Test reliability explained

What reliability measures

Reliability tells you how much of the difference between people’s scores comes from real differences in skill, rather than luck, wording or the particular questions they got. It is reported as a number from 0 to 1.

KR-20 in one paragraph

KR-20 (Kuder–Richardson Formula 20) is the standard reliability measure for tests where each question is simply right or wrong. It compares how much individual questions vary with how much total scores vary. If questions tend to agree with each other (people who get one right tend to get others right), KR-20 is high. Cronbach’s alpha is the same idea for questions with partial marks.

What is a good score?

• 0.90 and above: high-stakes exams and certification.
• 0.80–0.90: good for most graded exams and hiring decisions.
• 0.70–0.80: acceptable for screening and classroom quizzes.
• Below 0.70: use results with care, and improve the test.

Want a quick estimate from your own results? Try the reliability calculator.

Five ways to improve reliability

1. Add good questions. Longer tests are more reliable, as long as the new questions measure the same skill.
2. Remove confusing questions, especially ones that strong candidates get wrong more often than weak ones.
3. Replace questions everyone gets right or wrong. They don’t help tell people apart.
4. Fix weak distractors. Wrong options nobody picks make questions easier to guess.
5. Keep the test focused on one skill area; split very different skills into separate tests.

Examopotamia calculates KR-20 for every test and grades each question’s health automatically, so you can see exactly which questions to fix.