Choosing the right sample size for a baseline survey

Sample size is one of the first questions in any baseline or endline survey, and one of the most frequently misunderstood. Here is the practical version.

The four inputs

To calculate a sample size you need:

1. Confidence level. Usually 95%. It describes how often the true value would fall inside your interval if you repeated the survey.
2. Margin of error. How precise you need the estimate to be. Plus or minus 5 percentage points is common; 10 is often enough for project monitoring.
3. Expected proportion. If you expect about half the population to have a characteristic, use 50%. That is the most conservative choice.
4. Population size. If the population is small (a few thousand or fewer), the finite population correction reduces the required sample.

Finite versus non-finite

For large populations, the standard formula is enough. For smaller populations, applying the finite population correction avoids over-sampling. Our [sample size calculator](/product/sample-size-calculation) applies both formulas automatically.

What the number does not tell you

Sample size does not fix a biased sampling frame. If the list you draw from excludes part of the population, a large sample will simply give you a precise wrong answer.

It also does not account for non-response. If you expect 10% of selected households to be unavailable, inflate the sample accordingly.

Document your choices

Record the parameters you used and why. When a donor or reviewer asks why the sample was 383 rather than 400, a short methods note is worth more than a lengthy defence.

Budget for quality

A smaller sample with strong supervision, back-checks and clean data entry is usually better than a large sample collected quickly. Decide what precision you actually need, then protect the quality of the fieldwork.