata, survey in R).

1Census list ofvillagesStage 1: PPS sample ofvillages2Household listingStage 2: samplehouseholds3Eligible adultsStage 3: one perhousehold, at random4MeasurementInterview, bloodpressure, samples
A three-stage survey. Each stage's selection probability enters the weight.

Data quality

  • Standard operating procedures for every measurement (blood pressure: seated for 5 minutes, correct cuff size, three readings).
  • Electronic forms with range checks, skip logic and timestamps; GPS only if the ethics approval allows it and privacy risk is managed.
  • Supervision: field editing, spot checks, re-interview of 5 to 10% of households.
  • Missing data and response rate: record reasons for refusal, absent, unable; callbacks (at least 3 visits at different times).
  • Back-up daily, encrypt devices, and protect identifiers.

Common weaknesses

Sample size calculated for the whole sample but subgroups analysed; no allowance for clustering; convenience sampling described as 'random'; interviewers not trained; no pilot; low response rate ignored; questionnaires not validated or translated; weights omitted.

Keep a study log

Record every decision (changes to the protocol, villages replaced, equipment failures). It protects the validity of the analysis and makes the methods section easy to write.

Summary

  • A survey starts with a clear objective, defined population and a sampling frame.

  • Sample size needs the expected prevalence, precision, design effect and non-response.

  • Multistage cluster sampling needs weights and variance estimation that account for clustering.

  • Pilot, train, supervise and use electronic forms with checks.

  • Report with STROBE and discuss response rate and limitations.

Check your understanding

Question 1
1

A basic sample of 323 with a design effect of 2 requires:

Question 2
2

Probability proportional to size sampling gives larger villages:

Question 3
3

Design weights are the inverse of:

Question 4
4

Allowing 10% non-response for a required 646, enrol:

Question 5
5

Which step precedes full data collection?

Question 6
6

Ignoring clustering in the analysis leads to:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Maternal, newborn and child health and nutritionFinding and weighing the evidence