Prelims Marks Math: How Many Questions Should You Actually Attempt?
Negative marking turns attempt strategy into arithmetic. Here is the expected-value math behind a guess, how to compute your net score, when to skip, and how to build a buffer over the cutoff — across any PSC marking scheme.
'Should I attempt this one or leave it?' is the single most repeated decision in a prelims exam — you make it 100+ times in two hours. Most aspirants answer it by gut. With negative marking, gut is expensive. The right answer is arithmetic, and once you've internalized it, the decision takes a fraction of a second.
The expected value of a guess
Take a typical scheme: +2 for correct, −0.66 (one-third) for wrong, four options. If you guess blindly, you're right 1 time in 4. Expected marks = (1/4 × +2) + (3/4 × −0.66) = 0.5 − 0.5 = roughly 0. A blind four-option guess is break-even — it neither helps nor hurts you on average.
This is the key insight: under one-third negative marking with four options, a pure blind guess is break-even. The moment you can eliminate even one option, the math tips in your favor and you should attempt.
Elimination changes everything
Eliminate 0 options (blind guess): expected value ≈ 0 — coin flip, no edge.
Eliminate 1 option (1 of 3): expected value turns clearly positive — attempt.
So the real rule isn't 'never guess.' It's: skip only when you can't eliminate a single option. If you've ruled out even one, the expected value is positive and leaving it blank costs you marks over the length of the paper.
Marking schemes vary by exam — check yours. Some PSCs use −0.25, some −0.33, some have no negative marking at all (where you should attempt everything). Recompute the break-even for your scheme before exam day; don't assume the UPSC numbers apply.
Net score, not raw attempts
Aspirants brag about attempting 95/100. Meaningless. What clears the cutoff is net score = (correct × marks) − (wrong × penalty). Attempting 95 with 60% accuracy can score lower than attempting 80 with 75% accuracy. Track net score in every mock, not attempt count.
Building a buffer over the cutoff
Cutoffs move year to year, so don't aim to just clear last year's number — aim 10–15% above it. That buffer absorbs a hard paper, a normalization swing, and your own exam-day nerves. Use educated-guess marks (eliminate-then-attempt) to build the buffer; that's exactly where the expected-value edge compounds.
PSCPrep.ai mocks report net score and accuracy, not just attempt count — so you can see whether your guessing is adding marks or quietly bleeding them, and calibrate your skip threshold before the real exam.
Frequently Asked Questions
Should I guess in prelims with negative marking?
Guess only when you can eliminate at least one option. Under a typical one-third penalty with four options, a blind guess is break-even (expected value ≈ 0), but eliminating even one option makes the expected value positive. If you can't rule out a single option, skip it.
How do I calculate the expected value of an attempt?
Expected value = (probability correct × marks for correct) + (probability wrong × negative penalty). For +2/−0.66 with four options and a blind guess: (1/4 × 2) + (3/4 × −0.66) ≈ 0. Recompute it with your exam's actual marking scheme.
Is a higher number of attempts always better?
No. What clears the cutoff is net score — (correct × marks) minus (wrong × penalty) — not raw attempts. Attempting more questions at low accuracy can score lower than attempting fewer at high accuracy. Track net score in every mock.
How much above the cutoff should I aim?
Aim roughly 10–15% above the previous year's cutoff. Cutoffs shift with paper difficulty and normalization, so a buffer protects you against a hard paper and exam-day nerves rather than scraping the exact line.
Top current affairs + exam tips, every Monday morning.
Turn reading into marks — attempt PYQs online
Every supported State PCS exam has free public pages to preview how previous-year papers are organized, plus topic-weightage trend snapshots derived from the same corpus. Log in to run timed attempts and unlock full post-attempt analytics.