Chapter 5 — Probability

What is an experiment?
In probability a real or imagined action that leads to one of several possible results is called an experiment. Examples are tossing a coin, rolling a die, drawing a card, or measuring how many emails arrive in an hour. Each time you perform the experiment you observe a single result called an outcome.

Sample space and outcomes
The sample space, denoted by S, is the collection of all possible outcomes. For a single die roll S = {1,2,3,4,5,6}. For two tosses of a coin S = {(H,H),(H,T),(T,H),(T,T)} where order matters and ordered pairs are used. Writing the sample space clearly is the first step in any probability problem because it makes counting and reasoning exact.

Events and notation
An event is any subset of the sample space. If S is the set of outcomes, an event A is A ⊂ S. Events can be simple (one outcome) or compound (many outcomes). Notation: union A ∪ B means A or B; intersection A ∩ B means both A and B; complement A' means outcomes not in A; empty event ∅ has no outcomes; certain event S always occurs. Representing events with set notation helps manipulate and combine them algebraically.

Operations and properties
Set operations mirror logical relations: A ∪ B corresponds to 'A or B', A ∩ B to 'A and B', and A' to 'not A'. Disjoint or mutually exclusive events have A ∩ B = ∅; they cannot happen together. De Morgan's laws — (A ∪ B)' = A' ∩ B' and (A ∩ B)' = A' ∪ B' — are useful when working with complements. Practise drawing Venn diagrams to visualise these relations.

Ordered vs unordered outcomes
When experiments have multiple stages, decide whether order matters. For two coin tosses, (H,T) differs from (T,H) if you care about which toss was heads. If you only care about how many heads, treat outcomes as counts and use combinatorics. Mixing these views incorrectly causes counting mistakes, so always specify whether outcomes are ordered.

Continuous vs discrete sample spaces (brief)
Sample spaces can be finite, countably infinite (like {0,1,2,...}) or uncountable (like all real numbers in an interval when measuring lengths). Class 12 focuses mainly on finite and countable discrete spaces; continuous models come in higher study but the idea of events-as-subsets remains the same.

Practice tips
On every problem: list S, define events using set notation, and check whether outcomes are equally likely. Always ask whether sampling is with or without replacement and whether order matters; these determine counting and probabilities later.

Classical idea: equally likely outcomes
The classical definition of probability applies when a finite sample space S has outcomes that are equally likely. In that situation the probability of an event A is the number of favourable outcomes divided by the total number of outcomes: P(A) = |A|/|S|. This simple rule is ideal for fair coins, dice and cards when symmetry makes outcomes equally likely. Always verify the equally-likely assumption before using this rule.

Limitations of classical view
Many real-world situations do not have equally likely outcomes: a biased coin, non-uniform spinner, or events with weights require a more general approach. Even when outcomes are listed, different outcomes may carry different probabilities. For such cases assign probabilities p(ω) to each outcome ω and compute P(A) as the sum over ω in A of p(ω).

Kolmogorov axioms
Modern probability relies on three axioms formulated by Kolmogorov: (1) Non-negativity: For any event A, P(A) ≥ 0. (2) Normalisation: P(S) = 1. (3) Countable additivity: For any countable sequence of pairwise disjoint events A1, A2, ..., we have P(∪ Ai) = Σ P(Ai). These axioms give a rigorous foundation and lead to useful derived rules.

Derived rules from axioms
From the axioms follow practical formulae: P(∅)=0; P(A') = 1 − P(A); for two events A and B, P(A ∪ B) = P(A) + P(B) − P(A ∩ B). If A and B are disjoint then P(A ∪ B) = P(A) + P(B). These relations are used constantly in calculations and proofs.

Assigning probabilities in finite spaces
When outcomes are not equally likely, assign non-negative numbers p(ω) to each outcome ω with the sum over S equal to 1. Then for any event A, P(A) = Σ_{ω∈A} p(ω). This approach handles biased mechanisms like loaded dice or sensors with different reliability.

Working with infinite sample spaces
For countably infinite sample spaces one still assigns p(ω) with sum 1; but ensure convergence of sums. For uncountable sample spaces (not central to class 12) one must use measures; the axioms still serve as guide but require more advanced tools.

Problem-solving routine
Step 1: define S. Step 2: check equally-likely assumption. Step 3: use classical formula or assign probabilities to outcomes. Step 4: apply axioms-derived identities and simplify. Explicitly stating these steps in exams earns clarity marks.

Counting is central to probability
When outcomes are equally likely, probability reduces to counting favourable outcomes and dividing by total outcomes. Accurate counting is therefore essential. Permutations count ordered arrangements; combinations count unordered selections. Recognising which applies prevents mistakes.

Permutations: order matters
When order is important — for example, arranging books on a shelf or the sequence of winners — use permutations. The number of ways to choose and arrange r objects from n distinct ones is nPr = n(n−1)(n−2)...(n−r+1) = n!/(n−r)!. When r = n, the number of arrangements is n! . For example, seating 5 students in 5 chairs gives 5! = 120 arrangements. Remember to consider whether rotations count as distinct; for circular arrangements divide by n to account for rotations if positions are considered equivalent.

Combinations: order does not matter
When only the set of chosen objects matters, not the order, use combinations. The number of ways to choose r items from n distinct items without order is nCr = n!/(r!(n−r)!). Examples include choosing committee members, selecting cards, or picking lotto numbers. A quick check: nCr = nPr / r! because each unordered selection corresponds to r! ordered arrangements.

Selections with repetition
If repetition is allowed, formulas change. The number of r-length sequences from n symbols with repetition allowed is n^r (order matters). For combinations with repetition (selecting r items from n types where order doesn't matter), the count is C(n+r−1, r). Decide early if repetition is permitted, because it changes counts drastically.

Using counting in probability
For equally likely finite sample spaces, P(A) = favourable/total. For example, drawing 2 kings from 52 cards without order uses combinations: C(4,2)/C(52,2). If order matters (first and second draw recorded), use permutations or product rule. For multi-stage experiments build a tree diagram and multiply along branches when probabilities are conditional.

Advanced counting ideas
In some probability problems you need to count arrangements avoiding a pattern (inclusion-exclusion), or count ways to distribute identical objects to distinct boxes (stars-and-bars). Inclusion-exclusion helps count unions of overlapping events by alternately adding and subtracting counts of intersections.

Practical problem steps
1. List the sample space type and size for small cases. 2. Determine whether order and repetition matter. 3. Choose the counting formula. 4. Compute favourable and total counts and form the probability ratio. Practise with concrete examples to build confidence in choosing the correct formula.

Understanding conditional probability
Conditional probability measures how likely an event A is, given that another event B has occurred. It updates the sample space: once you know B happened, only outcomes in B remain possible. Formally, if P(B) > 0, P(A|B) = P(A ∩ B) / P(B). This basic definition is used extensively in sequential experiments and dependent events.

From definition to multiplication rule
Rearrange the conditional definition to get the multiplication rule: P(A ∩ B) = P(B) P(A|B). Since multiplication is symmetric if you swap A and B with conditional reversed, also P(A ∩ B) = P(A) P(B|A). This rule generalises to multiple events: for three events, P(A ∩ B ∩ C) = P(A) P(B|A) P(C|A ∩ B), and so on. Use this chain rule for probabilities in multi-stage experiments and to compute joint probabilities from conditional probabilities.

Sequential experiments and trees
Probability trees are a helpful visual for multi-stage problems: each branch is labelled with a conditional probability and joint probabilities are found by multiplying along branches. Sum branch probabilities to get probabilities of compound events. Trees clarify conditioning and avoid mistakes about which probabilities are conditional on previous stages.

Dependent vs independent events
If P(A|B) = P(A) (for P(B)>0), then A and B are independent: knowing B gives no information about A. Independence simplifies calculations since P(A ∩ B) = P(A)P(B). But many problems, especially without replacement sampling, create dependence and require conditional probabilities to account for changing composition of the sample space.

Common contexts
Conditional probability appears in drawing without replacement, diagnostic testing, reliability of systems where component failure affects others, and predictive tasks. For example, in a deck of cards the chance of drawing an ace on the second draw depends on whether the first draw was an ace when there is no replacement.

Calculation tips
Always check that the conditioning event has positive probability. When calculating P(A|B), list the intersection outcomes explicitly if possible and divide by the number of outcomes in B when outcomes are equally likely. For non-equally-likely outcomes, sum up the outcome probabilities in numerator and denominator separately. Label conditional probabilities clearly when working with multiple stages to avoid confusion.

Exam strategy
Write the formula P(A|B)=P(A∩B)/P(B) and show substitution steps, or draw a tree, and calculate joint probabilities by multiplication. Demonstrating the chain rule for multi-stage events often earns method marks even if final arithmetic requires a calculator.

Reversing conditionals: why Bayes matters
Often we observe evidence E and wish to update the probability of a hypothesis H. Bayes' theorem gives a principled way to compute P(H|E) using the known likelihood P(E|H) and prior probability P(H). It formalises how to revise beliefs in light of new data and is central to diagnostic reasoning, decision making and many applied problems.

Law of total probability
If H1, H2, ..., Hn partition the sample space (i.e., are mutually exclusive and exhaustive), then for any event E, P(E) = Σ_{k=1}^n P(E|H_k) P(H_k). This expresses the overall chance of evidence E by summing contributions from each possible cause weighted by its prior probability. The denominator in Bayes' formula is exactly this total probability and ensures the posterior probabilities sum to one.

Bayes' theorem formula
For a hypothesis Hi, Bayes' theorem is P(Hi|E) = P(E|Hi) P(Hi) / Σ_{k=1}^n P(E|Hk) P(Hk). The numerator is the likelihood-weighted prior; the denominator normalises across all hypotheses. Interpretationally, we start with prior beliefs P(Hi), observe evidence E summarised by likelihoods P(E|Hi), and compute posterior beliefs P(Hi|E).

Worked approach and care
Step 1: identify mutually exclusive hypotheses that cover all possibilities. Step 2: state priors P(Hi). Step 3: obtain likelihoods P(E|Hi). Step 4: compute P(E) by total probability. Step 5: apply Bayes' formula. A common pitfall is confusing P(E|H) with P(H|E); Bayes precisely relates the two but requires priors. When priors are small, even a test with good sensitivity and specificity can yield a low posterior.

Applications and examples
Bayes' theorem explains medical test interpretation: sensitivity (true positive rate) and specificity (true negative rate) combine with disease prevalence to give the post-test probability. It is used in forensic science to weigh evidence, in spam filters to update probability an email is spam given features, and in machine learning for naive Bayes classifiers (which assume conditional independence of features given the class).

Practical calculation tips
Use a table listing each hypothesis, its prior, likelihood P(E|Hi) and the product prior×likelihood. Sum these products to get P(E). Then divide each product by the sum to obtain posteriors. Showing the table in exams clarifies work and reduces arithmetic errors. For multi-stage processes, drawing a tree with branch weights equal to prior×likelihood makes the denominator and numerators visually clear.

Formal definition of independence
Two events A and B are independent if knowing that one occurred does not change the probability of the other. Formally A and B are independent if P(A ∩ B) = P(A) P(B). Equivalently, if P(B) > 0, independence means P(A|B) = P(A). This definition extends to random variables and collections of events as well.

Pairwise vs mutual independence
For three events A, B, C, pairwise independence means each pair is independent: P(A∩B)=P(A)P(B), P(B∩C)=P(B)P(C), and P(A∩C)=P(A)P(C). Mutual (or complete) independence additionally requires P(A∩B∩C)=P(A)P(B)P(C). Mutual independence is stronger than pairwise independence; three events can be pairwise independent but not mutually independent. Always check higher-order intersections when more than two events are involved.

Consequences and useful facts
If A and B are independent then complements are also independent: A' and B are independent, A and B' are independent, and A' and B' are independent. For independent events the probability of intersections factorises, which greatly simplifies calculations for sequences of independent trials: P(∩_{i=1}^n A_i) = ∏_{i=1}^n P(A_i).

Testing independence in problems
To test independence compute P(A∩B) and compare with P(A)P(B). For empirical data we might test independence statistically, but in many textbook problems independence is assumed by symmetry or process design: e.g., outcomes of different fair coin tosses are independent; outcomes of draws without replacement are generally dependent because the first draw alters the pool.

Cautionary examples
Independence is a modelling assumption that must be justified. For example, the events 'student studies hard' and 'scores high' are not independent in practice. Also, independence of indicators does not imply independence of functions of those indicators unless special conditions hold. Remember that zero covariance (or correlation) does not imply independence except in special families like the joint normal distribution.

Problem-solving tips
Explicitly state when independence is assumed. In calculations, when independence holds, replace conditional probabilities by unconditional ones: P(A|B)=P(A). This reduces complexity, especially for sums of independent random variables where variances also add. If in doubt, compute P(A∩B) directly from the model to check whether it equals P(A)P(B).

Random variable concept
A random variable is a rule that assigns a numerical value to each outcome of an experiment. Discrete random variables take values from a countable set such as {0,1,2,...} or a finite set. We use capital letters like X to denote random variables and lowercase x to denote their values.

Probability mass function (PMF)
The PMF p_X(x) = P(X = x) gives the probability that X equals a particular value x. By definition p_X(x) ≥ 0 for all x and the sum over all possible x equals 1: Σ_x p_X(x) = 1. The PMF fully describes the distribution of a discrete random variable and allows computation of any probability by summation.

Support
The support of X is the set of x values with p_X(x) > 0. Listing the support is key in calculations; probabilities are sums over the support. For example, for the number of heads in three coin tosses, support = {0,1,2,3}.

Calculating probabilities from PMF
To find P(a ≤ X ≤ b) sum p_X(x) over x = a to b. For events like ‘X is even’, sum p_X(x) over all even x in the support. When dealing with joint discrete variables (X,Y), the joint PMF p_{X,Y}(x,y) replaces the single-variable PMF and must satisfy non-negativity and total sum one.

Representations
PMFs are often presented in tables for small supports, or by formulas for common distributions (Binomial, Poisson, Geometric). Plotting a bar chart (probability histogram) of p_X(x) helps visualise concentration of probability and identify modes and skewness.

Why PMFs matter
PMFs allow calculation of expectations, variances and probabilities of complex events. Understanding PMFs is the first step toward joint distributions, marginalisation, conditioning and transformations of random variables. A clear PMF also helps check model assumptions: the probabilities should sum to one and reflect known constraints.

Practical tips
Always state the support explicitly, check normalization, and use summation notation carefully. When asked for a distribution of a function of X, either compute the PMF of the new variable by summing pre-image probabilities or use expectation identities if only moments are required.

Expectation as average
The expectation (or mean) of a discrete random variable X, denoted E[X], is the long-run average value of X if the experiment were repeated many times. It is calculated as the weighted sum of values with their probabilities: E[X] = Σ_x x p_X(x). Expectation is linear: E[aX + b] = aE[X] + b and E[X + Y] = E[X] + E[Y] for any X and Y (linearity holds even if X and Y are dependent).

Variance measures spread
Variance Var(X) measures how much X typically deviates from its mean. It is defined as Var(X) = E[(X − μ)^2] where μ = E[X]. For computation it is often easier to use Var(X) = E[X^2] − (E[X])^2, where E[X^2] = Σ x^2 p_X(x). Variance has squared units; standard deviation σ = √Var(X) returns to original units and is often used to describe typical deviation.

Properties and useful rules
For a constant c, Var(c)=0. For a random variable X and constant a, Var(aX)=a^2 Var(X). If X and Y are independent then Var(X+Y)=Var(X)+Var(Y). If dependent, Var(X+Y)=Var(X)+Var(Y)+2Cov(X,Y) where covariance measures linear co-movement. Expectation of functions: E[g(X)] = Σ g(x) p_X(x) allows computation of moments and transforms.

Interpretation of moments
Mean provides a measure of centre while variance and standard deviation indicate dispersion. A small variance means values cluster near the mean; a large variance means values spread out. For skewed distributions mean and median differ; both should be reported when distribution is asymmetric.

Computational shortcuts
Use linearity to break E[X] into sums of simpler expectations when X is a sum of indicators. For example, the mean of a Binomial(n,p) equals sum of n Bernoulli(p) expectations: np. For variance, sum variances of independent Bernoulli variables to get np(1−p). These shortcuts simplify many exam problems and reduce arithmetic work.

Checking results
Variance must be non-negative; if a computed variance is negative there is an algebraic error. Compare computed mean with typical values in the support to see if it is reasonable. When combining variables, check units and signs on covariance terms.

Bernoulli trial basics
A Bernoulli trial is a single experiment with exactly two outcomes: success with probability p, and failure with probability q = 1 − p. The Bernoulli random variable X takes value 1 for success and 0 for failure. It is the simplest discrete distribution. Expectation E[X] = p and variance Var(X) = p(1 − p). Many problems build on Bernoulli trials by repeating them.

Binomial distribution from repeated trials
If n independent Bernoulli trials are performed, each with success probability p, the number of successes X follows the Binomial(n,p) distribution. The PMF is P(X = k) = C(n,k) p^k (1 − p)^{n − k} for k = 0,1,...,n. The combination term C(n,k) counts the unordered ways to choose which k trials are successes; p^k accounts for probabilities of those successes and (1−p)^{n−k} for the failures.

Mean and variance and intuition
Binomial mean E[X] = np because the expectation of a sum of independent Bernoulli variables adds up. Variance Var(X) = np(1 − p) follows from summing variances of independent Bernoullis. These results show how the expected number of successes scales with trials and how variability depends on both n and p.

Applications and model assumptions
Binomial models apply when: (1) the number of trials n is fixed, (2) trials are independent, (3) each trial has identical success probability p, and (4) each trial outcome is binary. Examples include number of heads in coin tosses, number of defective items in a sample when sampling with replacement from a large population, and number of students passing a test if all attempts are independent with the same success chance.

Calculations and approximations
For moderate n and p, compute probabilities directly using the PMF or tables. For large n, use normal approximation with mean np and variance np(1−p) after continuity correction when appropriate. For small p and large n such that λ = np is moderate, the Poisson approximation can be used: Binomial(n,p) ≈ Poisson(λ).

Practical exam advice
When solving Binomial problems, define X clearly, state n and p, and show substitution into the PMF for exact probabilities. For cumulative probabilities show the required sum. Use linearity of expectation for mean problems and variance sum rules to compute variances efficiently. State model assumptions to justify using the binomial distribution.

Geometric distribution — waiting time to first success
The geometric distribution models the number of independent Bernoulli trials needed to obtain the first success, when each trial has the same success probability p. There are two conventions: one counts the trial number of the first success (support {1,2,...}) with PMF P(X=k) = (1−p)^{k−1} p; another counts the number of failures before the first success (support {0,1,2,...}) with PMF P(Y=k) = (1−p)^k p. Clarify which convention is used in a question.

Key properties
Geometric distribution has the memoryless property: P(X > s+t | X > s) = P(X > t). This means that once you have observed s failures, the distribution of additional waiting time is the same as starting new. The mean waiting time (trial-number convention) is E[X] = 1/p and variance Var(X) = (1−p)/p^2. These results reflect how rarer successes (small p) produce larger expected waiting times and larger variability.

Negative binomial: generalising to r successes
The negative binomial distribution generalises the geometric by modelling the number of trials needed to achieve r successes. If X denotes the trial number on which the r-th success occurs, then for k ≥ r, P(X=k) = C(k−1, r−1) p^r (1−p)^{k−r}. Intuitively, choose r−1 successes among the first k−1 trials and require the k-th trial to be a success. When r = 1 this reduces to the geometric distribution. The mean is E[X] = r/p and variance Var(X) = r(1−p)/p^2.

Applications and modelling
Use geometric or negative binomial models for waiting-time problems: the number of trials until first success (geometric) or until r-th success (negative binomial). Examples include number of customers until first sale, number of inspections until r defective items found, and number of trials until r successes in reliability testing. Independence and constant success probability across trials are key assumptions.

Computational tips
Be careful whether the question asks for the trial number containing the r-th success or the number of failures before the r-th success; choose the appropriate PMF accordingly. Use the memoryless property to simplify conditional probability calculations for geometric distributions. For expected values, derive by summing or using known formulas; for variance use the standard expressions to save algebra time.

Exam preparation advice
Practice small examples by listing outcomes to build intuition. Show the binomial coefficient reasoning in answers: C(k−1,r−1) counts placements of previous successes. Writing the PMF and stating support and conventions clearly earns marks and prevents misunderstanding in multi-part questions.

Poisson distribution basics
The Poisson distribution models the number of events that occur in a fixed interval of time or space under specific assumptions: events occur independently, the average rate (mean number of events) λ is constant over the interval, and two events cannot occur exactly simultaneously in an infinitesimally small subinterval. The PMF is P(X=k) = e^{−λ} λ^k / k! for k = 0,1,2,... The parameter λ is both the mean and variance, which is a distinctive property.

When to use Poisson model
Use the Poisson model for rare events in a large population or when counting occurrences in time or space, for example, number of calls to a helpline per hour, number of accidents at an intersection per month, or number of defects per unit length. Key is approximate independence and a roughly constant average rate.

Poisson as limit of Binomial
The Poisson distribution arises as a limiting case of the Binomial(n,p) when n is large and p is small with np = λ fixed. In practice, if n ≥ 50 and p ≤ 0.1 (a rule of thumb) and np ≤ about 10–20, the Poisson approximation is often reasonable. This approximation greatly simplifies calculations because factorials and combinations in the Binomial formula are replaced by the simple Poisson PMF.

Properties and combinations
Poisson variables have additivity: if X ∼ Poisson(λ1) and Y ∼ Poisson(λ2) are independent then X + Y ∼ Poisson(λ1 + λ2). This property is useful for modelling counts coming from independent sources. The equality of mean and variance allows a quick check: if sample variance is approximately equal to sample mean, a Poisson model may fit; if variance is much larger, overdispersion suggests another model is needed.

Computational advice
Compute probabilities using the PMF or cumulative sums for tail probabilities. For P(X ≥ k) use 1 − P(X ≤ k−1). For moderate λ, tables or calculators help. When approximating Binomial by Poisson, compute λ = np and then use Poisson PMF values. When using Poisson to approximate real data, check if event independence and constant rate assumptions are reasonable.

Exam hints
State assumptions clearly when using the Poisson model. In approximation questions show calculation of λ and justify why approximation is appropriate. For compound processes, use additivity to combine rates and compute probabilities for total counts.

Joint PMF for pairs
When two discrete random variables X and Y are considered together, their behaviour is described by the joint probability mass function p_{X,Y}(x,y) = P(X = x, Y = y). The joint PMF must satisfy p_{X,Y}(x,y) ≥ 0 for all (x,y) and Σ_{x,y} p_{X,Y}(x,y) = 1. A joint PMF captures dependencies between X and Y and is the starting point for computing marginals, conditionals, expectations and covariances.

Marginal PMFs
Marginalising means summing out the other variable to get the distribution of one variable alone: p_X(x) = Σ_y p_{X,Y}(x,y) and p_Y(y) = Σ_x p_{X,Y}(x,y). Marginals are useful when only one variable's distribution is needed. If the joint PMF is given in a table, summing rows or columns gives marginals immediately.

Conditional PMFs
The conditional PMF of X given Y=y is p_{X|Y}(x|y) = p_{X,Y}(x,y) / p_Y(y) when p_Y(y) > 0. This expresses the distribution of X when it is known that Y took the value y. Conditional PMFs are indispensable in sequential or dependent modelling and are the discrete analogue of conditional densities in continuous cases.

Independence and factorisation
X and Y are independent if and only if p_{X,Y}(x,y) = p_X(x) p_Y(y) for all x,y. Independence implies that knowing Y gives no information about X and vice versa. In independence checks compute marginals and see if their product recovers the joint; a counterexample shows dependence.

Working with joint distributions
Tables provide a convenient format when supports are small. For larger supports, provide formulas or use indicator sums. Joint distributions allow computation of joint expectations E[g(X,Y)] = Σ_{x,y} g(x,y) p_{X,Y}(x,y), marginal expectations, and conditional expectations E[X|Y=y] = Σ_x x p_{X|Y}(x|y).

Applications
Joint PMFs model paired experiments such as two dice, number of successes in two groups, or correlated counts. They lead directly to covariance and correlation calculations and to computing distributions of sums via convolution. In exams, laying out a clear joint-table, computing marginals and then conditionals step-by-step demonstrates method and avoids algebraic errors.

Covariance: definition and meaning
Covariance measures the tendency of two random variables X and Y to vary together. It is defined as Cov(X,Y) = E[(X − E[X])(Y − E[Y])] = E[XY] − E[X]E[Y]. A positive covariance suggests that when X is above its mean Y tends also to be above its mean; a negative covariance suggests opposite movement. Covariance is measured in units that are the product of units of X and Y, so its magnitude is hard to interpret directly.

Correlation: standardised covariance
Correlation is a dimensionless measure obtained by dividing covariance by the product of standard deviations: Corr(X,Y) = Cov(X,Y) / (σ_X σ_Y). Correlation lies between −1 and 1. A correlation of 1 or −1 indicates a perfect linear relationship while correlation 0 indicates no linear association (but not necessarily independence).

Properties and algebra
Cov(X,X) = Var(X). For constants a,b,c,d, Cov(aX+b, cY+d) = ac Cov(X,Y). If X and Y are independent, then Cov(X,Y) = 0; however the converse is not always true. Variance of a sum uses covariance: Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y). For many independent components, covariances vanish and variances simply add.

Computation from joint PMF
To compute Cov(X,Y) from a joint PMF, find E[X], E[Y], and E[XY] where E[XY] = Σ_{x,y} xy p_{X,Y}(x,y). Then subtract E[X]E[Y] from E[XY]. For discrete small-support problems a table helps compute E[X], E[Y], and E[XY] cleanly by summing rows and columns with products weighted by probabilities.

Interpretation and limits
Correlation near ±1 indicates strong linear association but does not capture nonlinear relationships. A correlation of 0 means no linear association; variables may still be related nonlinearly. Covariance sign indicates direction of linear association; magnitude depends on units. Use correlation to compare association strength across different variable pairs.

Practical tips
When asked for covariance or correlation in exam problems, show calculations of E[X], E[Y], E[XY] explicitly and then compute Cov and Corr. State whether independence is assumed and use Var(X+Y) formulas if helpful. For sample data, note the difference between population covariance and sample covariance conventions but for theoretical PMFs use population definitions given above.

Why transform random variables?
Often we are interested not in X itself but in a function Y = g(X) such as Y = 2X + 1, Y = X^2, or Y = I{X > 0} (an indicator). Finding the distribution of Y or its expectation is a common task. Methods differ depending on whether g is one-to-one or many-to-one.

One-to-one transforms
If g is one-to-one on the support of X, then the PMF of Y is straightforward: p_Y(y) = p_X(g^{-1}(y)). Values map bijectively so probabilities transfer directly. Expectations transform by substitution: E[Y] = E[g(X)] = Σ_x g(x) p_X(x). This avoids constructing the PMF of Y explicitly when only the expectation is needed.

Many-to-one transforms
If multiple x values map to the same y, sum their probabilities: p_Y(y) = Σ_{x: g(x)=y} p_X(x). For example, if Y = X mod 2, then p_Y(0)=Σ_{even x} p_X(x). Carefully list pre-images of each y value and sum corresponding probabilities. This method scales well when the support of X is moderate and g compresses values.

Expectations of functions
General rule: E[g(X)] = Σ_x g(x) p_X(x) for a discrete X. Use this to compute moments like E[X^2] or functions like E[1/(1+X)] without finding full PMF of transformed variable. For joint variables, E[g(X,Y)] = Σ_{x,y} g(x,y) p_{X,Y}(x,y).

Transforms for sums and convolution
When Y = X + Z and X and Z are independent, the PMF of the sum is the convolution p_Y(s) = Σ_x p_X(x) p_Z(s − x). Convolution combines distributions and is used to find PMFs of totals of independent random variables, such as total number of successes across different groups.

Variance under transformation
For linear transforms Y=aX+b, Var(Y) = a^2 Var(X). For nonlinear transforms apply Var(Y) = E[Y^2] − (E[Y])^2, computing E[Y^2] by summation. Sometimes approximations or Taylor expansions (beyond Class 12) are used to approximate variance of nonlinear transforms, but direct summation suffices for discrete finite supports.

Practical exam tips
When asked for PMF of Y write the mapping clearly, list each y value and its pre-image set, and compute p_Y(y) by summation. For expectations of g(X) state and use E[g(X)] = Σ g(x) p_X(x). For sums, show a small convolution table or use generating functions if allowed, and always state independence when using convolution simplifications.

Distribution of a sum
The distribution of the sum S = X + Y of two discrete random variables is found by summing the joint probabilities of all pairs (x,y) such that x + y = s. If X and Y are independent, the formula simplifies to convolution of their PMFs: p_S(s) = Σ_x p_X(x) p_Y(s − x). Convolution is a discrete sliding-sum that accumulates probability mass along diagonals in the joint probability table.

When independence helps
Independence is key: convolution p_S(s) = Σ_x p_X(x) p_Y(s − x) holds when X and Y are independent. If dependent, you must use the joint PMF: p_S(s) = Σ_x p_{X,Y}(x, s−x). For many practical problems variables are independent (separate experiments) and convolution is the standard tool.

Moments of sums
Expectations add: E[X+Y] = E[X] + E[Y]. For independent variables variances add too: Var(X+Y) = Var(X) + Var(Y). For dependent variables add covariance: Var(X+Y) = Var(X) + Var(Y) + 2Cov(X,Y). These rules are frequently used to compute means and variances of totals without computing full distribution.

Examples and special cases
Sums of independent Binomial variables with the same success probability p produce another Binomial with parameter equal to the sum of the n's: Binomial(n1,p) + Binomial(n2,p) = Binomial(n1+n2,p). Similarly, sums of independent Poisson(λ1) and Poisson(λ2) are Poisson(λ1+λ2). The sum of independent Bernoulli variables is Binomial. Such closure properties are very useful in modelling and problem solving.

Computational approach
To compute p_S(s) by hand for small supports, form a table of p_X(x) along rows and p_Y(y) along columns, fill the product table p_X(x)p_Y(y) if independent, and sum the products along diagonals where x+y equals s. This tabular method reduces errors and clarifies the required summations.

Generating functions
Probability generating functions turn convolution into multiplication: G_{X+Y}(s) = G_X(s) G_Y(s) for independent non-negative integer-valued X and Y. While PGFs are advanced, they offer algebraic shortcuts for sums of many independent variables and are introduced elsewhere in the unit.

Exam tips
State independence before using convolution. For sums of many identical independent components use moment formulas to get mean and variance quickly. When asked for full PMF do the diagonal sums carefully and show the table or algebraic convolution steps to earn method marks.

Law of Large Numbers (LLN) — stability of averages
The law of large numbers says that the sample average of many independent identically distributed random variables converges to the common expected value as the number of observations increases. Informally, if you repeat an experiment many times and compute the average of outcomes, this average will be close to the theoretical mean with high probability when the sample size is large. For example, the proportion of heads in many coin tosses will be close to 0.5. LLN explains why frequencies stabilise and supports empirical estimation of probabilities.

Types of convergence (brief)
There are technical forms of LLN: weak and strong law. The weak law (sufficient for most class 12 purposes) states convergence in probability; the strong law says convergence almost surely, a stronger statement. For examinations you need the intuitive statement: averages settle down to the expected value for large samples, provided independence and identical distribution with finite mean hold.

Central Limit Theorem (CLT) — shape of sums
The central limit theorem says that the sum (or average) of many independent identically distributed random variables with finite mean μ and variance σ^2, when properly centred and scaled, approximately follows a normal distribution for large sample sizes. Specifically, (Σ_{i=1}^n X_i − nμ) / (σ√n) tends toward the standard normal distribution as n grows. CLT explains why normal distributions appear so widely in nature and justifies normal approximations for sums and sample means.

Applications and approximations
CLT is used to approximate Binomial(n,p) by Normal(np, np(1−p)) when n is large and p not near 0 or 1. Use continuity correction for discrete-to-continuous approximation: P(X ≤ k) ≈ Φ((k + 0.5 − μ)/σ). For averages, use Normal(μ, σ^2/n) to find approximate probabilities about sample means. These approximations simplify calculations and provide intuition for sampling variability.

Limitations and conditions
Both LLN and CLT require independence (or weak dependence) and finite variance. CLT is asymptotic — how large n must be depends on skewness of the original distribution. For heavy-tailed distributions or dependent data, approximations may be poor and different limit results may apply. Also CLT speaks about distributional shape, not exact probabilities for small n.

Practical significance
LLN gives practical confidence that empirical averages approximate true expectations as sample size grows. CLT allows use of normal tools (confidence intervals, hypothesis tests) as approximations in many applied problems. For exam questions demonstrate understanding by stating assumptions, showing standardisation steps, and applying normal approximations only when conditions are reasonable.

Translating reality into a model
Modelling with probability means choosing a mathematical structure that captures the essential randomness of a real situation. Start by identifying outcomes, deciding whether trials are independent, whether events are rare, and whether order matters. These choices guide selection among Binomial, Poisson, geometric or joint-distribution models. Good modelling balances simplicity and realism so answers are interpretable and calculations are feasible.

Quality control and manufacturing
In manufacturing, the number of defective items in samples can often be modelled by a Binomial distribution if sampling is with replacement or the batch is large. When defects are rare, Poisson approximation simplifies calculations. Models help set acceptance criteria for lots, estimate defect rates and compute risk of shipping defective batches. Clearly state assumptions: independence of defects and constant defect probability are typical but must be justified.

Reliability and waiting times
Geometric and negative binomial distributions model waiting times until one or several failures or successes. Poisson processes model arrival rates: calls to a call centre, emails per hour, or decay events. Use these to estimate expected waiting time, probability of zero arrivals, or to schedule resources. The key is identifying stationarity (constant rate) and independence of increments for Poisson models.

Medical testing and diagnostics
Bayes' theorem is central in diagnostics: test results update prior probabilities of disease. Sensitivity and specificity combine with disease prevalence to produce posterior probability. Model carefully to include false positives and negatives; communicate results in percentages or natural frequencies to avoid misunderstandings. In screening programmes small prevalence can lead to many false positives even with high test accuracy.

Games, risk and decision-making
Probability models are used to compute expected gains or losses in games and investments. Expectation gives fair game values and variance measures risk. Understanding independence prevents common fallacies like gambler's fallacy; randomness produces streaks but these do not change underlying probabilities in independent trials.

Checking models against data
Always compare model predictions (mean, variance, shape) with observed data. If sample variance greatly exceeds mean for count data, Poisson may be unsuitable (overdispersion) and a different model is needed. Simple diagnostic checks and plotting empirical counts against theoretical PMFs help detect model mismatch. Refine models by relaxing independence or allowing varying rates when necessary.

Communication and assumptions
State model assumptions explicitly in answers, describe the chosen distribution and justify why it fits, compute required probabilities, and discuss limitations. Good practice in exams and real life is to present both numeric results and a brief interpretation in context.

Motivation for generating functions
Probability generating functions (PGFs) provide a compact algebraic representation of the PMF of a non-negative integer-valued random variable. The PGF for X is G_X(s) = E[s^X] = Σ_{k=0}^∞ p_X(k) s^k where p_X(k)=P(X=k). PGFs encode the entire distribution in a power series and turn convolution (sums of independent variables) into multiplication. They are powerful tools for deriving moments and studying sums of independent random variables.

Basic properties
Key properties include G_X(1)=1 because probabilities sum to 1. The derivative at s=1 gives the first moment: G'_X(1) = E[X]. Higher derivatives give factorial moments: G''_X(1) = E[X(X−1)], and so on. These relations allow moments to be computed by differentiation without summing over the PMF directly, which can be efficient for distributions with closed-form PGFs.

Convolution becomes multiplication
If X and Y are independent non-negative integer-valued variables, then the PGF of their sum is the product: G_{X+Y}(s) = G_X(s) G_Y(s). This property mirrors how characteristic functions and moment-generating functions behave and simplifies finding distributions of sums: multiply PGFs and, if needed, expand the resulting power series to read off coefficients as probabilities.

Common PGFs
Many standard distributions have simple PGFs: Bernoulli(p) has G(s)=1−p+ps; Binomial(n,p) has G(s)=(1−p+ps)^n by independence of n Bernoulli trials; Poisson(λ) has G(s)=exp(λ(s−1)). These closed forms allow quick derivation of means and variances: differentiate and evaluate at s=1. For example differentiating G(s) for Poisson gives E[X]=λ and Var(X)=λ directly.

Applications and limits
PGFs are especially useful when dealing with sums of many independent variables, branching processes, or deriving distributions of compound sums. In Class 12, use PGFs to show additivity of Poisson and to derive Binomial moments. While full inversion from PGF to PMF requires series expansion, for many exam problems it suffices to present PGF identities and extract moments.

Practice tips
Compute PGFs for simple distributions, practice differentiating to get E[X] and E[X(X−1)], and use PGF multiplication to combine independent distributions. In answers, state assumptions (non-negative integer-valued and independence) before using PGF properties to avoid misuse.

Recognise the problem type
ICSE/ISC probability questions commonly fall into categories: counting and combinatorics with equally likely outcomes, conditional probability and Bayes' theorem, binomial and Poisson computations, expectations and variances, joint distributions and covariance, and distribution of sums. Quickly identifying which category a question belongs to guides your approach and saves time in exams.

Structured problem-solving steps
Adopt a clear routine: (1) Define the experiment and write the sample space S or state clearly why S is large. (2) Specify the random variable(s) and their support. (3) Decide whether outcomes are equally likely; if so use counting; if not, assign outcome probabilities. (4) Choose the correct distribution or counting formula and compute numerators and denominators separately. (5) Show intermediate steps: marginals, conditionals, E[X], E[X^2] when needed. Clear structure gains method marks even when arithmetic is tricky.

Counting vs distribution approach
For finite equally likely problems, combinatorics (nCr, nPr) is commonly fastest. For repeated independent trials, use Binomial or Geometric. For rare events or large populations use Poisson approximation. Always justify model assumptions briefly: independence, identical trials, rarity, replacement vs no replacement.

Conditional and Bayes' questions
For conditional problems draw a probability tree and label branch probabilities. Use the law of total probability to compute denominators for Bayes' theorem and present numerator/denominator clearly. A small table of hypotheses, priors, likelihoods and weighted terms reduces errors and clearly shows the method.

Expectations and variances in exams
Use linearity of expectation to simplify calculations: E[Σ X_i] = Σ E[X_i] requires no independence. For variance, if independence holds, Var(Σ X_i) = Σ Var(X_i). Show intermediate E[X] and E[X^2] values rather than skipping steps. For sums of indicators use known Binomial results to avoid heavy summation.

Presentation and common errors
Write symbols clearly, box final answers and state units or probability ranges. Common mistakes are swapping P(A|B) and P(B|A), forgetting to adjust counts for order, and assuming independence without justification. Check that probabilities lie between 0 and 1 and that totals of PMFs sum to 1 where applicable.

Exam practice tips
Solve past papers, time yourself, and learn to decide quickly whether to compute exact sums or use approximations (Poisson or Normal). Show reasons for approximations and always state assumptions; this earns method marks and demonstrates understanding even when arithmetic is approximate.