Bernoulli Trials and the Binomial Distribution

Bernoulli trials have two labelled outcomes, and the binomial distribution counts successes in a fixed number of independent trials with the same success probability. In brief: the number of heads in four independent fair coin tosses is binomial with n = 4 and p = 0.5. The probability of exactly two heads is 6 × 0.5⁴ = 0.375.

The formula combines the probability of one success-failure sequence with the number of sequences giving the requested count. Before using it, check that the experiment satisfies the model’s assumptions. This guide develops the distribution, calculates exact and cumulative events and distinguishes a fixed trial count from experiments that stop after a success.

What is a Bernoulli trial?

A Bernoulli trial records whether a specified event occurs, labelled success, or does not occur, labelled failure. These labels are mathematical categories rather than judgments of quality. A success might be a head, a selected colour or an item meeting a stated condition.

The success probability is p and the failure probability is 1 − p, with 0 ≤ p ≤ 1. An indicator variable takes value 1 for success and 0 for failure. This numerical coding allows success counts to be formed by adding indicators across trials.

An experiment with many physical outcomes can still be coded as Bernoulli if it is reduced to one event versus its complement. Rolling a die and recording whether it shows 6 gives success probability 1/6 under a fair-die model, while all five other faces are grouped as failure.

What conditions are required for a binomial distribution?

The binomial model requires a fixed number n of trials, two outcomes per trial, independent trials and a common success probability p. The random variable X counts the successes among those n trials. Its possible values are the integers 0 through n.

Independence means earlier outcomes do not change the probability of later outcomes under the model. Drawing with replacement and adequate randomisation can support this assumption. Drawing without replacement from a finite collection generally changes success probabilities and creates dependence, so it is not exactly binomial.

The stopping rule matters. Tossing a coin exactly ten times gives a fixed trial count, while tossing until the third head does not. The latter experiment has a random number of trials and requires a different distribution, even though each individual toss can still be a Bernoulli trial.

What is the binomial probability formula?

For X following a binomial distribution with parameters n and p, P(X = k) = C(n, k)p^k(1 − p)^{n − k}, for k = 0, 1, …, n. The coefficient C(n, k) counts the ways to place k successes among n trial positions.

A particular sequence with k successes has probability p^k(1 − p)^{n − k} by independence. Every arrangement with the same count has that probability, and the arrangements are mutually exclusive. Multiplying by their number gives the complete probability of exactly k successes.

The combination coefficient is C(n, k) = n!/[k!(n − k)!], where factorial multiplies positive integers down to 1 and 0! = 1. At p = 0 or p = 1, use the direct interpretation: all trials fail or all succeed with certainty, avoiding ambiguous endpoint powers in a mechanical calculation.

How do you calculate the probability of exactly two successes?

Suppose five independent trials each have success probability 0.2. For exactly two successes, n = 5, k = 2 and 1 − p = 0.8. There are C(5, 2) = 10 possible success-position arrangements.

The probability of each arrangement is 0.2² × 0.8³ = 0.04 × 0.512 = 0.02048. Multiplying by 10 gives P(X = 2) = 0.2048, or 20.48%. Omitting the combination factor would calculate just one sequence, not every sequence with two successes.

For example, SSFFF and FSFSF are distinct sequences included in the same count event. Their order differs, but both contain two successes and three failures. The binomial coefficient accounts for all such orders without requiring a large tree to list each one separately.

1. Check the fixed-count, independence and constant-probability assumptions.
2. Identify n, p and the requested count k.
3. Calculate the combination coefficient.
4. Multiply by p to the success count and 1 − p to the failure count.
5. Check that the resulting probability lies between zero and one.

How do you calculate at most, at least and between probabilities?

At most k means X ≤ k, so add the probabilities for 0 through k. At least k means X ≥ k, and it is often easier to use 1 − P(X ≤ k − 1). The boundary changes by one because the count is integer-valued.

For five trials with p = 0.2, P(X = 0) = 0.8⁵ = 0.32768 and P(X = 1) = 5 × 0.2 × 0.8⁴ = 0.4096. Therefore P(X ≥ 2) = 1 − 0.32768 − 0.4096 = 0.26272.

Exactly two is different from at least two, and more than two means X ≥ 3 rather than X ≥ 2. For a bounded event such as 2 ≤ X ≤ 4, add the probabilities at 2, 3 and 4. Translate the wording into count symbols before using a table or calculator.

What are the mean and standard deviation of a binomial count?

The expected success count is E(X) = np. The variance is Var(X) = np(1 − p), and the standard deviation is √[np(1 − p)]. These formulas describe the distribution across repeated sets of n trials, not a guaranteed count in one set.

For n = 100 and p = 0.2, the expected count is 20, the variance is 16 and the standard deviation is 4. Individual counts may differ from 20. The standard deviation measures the modelled spread of those counts around the expectation.

For the observed success proportion X/n, the mean is p and the standard deviation is √[p(1 − p)/n]. Increasing n makes the count’s absolute spread larger in typical nondegenerate cases but makes the proportion more concentrated. This distinction explains why larger samples can give more stable percentage estimates even while raw count deviations grow.

How do n and p affect the binomial distribution?

When p = 0.5, the distribution is symmetric around n/2. Small p values tend to concentrate successes near zero, often producing right skew, while p near 1 concentrates counts near n. The exact shape also depends on n, and the endpoints p = 0 and p = 1 collapse to a single certain count.

For suitable large n with neither expected successes nor expected failures too small, a normal approximation may be useful. It is an approximation, not the definition of the distribution, and continuity correction can improve count-event approximations. Exact binomial calculation remains preferable when feasible and when tail accuracy matters.

The model appears in repeated inspection, independent response models and simulated trials. Its background is a sum of independent 0–1 indicators. Before applying it to real observations, examine whether the trial probabilities are stable and whether shared conditions create dependence that the simple model omits.

What mistakes should you avoid with binomial probabilities?

Do not use a binomial formula solely because the outcome is a count. Counts from sampling without replacement, varying success probabilities or a success-based stopping rule can follow different distributions. The assumptions determine the model, while the formula only calculates within that model.

Keep the success definition consistent. If success is a failure of a component, p is the component-failure probability despite the everyday wording. Replacing p by 1 − p halfway through changes what X counts and can reverse the interpretation of an at least event.

Finally, include the combination factor for an unspecified order, handle inclusive endpoints carefully and retain enough numerical precision in cumulative sums. A complete distribution should sum to 1, and complementary event calculations should agree. These checks help distinguish a correct single-path probability from the required count probability.