Negative marking: a calm way to think about guessing versus skipping

You're halfway through a paper. A question has four options. You've read it twice. Two options look wrong, two look plausible, and you genuinely don't know which is right. The clock is running, and a voice in your head says if I'm wrong I lose marks, so leave it. Another voice says but I might be right, and every mark counts.

Most of us settle this by mood. If the last three guesses went badly, we skip. If we're feeling lucky, we guess. That's understandable, and it's also a very noisy way to treat something worth many marks over a whole paper.

There is a calm, almost boring way to think about it, and it comes from one idea: expected value. This article walks through it with a worked example, then covers the two practical skills that matter more than the maths: eliminating options and calibrating how sure you really are.

First, a warning about the real scheme

Everything numerical below uses a hypothetical scoring scheme, chosen only to make the arithmetic clear: +4 for a correct answer, -1 for a wrong answer, 0 for an unanswered question, with four options per question.

That is an example, not a statement of how the exam is marked. The marking scheme, the number of options, whether unanswered questions score zero, and any other rules are set by the examination authority and may change. Check the current official notices from NBEMS and NMC for the real scheme before you build any strategy on it. If you hear a scheme described in a group chat or on social media, treat it as unconfirmed until you've seen it in an official document.

The good news is that the method works for any scheme. You just replace the numbers.

What expected value means

Expected value is the average result you'd get if you could make the same decision many, many times. It doesn't tell you what will happen on one question. It tells you which choice is better on average, which is what matters when you're making dozens of such decisions in one paper.

For a guess, the expected value is:

(chance you're right x marks for right) + (chance you're wrong x marks for wrong)

Skipping has an expected value of zero in our hypothetical scheme. So the question becomes: is the expected value of guessing above or below zero?

A worked example

Use our hypothetical +4 / -1 scheme with four options.

Case 1: a pure guess, nothing eliminated. Chance of being right: 1 in 4, or 0.25. Chance of being wrong: 0.75. Expected value = (0.25 x 4) + (0.75 x -1) = 1.00 - 0.75 = +0.25.

So under this particular scheme, even a blind guess is slightly positive on average. Not by much, and the swings are large, but the sign is positive.

Case 2: one option eliminated with confidence. Three options remain, so the chance of being right is 1 in 3, about 0.33. Expected value = (0.33 x 4) + (0.67 x -1) = 1.33 - 0.67 = about +0.67.

Case 3: two options eliminated with confidence. Two remain, so the chance is 0.5. Expected value = (0.5 x 4) + (0.5 x -1) = 2.00 - 0.50 = +1.50.

Case 4: you think you're 90% sure. Expected value = (0.9 x 4) + (0.1 x -1) = 3.60 - 0.10 = +3.50.

Now the useful part, the break-even point. For any scheme, guessing is better than skipping when your chance of being right is above:

penalty / (reward + penalty)

In the hypothetical scheme that's 1 / (4 + 1) = 0.2, or 20%. If you honestly think you have more than a one-in-five chance, guessing is favourable on average. If the real scheme has a different reward or penalty, the break-even moves. A heavier penalty makes it higher. A lighter one makes it lower. Plug in the real numbers once you've read the official notice, and keep the result on a sticky note.

Why "it's positive on average" isn't the whole story

It would be dishonest to stop here and say "always guess". Three caveats are worth taking seriously.

Averages need many trials. An expected value of +0.25 per blind guess means that over a hundred blind guesses you'd expect to gain about 25 marks, but on any handful you may well lose. If you only have a few such questions, luck dominates.

Your probability is a guess too. The calculation only works if the chance you assign is reasonable. If you tell yourself you've eliminated two options but one of your eliminations was wrong, your real chance is far below 50%. This is the single biggest source of error in guessing strategy, and it's why calibration matters (more on that below).

The cost of time is real. Agonising for three minutes over a coin-flip question to gain a fraction of a mark on average is a bad trade. If the maths says "slightly positive", don't spend long on it. Make a choice and move on.

So the sensible reading is: under a scheme like our example, guessing between a smaller number of options is clearly favourable on average, a blind guess is marginally so, and the real danger isn't the penalty itself. It's overestimating your own certainty.

Elimination: how to turn a guess into an informed answer

The way to raise your chance of being right is to remove options you can justify removing. A few ways to do it honestly.

Remove what's definitely wrong first. Read each option and ask whether you know a reason it can't be right. A drug that's contraindicated in the stated condition. An organism that doesn't cause that infection. A value that's physiologically impossible. Cross it off, and write down the reason in your head in one phrase. If you can't state a reason, don't eliminate it.

Look for the pair. Often two options are close to each other, and one of them has to be right if the question is well written. Decide what feature distinguishes the pair, then check which feature the stem mentions.

Check the stem for the odd word. Words like "except", "not", "most likely" and "first" change what you're looking for. A surprising number of wrong answers come from answering the question you expected, not the one asked.

Use the extremes with care. Options containing "always" or "never" are sometimes wrong in medicine, because biology has exceptions. But this is a tendency, not a rule. Treat it as a nudge, not a proof.

Use the clinical picture, not the keyword. One buzzword in the stem can point you to a single option while the rest of the picture points elsewhere. Check that the whole scenario fits.

Don't eliminate by pattern. Don't reason that "the last three were B, so this isn't B", or that the longest option is usually correct. Answer patterns in a paper aren't something you can rely on, and trying to exploit them will cost you marks.

The aim is not to trick the paper. It's to use what you partly know. A partial knowledge guess, where you've legitimately removed one or two options, is a very different thing from a blind one.

Calibration: knowing how sure you really are

Calibration means your feeling of confidence matches how often you turn out to be right. Someone who says "90% sure" and is right nine times out of ten is well calibrated. Someone who says "90% sure" and is right six times out of ten is overconfident, and under a penalty scheme that's the person who gets hurt.

Most of us are not well calibrated by default. We tend to feel more certain about things we've read recently, and about options that sound familiar. The fix is practice, with a simple record.

Step 1: label every answer in practice. When you do practice questions, mark each answer with a confidence label before you see the result. Use three: sure, leaning, and guess. Resist the urge to use "sure" for everything.

Step 2: tally after each set. For each label, count how many were right. You'll get a small table: of the "sure" answers, how many were correct? Of the "leaning" ones? Of the "guesses"?

Step 3: read the pattern. If your "sure" answers are right nearly every time, great: trust that label. If your "sure" answers are wrong a lot, you're overconfident, and you should treat "sure" as "leaning". If your "guesses" are right more often than you expect, you know more than you give yourself credit for, and you may be skipping too much.

Step 4: set your own rule. Based on your tally, decide which labels you'll answer and which you'll skip, once you know the real scheme. Your rule is based on your own data, not on mood.

Do this across several sets before you trust it, since a tiny sample proves nothing. A handful of questions in one label can swing wildly. You need a reasonable number of attempts before the percentages mean anything.

You can run this on any source of questions. If you'd like somewhere to do it, NextPrep has free practice questions with explanations of every option, and you can note your confidence in your own tally sheet next to them. The method is what matters, not the tool.

Common mistakes

Skipping everything you're not certain about. Under a scheme like our example, that gives away expected marks. Certainty isn't the bar. Breaking even is.

Guessing everything without thinking. Blind guessing is only marginally favourable under a mild penalty, and under a harsher one it can be unfavourable. It also trains you to stop reading carefully.

Changing answers late out of anxiety. Revisit an answer if you've noticed a specific reason, such as a misread word or a fact you've now recalled. Don't change it just because you feel nervous. Your first answer isn't always right, but a change driven by a concrete reason is usually better than one driven by a feeling.

Letting one wrong guess affect the next decision. Each question is a separate decision. A wrong answer a minute ago says nothing about the next one. This is the gambler's mistake, and a few stressed minutes after a loss can produce a run of bad choices.

Treating the maths as a formula for the whole paper. The calculation helps with the grey-zone questions. It doesn't replace knowing the subject.

A simple in-exam routine

Having a routine means you decide in seconds, not minutes.

1. Read the stem and the question line. Know what is being asked.
2. Answer in your head before looking at the options, if you can. If an option matches, take it.
3. If not sure, eliminate with stated reasons.
4. Count what's left and label your confidence: sure, leaning, guess.
5. Apply your rule. If the real scheme is one where your break-even is low and you have two or three options left, you'll usually answer. If you have no basis at all, decide ahead of time what you do with those questions and stick to it.
6. Mark and move on. If you want to return, note it, but don't let one question eat the time of five.

Rehearse this in your mocks. By the time you sit the real paper it should feel automatic.

The short version

• Expected value is the average result of a decision repeated many times: chance right x reward + chance wrong x penalty.
• The break-even chance is penalty divided by (reward plus penalty). In our hypothetical +4 / -1 example it is 20%. Check the real scheme in the official notice and recompute.
• Averages need many trials, and your own probability estimates may be too generous.
• Raise your chance by eliminating options you can give a reason for.
• Calibrate: label confidence in practice, tally, and trust your own data.
• Decide your policy in advance, and don't let one wrong guess steer the next.

What to do this week

1. Find the official notice from NBEMS or NMC for the marking scheme and write it down, with the date you checked it. Compute your own break-even point from it.
2. In your next two practice sets, label every answer as sure, leaning or guess before checking.
3. Make the tally table: how many of each label were right? Write down what it says about you.
4. Practise elimination out loud on five questions you got wrong, naming the reason you could have removed each option.
5. Write a one-line end-of-paper policy for what you'll do with leftover questions, and test it in your next timed mock.