AFCAT – Marking Scheme
Advanced Study Notes (MIT‑level, first‑principles treatment)
Audience: Undergraduate students in applied mathematics, statistics or engineering who wish to master the quantitative underpinnings of the AFCAT exam strategy.
Goal: Provide a rigorous, analytical view of the AFCAT marking scheme, covering probability, decision theory, optimization, and edge‑case handling.
Structure
- Exam‑Level Overview – scoring rules, question types, raw vs. normalized marks.
- Mathematical Modelling – formal probability model, expected score, variance.
- Decision‑Theoretic Optimisation – Bayesian updating, dynamic programming.
- Edge Cases & Practical Nuances – incomplete answers, tie‑breakers, time‑management.
- Challenging Problem – “Optimal answering strategy under a mixed‑difficulty distribution.”
- Key Take‑aways & Cheat‑Sheet – quick formula reference.
1. Exam‑Level Overview
| Feature | Detail |
|---|---|
| Exam | AFCAT (Air Force Common Admission Test) – 120 multiple‑choice questions (MCQs). |
| Question Types | All MCQs are of single‑correct type ( 1 correct, 3 distractors). |
| Marking Scheme | • Correct answer: +4 marks • Wrong answer: –1 mark • Unattempted: 0 marks |
| Maximum Score | 120 × 4 = 480 marks |
| Minimum (theoretical) | 0 (if all answers are wrong; but due to negative marking, worst‑case could be –120). |
| Cut‑off & Scaling | After raw score → percentile → scaled score (100‑point scale). The raw‑to‑scaled conversion is non‑linear and depends on the overall difficulty of the paper. |
| Tie‑Breakers | In case of identical raw scores, the time taken to answer may be used; thus, answering faster can give an advantage. |
Why does the marking scheme matter?
It defines the risk associated with guessing. Understanding the expected value of a random guess is the cornerstone of any optimal strategy.
2. Mathematical Modelling
2.1 Probability Space
Assume a candidate faces a question (Q).
Let the a priori probability that the candidate knows the correct answer be (p).
Define random variables:
- (X_Q = \begin{cases} +4, & \text{if answered correctly (probability }p)\ -1, & \text{if answered incorrectly (probability }1-p) \end{cases})
If the candidate does not answer, (X_Q = 0).
2.2 Expected Score per Question
[ E[X_Q] = 4p + (-1)(1-p) = 5p - 1 ]
Interpretation
| (p) | (E[X_Q]) | Action |
|---|---|---|
| (p > 0.2) | (> 0) | Answer |
| (p < 0.2) | (< 0) | Skip |
| (p = 0.2) | (0) | Indifferent (tie‑breaker may sway decision) |
Thus, the critical probability is (p_c = 0.2).
Answering a question is rational only if you are more than 20 % confident.
2.3 Variance
[ \text{Var}[X_Q] = (4)^2 p + (-1)^2 (1-p) - (5p-1)^2 = 16p + (1-p) - (5p-1)^2 ]
Simplify:
[ \text{Var}[X_Q] = 17p - (25p^2 - 10p + 1) = -25p^2 + 27p - 1 ]
This quadratic opens downward; maximum variance occurs near (p = \frac{27}{50} \approx 0.54).
2.4 Aggregate Score
Let (N) be the number of answered questions.
If we assume independence (a simplification; real papers have correlated difficulty), the total score (S) is:
[ S = \sum_{i=1}^{N} X_{Q_i} ]
Then
[ E[S] = \sum_{i=1}^{N} (5p_i - 1), \qquad \text{Var}[S] = \sum_{i=1}^{N} (-25p_i^2 + 27p_i - 1) ]
If we treat all (p_i = p):
[ E[S] = N(5p-1), \quad \text{Var}[S] = N(-25p^2+27p-1) ]