🎓 SBI PO – Foundation & Exam Strategy
📌 Topic: Marking Scheme
An MIT‑level deep dive: first‑principles, rigorous reasoning, edge cases, and a challenging problem.
Table of Contents
- Overview of the SBI PO Marking Scheme
- Mathematical Modelling of Scoring
- Decision‑Theory Analysis (Expected Value & Risk)
- Edge Cases & Nuanced Situations
- Practical Exam‑Day Strategy
- Worked Example
- Advanced Problem & Solution
<a name="overview"></a>
1. Overview of the SBI PO Marking Scheme
| Question Type | No. of Questions | Marks per Correct Answer | Marks per Incorrect Answer | Marks for Unanswered |
|---|---|---|---|---|
| Multiple Choice Questions (MCQ) | 30 | +4 | –1 | 0 |
| Multiple True/False (MTF) | 10 | +4 | –1 | 0 |
| Numerical Questions (Numeric) | 10 | +4 | –1 | 0 |
| Matrix‑Based (Matrix Match) | 5 | +4 | –1 | 0 |
| Comprehension (Verbal) | 5 | +4 | –1 | 0 |
Total Marks: 200
Negative Marking: −1 for every incorrect attempt (no penalty for skipping).
Time: 3 hours (180 minutes).
Note: The exact distribution may vary slightly each year, but the scheme above reflects the current structure.
<a name="model"></a>
2. Mathematical Modelling of Scoring
Let us formalise the scoring process.
2.1 Variables
| Symbol | Meaning | Domain |
|---|---|---|
| (n) | Total number of questions in a section | integer (e.g. 30) |
| (c) | Number of questions attempted correctly | (0 \le c \le n) |
| (w) | Number of questions attempted incorrectly | (0 \le w \le n - c) |
| (u) | Number of questions left unanswered | (u = n - c - w) |
| (M) | Total marks obtained | integer |
2.2 Scoring Formula
[ M = 4c - w ]
Because (u) does not affect the score (0 mark), we can drop it.
2.3 Probability Model
Assume the candidate’s skill for a question is quantified by a probability (p \in [0,1]) of answering correctly, independent across questions. Under this model:
- (c \sim \text{Binomial}(n, p))
- (w = n - c)
Expected marks:
[ \mathbb{E}[M] = \mathbb{E}[4c - w] = 4np - (n - np) = 5np - n ]
Key Insight: The expected score per question is
[ \mathbb{E}[\text{Mark per Q}] = 5p - 1 ]
Thus:
- If (p > 0.2), expected marks per question positive.
- If (p < 0.2), expected marks per question negative.
This gives a crisp threshold for whether to attempt or skip a question.
<a name="decision"></a>
3. Decision‑Theory Analysis (Expected Value & Risk)
3.1 Optimal Decision Rule
For each question, compare:
- Attempt: Expected value (E_{\text{attempt}} = 5p - 1)
- Skip: Expected value (E_{\text{skip}} = 0)
Optimal Rule:
Attempt if (p > 0.2); skip otherwise.
Why 0.2?
The break‑even point where the expected value of attempting equals zero.
3.2 Risk Adjustment
Real exam takers often incorporate a risk‑aversion factor ( \lambda \in [0,1] ):
[ E_{\text{attempt}}^{\lambda} = \lambda (5p - 1) ]
If a candidate is risk‑averse, they might set ( \lambda < 1) and therefore attempt less aggressively.
3.3 Bayesian Updating
During the exam, a candidate can update their estimate of (p) using observed successes/failures:
[ p_{\text{new}} = \frac{c_{\text{obs}} + \alpha}{n_{\text{obs}} + \alpha + \beta} ]
where (\alpha,\beta) are prior hyperparameters (e.g. (\alpha=\beta=1) for a uniform prior). This yields a posterior estimate for future decisions.
<a name="edge"></a>