Every coaching centre has this argument at some point: should the test series run -0.25 per wrong answer, or -1? It usually gets settled by whichever entrance exam the batch is preparing for, and then nobody looks at it again.
The argument is worth having, but it is usually had about the wrong number. This article takes one batch of four students, keeps every answer identical, and changes only the penalty. The batch average moves 7.5 marks, the number of students passing falls from three to one, and one student drops from second place to last without answering a single question differently.
-0.25 and -1 can be exactly the same penalty
A penalty on its own means nothing until you know what the question was worth. What actually governs the effect is the ratio between them.
| Scheme | Marks per question | Deducted per wrong | Penalty ratio |
|---|---|---|---|
| Common school MCQ paper | 1 | 0.25 | 25% |
| Common entrance pattern | 4 | 1 | 25% |
| Half-mark penalty | 2 | 0.5 | 25% |
| A quarter of a 4-mark question | 4 | 0.25 | 6.25% |
The first three rows are the same policy written three ways. The fourth is a genuinely different one.
So a centre that switches from "-0.25" on a 1-mark paper to "-1" on a 4-mark paper has changed nothing at all, and a centre that keeps "-0.25" while moving from 1-mark to 4-mark questions has quietly made its papers four times more forgiving. Both of those happen, and both are invisible in the sentence people actually say out loud, which is "we do minus point two five".
For the rest of this article the paper is 30 questions, 4 marks each, 120 total - and the penalty is quoted as a percentage of the question, because that is the number that carries meaning between papers.
One batch, four students
Four students, one test series paper. What separates them is not only how much they know but how much they are willing to attempt.
| Student | Correct | Wrong | Left blank | Approach |
|---|---|---|---|---|
| Aarav | 18 | 12 | 0 | Attempts everything, strong |
| Kabir | 16 | 14 | 0 | Attempts everything, guesses freely |
| Diya | 15 | 3 | 12 | Selective, high accuracy |
| Meera | 13 | 1 | 16 | Very cautious |
Note that Kabir knows more than Diya - 16 correct against 15 - and Diya knows more than Meera. Hold on to that, because the ranking is about to stop agreeing with it.
The same answers under three penalties
Nobody changes an answer. Only the setting changes. Passing is 50%, so 60 out of 120.
| Student | No penalty | -0.25 (6.25%) | -1 (25%) |
|---|---|---|---|
| Aarav | 72.00 | 69.00 | 60.00 |
| Kabir | 64.00 | 60.50 | 50.00 |
| Diya | 60.00 | 59.25 | 57.00 |
| Meera | 52.00 | 51.75 | 51.00 |
| Batch average | 62.00 | 60.13 | 54.50 |
| Students passing | 3 of 4 | 2 of 4 | 1 of 4 |
Two things are worth separating here, because they get confused constantly.
The average falling is not a finding. Going from 62 to 54.5 tells you nothing about the batch - you subtracted marks, so the total went down. If you switch penalty mid-series and then compare this month's average to last month's, you are looking at your own settings change, not at your students. That comparison is only meaningful when the penalty is held constant across the papers being compared.
The pass rate collapsing is a finding. Three students passing became one, and if your batch reporting is built on "how many cleared the cut-off", a penalty change rewrites that number for reasons that have nothing to do with preparation.
The batch did not change. The ranking did.
This is the part that actually matters, and it is invisible in the average.
| Rank | No penalty | -0.25 (6.25%) | -1 (25%) |
|---|---|---|---|
| 1 | Aarav | Aarav | Aarav |
| 2 | Kabir | Kabir | Diya |
| 3 | Diya | Diya | Meera |
| 4 | Meera | Meera | Kabir |
Kabir goes from second to last. He is not a weaker student at -1 than he was at zero; he answered exactly the same questions exactly the same way. Meera, who left more than half the paper blank and got 13 questions right, now finishes above him.
That is the real decision hiding inside the penalty setting. At a low ratio you are mostly measuring how much a student knows. At a high ratio you are also measuring how well they judge what they do not know. Both are legitimate things to test - selection exams deliberately test the second - but they are different things, and a centre should know which one its rank list is reporting.
It is also the reason a rank list is not comparable across papers with different penalties. Two mock tests, same syllabus, same batch, different setting: the rank order is not measuring the same quantity twice.
The usual -1 still rewards guessing
The stated purpose of negative marking is to stop blind guessing. It is worth checking whether the common setting does that.
On a 4-option question, a blind guess is right one time in four. On a 4-mark question with a 25% penalty, that is 4 marks a quarter of the time and 1 mark lost three quarters of the time:
Positive. A student who guesses every question they cannot do gains a quarter of a mark each time, on average. The standard scheme discourages guessing without actually making it a bad idea - and a student who has worked this out has a real edge over one who has been told to leave doubtful questions blank.
The penalty at which a blind guess is genuinely neutral:
For a 4-mark, 4-option question that is 4 ÷ 3 = 1.33 marks, or 33.3% - a third of the question, not a quarter. For a 5-option question it is a quarter. For a true or false question it is the whole mark, which is why penalising two-option questions rarely works.
| Penalty | Value of one blind guess | Effect |
|---|---|---|
| None | +1.00 | Guessing is strictly correct |
| 6.25% (-0.25 of 4) | +0.81 | Guessing is still clearly worth it |
| 25% (-1 of 4) | +0.25 | Guessing still pays, slightly |
| 33.3% (-1.33 of 4) | 0.00 | Guessing is neutral |
None of which means 25% is the wrong choice. If you are running a series for an exam that uses 25%, you should use 25%, because the job is to rehearse the real paper. It means that "we use negative marking so students do not guess" is not accurate, and if eliminating the value of guessing is the actual goal, the number that does it is a third.
One caveat on the arithmetic: this is the value of a genuinely blind guess. A student who has eliminated one option is choosing between three, and the sums move in their favour immediately. Educated guessing survives every one of these schemes, which is usually the intent.
Two things people assume that are not true
A blank answer is not a wrong answer
A penalty applies to a question the student answered incorrectly. It does not apply to one they left alone. Meera left 16 questions blank and lost nothing for them; her only deduction came from the one question she attempted and got wrong.
This sounds obvious written down, and students get it wrong constantly - which is a coaching problem, not a software problem. Before a test series that introduces a penalty, it is worth stating explicitly that skipping is free, because a student who believes otherwise will attempt everything defensively and end up exactly where Kabir did.
The score stops at zero
A student who gets 1 right and 29 wrong on this paper has earned 4 marks and been penalised 29, which is -25. What gets recorded is 0.
That floor is the right behaviour - a negative mark on a report card is not meaningful to a parent - but be aware of what it does to your data. Everyone below the floor is stored as the same number, so the bottom of your distribution is compressed, and a batch average that includes several floored papers understates how far behind those students actually are. If you are tracking batch averages over a series, it is worth knowing how many zeros are real zeros.
Choosing a number
Three questions, in order:
- Is this paper rehearsing a specific exam? If yes, the decision is already made - copy that exam's ratio exactly, including how it treats blanks. Everything below is irrelevant. Match the pattern rather than improving on it.
- What do you want the rank list to reward? Knowledge alone points to a low ratio or none. Knowledge plus judgement about what to attempt points to 25% or more. There is no neutral setting, so pick deliberately.
- Will you hold it constant for the whole series? This matters more than the value you pick. A consistent 25% across ten papers gives you a comparable trend. Changing it at paper six means papers one to five cannot be compared with six to ten, and no analysis will warn you.
Then write the number on the paper, in the ratio form - "4 marks per question, 1 mark deducted per wrong answer, blanks not penalised". Students who know the exact terms make better decisions during the exam, and that is a skill the series is supposed to be building.
Set it once, keep it constant
The penalty is a small setting with a long reach. It moves the average, it moves the pass rate, and it quietly reorders your rank list in a way that no report will flag, because every number involved still looks perfectly reasonable.
Teemly stores negative marking as a percentage of each question's marks rather than as a fixed deduction, which is the same thing this article argues for: reweight a question from 4 marks to 8 and the penalty scales with it, so the policy survives an edit to the paper. Blanks are never penalised, and the recorded score floors at zero. If you are working through how test results roll up more broadly, our guide to calculating a term percentage when tests have different total marks covers the arithmetic downstream of this one, and choosing an online test platform for coaching centres covers what else to look for. You can see how the settings work on our assessment software for coaching institutes page.