CAT Accuracy vs Attempts: What Mix Produces a Strong Percentile?

Attempts have no meaning without accuracy. Under the conservative all-MCQ assumption, net score = 3×correct − wrong. TITA/non-MCQ wrong answers do not carry the MCQ −1 penalty.

The easiest way to see the trade-off is with a simple example. Suppose two students attempt the same section. Student A answers 20 questions with 18 correct and 2 wrong MCQs: the net is 52. Student B answers 24 with 18 correct and 6 wrong MCQs: the net is 48. Four extra attempts produced a lower score because the added attempts were low quality.

Use three attempt buckets

Before a mock, do not decide “I must attempt 45.” Instead classify questions while solving: high confidence, workable but time-consuming, and low confidence. High-confidence questions should dominate the first pass. Workable questions become second-pass candidates. Low-confidence MCQs need a strong reason to be attempted because a random guess can reduce the net score.

TITA changes the arithmetic because an incorrect TITA answer does not carry the usual MCQ −1. That does not mean you should waste time on every TITA. Time still has an opportunity cost: three minutes spent on a hopeless TITA can prevent you from solving an easier MCQ elsewhere.

What to measure after a mock

Track attempts, correct, wrong MCQs, TITA attempts, accuracy, and net marks gained from the final ten attempts. That last number is revealing. If the final ten attempts consistently add little or negative value, your problem is over-attempting. If you finish with high accuracy but leave several easy questions, your problem may be excessive caution or poor pacing.

For a 99-percentile planning score around 88: 30/30 correct gives 90; 31 correct out of 35 gives 89; 32 correct out of 40 gives 88; 34 correct out of 48 gives 88 under the conservative all-wrong-MCQ assumption.

The objective is to find the attempt zone where your marginal question is still worth attempting. That zone differs by section and by student. Build it from evidence, then retest it under different mock difficulty levels.