Negative marking takes points away for a wrong choice, so a student who guesses risks what a blank would have kept. It also goes by formula scoring and the correction for guessing; counting right answers alone is number-right scoring. This guide is for a teacher, lecturer or trainer deciding whether to switch the penalty on.
EdProcto, which publishes this guide, has "Negative marking" as one switch among a quiz's settings, off until you turn it on. A section below says exactly what it does to a multiple choice mark.
How does negative marking work?
Negative marking gives the question's points for the right choice, nothing for a blank, and takes a fraction away for a wrong choice. In the classic correction the fraction is one over the number of wrong options, which makes a blind guess worth nothing on average: with four options, one guess in four wins a point and three lose a third each.
The table is our arithmetic, not a finding, for a question worth one point; its last row assumes every question is answered.
| Options | 2 | 3 | 4 | 5 |
|---|---|---|---|---|
| A wrong choice costs | 1 | 1/2 | 1/3 | 1/4 |
| A guess between the last two options expects | 0 | +0.25 | +0.33 | +0.38 |
| Pass mark with no penalty that asks what 60% asks with one | 80% | 73% | 70% | 68% |
Row two carries the catch: the penalty cancels only the blind guess. A student who can rule out one option gains on average by answering, and one who hears only "do not guess" gives marks away.
What a penalised score means, and why it can fall below zero
A score under the penalty estimates the share of the questions a student knew, assuming every other answer was a blind guess. Real students also half know things and hold wrong beliefs, so it is only an estimate.
It falls below zero when wrong choices outnumber right ones by more than the number of wrong options: on four options, more than three wrong for each right. Bad luck is enough. By our arithmetic, an exact binomial sum, a student guessing blind on 20 four-option questions finishes below zero about 41 times in 100. A negative total is a normal result of the method.
Does negative marking stop guessing?
It reduces guessing, and it reduces answering along with it. Betts et al. (2009) compared psychology examinations with and without the correction, across two cohorts and three papers. Without it students scored higher and left fewer questions unanswered, and on one paper those told there was no correction did better than those told there was one (abstract read).
The blanks are not all ignorance. Goldik (2008), in the editorial that announced the change at the European anaesthesia diploma, reports that candidates were omitting a quarter of its true, false or don't know questions on average. He cites earlier studies in which candidates omitted questions they stood a better than chance prospect of getting right.
Vanderoost et al. (2018) followed 683 master's students of medicine at a Belgian university through two real examinations of five-option questions. Each year group sat one paper with a quarter of a mark off for a wrong answer and the other under a scheme for striking out wrong options. Under the penalty the share of students leaving at least one question blank was 14.8 to 50.0 points higher, and scores did not differ significantly.
Who does a penalty for wrong answers disadvantage?
The penalty costs most to the student who knows something and will not risk it. Whether that falls on one gender is disputed, and the field studies split.
- Pekkarinen (2015): in Finnish university entrance examinations in economics and business, where wrong answers lost points and omitted answers scored zero, women omitted more questions than men, and more than would have given them their best chance of a place (abstract read).
- Akyol et al. (2022): a model fitted to Turkey's university entrance examination found women and high-ability candidates significantly more risk-averse. The effect on scores was small, and the authors conclude that a stiffer penalty makes the examination more precise (abstract read).
- Funk and Perrone (2016): more than 500 undergraduates on a microeconomics course in Spain sat a final examination in which, by random allocation, some questions carried a penalty. Risk aversion changed how students answered but not their scores, and women did relatively better on the penalised part (abstract of a discussion paper).
On how the penalty feels, two studies that asked students agree. Among 136 first-year and 40 second-year life-science students at Swansea, Bond et al. (2013) found no significant difference between men and women under a penalty, and students reported more stress choosing one answer than striking out wrong ones. In the Belgian study, 63% of the quarter of students who answered a questionnaire said choosing one answer under the penalty felt risky, and women reported more stress than men under both schemes (Vanderoost et al., 2018).
Our guide to the don't know option covers more studies of who omits.
What does a penalty do to reliability and validity?
A penalty can make scores steadier and, in the same stroke, make them measure something besides knowledge.
Burton (2004) analysed 13 negatively marked true or false tests in which students showed considerable misinformation, wrong beliefs held as knowledge. He argues that the penalty improves reliability by charging for misinformation as well as by discouraging guessing (abstract read). Akyol et al. (2022) reach the same verdict on precision.
Validity is the cost. Once a blank is safe and a wrong answer is not, the score depends partly on a student's willingness to risk a mark, which no syllabus teaches. The don't know guide sets out the studies that put figures on this trade, and test reliability explains the figure itself. We found no comparison from a school classroom.
When everyone answers, the penalty only changes the scale
If every student answers every question, the penalised score is the number-right score stretched: on four options 70% right becomes 60%, 85% becomes 80%, and nobody changes place. Goldik (2008) makes the same point about examinations that only rank. The table's last row is that conversion.
Which examinations dropped the penalty, and why?
Three large examinations dropped a penalty between 2008 and 2016, and each gave its reason in public.
| Examination | Penalty ended | Reason given |
|---|---|---|
| European Diploma of Anaesthesiology and Intensive Care, Part I | October 2008 | A quarter of questions omitted, and a don't know answer hid partial knowledge (Goldik, 2008) |
| Advanced Placement, US | May 2011 | Redesigned courses with fewer, more complex questions, best scored on the number right, as the College Board told Jaschik (2010) |
| SAT, US | Spring 2016 | A move to "the simpler, more transparent model" of points for right answers only (College Board, 2014) |
The diploma cites what cautious candidates leave unanswered; the College Board cites simpler scoring. Until 2011 Advanced Placement had taken off a quarter of a point on five-option questions and a third on four-option ones (Jaschik, 2010), the classic correction.
What can you use instead of negative marking?
Five alternatives have been studied, and the simplest is to count right answers and move the pass mark.
- A higher pass mark. Add the chance share: 60% becomes 70% on four options. Setting a pass mark you can defend gives a method for the starting figure.
- More questions. Luck shrinks as a quiz grows: see how many questions a quiz needs and how many options a question should have.
- A confidence rating. Confidence-based marking charges only the wrong answer a student claimed to be sure of, and EdProcto's multiple choice with confidence shape does it.
- A don't know option, on a diagnostic where nothing rides on the score.
- Striking out wrong options for part marks. At Swansea students scored three to four points higher that way than under the single-answer penalty (Bond et al., 2013); in Belgium, under a stricter scoring rule, no higher (Vanderoost et al., 2018).
When is a penalty still justified, and how do you tell students?
A penalty is still defensible where blind guessing would do real damage and the students have practised under the rule.
| Your situation | Use a penalty? |
|---|---|
| Practice, homework, a weekly quiz | No. Count right answers |
| A class test with a pass mark | No. Raise the pass mark by the chance share |
| A long true or false test that decides something | Defensible: half of blind guesses land, and it charges misinformation too (Burton, 2004) |
| A competitive selection test, candidates trained in the rule | Defensible: precision gained, little bias found (Akyol et al., 2022) |
| Students cannot skip, or have never met the rule | No. With every question answered it only rescales |
| You want to know what students are sure of | No. In EdProcto, use the confidence shape with "Confidence bonus" on |
What to tell students
- Say before the day, and again at the start, that wrong answers lose points, in one line: "Right, 1. Blank, 0. Wrong, minus a third."
- Say when to answer: "If you can rule out even one option, answer. Leave a blank only if you have no idea."
- Never say only "do not guess". What students were told changed their scores in Betts et al. (2009), and Goldik (2008) reports an older study in which calling the penalty severe made candidates omit more.
- Run one practice quiz under the rule first. EdProcto states the rule on the start screen of any quiz with the penalty on.
A worked example: one class marked with and without the penalty
Five students in an invented class sit 20 four-option questions worth a point each, with a pass mark of 60%. The same answer sheets were marked twice by running EdProcto's marking function, with "Negative marking" off and on.
| Student | Right, wrong, blank | Penalty off | Penalty on |
|---|---|---|---|
| A | 16, 4, 0 | 80% | 73.4% |
| B | 12, 0, 8 | 60% | 60% |
| C | 12, 8, 0 | 60% | 46.8% |
| D | 9, 4, 7 | 45% | 38.4% |
| E | 4, 16, 0 | 20% | −6.4% |
- Nobody gains. The class average falls from 53% to 42.4%, and the passes from three to two.
- B and C tie without the penalty and sit 13 points apart with it. The score cannot say whether B knew more or only risked less.
- E finishes below zero: 4 points won and 16 × 0.33 lost. EdProcto rounds each mark to the hundredth, so E has −6.4% where the exact formula gives −6.7%.
- The sheets would differ under another rule. With no penalty B would answer all 20 and expect two more right: 70%, the pass mark the first table gives for four options.
How does EdProcto apply negative marking?
EdProcto applies the classic correction to every question answered by choosing one option, and to nothing else. It is the best tool for a penalty you can explain, because the rule is one switch with one formula, the same for every student, and the marks are made on the server (why it is built this way). The studies above show why that matters: what students are told decides how a penalty behaves.
- Where. In the quiz's settings, under "Grading": "Negative marking", with the hint "A wrong choice loses points: one over the number of wrong options." It starts off, is on every plan, and no preset changes it.
- Which questions. Any question answered by picking one option, whether the prompt is text, a picture or a clip: multiple choice, true or false and the yes, no or maybe shape among them. Typed answers, numbers, matching, ordering, sorting, best and worst, and a choice that asks for its reason are never penalised.
- How much. A wrong choice loses the question's points divided by its number of wrong options, rounded to the hundredth: on one point, 1, 0.5, 0.33 and 0.25 for two to five options. A blank scores zero.
- Below zero. Nothing holds a question's mark or a total at zero. In the results a negative total shows as a negative percentage beside the student and counts in the average and the median. "Score distribution" runs from 0 to 100, and a score below zero stands in its first bar. Each question's "Got it" counts a penalised answer as a miss and no worse.
- Multiple answers. The switch changes nothing on a multiple-answer question. There "Partial credit" sets wrong ticks against right ones and stops at zero.
- With "Confidence bonus". The two losses add: on four options a wrong answer rated 5 loses 0.33 and 0.6, so 0.93 of a point.
- The pass mark. "Pass mark", which starts at 60%, is compared with the penalised percentage.
- Skipping and clearing. While the penalty is on, "Allow skipping a question" is held on, through the Exam preset too, and greyed with the reason: a student can always move on from a blank. Pressing a chosen option again clears it, so a guess can be taken back.
- Telling students. With the penalty on, the start screen says: "Negative marking: a wrong choice loses points, one over the number of wrong options. A blank loses nothing, and pressing your choice again clears it."
Free takes thirty students a quiz, Tutor 100, Educator 500 and the full analysis; pricing has the figures. Whatever the rule, the score is yours to read and release.
Quick questions
What is negative marking in a multiple-choice test?
Negative marking takes points off for a wrong answer and nothing for a blank, classically one over the number of wrong options, so that blind guessing on multiple choice averages zero.
Should students guess when there is negative marking?
Yes, whenever they can rule out at least one option: the expected mark is then above zero. Only a guess with no idea at all is worth leaving blank.
Can a score be negative with negative marking?
Yes. By our arithmetic a blind guesser on 20 four-option questions finishes below zero about 41 times in 100. EdProcto shows such a total as a negative percentage in its results.
Is negative marking unfair to women?
The studies split. Pekkarinen (2015) found women omitting too many questions in Finnish entrance examinations; Funk and Perrone (2016) found no harm in a Spanish university examination.
References
- Akyol, P., Key, J., & Krishna, K. (2022). Hit or miss? Test taking behavior in multiple choice exams. Annals of Economics and Statistics, (147), 3–50. https://doi.org/10.2307/48684785
- Betts, L. R., Elder, T. J., Hartley, J., & Trueman, M. (2009). Does correction for guessing reduce students' performance on multiple-choice examinations? Yes? No? Sometimes? Assessment & Evaluation in Higher Education, 34(1), 1–15. https://doi.org/10.1080/02602930701773091
- Bond, A. E., Bodger, O., Skibinski, D. O. F., Jones, D. H., Restall, C. J., Dudley, E., & van Keulen, G. (2013). Negatively-marked MCQ assessments that reward partial knowledge do not introduce gender bias yet increase student performance and satisfaction and reduce anxiety. PLoS ONE, 8(2), Article e55956. https://doi.org/10.1371/journal.pone.0055956
- Burton, R. F. (2004). Multiple choice and true/false tests: Reliability measures and some implications of negative marking. Assessment & Evaluation in Higher Education, 29(5), 585–595. https://doi.org/10.1080/02602930410001689153
- College Board. (2014, March 5). The College Board announces bold plans to expand access to opportunity; redesign of the SAT [Press release]. https://newsroom.collegeboard.org/college-board-announces-bold-plans-expand-access-opportunity-redesign-sat
- Funk, P., & Perrone, H. (2016). Gender differences in academic performance: The role of negative marking in multiple-choice exams (CEPR Discussion Paper No. 11716). Centre for Economic Policy Research. https://ideas.repec.org/p/cpr/ceprdp/11716.html
- Goldik, Z. (2008). Abandoning negative marking. European Journal of Anaesthesiology, 25(5), 349–351. https://doi.org/10.1017/S0265021508003876
- Jaschik, S. (2010, August 9). AP eliminates guessing penalty. Inside Higher Ed. https://www.insidehighered.com/news/2010/08/10/ap-eliminates-guessing-penalty
- Pekkarinen, T. (2015). Gender differences in behaviour under competitive pressure: Evidence on omission patterns in university entrance examinations. Journal of Economic Behavior & Organization, 115, 94–110. https://doi.org/10.1016/j.jebo.2014.08.007
- Vanderoost, J., Janssen, R., Eggermont, J., Callens, R., & De Laet, T. (2018). Elimination testing with adapted scoring reduces guessing and anxiety in multiple-choice assessments, but does not increase grade average in comparison with negative marking. PLoS ONE, 13(10), Article e0203931. https://doi.org/10.1371/journal.pone.0203931