Poisson binomial distribution¶

Describe the success count obtained by summing independent Bernoulli variables whose success probabilities may differ, retaining the full heterogeneous probability vector.

Version: v1 · 2026-08-30 · History
Domain-specific #: 2501
Origin domain: probability theory
Subdomain: heterogeneous bernoulli sums
Aliases: Poisson's binomial distribution, Poisson–binomial law

Core Idea¶

The Poisson binomial distribution is the law of \(S=\sum_{i=1}^{n}X_i\) when \(X_i\) are independent Bernoulli variables with success probabilities \(p_i\) that need not be equal. Its support is \(\{0,1,\ldots,n\}\). The probability of \(k\) successes sums \(\prod_{i\in A}p_i\prod_{j\notin A}(1-p_j)\) over all size-\(k\) subsets \(A\). Equal probabilities reduce the law to the ordinary binomial distribution, but heterogeneity is retained in the general case.^[1]

Independence makes the probability-generating function factor as \(G(z)=\prod_{i=1}^{n}(1-p_i+p_i z)\); the coefficient of \(z^k\) is \(\Pr(S=k)\). The mean and variance are \(\sum_i p_i\) and \(\sum_i p_i(1-p_i)\). Direct subset enumeration is exponential, while polynomial multiplication, recurrences, dynamic programming, or discrete Fourier inversion compute the mass function more efficiently. Approximation by a Poisson or normal law requires separate conditions and discards some heterogeneity.^[2]

The distribution requires independent Bernoulli summands. Dependence changes the law even if all marginal probabilities are known. It is not a Poisson distribution, despite the name, and it is not an ordinary binomial unless all success probabilities coincide. A weighted sum with nonunit weights has a different support and distribution. Computational formulas also have stability regimes: odds-based recurrences can fail when a probability equals one, and naive Fourier implementations need normalization and roundoff checks.^[3]

Structural Signature¶

Trial index. A finite collection identifies the individual Bernoulli components.
Binary outcomes. Each component contributes zero or one to the count.
Heterogeneous probabilities. Each trial carries its own parameter between zero and one.
Independence. The joint law factors across components and enables the generating product.
Success count. The sum forgets which trials succeeded and retains only how many.
Mass function. Coefficients assign a probability to every integer from zero through n.
Moments. Additive mean and variance summarize central tendency and dispersion.
Computation method. Recurrence, convolution, or Fourier evaluation produces numerical probabilities under stated precision.

What It Is Not¶

Not an ordinary binomial distribution. That special case imposes one common success probability.
Not a Poisson distribution. The latter is an unbounded count law and usually an approximation here.
Not a dependent Bernoulli sum. Marginal probabilities alone do not determine the count law under dependence.
Not a weighted Bernoulli sum. Nonunit contributions produce a different support and coefficient problem.
Not a conditional Bernoulli distribution. Conditioning on a total success count concerns configurations given S, not the marginal law of S.
Not one numerical algorithm. The mathematical distribution is independent of the chosen evaluator.

Scope of Application¶

The abstraction is literal wherever practitioners can identify the same constitutive roles, apply the same boundary tests, and obtain the same kind of output. The following habitats are uses of Poisson binomial distribution itself, not metaphors based only on resemblance.

Reliability. Counting functioning components with unequal independent reliabilities.
Survey and auditing. Modeling heterogeneous independent inclusion or error indicators.
Risk aggregation. Counting occurrences across exposures with unequal probabilities under an independence model.
Voting models. Counting independent approvals with voter-specific probabilities.
Exact testing. Computing a heterogeneous-null count distribution without replacing probabilities by their mean.
Approximation analysis. Quantifying when binomial, normal, or Poisson substitutions are adequate.

Clarity¶

A clear account of Poisson binomial distribution must preserve the recognition invariant stated in the Core Idea rather than rely on the title alone. List the probability vector and justify independence at the modeled level. State whether the desired object is the count law, a conditioned configuration law, or an approximation. Handle probabilities equal to zero or one explicitly before using odds recurrences. Validate numerical probabilities for nonnegativity, unit sum, and agreement with analytic moments. These declarations are not editorial extras: each changes what observations count, which transformations are licensed, and what conclusion can be drawn. A reader should be able to reconstruct the input, the operative rule, the output, and at least one defeater from the account without consulting an implementation or guessing an unstated convention.

Manages Complexity¶

Poisson binomial distribution manages complexity by replacing a diffuse field of observations or possible operations with a bounded role structure: trial index supplies a finite collection identifies the individual Bernoulli components.; binary outcomes supplies each component contributes zero or one to the count.; heterogeneous probabilities supplies each trial carries its own parameter between zero and one.; independence supplies the joint law factors across components and enables the generating product.; success count supplies the sum forgets which trials succeeded and retains only how many.. The compression is useful because it localizes disagreement. One can ask whether the input was properly formed, whether a constitutive relation held, whether an alternative explanation defeats the inference, or whether the output was overinterpreted. The same compression can mislead when its discarded detail is exactly what the decision requires. A reference-grade use therefore reports both the invariant retained and the information intentionally lost.

Abstract Reasoning¶

Define one Bernoulli indicator per trial and record its success probability.
Assess whether conditional or unconditional independence is the defensible assumption.
Form the generating polynomial or initialize an equivalent convolution recurrence.
Extract coefficients for the required probabilities while tracking numerical error.
Check that masses sum to one and reproduce the theoretical mean and variance.
Compare with the equal-probability binomial special case and any proposed approximation.
Interpret the output as a model-conditioned count law, not proof that real trials are independent.
Test the candidate interpretation against the nearest named confusable rather than accepting a shared surface feature.
State the conclusion at the same scope as the source conditions, and retain uncertainty or nonuniqueness where the construct does not remove it.

Knowledge Transfer¶

The strict upward abstraction is Probability. Poisson Binomial Distribution instantiates Probability because it assigns normalized likelihoods to every possible success count under an explicitly stated random model. Within heterogeneous bernoulli sums, the full mechanism transfers literally when the same roles and boundary tests recur. Beyond that domain, only the parent-level skeleton should travel. Reusing the label Poisson binomial distribution after removing its constitutive vocabulary would hide a change of mechanism behind an analogy. The honest transfer rule is therefore two-stage: recognize the domain-specific pattern first, then lift only the parent relation that remains invariant under a substrate change.

Examples¶

Canonical¶

Three independent components operate with probabilities \(0.2\), \(0.5\), and \(0.9\). Their generating polynomial is \((0.8+0.2z)(0.5+0.5z)(0.1+0.9z)\). The coefficient of \(z^2\) is \(0.2\cdot0.5\cdot0.1+0.2\cdot0.5\cdot0.9+0.8\cdot0.5\cdot0.9=0.46\). Replacing the vector by its average would create a binomial approximation, not the exact law.

Mapped back: input and conventions → constitutive role test → bounded output → explicit interpretation and defeater check.

Applied / In Practice¶

A portfolio contains many exposures whose event indicators share an unobserved common environment. Feeding only marginal event probabilities into a Poisson-binomial calculator yields precise-looking values but uses a false independence model. Conditioning on the environment may recover independence, or a dependent model may be required. The boundary check matters more than selecting a faster coefficient algorithm.

Mapped back: field observation or problem → candidate recognition → confusable and limit checks → appropriately scoped conclusion.

Structural Tensions¶

T1: Heterogeneity versus compression. Replacing the vector by its mean simplifies calculation but changes the law. Diagnostic: Compare exact and binomial masses in the tails relevant to the decision.
T2: Independence versus shared causes. Common environments can create dependence invisible in marginal probabilities. Diagnostic: Name the conditioning variables under which factorization is asserted.
T3: Exact formula versus feasible computation. Subset summation is transparent but exponential. Diagnostic: Use coefficient methods and verify them on small enumerated cases.
T4: Approximation speed versus tail accuracy. Poisson or normal approximations can distort rare-count probabilities. Diagnostic: Bound or empirically check error in the target tail.
T5: Mathematical law versus parameter uncertainty. Known p-values are assumed even when they are estimated. Diagnostic: Propagate estimation uncertainty separately from count randomness.
T6: Autonomy versus generic probability. Probability supplies uncertainty quantification; this law adds heterogeneous independent Bernoulli aggregation. Diagnostic: Remove the summand structure and test whether the named distribution remains.

Structural–Framed Character¶

Poisson binomial distribution is highly structural under its model, while selection of trials, independence, and estimated probabilities are framed empirical assumptions. The five framing criteria point in a consistent direction. Evaluative weight is limited to whether the defining conditions are met, not whether the outcome is desirable. Human practice matters to the extent that experts choose conventions, instruments, or reporting thresholds, but those choices do not make every verdict arbitrary. Institutional history explains the name and standard use; it does not replace the recognition rule. The operative vocabulary travels within the home field and closely adjacent subfields, while transfer farther away requires translation to the parent prime. Thus recognition remains disciplined even where interpretation is defeasible.

Structural Core vs. Domain Accent¶

What is skeletal. Poisson Binomial Distribution instantiates Probability because it assigns normalized likelihoods to every possible success count under an explicitly stated random model. This is the part that can be expressed without the candidate's specialist nouns.

What is domain-bound. The domain accent is a finite vector of unequal Bernoulli probabilities, independence, a unit-weight sum, generating-function coefficients, and count-law computation. Remove those elements and the result is no longer Poisson binomial distribution; it is only the parent relation or a loose analogy.

Why this does not clear the prime bar. The name does not recur with unchanged diagnostics across three independent domains. What transfers is already represented by prime:probability. The candidate remains autonomous because its in-domain recognition rule, failure modes, and consequences are stable, but its vocabulary and interventions do not float free of the home substrate.

Poisson Binomial Distribution instantiates Probability because it assigns normalized likelihoods to every possible success count under an explicitly stated random model.

The prospective workspace queue contains one strict upward edge to prime:probability. No live DAG mutation is authorized.

Relationships to Other Abstractions¶

Current abstraction Poisson binomial distribution Domain-specific

Parents (1) — more general patterns this builds on

Poisson binomial distribution is a kind of Probability Prime

Poisson Binomial Distribution instantiates Probability because it assigns normalized likelihoods to every possible success count under an explicitly stated random model.

Hierarchy paths (2) — routes to 2 parentless roots

Poisson binomial distribution → Probability → Measure → Aggregation → Micro Macro Linkage

Show alternative path (1)

Neighborhood in Abstraction Space¶

Poisson binomial distribution sits in a sparse region of the domain-specific corpus (90^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Delaporte Distribution — 0.80
Negative Hypergeometric Distribution — 0.80
Cumulant — 0.78
Schuette–Nesbitt formula — 0.77
Mixture Distribution — 0.77

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Binomial distribution. Uses identically distributed independent Bernoulli trials.
Poisson distribution. An unbounded count law often used as a small-probability approximation.
Bernoulli distribution. Describes one binary trial rather than their heterogeneous count.
Conditional Bernoulli distribution. Describes which trials succeed conditional on a fixed total.
Generalized Poisson binomial distribution. May allow weighted or multivalued summands under broader conventions.
Beta-binomial distribution. Introduces parameter mixing and dependence/overdispersion rather than fixed heterogeneous probabilities.

References¶

[1] Wang, Y. H. (1993). 'On the Number of Successes in Independent Trials.' Statistica Sinica 3(2), 295–312. https://www3.stat.sinica.edu.tw/statistica/j3n2/j3n23/j3n23.htm registry ↩

[2] Chen, S. X., and Liu, J. S. (1997). 'Statistical Applications of the Poisson-Binomial and Conditional Bernoulli Distributions.' Statistica Sinica 7(4), 875–892. registry ↩

[3] Biscarri, W., Zhao, S. D., and Brunner, R. J. (2018). 'A Simple and Fast Method for Computing the Poisson Binomial Distribution Function.' Computational Statistics & Data Analysis 122, 92–100. https://doi.org/10.1016/j.csda.2018.01.007 registry ↩