Skip to content
Scintria
ToolsGrant writerFunding
Start free
ScintriaStart free
Simulation Lab

Survey precision

A cluster survey estimating a prevalence. People in the same village or school are alike, and some people do not respond, who may differ from those who do. See the real margin of error, and the bias that no sample size can fix.

%
%
More options

Often 0.01–0.1 for health surveys; higher for things shared by a community, like water source.

points

Margin of error

±5.4

Percentage points, 95% interval. A simple random sample of the same size would give ±4.1.

Interval holds the truth

95%

Of simulated surveys; it should be 95%.

Design effect

1.75

Effective sample size

274

Bias from non-response

-0.0

Respondents

480

Estimates across simulated surveys

253035TruthEstimated prevalence (%)

True prevalence 30.0%; respondents' prevalence 30.0%. Teal bars are within the margin of error of the truth.

For your protocol

A survey of 30 clusters with 20 people approached in each (80% expected response, about 480 respondents), assuming a prevalence of 30.0% and an intraclass correlation of 0.05, has a design effect of 1.75 (effective sample size 274) and a 95% margin of error of about ±5.4 percentage points. Simulation confirms the 95% interval contains the true value 95% of the time.

How this simulation works

Each simulated survey samples the clusters; each cluster's own prevalence is drawn from a beta distribution whose spread gives the intraclass correlation, a random share of the people approached respond, and each respondent is a case or not. Non-respondents can have a different prevalence from respondents, so the respondents' prevalence is set to keep the population prevalence where you put it. The estimate is the respondents' proportion, with a cluster-robust (ratio-estimator) 95% interval. The design effect is 1 + (m − 1) × ICC, where m is the number of respondents per cluster. 2,000 surveys per setting.