One-third negative marking is the most misunderstood number in UPSC preparation. Half the hall over-fears it and attempts 60 questions, leaving winnable marks on the table. The other half ignores it and sprays guesses. Both lose to aspirants who understand a little expected-value arithmetic.
The maths, once and for all
Each GS-I question: +2 for correct, −0.667 for wrong. If you guess among n equally likely options:
| Situation | Expected value per question |
|---|---|
| Blind guess (4 options) | (¼ × 2) − (¾ × 0.667) = 0.0 |
| Eliminate 1, guess among 3 | (⅓ × 2) − (⅔ × 0.667) = +0.22 |
| Eliminate 2, guess between 2 | (½ × 2) − (½ × 0.667) = +0.67 |
| Certain | +2.0 |
Read that table again — it contains the entire strategy:
- Blind guessing is neutral in expectation — but you're not playing for expectation, you're playing for a cutoff, and variance can sink you. Skip true blind guesses.
- Eliminating even one option makes guessing profitable. Eliminating two makes it a third of a certain answer — at scale, that's the exam.
- Twenty eliminate-two guesses ≈ +13 marks expected. Recent cutoffs sit around 75–93/200. The eliminate-and-attempt zone is where Prelims is won.
The four decision rules
Apply in order, per question, in seconds:
- Certain? Answer. (Target: your "certain" accuracy should be >95% — if "certain" questions come back wrong in practice, your certainty is miscalibrated and that's the first thing to fix.)
- Can I kill two options? Attempt, always. This is +0.67 EV territory.
- Can I kill one option? Attempt if your tracked accuracy in this zone is decent (see training below); skip if the question is in your historically weak subject.
- Nothing eliminable? Skip without guilt. Mark for the second pass — later questions sometimes leak hints.
The common catastrophic behaviours are both violations of rule 4's spirit: ego-attempting ("I studied this chapter, I must answer") and fear-skipping ("I'm not 100% sure"). The paper punishes emotion in both directions.
Explore Improvam, then choose the practice resources that fit your exam and study routine.
Elimination technique is learnable
Elimination isn't luck; it's pattern literacy built from PYQs:
- Extreme words lose. Statements with "only", "all", "always", "completely" are wrong far more often than not — governance and economics rarely admit absolutes.
- Inverted rules. UPSC loves flipping a provision ("cannot join a party within six months" vs the true "may"). If a statement feels like a rule you know but reversed, trust the alarm.
- Precision beats grandeur in "best describes". The boring, technically-scoped option outperforms the sweeping one.
- Pair logic. Options often pair statements such that disproving one statement kills two options at once — find the statement you're surest about and let it do the demolition.
Each trick is worth marks only if drilled until automatic — which means hundreds of statement-based MCQs with post-answer review, not reading about tricks in articles like this one.
Train it before the hall
Calibration — knowing how sure you actually are — is the trainable core. The loop: attempt daily timed MCQs under negative marking, tag your confidence before seeing the answer, and compare tags to outcomes weekly. Most aspirants discover they're overconfident in their strong subject and underconfident in their weak one — exactly the information that turns rules 2 and 3 from theory into marks.
FAQ
How many questions should I attempt in GS-I? Let the rules decide the number, not a target. Applied honestly they typically produce 75–90 attempts for a well-prepared aspirant. Fixing "I will attempt 85" in advance reverses cause and effect.
Is skipping 20+ questions acceptable? If they're genuine rule-4 skips, yes — 20 skips means 80 attempts, plenty against any recent cutoff. Skipping from fear on eliminate-two questions is what's unacceptable.
Does CSAT need a different strategy? Same arithmetic (−0.833 per wrong there), simpler goal: ~28 net-correct to qualify. Bank your easiest 40 questions and ignore the rest — see the CSAT pillar.
My accuracy on 50-50 guesses feels below 50%. Possible? Yes — it means your "elimination" is deleting the right option via a misremembered fact. That's a knowledge bug, not a luck bug; your review log will show which subject is generating it.
Related reading: PYQ analysis 2021–2024 · Prelims pillar page · Solve real papers