Binomial Distribution - College Statistics

SECTION 1

Historical Context & Motivation

Long before the formal apparatus of probability theory existed, scholars grappled with a deceptively simple question: if an event has some fixed chance of occurring on each attempt, how likely is it that the event occurs exactly k times out of n attempts? The answer to this question required centuries of mathematical development—from the earliest combinatorial reasoning of medieval mathematicians to the rigorous probability calculus of the Enlightenment. The binomial distribution emerged from this lineage as one of the most fundamental discrete probability distributions, and it remains indispensable in fields ranging from quality control to clinical trials.

1654 • Pascal & Fermat Correspondence

Blaise Pascal and Pierre de Fermat exchanged letters on the 'problem of points,' laying the combinatorial groundwork for counting outcomes in repeated trials and formalizing the idea of expected value.

1713 • Bernoulli's Ars Conjectandi

Jacob Bernoulli's posthumous masterpiece introduced the formal concept of Bernoulli trials—independent experiments with exactly two outcomes—and derived the binomial probability formula. His Law of Large Numbers connected empirical frequency to theoretical probability.

1733 • De Moivre's Normal Approximation

Abraham de Moivre demonstrated that for large n, the binomial distribution is well-approximated by a bell-shaped curve, anticipating the central limit theorem and establishing the Gaussian approximation to the binomial.

1837 • Poisson's Limit Theorem

Siméon Denis Poisson showed that when n is large and p is small, the binomial distribution converges to what we now call the Poisson distribution, expanding the family of models available for rare events.

1900s • Modern Statistical Practice

The binomial distribution became a cornerstone of hypothesis testing, quality control (acceptance sampling), clinical trial design, and machine learning classification metrics throughout the twentieth century.

The central question the binomial distribution answers is straightforward yet powerful: given a process with a constant probability of success on each independent trial, what is the probability of observing exactly k successes in n trials? Understanding this distribution is the gateway to grasping more complex models such as the negative binomial, hypergeometric, and multinomial distributions.

SECTION 2

Core Principles & Definitions

The binomial distribution rests on a precise set of assumptions that collectively define what statisticians call a binomial experiment. When all four conditions below are satisfied, the random variable X counting the number of successes follows a binomial distribution. Violating any one of these assumptions means a different model—such as the hypergeometric or negative binomial—may be more appropriate.

Fixed Number of Trials (n)

The experiment consists of a predetermined, finite number of trials n. You decide how many times you will repeat the process before data collection begins.

Two Mutually Exclusive Outcomes

Each trial results in exactly one of two outcomes, conventionally labeled success and failure. These labels are arbitrary and need not carry positive or negative connotations.

Constant Probability (p)

The probability of success p remains the same from trial to trial. Consequently, the probability of failure is q = 1 − p on every trial.

Independence of Trials

The outcome of any one trial does not influence the outcome of any other trial. This means the joint probability of a sequence of outcomes equals the product of the individual probabilities.

✦

KEY TAKEAWAY

Think of the binomial distribution as a coin-flipping machine. You load the machine with a coin that lands heads with probability p, press the button n times, and ask: how many heads will I get? Each press is independent, the coin doesn't change between presses, and you decide the number of presses in advance. If any of those conditions fail—the coin wears down, the machine jams after a certain result, or you keep pressing until you see a head—you need a different distribution.

When a random variable X satisfies these conditions, we write X ~ Bin(n, p), indicating that X follows a binomial distribution with parameters n (number of trials) and p (probability of success). The support of the distribution is the set {0, 1, 2, …, n}.

SECTION 3

Visual Explanation — The PMF in Action

The probability mass function (PMF) of the binomial distribution assigns a probability to each possible number of successes. The shape of this PMF depends critically on the parameters n and p. When p = 0.5, the distribution is perfectly symmetric; when p deviates from 0.5, the distribution becomes skewed. The following diagram compares three binomial PMFs—all with n = 10—but with different values of p.

Three overlaid binomial PMFs with n = 10. The cyan bars (p = 0.2) peak near k = 2 and are right-skewed. The violet bars (p = 0.5) are symmetric about k = 5. The pink bars (p = 0.8) peak near k = 8 and are left-skewed—a mirror image of the p = 0.2 case.

Several visual features are worth noting. First, the mode—the most probable value of k—sits near np for each distribution. Second, the spread of the bars widens as p approaches 0.5, because the variance npq is maximized when p = q = 0.5. Third, observe the mirror symmetry: Bin(10, 0.2) reflected about k = 5 yields exactly Bin(10, 0.8). This is a general property—replacing every success with a failure and vice versa swaps p and q.

SECTION 4

Mathematical Framework

The binomial PMF can be derived from first principles. Consider n independent Bernoulli trials, each with success probability p. Any specific sequence containing exactly k successes and n − k failures has probability p^k(1 − p)^n−k by independence. The number of such sequences is the binomial coefficient C(n, k), yielding the PMF.

BINOMIAL PMF

P(X = k) = C(n, k) × p^k × (1 − p)^(n−k), k = 0, 1, 2, …, n

where C(n, k) = n! / [k!(n − k)!] is the binomial coefficient ("n choose k"), p is the probability of success on each trial, and n is the number of trials.

EXPECTED VALUE

E[X] = μ = n × p

The mean is derived by summing k × P(X = k) over all k, or more elegantly by noting that X = X₁ + X₂ + … + Xₙ where each Xᵢ ~ Bernoulli(p), so E[X] = n × E[Xᵢ] = np.

VARIANCE & STANDARD DEVIATION

Var(X) = σ² = n × p × (1 − p), σ = √[n × p × (1 − p)]

Since the Bernoulli random variables X₁, …, Xₙ are independent, Var(X) = n × Var(Xᵢ) = n × p(1 − p). The standard deviation σ measures the typical spread around the mean.

MOMENT GENERATING FUNCTION

M_X(t) = (1 − p + p × e^t)^n

The MGF is the product of n identical Bernoulli MGFs. Differentiating and evaluating at t = 0 produces the mean and higher moments. The MGF also provides a clean proof that the sum of independent binomials (with the same p) is again binomial.

💡

Derivation Insight

The binomial coefficient C(n, k) counts the number of ways to arrange k successes among n trials. This combinatorial factor is the bridge between the probability of a single specific outcome sequence and the total probability of obtaining exactly k successes regardless of ordering.

SECTION 5

Detailed Properties & CDF

Beyond the PMF, several properties of the binomial distribution are essential for applied work. The cumulative distribution function (CDF) gives the probability of observing at most k successes and is defined as F(k) = P(X ≤ k) = Σ from i = 0 to k of C(n, i) × pⁱ × (1 − p)ⁿ⁻ⁱ. There is no closed-form expression for this sum, so practitioners rely on tables, software, or the regularized incomplete beta function. The skewness of the distribution is (1 − 2p) / √[np(1 − p)], confirming the visual observation that the PMF is symmetric only when p = 0.5. The kurtosis is 3 + (1 − 6p(1 − p)) / [np(1 − p)], which approaches the Gaussian value of 3 as n grows.

The CDF of Bin(10, 0.5) is a step function. Each jump at integer k equals P(X = k), the corresponding PMF value. The filled circles indicate that F(k) includes the probability at k, consistent with the convention P(X ≤ k). The function rises most steeply near the mean (k = 5) where PMF values are largest.

Property	Formula / Value
Parameters	n ∈ {1, 2, 3, …}, p ∈ [0, 1]
Support	k ∈ {0, 1, 2, …, n}
Mean	np
Variance	np(1 − p)
Skewness	(1 − 2p) / √[np(1 − p)]
Mode	⌊(n + 1)p⌋ or ⌊(n + 1)p⌋ − 1
MGF	$(1 − p + pe^t)^n$

Key properties of the Bin(n, p) distribution

SECTION 6

Worked Example

A pharmaceutical company tests a new vaccine that has a 70% efficacy rate (i.e., 70% of vaccinated individuals develop immunity). In a clinical trial, 15 participants receive the vaccine. We wish to find the probability that exactly 12 of them develop immunity.

Vaccine Efficacy — Finding P(X = 12)

1
Step 1 — Verify Binomial Conditions

Check the four conditions: (1) Fixed number of trials: n = 15 participants. (2) Two outcomes per trial: immunity (success) or no immunity (failure). (3) Constant probability: p = 0.70 for each participant. (4) Independence: each participant's immune response is independent of others. All conditions are satisfied, so X ~ Bin(15, 0.70).
2
Step 2 — Identify Parameters and Target

We have n = 15, p = 0.70, q = 1 − 0.70 = 0.30, and we seek P(X = 12).

n = 15, p = 0.70, k = 12
3
Step 3 — Compute the Binomial Coefficient

C(15, 12) = 15! / (12! × 3!) = (15 × 14 × 13) / (3 × 2 × 1) = 2730 / 6 = 455.

C(15, 12) = 455
4
Step 4 — Evaluate the Probability Terms

Compute $p^k$ = $0.70^{12}$ ≈ 0.013841 and $q^{n−k}$ = $0.30^3$ = 0.027. The product of the probability terms is 0.013841 × 0.027 ≈ 0.0003737.

0.70¹² × 0.30³ ≈ 0.0003737
5
Step 5 — Multiply and State Result

P(X = 12) = C(15, 12) × $0.70^{12}$ × $0.30^3$ = 455 × 0.0003737 ≈ 0.1700. There is approximately a 17.0% probability that exactly 12 out of 15 vaccinated participants develop immunity.

P(X = 12) ≈ 0.1700 (17.0%)
6
Step 6 — Contextual Check

The mean of this distribution is μ = 15 × 0.70 = 10.5, and the standard deviation is σ = √(15 × 0.70 × 0.30) ≈ 1.775. The value k = 12 is roughly 0.85 standard deviations above the mean, so a probability of 17% is reasonable—neither too close to the center nor in the extreme tail.

SECTION 7

Strengths, Limitations & When to Use Alternatives

The binomial distribution is one of the most widely applied discrete models, but it is not universally appropriate. Understanding its limitations is as important as understanding its formula. The following table contrasts its strengths with situations where alternative distributions are preferable.

Strengths	Limitations	Alternative Model
Simple two-parameter model that is analytically tractable and computationally efficient	Assumes constant p; fails when success probability changes across trials	Beta-binomial or logistic regression for varying p
Exact probabilities available for any n and k via the PMF	Requires independence; sampling without replacement from finite populations violates this	Hypergeometric distribution for sampling without replacement
Well-understood approximations (normal, Poisson) for large n	Only models count of successes in fixed n; cannot model number of trials until r-th success	Negative binomial (or geometric for r = 1)
Foundation for exact binomial tests, confidence intervals, and power analysis	Restricts outcomes to two categories; does not handle multiple outcome types	Multinomial distribution for more than two categories

Binomial distribution: strengths, limitations, and alternatives

🎯

CHOOSING THE RIGHT MODEL

Think of the binomial distribution as the 'default template' for counting successes. Just as an engineer might start with a linear model and then add complexity, a statistician often starts with the binomial and checks whether the data's variance, dependence structure, or outcome space requires upgrading to a more flexible distribution. If the observed variance exceeds np(1 − p), this overdispersion is a strong signal that the binomial's independence or constant-p assumption is violated.

SECTION 8

Connection to the Normal & Poisson Distributions

Two classical approximation results connect the binomial distribution to continuous and other discrete distributions. These approximations are not merely computational shortcuts—they reveal deep structural relationships in probability theory and are foundational to understanding the central limit theorem.

Feature	Normal Approximation	Poisson Approximation
Regime	n large, p not too close to 0 or 1; commonly np ≥ 5 and n(1−p) ≥ 5	n large, p small, np = λ moderate (say λ < 20)
Approximating distribution	N(np, np(1 − p))	Poisson(λ = np)
Continuity correction	Yes: P(X ≤ k) ≈ Φ((k + 0.5 − np) / √[np(1−p)])	Not needed (both discrete)
Theoretical basis	De Moivre–Laplace theorem (special case of CLT)	Poisson limit theorem: lim as n→∞, p→0 with np = λ
Primary use	Hypothesis testing, confidence intervals for proportions	Modeling rare events: defects, mutations, accidents

Normal vs. Poisson approximations to the binomial

Looking forward, the binomial distribution serves as a springboard to more advanced topics. In Bayesian statistics, pairing a binomial likelihood with a Beta prior yields a Beta posterior for the unknown success probability, forming the Beta-binomial conjugate family. In generalized linear models, the binomial distribution is the response distribution underlying logistic regression. And in information theory, the entropy of a Bernoulli trial H(p) = −p log p − (1 − p) log(1 − p) governs the efficiency of binary coding schemes. These connections underscore that mastering the binomial is not merely an exercise in computing probabilities—it is an investment in the conceptual infrastructure of modern statistics.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A researcher draws 5 cards one at a time from a standard 52-card deck without replacement and records whether each card is a heart. She claims the number of hearts follows a Bin(5, 0.25) distribution. Identify the specific binomial assumption that is violated and explain why.

Show answer

The independence assumption is violated. When cards are drawn without replacement, the probability of drawing a heart on any trial depends on which cards were drawn previously. For example, if the first card is a heart, only 12 hearts remain among 51 cards, so P(heart on trial 2 | heart on trial 1) = 12/51 ≈ 0.235, not 0.25. The correct model is the hypergeometric distribution with N = 52, K = 13, and n = 5. The binomial could serve as an approximation if the population were much larger relative to the sample size (the '10% rule'), but with 5 draws from 52 cards, the dependence is non-negligible.

Question 2: BASIC CALCULATION

A fair six-sided die is rolled 8 times. Let X be the number of times a '6' appears. Calculate P(X = 2) and find the mean and standard deviation of X.

Show answer

Here X ~ Bin(8, 1/6). The PMF gives P(X = 2) = C(8, 2) × (1/6)² × (5/6)⁶ = 28 × (1/36) × (15625/46656) = 28 × 15625 / 1679616 ≈ 0.2605. The mean is μ = np = 8 × (1/6) ≈ 1.333. The variance is σ² = np(1 − p) = 8 × (1/6) × (5/6) ≈ 1.111, so the standard deviation is σ ≈ 1.054. A result of k = 2 is less than one standard deviation above the mean, consistent with its fairly high probability of about 26%.

Question 3: INTERMEDIATE

In a manufacturing process, 4% of circuit boards are defective. A quality inspector randomly selects 20 boards. Find the probability that at most 2 boards are defective. Should the inspector use the exact binomial or the Poisson approximation? Justify your choice and compute both answers.

Show answer

Let X ~ Bin(20, 0.04). We need P(X ≤ 2) = P(X = 0) + P(X = 1) + P(X = 2). Exact binomial: P(X = 0) = 0.96²⁰ ≈ 0.4420, P(X = 1) = C(20,1) × 0.04 × 0.96¹⁹ ≈ 20 × 0.04 × 0.4604 ≈ 0.3683, P(X = 2) = C(20,2) × 0.04² × 0.96¹⁸ ≈ 190 × 0.0016 × 0.4796 ≈ 0.1458. So P(X ≤ 2) ≈ 0.9561. For the Poisson approximation, λ = np = 0.8. P(X = 0) = e⁻⁰·⁸ ≈ 0.4493, P(X = 1) = 0.8e⁻⁰·⁸ ≈ 0.3595, P(X = 2) = 0.32e⁻⁰·⁸/2 ≈ 0.1438. So the Poisson gives P(X ≤ 2) ≈ 0.9526. The Poisson approximation is appropriate because n is large (20) and p is small (0.04), with λ = 0.8 well within the typical regime. The two results differ by only about 0.004.

Question 4: APPLIED

An airline knows that 8% of passengers who book a seat do not show up. For a flight with 200 seats, the airline sells 215 tickets. Use the normal approximation with continuity correction to estimate the probability that more than 200 passengers show up (resulting in overbooking). State any assumptions you make.

Show answer

Let X = number of passengers who show up, where each passenger independently shows up with probability p = 0.92. Then X ~ Bin(215, 0.92). We want P(X > 200) = P(X ≥ 201). The mean is μ = 215 × 0.92 = 197.8 and the standard deviation is σ = √(215 × 0.92 × 0.08) = √(15.824) ≈ 3.978. Using the normal approximation with continuity correction: P(X ≥ 201) ≈ P(Z ≥ (200.5 − 197.8) / 3.978) = P(Z ≥ 0.6787) ≈ 1 − Φ(0.6787) ≈ 1 − 0.7513 ≈ 0.2487. There is approximately a 24.9% chance of overbooking. Key assumptions: each passenger's no-show decision is independent of others, and the 8% no-show rate is constant. The normal approximation is valid because np = 197.8 > 5 and n(1−p) = 17.2 > 5.

Question 5: CRITICAL THINKING

Prove that if X ~ Bin(n, p) and Y ~ Bin(m, p) are independent, then X + Y ~ Bin(n + m, p). Use the moment generating function approach. Then explain intuitively why the same-p requirement is essential—what goes wrong if X ~ Bin(n, p₁) and Y ~ Bin(m, p₂) with p₁ ≠ p₂?

Show answer

The MGF of X is

M_X

(t) =

(1 − p + pe^t)^n

and the MGF of Y is

M_Y

(t) =

(1 − p + pe^t)^m

. Since X and Y are independent,

M_{X+Y}

(t) =

M_X

(t) ×

M_Y

(t) =

(1 − p + pe^t)^n

(1 − p + pe^t)^m

(1 − p + pe^t)^{n+m}

. This is the MGF of Bin(n + m, p), and by the uniqueness theorem for MGFs, X + Y ~ Bin(n + m, p). The same-p requirement is essential because if p₁ ≠ p₂, the product of MGFs becomes

(1 − p₁ + p₁e^t)^n

(1 − p₂ + p₂e^t)^m

, which cannot be factored into

(1 − p + pe^t)^{n+m}

for any single p. Intuitively, a binomial distribution requires that every constituent trial have the same success probability. When p₁ ≠ p₂, the combined trials are heterogeneous, and the sum follows a Poisson-binomial distribution—a generalization whose PMF requires convolution rather than a single binomial formula.

SUMMARY

Binomial Distribution — Summary

Summary

The binomial distribution models the count of successes in n independent trials, each with a constant success probability p. Its PMF, P(X = k) = C(n, k) × $p^k$ × $(1 − p)^{n−k}$ , combines a combinatorial counting factor with the probability of any specific sequence of k successes and n − k failures. The mean is μ = np and the variance is σ² = np(1 − p).

Verification of the four assumptions—fixed n, two outcomes, constant p, and independence—is essential before applying the model. For large n, the normal approximation (with continuity correction) simplifies calculations, while for large n and small p the Poisson approximation is appropriate. When assumptions are violated, alternatives such as the hypergeometric, negative binomial, or beta-binomial distributions should be considered. The binomial distribution underpins exact tests for proportions, logistic regression, and Bayesian conjugate analysis, making it one of the most consequential building blocks in probability and statistics.