SmartCalc
FinancialHealthHome & EngineeringMathematicsTime & DateConvertersEducationGeneratorsGeneral
  1. Home
  2. /Mathematics
  3. /Intervals & sampling design
  4. /Design Effect Calculator

Design Effect Calculator

Free design effect calculator: turn the average cluster size and the intraclass correlation into the design effect, the sample to field and clusters to visit.

This calculator turns a clustered survey design into the two numbers a survey plan needs: the design effect the design carries and the sample size that has to be fielded to keep the precision of a simple random sample. Enter the average number of interviews per cluster, the intraclass correlation and the sample size a simple random sample would need.

The design effect, or DEFF, is the variance of your estimate under the real design divided by the variance under a simple random sample of the same size. Interviewing several people in the same village, school or clinic brings back less new information than spreading the same number of interviews at random, because neighbours resemble each other. A design effect of 2 means the survey carries twice the variance, so 1,000 interviews buy the precision of only 500.

For one-stage cluster sampling the standard formula is 1 + (m - 1) x rho, where m is the average cluster size and rho the intraclass correlation, the share of the total variation that sits between clusters rather than inside them. With 20 interviews per cluster and rho of 0.05 the design effect is 1.95, close to the doubling that survey manuals usually assume.

The results are a field plan: the design effect itself, the number of interviews to field (a simple random sample requirement multiplied by the design effect and rounded up), the number of clusters that costs, and how much bigger the sample is as a percentage. Set the cluster size to 1 or rho to 0 and the design effect collapses to 1, the reminder that clustering only ever costs precision.

Which rho to use is the hard part and no formula settles it: take it from a pilot, an earlier survey of the same population or published estimates, and prefer the upper end of the plausible range. The formula assumes equal cluster sizes, a large number of clusters and no finite population correction, and it cannot express a stratification that makes a design more efficient than simple random sampling.

Your inputs

Average number of interviews collected in one village, school, clinic or other cluster. Decimal averages are allowed.
01
How similar people inside one cluster are, from 0 to 1. Survey manuals commonly plan with 0.05; the larger it is, the more clustering costs.
The sample size a simple random sample would need for your precision target, before any adjustment for clustering.
Design effect
1.9500000000000002
Sample to field
780
Clusters to visit
39
Sample inflation
95%

The large figure is the design effect. All derived results use its unrounded calculation value. Interviews and clusters are rounded up, and the effective sample of the plan is the fielded sample divided by the design effect.

Frequently Asked Questions

What is a design effect?

It is the ratio of the variance of your estimate under the survey design you actually use to the variance of the same estimate from a simple random sample of the same size. Leslie Kish introduced it in 1965 and it is still the standard way to describe what clustering, weighting and stratification cost. A value of 2 means twice the variance, so 1,000 interviews are worth 500 simple random ones; below 1 the design is more efficient than simple random sampling.

What intraclass correlation should I use?

Take it from a pilot, an earlier survey of the same population or estimates published for the same indicator rather than from a guess. Health and household surveys commonly land between 0.01 and 0.15, and inside one village or classroom the value is usually positive, which is exactly why clustering hurts. The WHO planning figure of 2.0 for vaccination coverage corresponds to rho of 0.05 with about 21 interviews per cluster. When the choice is uncertain, plan with the largest plausible rho.

How do I get the effective sample size, or reuse a published design effect?

The effective sample size is the fielded sample divided by the design effect: 1,000 interviews with a design effect of 2 carry the precision of 500. To reuse a design effect reported by someone else, convert it into rho first with rho = (DEFF - 1) / (m - 1); a reported DEFF of 1.4 with 11 interviews per cluster is rho of 0.04, and entering that pair here reproduces the inflated sample size.

Related Calculators

Confidence Interval Calculator

Free confidence interval calculator: enter the sample mean, standard deviation and size to get the margin of error and 90%, 95% or 99% interval instantly.

→

SmartCalc.top — Free online calculators. No registration required.

© 2026 SmartCalc.top