Skip to content
Omnikalc

Design effect calculator: how much larger must a cluster sample be?

Enter the simple random sample size, the average cluster size and the intracluster correlation to get the design effect and the sample size with clusters.

Updated on
Your inputs

Value out of range

Value out of range

Value out of range

Results

Total sample size needed with clustering

780

Design effect (DEFF)
1.950
Noticeable clustering effect
Number of clusters needed
39
Effective sample size if you only recruit the simple random sample
205.1
Inflation of the standard error
1.396

Surveys and trials often sample whole groups, such as villages, schools or clinics, instead of individuals. The design effect calculator takes the sample size under simple random sampling, the average cluster size and the intracluster correlation, and gives the design effect, the total sample size with clustering, the number of clusters, the effective sample size and the inflation of the standard error.

How the calculation works

  • Start from the sample size n you would need if people were picked at random one by one, for example from a margin-of-error calculation.
  • Estimate the intracluster correlation ρ, which is how alike people in the same cluster are, and the average cluster size m.
  • The design effect is DEFF = 1 + (m − 1) × ρ. It is the factor by which clustering inflates the variance of an estimate.
  • The sample you need with clusters is n × DEFF, the number of clusters is that total divided by m (rounded up), and the effective sample size of a clustered sample is its size divided by DEFF.

Example

A survey needs 400 people under simple random sampling. It will be run in clusters of 20 people with an intracluster correlation of 0.05. DEFF = 1 + 19 × 0.05 = 1.95, so the sample must grow to 400 × 1.95 = 780 people in 39 clusters. If only 400 people were recruited in clusters, the effective sample size would be 205, and the standard error would be 1.396 times larger.

Why it matters

Ignoring clustering gives a sample that is too small and confidence intervals that are too narrow. The design effect is the standard fix, and it appears in the plan of almost every household survey and cluster-randomized trial.

Good to know

The formula assumes clusters of similar size and a single level of clustering. With unequal sizes the design effect is a little larger, and with several levels of sampling the effects multiply. The ICC is usually not known precisely, so treat the result as an estimate and plan for some margin.

The sample size calculator gives the simple random sample, and the margin of error calculator shows how precision depends on it.

Frequently asked questions

What is the design effect?

It is the ratio of the variance of an estimate from a complex sample, such as a cluster sample, to the variance from a simple random sample of the same size. A design effect of 2 means you need twice as many people to get the same precision.

What is the intracluster correlation?

It measures how similar the answers of people from the same cluster are compared with people from different clusters. It is often small, from 0.01 to 0.1 for health surveys, but even a small value matters when clusters are large, because it is multiplied by the cluster size minus one.

Should I use many small clusters or a few big ones?

More, smaller clusters give a smaller design effect and more precision for the same number of people. Large clusters are cheaper to survey because of travel and set-up costs, so the choice is a balance between cost and precision.

Where do I find a value for the ICC?

Use values from earlier studies on the same outcome and population, or from a pilot. Published tables exist for many health, education and social outcomes. If unsure, run the calculation with a range of values to see how sensitive the sample size is.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.