- Foundation
- About 14 min
- 5 practice questions
Writing your own answering policy and testing it on recorded papers
Record your confidence on each question in practice, replay different policies on the record, and choose a policy with evidence, remembering that one paper is a small sample.
If something looks wrong, please tell us.
The best answering policy depends on you: your accuracy when you are sure, when you have narrowed the options, and when you have no idea. You can find out in practice. Mark each answer with how sure you were, and after the paper work out what different policies would have scored. All figures here are from an invented paper and an invented scheme (+1 for right, -0.25 for wrong); your real results and your real instructions will differ.
Step 1: record confidence
While practising, tag each answer: S (sure), N (narrowed to two or three options) or B (blind). A tag takes a second. After marking, count for each tag how many answers there were and how many were right.
Worked example 1: Mehul's record
| Tag | Answers | Right | Hit rate |
|---|---|---|---|
| S (sure) | 24 | 22 | 92% |
| N (narrowed) | 10 | 6 | 60% |
| B (blind) | 6 | 1 | 17% |
- Policy A (answer everything): right = 22 + 6 + 1 = 29; wrong = (24 - 22) + (10 - 6) + (6 - 1) = 11; net = 29 - 11 x 0.25 = 26.25.
- Policy B (skip blind ones): right = 22 + 6 = 28; wrong = 2 + 4 = 6; net = 28 - 6 x 0.25 = 26.5.
- Policy C (answer only the sure ones): right = 22; wrong = 2; net = 22 - 2 x 0.25 = 21.5.
- Result: B scored 26.5, A scored 26.25 and C scored 21.5. Skipping the blind answers helped a little (0.25 marks), while skipping the narrowed answers as well was clearly worse (by 5 marks).
Step 2: do not trust one paper too much
The difference between A and B is a quarter of a mark. That is too small to prove anything: one different blind guess would reverse it. The bigger finding is that narrowed answers earned their place, with 6 right out of 10. Record several papers and add up the tags before settling on a policy. A change in the policy that wins on three papers is more convincing than one that wins on one.
Worked example 2: Mehul checks the rule
- Step 1: in this invented scheme, the break-even hit rate is 0.25 / (1 + 0.25) = 0.2, or 20%.
- Step 2: his hit rates are 92% (sure), 60% (narrowed) and 17% (blind). The first two are well above 20%; the third is below it.
- Step 3: this agrees with the replay: narrowed answers pay, blind ones do not seem to, on this paper.
- Step 4: he writes the policy: answer sure questions, answer narrowed ones, leave blind ones for the second pass and decide then.
- Result: his policy is based on his own numbers, and he will test it again on the next two papers.
Writing the policy down
- Say which tags you answer, which you skip and which you revisit.
- Say what time you will spend on a question before moving on.
- Review the policy after every two or three papers, not after every one.
Tag your confidence, replay the policies on the record, and trust a result only when it repeats.
Key takeaways
- Record S, N or B for each answer in practice.
- Replay different policies to see what each would have scored.
- One paper is a small sample; look for results that repeat.
- Write the policy down and review it every few papers.
Check your understanding
-
1. In an invented scheme (+1, -0.25), a student has 20 right and 8 wrong. What is the net score?
- 18
- 12
- 20
- 16
Show answer
Answer: A. 20 - 8 x 0.25 = 20 - 2 = 18.
-
2. Of 10 narrowed answers, 6 are right. What is the hit rate?
- 6%
- 40%
- 60%
- 16%
Show answer
Answer: C. 6 divided by 10 is 0.6, which is 60%.
-
3. In the invented scheme (+1, -0.25), the break-even hit rate is
- 25%
- 20%
- 75%
- 80%
Show answer
Answer: B. Break-even = penalty / (mark + penalty) = 0.25 / 1.25 = 0.2, which is 20%.
-
4. Why should one paper not settle your answering policy?
- Papers are never recorded
- Policies cannot be tested
- Hit rates cannot be measured
- One paper is a small sample
Show answer
Answer: D. A single paper has few answers in each tag and chance can change the result. Results that repeat across papers are more trustworthy.
-
5. A student's blind answers have a 15% hit rate with a 20% break-even. In this scheme, blind answers
- Gain marks on average
- Lose marks on average
- Break even exactly
- Cannot be counted
Show answer
Answer: B. A hit rate below break-even loses marks on average. The rule is only as good as the measured hit rate, so keep recording it.
Sign in to keep track of the lessons you finish.