GATE negative marking: when guessing an MCQ is worth it

GATE has three types of question, and only one of them can cost you marks.

No question type gives partial marks. As the GATE 2027 information brochure puts it: “There is no partial marking for any question.” An unanswered question always scores zero.

Pick the question type and how many options you can rule out. The calculator compares the marks you can expect from answering with the zero you get for leaving it blank.

For an MCQ worth m marks, if your chance of being right is p, the expected marks from answering are p × m − (1 − p) × m/3, which simplifies to m(4p − 1)/3. That is zero when p is exactly 1/4, positive above it and negative below it. With four options, any option you can rule out pushes you above 1/4.

Averages hide the risk on a single question: one blind guess on a 2-mark MCQ either gains 2 marks or loses 2/3. Over many such questions the gains and losses even out, which is why the expected value is the right guide for the paper as a whole. The catch is honesty about p: if you feel 50% sure but are usually wrong in that situation, your real chance is lower. Your practice reports show how often your guesses land.

For MSQ and NAT the expected value can never be negative, so an answer always beats a blank. A completely blind MSQ answer has a 1 in 15 chance (four options give 15 possible non-empty combinations), worth 1/15 of the marks on average, which is still more than zero.

Counted across the 71 official papers in our archive: since MSQ arrived in 2021, an average of 50.1 marks per paper (between 38 and 70) have been MCQs, 13.1 MSQ and 36.7 NAT. If every MCQ in a paper were answered wrong, the penalty would average 16.7 marks. Before 2021 MCQs took a larger share: in the CS papers from 2014 to 2019 they averaged 63.5 marks.

The mix differs a lot by subject. Electronics and Electrical papers lean on MCQs; Mechanical, Aerospace and Civil papers have the most NAT questions; Computer Science and Data Science have the most MSQs.

GATE 2024 CS Set 2, Q37 is a 2-mark MCQ. Rule out options below and watch the value of a guess change.

The range 10.12.2.0 to 10.12.3.255 holds 512 addresses. Option B, a /24, holds only 256. Option C covers 10.12.0.0 to 10.12.3.255, which is twice too large. Option D is not a valid /22 block, because a /22 must start where the third number is a multiple of 4. That leaves (A), 10.12.2.0/23, the official answer. Even if you only knew that a /24 holds 256 addresses, ruling out B alone makes a guess worth +0.22 marks on average. See the whole paper

The marking rules are from the GATE 2027 Information Brochure (IIT Madras), section 7.2.1, pages 22 and 23, and have been the same in every paper since MSQs were introduced in 2021. The question-type figures count every question in the 71 official papers in the Lemyte archive, at its official marks: CS papers from 2014, other subjects from 2020 (Data Science from 2024, when the paper began). Each paper adds up to 100 marks. “Most MCQs could cost” assumes every MCQ in a paper is answered wrong; it is a ceiling, not a typical loss.

In the exam: answer every MSQ and NAT question, even with a rough estimate. For an MCQ, guess when you can rule out at least one option. With nothing ruled out, a guess and a blank are worth the same on average, so spend the time on questions you can solve. Before the exam: practise under real marking, then check how many marks your report shows lost to negative marks, and whether your “50% sure” guesses are really right half the time.