What is CAT's exact negative marking scheme?

+3 marks for a correct MCQ answer, -1 for an incorrect MCQ answer, and 0 for a question left unattempted. Non-MCQ (Type-In-The-Answer) questions carry no negative marking at all - a wrong answer there costs nothing beyond the time spent on it. This applies uniformly across VARC, DILR and QA.

Is a blind guess worth attempting on CAT?

On a standard four-option MCQ, a genuinely blind guess has an expected value of exactly zero - neither a gain nor a loss in theory. In practice, treat it as a reason to skip rather than attempt, because real wrong options are designed to attract a specific mistake, which pushes actual blind-guess accuracy below the true-random 25% the zero assumes.

Why is CAT's guessing math different from NEET's?

CAT's +3/-1 scheme puts the breakeven point for a blind guess at exactly four options, which is CAT's own standard MCQ format - so a blind guess is worth zero. NEET's +4/-1 scheme puts breakeven at five options, one more than NEET's own four-option format, which is why a blind NEET guess carries a small positive expected value of +0.25 that CAT's does not.

Should I attempt a CAT MCQ if I can eliminate just one option?

Yes. Eliminating one of four options moves the expected value from a breakeven zero to +0.33, which is comfortably worth attempting. Eliminating two options moves it to +1.00 - a full mark of expected value from the same question purely through partial knowledge.

What if I can eliminate two options out of four?

The expected value of guessing between the remaining two is +1.00, using CAT's +3/-1 scheme. This is one of the strongest positive-EV situations in the whole framework and should be attempted without hesitation.

Should I ever skip a CAT Non-MCQ (TITA) question I have attempted to solve?

Only if you have genuinely no answer to write down. Since Non-MCQ questions carry no negative marking, any derived answer - even from an uncertain or partially completed calculation - has non-negative expected value. The only real cost of attempting is the time already spent working through it.

Does eliminating options always help on CAT?

Yes, monotonically. Each option eliminated from the guessing pool raises the expected value of the remaining guess, from a breakeven zero with no elimination up to a certain +3 once every wrong option has been ruled out. There is no elimination level at which attempting becomes worse than skipping.

What happens if a CAT question has five or six options instead of four?

The expected value of a blind guess turns negative - -0.20 at five options and -0.33 at six, using the same +3/-1 formula. This is uncommon in CAT's standard format but worth knowing: the breakeven-at-four-options result is specific to CAT's typical question format, not a universal constant.

Why might a "zero expected value" blind guess actually lose me marks?

Because the zero assumes your wrong guesses are spread randomly across every wrong option. Real distractors are written to catch a specific, common mistake, so a test-taker's actual blind-guess accuracy is often below the true-random 25% the zero assumes. At 15% accuracy, for example, each such guess costs about -0.4 marks in expectation rather than zero.

How is the expected-value formula for guessing derived?

For k equally likely remaining options under CAT's +3/-1 scheme, EV = 3 × (1/k) − 1 × ((k−1)/k), which simplifies to (4 − k)/k. Setting this to zero and solving gives k = 4, the breakeven point - which happens to match CAT's own standard four-option MCQ format exactly.

Does Karma Yogi track guessing accuracy specifically?

Not at that level of detail. Logged mocks store each section's score, attempts and correct count, which is enough to compute overall accuracy per section per mock, but nothing in the current data model distinguishes a confidently answered question from a partially eliminated guess or a true blind guess.

Should this expected-value math change how I approach VARC differently from QA?

The math itself is section-agnostic - it only depends on how many options you can rule out, not which section the question is in. What differs by section is how reliably you can actually eliminate options: a QA calculation error is often easier to rule out definitively than a VARC inference option built to sound plausible, which is a preparation question rather than a math one.

The Journal

CAT Preparation

10 September 2026
9 min read

Anish Guruvelli

Karma Yogi

CAT Negative Marking Strategy: The Real Expected-Value Math

CAT's +3/-1 marking means a genuinely blind guess on a standard four-option MCQ has an expected value of exactly zero - not the small positive edge NEET's +4/-1 gives you. Eliminate even one option and the calculus flips hard in your favour. Here is the arithmetic worked properly, section by section.

On a standard four-option CAT MCQ, a genuinely blind guess is worth exactly zero marks in expectation - not a small positive edge, unlike NEET's +4/-1 scheme. CAT awards +3 for a correct MCQ answer, -1 for an incorrect one, and nothing at all for skipping - and, distinctly, carries no negative marking on Non-MCQ (TITA) questions. Run the arithmetic on a pure guess among four options: a 25% chance of +3 and a 75% chance of -1 gives (0.25 × 3) + (0.75 × -1) = 0.75 - 0.75 = 0.00. That is a genuinely different number from NEET's own well-known +0.25, and the difference is entirely down to CAT's 3:1 penalty ratio against NEET's 4:1.

What follows is the full expected-value table by number of eliminated options, why CAT's marking scheme happens to make a blind guess exactly breakeven rather than positive or negative, how fragile that theoretical number is once real distractors are involved, and how the calculation is completely different for CAT's Non-MCQ questions.

CAT's marking scheme, precisely

Every section - VARC, DILR and QA - carries the same rule: +3 for a correct MCQ, -1 for an incorrect MCQ, 0 for an unattempted question, and no penalty at all on Non-MCQ (Type-In-The-Answer) questions, where you key in a numeric or text answer rather than choosing from options. The full pattern is in this site's pattern and syllabus post; this piece is specifically about what that scheme implies for the attempt-or-skip decision.

The general formula

For a question with k remaining options you genuinely cannot distinguish between, the expected value of guessing among them under CAT's +3/-1 scheme is:

EV(k) = 3 × (1/k) − 1 × ((k−1)/k) = (4 − k) / k

Solve for the breakeven point - where EV = 0 - and you get k = 4. That is not a coincidence worth glossing over: CAT's standard MCQ carries exactly four options, so the marking scheme is calibrated so that a truly uninformed guess on the typical question is worth precisely nothing, neither a gift nor a trap. Compare that to NEET's +4/-1 scheme, where the same algebra (EV(k) = (5 − k)/k) puts breakeven at k = 5 - one option higher than NEET's own actual four-option format carries, which is exactly why a blind NEET guess has a real, if small, positive edge of +0.25. CAT and NEET look like near-identical marking schemes at a glance. They are not: the one-mark difference in the penalty changes a blind guess from a genuine coin-flip-neutral bet to a small structural advantage.

Expected value by options eliminated

Options remaining (k)	What this represents	CAT EV (+3/-1)	NEET EV (+4/-1), for comparison
1	You know the answer - not a guess	+3.00	+4.00
2	Eliminated 2 of 4	+1.00	+2.00
3	Eliminated 1 of 4	+0.33	+1.00
4	No elimination - blind guess on a standard MCQ	0.00 (breakeven)	+0.25
5	A rare 5-option question	-0.20	0.00 (breakeven)
6	A rare 6-option question	-0.33	-0.17

Read the k = 4 row as the headline finding: CAT does not reward pure guessing the way NEET does, even marginally. It also does not punish it. Every other row shows why elimination is worth doing even when you can't get all the way to a confident answer: removing a single option out of four takes you from a breakeven bet to a genuinely good one (+0.33), and removing two takes you to a very good one (+1.00) - a full mark of expected value for the same question, purely from partial knowledge.

Why the theoretical zero is optimistic, not pessimistic

The NEET negative-marking analysis on this site makes an important point that applies here with even more force: the theoretical EV assumes your wrong answers are randomly distributed across the wrong options, and a well-written distractor is specifically designed to make sure they are not. A CAT VARC inference option that repeats the passage's exact wording with one word swapped, or a QA option that matches a calculation one step short of completion, is built to catch a specific, common wrong instinct - not to sit there as random noise.

That matters more for CAT than for NEET, precisely because CAT's breakeven point is zero rather than positive. If your blind-guess accuracy on truly uncertain questions runs below the true-random 25% - say, because you're consistently drawn to the same well-designed trap - the arithmetic turns actively negative: EV = 3p − (1 − p) = 4p − 1, which is negative for any accuracy below 25%. At an accuracy of 15%, each such guess costs 4(0.15) − 1 = -0.4 marks in expectation. Run that across just five genuinely blind attempts in a section and the theoretical breakeven of 0 becomes a real loss of 2 marks - a meaningfully worse outcome than simply skipping all five.

A practical decision framework

Eliminated at least one of four options? Attempt it. Even a single elimination (+0.33) is comfortably worth the risk, and two eliminations (+1.00) should be attempted without hesitation.
Eliminated nothing - a genuine blind guess on a four-option question? Skip it, despite the theoretical breakeven. Zero expected value under an assumption of true randomness is not a reason to attempt when your actual wrong-answer pattern is unlikely to be random, for the reason above.
Non-MCQ (TITA) question with any derived answer, however rough? Always attempt it. There is no penalty for a wrong Non-MCQ answer, so the only cost of attempting is the time already spent - the expected value of writing down any calculated answer is never negative, unlike an MCQ guess.

That third point is worth stating plainly because it is a different kind of decision from the first two, not just a smaller version of it. MCQ elimination is a probability calculation; Non-MCQ attempt is a pure time-cost calculation, since there is no options list to eliminate from and no downside to a wrong numeric answer beyond the minutes already spent computing it.

What this site's own mock data can and cannot show

Honestly: this piece was written without a live connection to any Karma Yogi account's mock data, so no real guessing-accuracy breakdown is presented here as evidence. It is also worth being precise about what the product could and could not show even with a connection - a logged mock stores each section's score, attempted count, correct count and percentile, which is enough to compute overall accuracy (correct ÷ attempted) per section per mock, but nothing in that schema currently distinguishes a question answered from genuine certainty, from one answered via partial elimination, from one answered as a true blind guess. Accuracy trending up or down across mocks is a real, trackable signal; isolating specifically how well your blind guesses are doing is not something the current data model captures, and this post does not pretend otherwise.

Where this is weak

The model assumes exactly four options and a single fixed marking scheme. A rare question with a different option count, or a marking-scheme change in a future CAT cycle, changes every number in the table above - always confirm the current year's scheme against the official notification before relying on it.
"Blind guess" and "true randomness" are theoretical constructs. No study measured actual CAT candidates' guess accuracy by elimination count for this piece; the 15%-accuracy illustration above is a plausible scenario used to show sensitivity, not a measured figure.
This is pure expected-value reasoning, which is about the long run, not any single question. A positive expected value does not guarantee a correct answer on the one question in front of you - it describes what happens on average across many similar decisions.
No live mock data was available for this piece, as stated above. Everything here is mechanics, not measurement of an actual candidate's guessing behaviour.
The framework says nothing about time cost, beyond the Non-MCQ point above. Deciding whether a question is worth the seconds needed to even attempt elimination is a separate, real constraint this post does not model.

Sources

CAT marking scheme (+3 correct MCQ, -1 incorrect MCQ, no penalty on Non-MCQ, uniform across VARC/DILR/QA) as established in this site's pattern and syllabus post and the 3-month vs 6-month study plan post.
NEET's +4/-1 marking scheme and its own expected-value analysis, for comparison, from this site's NEET negative-marking post.
All expected-value figures above are original arithmetic for this post, derived from the stated marking schemes rather than drawn from any external source.

Turning "eliminate at least one option" into an instant, in-exam instinct takes the same thing every skill in this framework needs - real timed repetition, tracked afterward rather than guessed at. Karma Yogi logs full CAT mocks with per-section scores, attempts and accuracy over time, so a rising or falling attempt-versus-accuracy trend is visible rather than assumed. Start tracking free.

End of essay

- Anish Guruvelli

Common questions

What is CAT's exact negative marking scheme?: +3 marks for a correct MCQ answer, -1 for an incorrect MCQ answer, and 0 for a question left unattempted. Non-MCQ (Type-In-The-Answer) questions carry no negative marking at all - a wrong answer there costs nothing beyond the time spent on it. This applies uniformly across VARC, DILR and QA.
Is a blind guess worth attempting on CAT?: On a standard four-option MCQ, a genuinely blind guess has an expected value of exactly zero - neither a gain nor a loss in theory. In practice, treat it as a reason to skip rather than attempt, because real wrong options are designed to attract a specific mistake, which pushes actual blind-guess accuracy below the true-random 25% the zero assumes.
Why is CAT's guessing math different from NEET's?: CAT's +3/-1 scheme puts the breakeven point for a blind guess at exactly four options, which is CAT's own standard MCQ format - so a blind guess is worth zero. NEET's +4/-1 scheme puts breakeven at five options, one more than NEET's own four-option format, which is why a blind NEET guess carries a small positive expected value of +0.25 that CAT's does not.
Should I attempt a CAT MCQ if I can eliminate just one option?: Yes. Eliminating one of four options moves the expected value from a breakeven zero to +0.33, which is comfortably worth attempting. Eliminating two options moves it to +1.00 - a full mark of expected value from the same question purely through partial knowledge.
What if I can eliminate two options out of four?: The expected value of guessing between the remaining two is +1.00, using CAT's +3/-1 scheme. This is one of the strongest positive-EV situations in the whole framework and should be attempted without hesitation.
Should I ever skip a CAT Non-MCQ (TITA) question I have attempted to solve?: Only if you have genuinely no answer to write down. Since Non-MCQ questions carry no negative marking, any derived answer - even from an uncertain or partially completed calculation - has non-negative expected value. The only real cost of attempting is the time already spent working through it.
Does eliminating options always help on CAT?: Yes, monotonically. Each option eliminated from the guessing pool raises the expected value of the remaining guess, from a breakeven zero with no elimination up to a certain +3 once every wrong option has been ruled out. There is no elimination level at which attempting becomes worse than skipping.
What happens if a CAT question has five or six options instead of four?: The expected value of a blind guess turns negative - -0.20 at five options and -0.33 at six, using the same +3/-1 formula. This is uncommon in CAT's standard format but worth knowing: the breakeven-at-four-options result is specific to CAT's typical question format, not a universal constant.
Why might a "zero expected value" blind guess actually lose me marks?: Because the zero assumes your wrong guesses are spread randomly across every wrong option. Real distractors are written to catch a specific, common mistake, so a test-taker's actual blind-guess accuracy is often below the true-random 25% the zero assumes. At 15% accuracy, for example, each such guess costs about -0.4 marks in expectation rather than zero.
How is the expected-value formula for guessing derived?: For k equally likely remaining options under CAT's +3/-1 scheme, EV = 3 × (1/k) − 1 × ((k−1)/k), which simplifies to (4 − k)/k. Setting this to zero and solving gives k = 4, the breakeven point - which happens to match CAT's own standard four-option MCQ format exactly.
Does Karma Yogi track guessing accuracy specifically?: Not at that level of detail. Logged mocks store each section's score, attempts and correct count, which is enough to compute overall accuracy per section per mock, but nothing in the current data model distinguishes a confidently answered question from a partially eliminated guess or a true blind guess.
Should this expected-value math change how I approach VARC differently from QA?: The math itself is section-agnostic - it only depends on how many options you can rule out, not which section the question is in. What differs by section is how reliably you can actually eliminate options: a QA calculation error is often easier to rule out definitively than a VARC inference option built to sound plausible, which is a preparation question rather than a math one.

Keep Reading

My 20 CAT Mocks: 41 to 91, With Every Crash Left In

Three months of real mock scores, section by section, including the mock where I scored 40 two days after scoring 96. Not a success story - a log of what the aggregate number hid.

Read

Self Study· 5 min read

CAT Preparation Without Coaching: What I Built to Replace It

I haven't enrolled in a single coaching institute for CAT. What I did instead was write a month-by-month plan, buy standalone mock series, and build my own tracker when nothing on the market did what I needed. 340 logged sessions later, here is what actually replaced coaching and what didn't.

Read

CAT Selection Process· 6 min read

CAT Selection Process: WAT, GD, and Personal Interview

I haven't sat a WAT-GD-PI round myself - my CAT is in November, so any interview process is still months off if it happens at all. What follows is what the process actually involves and how I'm planning to prepare for it, not a memoir of a round I haven't lived yet.

Read

Mock Analysis· 7 min read

How to Analyse CAT Mocks: 5 Steps, One 96-40-94 Week

On 29 July I scored 96. On 30 July, 40. On 1 August, 94. Same person, four days, three mocks. Here is the actual analysis framework, worked through that stretch and one sectional test, rather than asserted in the abstract.

Read