Core Concepts
Precision
Definition
The share of documents marked responsive that were actually responsive.
Why it matters
It affects cost and reviewer burden more than completeness.
In practice
Precision answers the purity question: of the documents marked responsive, by a model or by reviewers, what share actually are?
Low precision costs money, because reviewers spend time on documents that turn out not to matter, and it can mean producing material that did not need to be produced. It says nothing about what was missed.
Precision and recall usually trade off. Lowering the cutoff score finds more responsive documents (higher recall) but lets in more non-responsive ones (lower precision).
Formula
Precision = true positives ÷ (true positives + false positives)
Worked example
The model flags 1,000 documents as responsive. Reviewers confirm 850 of them and reject 150. Precision is 850 ÷ 1,000, or 85%.
Illustrative numbers, not from a real matter.
Compare with
- Recall: Recall measures completeness. A review can have high precision and still miss most responsive documents.
Work it out
- Precision, Recall & F1 calculator · How accurate are the model or reviewer calls?