Sampling & Statistics

Sample Size

Definition

The number of documents drawn for a statistical estimate, set by the target precision, confidence level, and how rare the trait is.

Why it matters

There is no universal number; it follows from the assumptions you state up front.

In practice

Sample size is set by three choices: how precise the estimate needs to be (the margin of error), how confident you want to be (the confidence level), and what you expect the proportion to be. For large collections, the size of the collection barely matters.

When the proportion is unknown, 50% is the cautious assumption, because it needs the largest sample. When responsive documents are rare, a fixed margin in percentage points may not be precise enough, and teams size the sample to find a minimum number of responsive documents instead.

Formula

n = z² × p × (1 − p) ÷ E², then n ÷ (1 + (n − 1) ÷ N) for a population of N

Worked example

For 95% confidence (z = 1.96), a ±2.5% margin and p = 50%: n = 3.8416 × 0.25 ÷ 0.000625, about 1,537. For a 20,000-document population, the correction brings it to about 1,428.

Illustrative numbers, not from a real matter.

Compare with

  • Margin of Error: The margin is an input to sample size, and the result you get back after sampling.

Work it out

← Back to glossary