The Complete Statistical Guide to Binomial Distribution: Bernoulli Trials, Probability Mass Functions & Large-Sample Normal Approximations
In mathematical statistics, probability theory, clinical pharmacology, quality control engineering, and financial actuarial science, the Binomial Distribution is the foundational discrete probability distribution used to calculate the likelihood of observing exactly k successes across n independent, identical binary experiments (termed Bernoulli Trials). Explore our comprehensive Advanced Probability & Mathematics Suite for all probability tools.
Whether modeling the probability of getting at least 7 heads out of 10 coin tosses, assessing the reliability of an aerospace launch vehicle containing redundant electrical circuits, determining whether a pharmaceutical vaccine surpasses a 70% efficacy threshold in clinical trial cohorts, or managing industrial defect acceptance quality limits (AQL), the binomial distribution provides exact, closed-form combinatorial probabilities. For continuous statistical modeling and bell-curve Z-scores, explore our Standard Deviation Calculator, or examine dataset medians with our Median & IQR Calculator.
1. Binomial Probability Mass Function (PMF):
P(X = k) = &binom;{n}{k} × pk × (1 - p)n - k = [ n! ÷ (k! × (n - k)!) ] × pk × qn - kWhere
n is number of trials, k is number of successes, p is probability of success, and q = 1 - p is probability of failure.2. Cumulative Distribution Function (CDF - At Most k Successes):
P(X ≤ k) = ∑i=0k &binom;{n}{i} × pi × (1 - p)n - i3. Tail Cumulative Probability (At Least k Successes):
P(X ≥ k) = 1 - P(X ≤ k - 1) = ∑i=kn &binom;{n}{i} × pi × (1 - p)n - i4. Expected Value (Mean):
μ = E[X] = n × p5. Variance:
σ² = Var(X) = n × p × (1 - p) = npq6. Standard Deviation:
σ = √(n × p × (1 - p)) = √(npq)7. Skewness & Kurtosis:
Skewness γ₁ = (1 - 2p) ÷ √(npq) • Excess Kurtosis γ₂ = (1 - 6pq) ÷ (npq)
💡 The 4 Strict Bernoulli Trial Conditions (The BINS Criteria)
For a probability experiment to qualify strictly as a Binomial Distribution, all four of the following BINS criteria must be unconditionally satisfied:
- B – Binary Outcomes: Each individual trial must result in exactly one of two mutually exclusive categories: Success or Failure (e.g. Pass/Fail, Heads/Tails, Defective/Non-Defective, Win/Loss).
- I – Independent Trials: The outcome of any given trial must have zero mathematical influence on the outcome of any subsequent trial. Sampling must be conducted with replacement (or from an essentially infinite population where sample size
n ≤ 5%of total populationN). - N – Number of Trials is Fixed: The total number of trials
nmust be set in advance before the experiment begins (contrast with Geometric or Negative Binomial distributions where trials continue until a set number of successes occur). - S – Same Probability of Success: The probability of success
pmust remain perfectly identical and invariant on every single trial (0 ≤ p ≤ 1).
Comprehensive Comparative Matrix: Probability Distributions Evaluated
Selecting the appropriate probability distribution model depends upon discrete vs. continuous variables, sampling with or without replacement, and whether event occurrences are bounded:
| Distribution Model | Discrete / Continuous | Underlying Sampling Conditions | Key Mathematical Parameters | Optimal Real-World Applications |
|---|---|---|---|---|
| Binomial Distribution | Discrete counts (k ∈ [0, n]) | Fixed n trials, constant p, independent with replacement. |
n (trials), p (success rate) |
Coin flips, medical drug recovery rates, sales conversion rates, quality control batch testing. |
| Hypergeometric Distribution | Discrete counts (k ∈ [0, n]) | Fixed n draws, WITHOUT replacement from a finite population N with K successes. |
N (population), K (successes), n (draws) |
Lottery jackpot ticket drawings, poker card hands (dealing without replacement), wildlife capture-recapture tagging. |
| Poisson Distribution | Discrete counts (k ∈ [0, ∞)) | Continuous time or space interval, events occur independently at a constant average rate. | λ (average event rate per interval) |
Call center incoming phone arrivals per hour, server network packet traffic, radioactive decay particle counts. |
| Negative Binomial | Discrete counts (trials until r successes) | Independent Bernoulli trials repeated until a predetermined number r of successes occurs. |
r (target successes), p (success rate) |
Drilling for oil until finding 3 productive wells, sales prospecting until landing 5 closed deals. |
| Normal (Gaussian) | Continuous real numbers (x ∈ (-∞, +∞)) | Symmetrical bell curve resulting from the Central Limit Theorem across aggregated variables. | μ (mean), σ (standard deviation) |
Human height distributions, standardized IQ exam testing, measurement sensor instrumentation noise. |
Step-by-Step Worked Example: Clinical Trial Vaccine Efficacy
Suppose an infectious disease research laboratory administers a novel antiviral therapy to n = 10 infected test patients. Historical laboratory data establishes that the single-patient recovery probability is p = 0.70 (70%).
The lead pharmacologist wishes to determine:
- The probability that exactly
k = 8patients recover:P(X = 8). - The probability that at least 8 patients recover:
P(X ≥ 8). - The expected number of recovered patients and the standard deviation.
&binom;{10}{8} = 10! ÷ [8! × (10 - 8)!] = (10 × 9) ÷ (2 × 1) = 45Step 2: Solve for Exact Probability P(X = 8)
P(X = 8) = 45 × (0.70)⁸ × (1 - 0.70)¹⁰⁻⁸ = 45 × (0.70)⁸ × (0.30)²P(X = 8) = 45 × 0.057648 × 0.090 = 0.233474 (23.35%)Step 3: Calculate Cumulative Probability for At Least 8 Recoveries P(X ≥ 8)
P(X ≥ 8) = P(X = 8) + P(X = 9) + P(X = 10)•
P(X = 9) = &binom;{10}{9} × (0.70)⁹ × (0.30)¹ = 10 × 0.040354 × 0.30 = 0.121061 (12.11%)•
P(X = 10) = &binom;{10}{10} × (0.70)¹⁰ × (0.30)⁰ = 1 × 0.028248 × 1 = 0.028248 (2.82%)P(X ≥ 8) = 0.233474 + 0.121061 + 0.028248 = 0.382783 (38.28%)Step 4: Compute Distribution Mean, Variance and Standard Deviation
•
Expected Recoveries (μ) = n × p = 10 × 0.70 = 7.00 patients•
Variance (σ²) = n × p × q = 10 × 0.70 × 0.30 = 2.10•
Standard Deviation (σ) = √2.10 = 1.4491 patients
Large-Sample Normal & Poisson Approximations to the Binomial
When the number of trials n becomes extremely large (e.g. n = 1,000 or n = 100,000), calculating exact combinatorial factorials n! directly is computationally expensive. Statisticians employ two powerful asymptotic approximations:
1. The De Moivre-Laplace Normal Approximation
By the Central Limit Theorem, if both np ≥ 5 and n(1 - p) ≥ 5 (or ≥ 10 for stricter precision), the discrete binomial distribution converges toward a continuous Normal distribution with mean μ = np and standard deviation σ = √(npq).
Because a continuous distribution is approximating discrete integer bars, one must apply a Continuity Correction Factor (±0.5):
P(X = k) ≈ P(k - 0.5 ≤ Y ≤ k + 0.5) = Φ[ (k + 0.5 - np) ÷ √(npq) ] - Φ[ (k - 0.5 - np) ÷ √(npq) ]P(X ≤ k) ≈ Φ[ (k + 0.5 - np) ÷ √(npq) ]P(X ≥ k) ≈ 1 - Φ[ (k - 0.5 - np) ÷ √(npq) ]
2. The Poisson Approximation for Rare Events (Law of Small Numbers)
When the trial count n is very large (n ≥ 100) but the success probability p is extremely rare (p ≤ 0.05) such that λ = np remains moderate (np ≤ 10), the binomial distribution is extraordinarily well approximated by the Poisson Distribution:
P(X = k) ≈ [ e-λ × λk ] ÷ k!, where λ = n × p
Real-World Applications of the Binomial Distribution
Binomial modeling is an essential operational tool across engineering, medicine, and business:
1. Industrial Quality Control & Acceptance Sampling (AQL)
In manufacturing supply chains (ISO 2859-1 standards), quality engineers inspect a random sample of n = 80 electronic microchips from a shipping container of 10,000 units. Using the binomial distribution with an acceptable quality limit of p = 0.015 (1.5%), the inspection protocol determines the critical rejection number c (e.g. reject entire shipment if defects X > 3).
2. Financial Credit Default & Loan Portfolio Risk
Commercial banks evaluate small business loan portfolios. For a pool of n = 200 independent retail loans each having an annualized historical default probability of p = 0.03 (3%), actuarial risk officers calculate P(X ≥ 10) to ensure loan loss reserves can withstand severe default clusters.
3. Genetics & Mendelian Inheritance Ratios
In genetic biology, heterozygous crossbreeding (Aa × Aa) yields a dominant phenotype with probability p = 0.75 and a recessive phenotype with probability q = 0.25. For a litter of n = 8 offspring, geneticists use binomial probability to calculate the likelihood of observing exactly 2 recessive phenotypes (P(X = 2) = &binom;{8}{2}(0.25)²(0.75)⁶ = 31.15%).
Frequently Asked Questions (FAQ)
A binomial distribution models the number of successes in a fixed sequence of independent Bernoulli trials. The 4 required conditions (acronym BINS) are:
- Binary outcomes: Each trial has only two mutually exclusive results: Success or Failure.
- Independent trials: The outcome of one trial does not affect any other.
- Number of trials is fixed: The sample size
nis determined before the experiment begins. - Same probability: The probability of success
premains constant on every trial.
The Binomial PMF formula is:
P(X = k) = nCk × p^k × (1 - p)^(n - k)
where nCk = n! / [k!(n - k)!] is the binomial coefficient, p is the success probability, (1 - p) is the failure probability, and k is the target success count.
The Probability Mass Function (PMF) calculates the exact probability of obtaining precisely k successes: P(X = k).
The Cumulative Distribution Function (CDF) calculates the probability of obtaining at most k successes: P(X ≤ k) = ∑ P(X = i) from i = 0 to k.
The parameters are calculated directly from trial count n and success probability p:
- Mean (Expected Value):
μ = n × p - Variance:
σ² = n × p × (1 - p) - Standard Deviation:
σ = √(n × p × (1 - p))
The Normal Approximation can be used when both np ≥ 5 and n(1 - p) ≥ 5 (some textbooks recommend ≥ 10 for higher accuracy). When applying the normal approximation, a continuity correction of ±0.5 must be used because a continuous smooth curve is approximating discrete integer counts.
The Binomial Distribution models sampling with replacement (or sampling from an infinite population) where probability p remains perfectly constant on every draw.
The Hypergeometric Distribution models sampling without replacement from a finite population where probability p changes after every consecutive draw (e.g. dealing playing cards from a 52-card deck).
No. In probability theory, the probability parameter p must strictly satisfy the boundary condition 0 ≤ p ≤ 1. A value of 0 means the event is impossible, while a value of 1 means the event is mathematically certain.
When p = 0.5 (such as a fair coin toss), the probability of success equals failure (p = q = 0.5), producing a perfectly symmetrical distribution with zero skewness (γ₁ = 0).
When p < 0.5, the distribution is positively skewed (right-tailed). When p > 0.5, the distribution is negatively skewed (left-tailed). However, as n → ∞, the distribution becomes increasingly symmetric regardless of p due to the Central Limit Theorem.