Probability

In short

Probability measures how likely an event is, on a scale from 0 (impossible) to 1 (certain). It is built on counting favourable outcomes against total equally likely outcomes, and then extended by the addition, multiplication, conditional and Bayes rules to multi-stage experiments. In Indian examinations it appears in the AICTE/UGC CBCS "Probability and Statistics" course of almost every engineering branch, in GATE Engineering Mathematics (most engineering papers), in CAT, and in the quantitative section of TCS NQT, Infosys, Wipro, Capgemini, Accenture and IBPS/SBI bank exams. On the job it drives reliability and quality engineering, insurance and actuarial pricing, credit risk, A/B testing and every machine-learning model. Expect two to four marks in a placement aptitude paper and at least one conceptual interview question in data, risk or analytics roles.

Explained

Basic terms and the classical definition

• Random experiment: an action whose outcome is not known in advance - tossing a coin, rolling a die, drawing a card.
• Sample space (S): the set of all possible outcomes. One coin: {H, T}; n coins: 2ⁿ outcomes; n dice: 6ⁿ; a pack of cards: 52.
• Event (E): any subset of S. Simple event = one outcome; compound = more than one.
• Classical (a-priori) probability: P(E) = (number of favourable outcomes) / (total number of equally likely outcomes). Valid only when all outcomes are equally likely.
• Empirical (relative-frequency) probability: P(E) = (times E occurred) / (number of trials), used when outcomes are not symmetric - machine failure rates, claim rates.
• Axioms: 0 ≤ P(E) ≤ 1; P(S) = 1; P(impossible event) = 0; for mutually exclusive events probabilities add.
• Complement: E′ = "E does not happen", P(E′) = 1 − P(E). This single idea solves most "at least one" questions.
• Odds: odds in favour of E = P(E) : P(E′) = a : b, so P(E) = a/(a+b); odds against = b : a. Banking and betting questions are phrased in odds.
• Exhaustive events: their union is S. Equally likely: none is preferred over another.
• A pack of 52 cards: 4 suits (spades and clubs black, hearts and diamonds red) × 13 cards; 12 face/court cards (J, Q, K); 4 aces; 26 red, 26 black.

Addition, multiplication, independence

• Addition rule: P(A ∪ B) = P(A) + P(B) − P(A ∩ B). If A and B are mutually exclusive (cannot occur together), P(A ∩ B) = 0 and P(A ∪ B) = P(A) + P(B).
• Three events (inclusion-exclusion): P(A∪B∪C) = ΣP − ΣP(pairs) + P(A∩B∩C).
• Conditional probability: P(A | B) = P(A ∩ B) / P(B), P(B) > 0 - the sample space shrinks to B.
• Multiplication rule: P(A ∩ B) = P(B) × P(A | B). For independent events P(A | B) = P(A), so P(A ∩ B) = P(A) × P(B).
• Mutually exclusive ≠ independent. Mutually exclusive events with non-zero probability are strongly *dependent*: if one happens the other cannot. Independence is about information, exclusivity about overlap.
• With replacement → trials independent, probabilities unchanged each draw. Without replacement → dependent; the denominator (and sometimes the numerator) falls by one each draw.
• At least one: P(at least one) = 1 − P(none) = 1 − Π(1 − p_i) for independent events.
• Pairwise vs mutual independence: three events can be independent in pairs yet not jointly independent; examiners use this as a trap in theory papers.

Counting-based probability (the placement workhorse)

• Most aptitude problems are combinatorics divided by combinatorics: P = favourable combinations / total combinations.
• Order does not matter (a committee, a handful of balls) → use combinations C(n, r) = n!/(r!(n−r)!). Order matters (arrangements, codes, ranks) → use permutations.
• Drawing r items from a bag of mixed colours: total = C(n, r); favourable = product of C(.) for each colour required.
• Dice: two dice give 36 outcomes - 6 doublets, sum 7 is the most likely (6 ways), sums 2 and 12 the least (1 way each). Sum ≥ 10 occurs in 6 ways.
• Coins: exactly r heads in n tosses = C(n, r)/2^n. For 3 coins, exactly 2 heads = 3/8.
• Cards: always state whether the draw is with or without replacement, and whether "ace" counts as a face card (conventionally it does not).
• Arrangement questions: probability that two particular people sit together among n in a row = 2(n−1)!/n! = 2/n.
• Derangement (letters in envelopes): number of ways no object goes to its own place, D_n = n!·Σ(−1)^k/k!. D_2 = 1, D_3 = 2, D_4 = 9, D_5 = 44.
• Birthday problem: with 23 people the chance that two share a birthday already exceeds 50% (about 50.7%) - a standard interview surprise.

Total probability and Bayes' theorem

• Partition: events B₁, B₂, …, B_n are mutually exclusive and exhaustive (two machines, three suppliers, two bags).
• Law of total probability: P(A) = Σ P(B_i) × P(A | B_i) - average the conditional probabilities, weighted by the prior probabilities.
• Bayes' theorem: P(B_k | A) = P(B_k)P(A | B_k) / Σ P(B_i)P(A | B_i). It reverses the conditioning: from "probability of a defect given machine A" to "probability it came from machine A given a defect".
• Prior → likelihood → posterior: the prior is what you believed before the evidence, the likelihood is how well each cause explains the evidence, the posterior is the updated belief.
• Base-rate neglect: when a condition is rare, even a very accurate test gives mostly false positives. This is the most asked applied Bayes question in analytics interviews (worked below).
• Typical exam settings: urns and transferred balls, defective items from several machines, medical screening, spam filters, two-children and Monty Hall puzzles.

Random variables, expectation and standard distributions

• Random variable (X): a number attached to each outcome. Discrete (counts) or continuous (measurements).
• Probability mass function p(x) for discrete X with Σp(x) = 1; probability density function f(x) for continuous X with ∫f(x)dx = 1 and P(X = a) = 0.
• Cumulative distribution function F(x) = P(X ≤ x), non-decreasing from 0 to 1.
• Expectation (mean) E(X) = Σ x·p(x); Variance Var(X) = E(X²) − [E(X)]²; standard deviation σ = √Var.
• Linearity: E(aX + b) = aE(X) + b, Var(aX + b) = a²Var(X); E(X + Y) = E(X) + E(Y) always, Var(X + Y) = Var(X) + Var(Y) only if independent.
• Bernoulli: one trial, success probability p; mean p, variance p(1−p).
• Binomial B(n, p): number of successes in n independent identical trials; mean np, variance npq. Used for defectives in a sample, guessed MCQs, coin tosses.
• Geometric: number of trials until the first success; P(X = k) = q^k−1p, mean 1/p - so on average 6 rolls to get a six.
• Poisson(λ): rare events in a fixed interval; mean = variance = λ. Calls per minute at a helpline, flaws per metre of cloth, accidents per month. Binomial tends to Poisson when n is large, p small, np = λ.
• Uniform, exponential: exponential models time between Poisson events, mean 1/λ, and is memoryless - the basis of constant-failure-rate reliability (MTBF).
• Normal N(μ, σ²): bell-shaped, symmetric; about 68.27%, 95.45% and 99.73% of the area lies within 1σ, 2σ and 3σ of the mean. Standardise with z = (x − μ)/σ and read standard normal tables. Six Sigma quality targets derive from this.
• Central limit theorem: the mean of a large sample is approximately normal whatever the parent distribution - why normal tables dominate quality control and polling.
• Limitations: classical probability needs symmetry; binomial needs constant p and independence; Poisson needs rare, independent events; normal is poor for strongly skewed data such as incomes or insurance claims.

Exam practice, Indian standards and quick techniques

• In aptitude papers probability questions are usually solvable in under 90 seconds - identify the sample space size first, then count favourables; use the complement whenever the phrase "at least" or "not" appears.
• Answers in objective tests are expected as simplified fractions; in GATE numerical-answer type questions as decimals to the stated number of places.
• Acceptance sampling: IS 2500 (sampling inspection procedures, based on the ISO 2859 series) fixes sample sizes and acceptance numbers for a given AQL - a direct industrial use of the binomial/hypergeometric distribution. Verify the current revision before quoting clause numbers.
• Six Sigma: a process centred with 1.5σ shift gives about 3.4 defects per million opportunities; 3σ control limits in SPC charts come from the normal distribution.
• Reliability: R(t) = e^−λt for constant failure rate; series system reliability = ΠR_i, parallel = 1 − Π(1 − R_i).
• Actuarial and insurance: IRDAI-regulated pricing and the Institute of Actuaries of India (IAI) CS1/CM1 syllabus rest on probability distributions and expected values.
• Probability is mandatory in GATE for CS, ME, CE, EE, EC, PI and the Data Science & AI (DA) paper, and in JAM/MSc statistics syllabi.

Key points

• Probability range - 0 ≤ P(E) ≤ 1; a negative value or one above 1 means an arithmetic error.
• Classical formula - favourable outcomes ÷ total equally likely outcomes; the "equally likely" condition is essential.
• Complement rule - P(not E) = 1 − P(E); the fastest route for "at least one" problems.
• Addition rule - P(A∪B) = P(A) + P(B) − P(A∩B); subtract the overlap only once.
• Mutually exclusive - cannot occur together, P(A∩B) = 0; not the same as independent.
• Independent - P(A∩B) = P(A)·P(B); one event gives no information about the other.
• Conditional probability - P(A|B) = P(A∩B)/P(B); the sample space becomes B.
• With vs without replacement - replacement keeps trials independent; without replacement the denominator shrinks each draw.
• Odds a:b - probability a/(a+b); convert before using any formula.
• Total probability - weight each conditional probability by its prior and add.
• Bayes' theorem - reverses conditioning; rare causes stay unlikely even after strong evidence.
• Binomial - r successes in n trials, C(n,r)p^r q^n−r, mean np, variance npq.
• Poisson - rare events, P(X=k) = e^−λλ^k/k!, mean = variance = λ.
• Geometric expectation - average trials to first success = 1/p (6 rolls for a six).
• Expected value - Σ(value × probability); the fair price of a game or a lottery ticket.
• Normal distribution - 68-95-99.7 rule; z = (x − μ)/σ for table lookup.
• Derangement - no item in its own place: D₄ = 9, so P = 9/24 = 3/8 for four letters in four envelopes.
• Two dice - 36 outcomes, sum 7 most likely (6/36), 6 doublets, sum ≥ 10 in 6 ways.

Formulas and examples

Example 1 - two cards from a pack of 52 (without replacement). Total selections n(S) = C(52,2) = 52×51/2 = 1,326.

• Both kings: C(4,2) = 6 → P = 6/1326 = 1/221.
• Both of the same suit: 4 × C(13,2) = 4 × 78 = 312 → P = 312/1326 = 4/17.
• At least one ace: P = 1 − C(48,2)/C(52,2) = 1 − 1,128/1,326 = 198/1,326 = 33/221 ≈ 0.149.

Example 2 - bag of 5 red, 4 blue, 3 green balls; 3 drawn at random. n(S) = C(12,3) = 220.

• All of one colour: C(5,3) + C(4,3) + C(3,3) = 10 + 4 + 1 = 15 → P = 15/220 = 3/44.
• One of each colour: 5 × 4 × 3 = 60 → P = 60/220 = 3/11.
• Exactly two red: C(5,2) × C(7,1) = 10 × 7 = 70 → P = 70/220 = 7/22.

Example 3 - Bayes' theorem, screening test. A disease affects 1% of a population. The test detects it in 99% of patients who have it (sensitivity) and wrongly flags 5% of healthy people (false positive). A person tests positive - what is the chance they have the disease?

• P(positive) = 0.01 × 0.99 + 0.99 × 0.05 = 0.0099 + 0.0495 = 0.0594.
• P(disease | positive) = 0.0099 / 0.0594 = 1/6 ≈ 16.7%.
• Interpretation: five of every six positives are false, because the disease is rare - the base rate dominates.

Example 4 - binomial and expectation. A fair die is rolled 5 times; X = number of sixes.

• P(exactly two sixes) = C(5,2)(1/6)²(5/6)³ = 10 × (1/36) × (125/216) = 1,250/7,776 = 625/3,888 ≈ 0.1608.
• P(no six) = (5/6)⁵ = 3,125/7,776 ≈ 0.4019, so P(at least one six) ≈ 0.5981.
• E(X) = np = 5 × 1/6 = 0.833, Var(X) = npq = 5 × (1/6)(5/6) = 25/36 ≈ 0.694.
• For one roll of the die, E = (1+2+3+4+5+6)/6 = 3.5 and Var = 91/6 − 3.5² = 15.1667 − 12.25 = 2.9167.

Example 5 - independent workers and expected value. Three students solve a problem independently with probabilities 1/2, 1/3, 1/4.

• P(none solves) = (1/2)(2/3)(3/4) = 6/24 = 1/4 → P(at least one solves) = 3/4.
• P(exactly one solves) = 6/24 + 3/24 + 2/24 = 11/24; P(all three) = 1/24.
• A lottery sells 1,000 tickets at ₹50 each, with one prize of ₹20,000 and ten prizes of ₹1,000. Expected prize per ticket = (20,000 + 10,000)/1,000 = ₹30, so the expected net result is 30 − 50 = −₹20 per ticket.

Where it is used

• Quality and manufacturing: acceptance sampling under IS 2500 / ISO 2859, SPC control charts and Six Sigma DPMO at Maruti, Tata Motors, Bosch, Sundaram Fasteners; defect rates are binomial or Poisson.
• Insurance and actuarial work: LIC, ICICI Lombard, HDFC Life and IRDAI-regulated pricing; expected claim amounts, mortality tables, IAI actuarial exams.
• Banking, credit and fraud: probability of default in RBI-mandated credit-risk models, Bayesian scorecards at CIBIL/CRIF, UPI fraud scoring at NPCI.
• Data science and analytics roles: Naive Bayes classifiers, logistic regression outputs read as probabilities, A/B testing and confidence intervals at Flipkart, Swiggy, Zomato, PhonePe.
• Reliability and maintenance engineering: MTBF, exponential failure models and redundancy design in power plants, Indian Railways rolling stock and ISRO mission assurance.
• Telecom, traffic and queueing: Poisson call arrivals and Erlang capacity planning at Jio and Airtel; Poisson arrival models for toll plazas and metro gates.

Interview questions

• Q: Difference between mutually exclusive and independent events? A: Mutually exclusive events cannot occur together (P(A∩B) = 0); independent events can occur together but neither changes the other's probability (P(A∩B) = P(A)P(B)).
• Q: Two dice are rolled and you are told at least one shows a six. What is the probability both show six? A: 11 outcomes contain at least one six and only one is (6,6), so the answer is 1/11, not 1/6.
• Q: How many times must you roll a die on average to get a six? A: Six - the number of trials to the first success is geometric with mean 1/p = 6.
• Q: A test detects 99% of cases of a disease affecting 1 in 100 people, and wrongly flags 5% of healthy people. You test positive - should you worry? A: Not immediately; Bayes gives only about a 1/6 chance of actually having it, because the condition is rare.
• Q: State and explain Bayes' theorem in one line. A: Posterior ∝ prior × likelihood - it converts P(evidence | cause) into P(cause | evidence) by weighting each possible cause by how likely it was beforehand.
• Q: When does the binomial distribution become Poisson? A: When n is large, p is small and np = λ stays moderate; the Poisson is then an accurate and far simpler approximation.
• Q: In the Monty Hall problem, should you switch doors? A: Yes - switching wins with probability 2/3 because the host's choice is not random and carries information about where the car is.
• Q: Four letters are placed at random in four addressed envelopes. Probability that none goes to the right envelope? A: D₄/4! = 9/24 = 3/8.
• Q: Why does the variance of a sum need independence but the mean does not? A: Expectation is linear for all random variables; variance picks up a 2·Cov(X,Y) term that vanishes only when the variables are uncorrelated.
• Q: What does an expected value of −₹20 on a lottery ticket mean? A: Over many repetitions the average loss per ticket is ₹20; it says nothing about any single ticket.
• Q: With only 23 people in a room, is a shared birthday likely? A: Yes - the probability is about 50.7%, because 23 people form 253 pairs, not 23 comparisons.
• Q: A bag has 3 white and 5 black balls; two are drawn without replacement. Probability both are white? A: (3/8) × (2/7) = 6/56 = 3/28.

Common mistakes

• Adding probabilities of events that can occur together and forgetting to subtract P(A ∩ B).
• Treating mutually exclusive events as independent (or multiplying when the question needs addition).
• Not reducing the denominator in draws without replacement, or reducing it when replacement is stated.
• Confusing P(A | B) with P(B | A) - the heart of every base-rate error in interviews.
• Reading "at least one" as "exactly one", instead of using 1 − P(none).
• Mixing permutations with combinations in the numerator and denominator of the same ratio, which silently changes the sample space.
• Using the binomial formula when trials are dependent (sampling without replacement from a small lot) - the hypergeometric distribution applies there.

Related topics

• Permutations and combinations · Data interpretation · Averages, mixtures and alligation · Ratio, proportion and partnership · Number system and divisibility · Percentages.