# How to set a pass mark or cut score you can defend

**How do I set a pass mark for an exam?** A defensible pass mark is the score a borderline competent candidate would be expected to get on this paper, not a habit such as 60% or 70%. Use a recognised method such as Angoff, with a trained panel of about ten judges rating every question, check the result against real scores, and keep a written record of each step.

Published 2026-10-09 by EdProcto. https://edprocto.com/guides/setting-a-pass-mark-you-can-defend

Most pass marks are inherited. Somebody chose 60% or 70% years ago, nobody wrote down why, and the figure has decided who gets the certificate ever since. Then a candidate who missed by one mark asks how the line was drawn, and there is nothing to say.

To set a pass mark you can defend, start from a person, not a percentage: the candidate who is only just competent. The mark is the score that person would be expected to reach on this paper, worked out by a published method and written down. Examiners call this standard setting. It falls to the boards, federations and professional bodies that EdProcto's page [for associations and boards](https://edprocto.com/solutions/associations) is written for.

EdProcto publishes this guide and cannot set your standard: that is a judgement for people with the authority to make it. Pass mark, cut score and passing score mean the same thing here.

## Why is 60% or 70% not a standard?

A fixed percentage is not a standard because what it demands changes with every paper. Sixty per cent of a hard paper and of an easy one are different amounts of knowledge.

Cohen-Schotanus and van der Vleuten (2010) measured this in historical data from two Dutch universities. Across 52 tests at Groningen marked against a conventional fixed 60%, the failure rate ran from 17% to 97%. Across 54 tests at Maastricht, where the mark was set against the cohort, the failure rate held at about 17%, but the score needed to pass moved between 15% and 46%. Neither says what a passing candidate can do.

### Relative or absolute?

A relative standard passes a share of the group. An absolute standard passes everyone who reaches a described level, so in principle all could pass or all fail (Mortaz Hejri and Jalili, 2014). A certificate says its holder can do something, not that they outscored whoever sat that day, so a certifying body wants an absolute standard.

## What should a pass mark describe?

A pass mark should describe the borderline candidate, whose knowledge is only just enough. Zieky and Perie (2006) put the line between the best performer who still deserves to fail and the weakest who still deserves to pass. The average candidate is the wrong picture, though an easier one for judges to imagine (Mortaz Hejri and Jalili, 2014).

Write the description before anyone reads a question: a short list of what the just-competent person does. For a new match official it might read: applies every law that stops or restarts play; may be slow on rare situations; never gets a safety decision wrong.

Then decide which mistake is worse. Any mark passes some who should fail and fails some who should pass, and reducing one error increases the other (American Educational Research Association et al., 2014). That choice is about values, and it belongs to the body.

## Which methods can you use to set a cut score?

Use a method with a published procedure, chosen by the kind of judgement your people can make well.

| Method | How the mark is found |
| --- | --- |
| Angoff | Each judge estimates, for every question, the chance that a borderline candidate answers correctly. The estimates are summed for each judge and the sums averaged. |
| Borderline group | Judges who know the candidates' work name the borderline ones, blind to scores. The cut score is that group's median. |
| Contrasting groups | Judges class candidates as qualified or not, blind to scores. The cut score is where the two are equally likely. |
| Hofstee | Judges name the lowest and highest pass mark and failure rate they could accept. The mark is found where those limits meet the real scores. |
| Cohen | No panel: the mark is 60% of the score at the 95th percentile, after a correction for guessing. |

The first four rows follow Zieky and Perie (2006). Cohen's is the relative method of Cohen-Schotanus and van der Vleuten (2010), as Taylor (2011) summarises it: an affordable compromise that anchors the mark on the best performers.

### Different methods give different marks

No method finds a true answer. Downing et al. (2006), working through five methods for practical examinations, conclude that different methods produce different passing scores, that there is no gold standard, and that every standard is finally a policy decision. So the choice is yours to justify.

Where the judges have not watched the candidates work, Zieky and Perie (2006) suggest a question-based method such as Angoff, checked against real scores.

## How do you run an Angoff panel?

Run it as a scripted meeting: a defined target, estimates made alone, a discussion and a calculation agreed in advance. The steps follow Zieky and Perie (2006), simplified.

1. **Fix the paper.** The panel rates the questions candidates will sit. If you are building it in EdProcto from your syllabus, starting on the [home page](https://edprocto.com/), finish editing first. For the items, see [how to write multiple-choice questions that measure](https://edprocto.com/guides/how-to-write-multiple-choice-questions-that-measure).
2. **Choose the judges.** People who know the subject and what a newly certified person must do, drawn from across the membership.
3. **Agree the borderline description,** then have the judges sit the paper before they see the key. Seeing the answers first makes questions look easier than they are.
4. **Round one, alone.** For each question: of 100 borderline candidates, how many would get this right? On multiple choice the figure should not fall below chance, 25 with four options.
5. **Show and discuss.** Where the figures lie within about ten points, move on. Where they do not, a high judge and a low judge explain. Nobody is pressed to agree.
6. **Round two.** Judges may change any figure.
7. **Calculate.** Sum each judge's figures, then combine the sums by the mean, the median or a trimmed mean, whichever you announced beforehand.

Steps 3, 5 and 6 make it a modified Angoff, and the additions are not neutral. In the meta-analysis by Hurtz and Auerbach (2003), a shared definition of the minimally competent candidate brought judges closer together, discussion tended to raise the cut score, and showing judges how real candidates had scored tended to lower it. So fix the design before the meeting.

## A worked Angoff example with five judges

These are second-round figures from five judges for six four-option questions. A real paper is longer; the arithmetic is the same.

| Question | A | B | C | D | E |
| --- | --- | --- | --- | --- | --- |
| 1 | 85 | 90 | 80 | 90 | 85 |
| 2 | 70 | 75 | 70 | 65 | 80 |
| 3 | 60 | 60 | 60 | 70 | 55 |
| 4 | 55 | 60 | 50 | 65 | 55 |
| 5 | 60 | 65 | 55 | 65 | 65 |
| 6 | 50 | 55 | 45 | 60 | 50 |
| Sum | 380 | 405 | 360 | 415 | 390 |

Divide a sum by 100 and you have that judge's expected score for the borderline candidate: 3.80, 4.05, 3.60, 4.15 and 3.90 marks out of 6. Those add up to 19.50, and 19.50 divided by 5 is 3.90. The panel's cut score is 3.9 of 6, or 65%.

### Reading the result

**The spread.** The judges lie between 3.60 and 4.15, about half a mark apart, and the median is also 3.90. Report the spread with the mark.

**Rounding.** Nobody scores 3.9. If the rule is "at or above the cut score", a candidate needs 4 of 6. That step is a sixth of the scale, one reason a short paper makes a poor certification exam. Write the rule down before the sitting.

**The forecasts.** Question 6 averages 52: the panel expects 52 in 100 borderline candidates to get it right. Keep these averages for the review after the exam.

So the mark is 65%, and a harder paper would have produced a lower one.

## How many judges do you need, and what should you record?

Aim for about ten judges, more when the paper is short, and record enough for a stranger to rebuild the mark. Zieky and Perie (2006) say research suggests 10 to 15. Mortaz Hejri and Jalili (2014) report studies supporting as few as 5 and as many as 20, with most advice near 10. Shulruf et al. (2016), simulating 4,900 panels, found about 15 judges enough for 80 questions or more, and at least 20 needed for fewer than 15 questions. A federation with six qualified people should use six and report how far apart they were.

Record at least:

- why this method was chosen;
- who the judges were and how they were picked;
- the borderline description and the training given;
- every judge's figure in every round, and any data they were shown;
- the calculation, the rounding rule and the spread;
- any later adjustment, by whom and why.

The list draws on Zieky and Perie (2006) and on Standard 5.21 of the Standards for Educational and Psychological Testing (American Educational Research Association et al., 2014). Write it, the primer advises, for whoever must defend the study after everyone present has forgotten it.

## What should you do after the exam, before confirming the mark?

Review the questions that misbehaved before any result is released. The panel's figures were forecasts; the answers are evidence.

When Downing (2005) had raters sort the questions on four medical school examinations into sound and flawed, 36% to 65% of each paper was flawed, and pass rates on the two kinds differed by a median of 3.5 percentage points and by as much as 35.

The panel reads four things:

| Figure in EdProcto's [results](https://edprocto.com/#results) | Worth a second look when |
| --- | --- |
| Difficulty | Far fewer candidates answered correctly than the panel forecast |
| Discrimination and point-biserial | Near zero or negative: stronger candidates did no better than weaker ones |
| Takers of each wrong option, with their mean score | One wrong option drew the high scorers |
| Cronbach's alpha | Low for the whole paper, so scores near the mark deserve caution |

If the panel withdraws a question, its Angoff estimates come out of the sum too. Zieky and Perie (2006) also ask for a reality check: if the forecasts do not describe how borderline candidates performed, admit that the method may not have worked. Norcini (2003) adds that pass rates should bear a sensible relation to other signs of competence.

The committee with the authority then confirms the mark. Until it does, the panel's figure is provisional (Zieky and Perie, 2006).

## What makes a pass mark defensible?

A pass mark is defensible when the process behind it was sound and the record shows it. Nobody can prove a cut score correct, as Zieky and Perie (2006) say outright, so the defence is never that the number is right. It is who decided, how, and on what evidence.

Kane (1994) names three kinds of evidence: that the procedure suited the purpose, that the results are consistent within themselves, and that they agree with outside measures. Procedure alone is thin support (Mortaz Hejri and Jalili, 2014), so keep the judges' spread and any comparison with other measures.

### Explaining it to a failed candidate

Say what the standard is, who set it and how: "The pass mark was 65%. A panel of our members set it before the sitting, by the Angoff method, against a written description of the minimum standard. Your score was 62%."

Rules for certification and licensing differ by country and sector. This guide is not legal advice.

## How does EdProcto fit a certification exam?

EdProcto is the best tool for a body that sets its own examination because, the panel aside, a defensible mark rests on two things: a paper written from the body's own standard, and item figures reviewed before results go out. It does both.

Paste the syllabus or rulebook into the quiz maker and Procto AI writes the questions from it, in any of 111 [templates](https://edprocto.com/templates), each yours to edit. The Exam preset sets no going back, the score held until you release it, answers and explanations not shown, and every browser measure on. Each [integrity](https://edprocto.com/integrity) measure is switched on by itself, and every one is named to the candidate first.

The pass mark is one setting, from 0 to 100%. It starts at 60 as a placeholder, so replace it with your panel's figure. Marks are made on the server, and under the Exam preset scores wait for your release, which leaves room for the review. The analysis gives every figure in the table above. A certificate on passing carries a code anyone can check.

Free carries the core templates, marking by rule and results question by question. Tutor opens all 111 templates and medium integrity. Educator adds the full analysis and certificates. The Institution licence is by a call: see [pricing](https://edprocto.com/pricing) and the page [for schools](https://edprocto.com/schools). EdProcto is not affiliated with any board or federation, and using it does not make an examination accredited.

## Quick questions

### What is a good pass mark for a certification exam?

None in general. The right figure is the score a just-competent candidate would get on that paper, worked out by a trained panel. In the example above it was 65%.

### What is the Angoff method in simple terms?

Each judge estimates, for every question, how many of 100 just-competent candidates would answer correctly. A judge's estimates add up to an expected score, and the average across judges is the cut score.

### Can the pass mark change after the exam?

Before results are released, yes: for a stated reason such as a withdrawn question, by the committee with the authority. Moving it after seeing who failed turns an absolute standard into a relative one.

### Where does EdProcto come in?

EdProcto runs the paper, applies the passing score you set, reports the figures your panel reviews and issues a certificate on passing. The page [for associations and boards](https://edprocto.com/solutions/associations) shows a sitting from syllabus to certificate.

## References

- American Educational Research Association, American Psychological Association, & National Council on Measurement in Education. (2014). *Standards for educational and psychological testing*. American Educational Research Association. [https://www.testingstandards.net/open-access-files.html](https://www.testingstandards.net/open-access-files.html)
- Cohen-Schotanus, J., & van der Vleuten, C. P. M. (2010). A standard setting method with the best performing students as point of reference: Practical and affordable. *Medical Teacher, 32*(2), 154–160. [https://doi.org/10.3109/01421590903196979](https://doi.org/10.3109/01421590903196979)
- Downing, S. M. (2005). The effects of violating standard item writing principles on tests and students: The consequences of using flawed test items on achievement examinations in medical education. *Advances in Health Sciences Education, 10*(2), 133–143. [https://doi.org/10.1007/s10459-004-4019-5](https://doi.org/10.1007/s10459-004-4019-5)
- Downing, S. M., Tekian, A., & Yudkowsky, R. (2006). Procedures for establishing defensible absolute passing scores on performance examinations in health professions education. *Teaching and Learning in Medicine, 18*(1), 50–57. [https://doi.org/10.1207/s15328015tlm1801_11](https://doi.org/10.1207/s15328015tlm1801_11)
- Hurtz, G. M., & Auerbach, M. A. (2003). A meta-analysis of the effects of modifications to the Angoff method on cutoff scores and judgment consensus. *Educational and Psychological Measurement, 63*(4), 584–601. [https://doi.org/10.1177/0013164403251284](https://doi.org/10.1177/0013164403251284)
- Kane, M. (1994). Validating the performance standards associated with passing scores. *Review of Educational Research, 64*(3), 425–461. [https://doi.org/10.2307/1170678](https://doi.org/10.2307/1170678)
- Mortaz Hejri, S., & Jalili, M. (2014). Standard setting in medical education: Fundamental concepts and emerging challenges. *Medical Journal of the Islamic Republic of Iran, 28*, Article 34. [https://pmc.ncbi.nlm.nih.gov/articles/PMC4153516/](https://pmc.ncbi.nlm.nih.gov/articles/PMC4153516/)
- Norcini, J. J. (2003). Setting standards on educational tests. *Medical Education, 37*(5), 464–469. [https://doi.org/10.1046/j.1365-2923.2003.01495.x](https://doi.org/10.1046/j.1365-2923.2003.01495.x)
- Shulruf, B., Wilkinson, T., Weller, J., Jones, P., & Poole, P. (2016). Insights into the Angoff method: Results from a simulation study. *BMC Medical Education, 16*, Article 134. [https://doi.org/10.1186/s12909-016-0656-7](https://doi.org/10.1186/s12909-016-0656-7)
- Taylor, C. A. (2011). Development of a modified Cohen method of standard setting. *Medical Teacher, 33*(12), e678–e682. [https://doi.org/10.3109/0142159X.2011.611192](https://doi.org/10.3109/0142159X.2011.611192)
- Zieky, M., & Perie, M. (2006). *A primer on setting cut scores on tests of educational achievement*. Educational Testing Service. [https://www.ets.org/Media/Research/pdf/Cut_Scores_Primer.pdf](https://www.ets.org/Media/Research/pdf/Cut_Scores_Primer.pdf)