Reed–Frost Model¶

A finite-population, discrete-generation epidemic model in which independent pairwise exposure risks combine into a binomial count of new cases while each infectious generation then exits.

Version: v1 · 2026-10-31 · History
Domain-specific #: 14565
Domain group: Applied Sciences & Engineering
Origin domain: Medicine & Healthcare
Subdomain: Infectious Disease Modeling → Medicine & Healthcare
Aliases: Reed Frost Chain Binomial Model

Core Idea¶

The Reed–Frost model is a stochastic rule for epidemic spread in a finite population, advancing one infectious generation at a time.^[1] At generation \(t\), let \(S_t\) be the number susceptible, \(I_t\) the number infectious, and \(p\) the common chance that one infectious person transmits to one susceptible person during that generation. The classic rule treats these pair opportunities as independent.^[1] A susceptible therefore escapes all \(I_t\) infectious people with probability \((1-p)^{I_t}\) and is newly infected with probability \(1-(1-p)^{I_t}\). The chance of escaping all current infectious people is a power of the single-pair escape chance; the next infection chance is its complement.^[1] Given the current counts, the next infectious count is a binomial draw over the susceptible people.^[1] The current infectious generation then leaves the infectious class.^[1] These are model commitments, not claims that contacts in an observed outbreak are literally uniform or independent.

In compact form, with \(q=1-p\),

\[ I_{t+1}\mid(S_t,I_t)\sim\operatorname{Binomial}\!\left(S_t,1-q^{I_t}\right),\qquad S_{t+1}=S_t-I_{t+1},\qquad R_{t+1}=R_t+I_t. \]

The last equality makes explicit the one-generation removal that the source's susceptible–infectious description also permits one to track implicitly.^[1] With no infectious people left, the escape power is one and the internal chain stops even if susceptible people remain.^[1] The model thus represents an ensemble of possible epidemic chains, including early extinction and different final sizes from the same initial state.^[1] Its distinctive rule is the escape-probability exponent equal to the current infectious count, coupled to a conditional binomial next generation; discrete time or stochasticity alone is broader.

How would you explain it like I'm…

Coin-Flip Germ Tag

Imagine a game of tag where each round, every kid who is 'it' gets a chance to tag each other kid, like flipping a coin. Kids who get tagged become 'it' next round, and the old 'it' kids sit down for good. Sometimes the game spreads to almost everyone, and sometimes it stops early, even when you start the same way.

Sickness Spreading by Rounds

The Reed–Frost Model is a way to imagine how a sickness spreads through a group, one 'generation' at a time. Each sick person has the same small chance of passing it to each healthy person who can still catch it, like separate coin flips. A healthy person stays well only if they dodge every sick person that round. Those who catch it become the sick ones next round, while the earlier sick ones are done. Because it's based on chance, the same starting group can end with a big outbreak or a tiny one, and it stops when no one is sick anymore.

Chain-Binomial Epidemic Model

The Reed–Frost model is a chance-based (stochastic) model of an epidemic in a fixed-size population, moving one generation of infection at a time. Each infectious person has the same probability p of infecting each susceptible person in a generation, and these pairings are treated as independent. A susceptible person escapes one infectious person with probability 1−p, so they escape all I of them with probability (1−p)^I, and get infected with probability 1−(1−p)^I. The number of new infections is then like flipping that weighted coin for every susceptible person, and the current infectious group is removed. If no one is infectious, the process stops, even if susceptible people remain. These are simplifying assumptions of the model, not claims that real contacts are uniform.

The Reed–Frost Model is a discrete-generation stochastic epidemic model for a finite population. With S_t susceptible, I_t infectious, and p the per-pair transmission probability per generation, pairwise transmission events are assumed independent. A susceptible therefore escapes infection with probability q^{I_t}, where q = 1 − p, and is infected with probability 1 − q^{I_t}. Conditional on current counts, the next infectious count is binomial: I_{t+1} ~ Binomial(S_t, 1 − q^{I_t}). Susceptibles are then updated by S_{t+1} = S_t − I_{t+1}, and the current infectious generation is removed, R_{t+1} = R_t + I_t. If I_t = 0, the escape probability is 1 and the chain stops even with susceptibles remaining. The model thus generates a distribution over epidemic chains, including early extinction and variable final sizes. Its defining feature is the escape exponent equal to the current infectious count coupled to a binomial next generation; discrete time or randomness alone does not make a model Reed–Frost, and the uniform independent mixing is a modeling assumption.

Structural Signature¶

The signature is finite \(S/I\) state → one-generation exposure → independent common pair chances → compounded escape → conditional binomial next generation → removal or extinction.^[1]

Finite susceptible and infectious state. \(S_t\) supplies the number of susceptible trials; \(I_t\) supplies the number of current sources of exposure.^[1] The removed class may be implicit in a fixed population. Without a specified susceptible set and current infectious count, the transition's trial number and exponent are undefined.
One-generation infectious turnover. Each current infectious person contributes exposure during a generation and then ceases to be infectious in the next.^[1] Retaining old infectious people would change both the state and the next risk calculation.
Homogeneous independent pair chance. The classic model assigns the same \(p\) to each infectious–susceptible pair and composes their opportunities independently.^[1] Heterogeneity or dependence may be modeled, but it no longer licenses the unmodified power rule merely by carrying the Reed–Frost name.
Escape-power composition. The probability of avoiding all \(I_t\) current infectious people is \(q^{I_t}\); its complement is the susceptible person's infection probability.^[1] Removing this composition removes the model's difference from a generic chain-binomial process and from Greenwood's fixed-positive-chance variant.
Conditional binomial update. Given current susceptible and infectious counts, the model draws a new infectious count from the susceptible trials with chance equal to one minus the compounded escape probability, and subtracts those cases from the susceptible count.^[1] Replacing the draw with only its mean discards the distribution of possible chains.
Internal stopping rule. Once a generation has no infectious people, its internal infection probability is zero and subsequent generations remain still.^[1] This condition does not state that every susceptible became immune, nor does it bar a new external introduction in a different model.

What It Is Not¶

It is not every chain-binomial epidemic model. Greenwood's version also advances in discrete generations and draws new cases binomially, but it gives a susceptible a constant positive infection chance when at least one infectious person is present.^[1] Reed–Frost changes that chance with the number of infectious people through \(1-(1-p)^I\).^[1]

It is not a deterministic epidemic curve. The conditional expectation \(S_t[1-(1-p)^{I_t}]\) can summarize the next draw at a known state, but substituting means for random states is an approximation.^[1] In general, the unconditional mean of the nonlinear product is not obtained by simply replacing \(S_t\) and \(I_t\) with their means. The difference matters when chance extinction and chain-to-chain variation are the questions.

It is not a measured law that every household or ship obeys. The classic transition assumes a closed finite population, common pair chance, one-generation infectivity, and independence.^[1] In the family common-cold fit, generation classification and contact homogeneity remain model or classification choices.^[2] In the ship simulation, log completeness and estimated immunity limit claims about the historical transmission record.^[3] An extended household analysis can add outside introductions and heterogeneous infectiousness while retaining an identifiable basic internal Reed–Frost component.^[4]

Scope of Application¶

The model belongs to mathematical epidemiology where an analyst can specify a bounded susceptible population, current infectious count, generation interval, and pair chance. It supports conditional next-generation probabilities, simulated chains, and final-size distributions under a stated classic rule.^[1] The common-cold study compares observed chain patterns with model expectations,^[2] while the ship study uses estimated inputs for scenario simulations.^[3] In each use, model fit and the data-generating process remain distinct.

One literal habitat is a classified household outbreak episode. Heasman and Reid grouped common-cold occurrences in five-person families into approximate generations and compared observed complete chains with classic Reed–Frost expectations.^[2] A second is a bounded voyage simulation: Paterson and colleagues used historical measles outbreaks on ships to parameterize stochastic Reed–Frost simulations of subsequent generations and outbreak duration.^[3] The first compares an observed chain classification with model expectations; the second asks what modeled trajectories would result from estimated starting conditions. Neither application establishes that the idealized pair assumptions were directly observed.

The scope narrows when a whole study adds an outside source or individual infectiousness differences. In the 1918 Baltimore household analysis, a basic internal Reed–Frost component could be fitted, but the analysis allowed outside-household infection and found a preferred heterogeneous extension.^[4] Its internal kernel is a related use; the whole extended study is not a closed, homogeneous classic instance.

Clarity¶

The model separates three questions often compressed into “infection risk.” What is the chance from one infectious–susceptible pair? That is \(p\). What is the chance for one susceptible exposed to \(I\) current infectious people? That is \(1-(1-p)^I\) under the independence assumption. How many new people become infectious? That is a binomial random count over \(S\) susceptible trials.^[1] Confusing the second probability with \(p\) erases the dependence on \(I\); confusing the third count with its mean erases the distribution of epidemic chains.

The state notation also clarifies termination. \(I=0\) stops internal spread without requiring \(S=0\).^[1] Conversely, observed case-free time does not prove that no outside introduction can occur later; an extended household study can explicitly include external exposure.^[4] That distinction is crucial when deciding whether a reported setting is the full classic model or an adaptation containing its internal transition.

Manages Complexity¶

The model compresses a potentially large array of pairwise opportunities into the current counts \((S_t,I_t)\) and a common \(p\). For each susceptible, the \(I_t\) pair escapes multiply to \(q^{I_t}\); the \(S_t\) resulting trials yield one next-generation distribution. Repeating that update gives a law over possible chains rather than a single prescribed curve. The compression is possible because the classic formulation imposes common pair chances, conditional independence, a fixed generation step, and one-generation removal.^[1]

That economy also exposes the cost of the assumptions. If the analyst needs different pair risks, correlated contacts, imported cases, or time-varying mixing, counts and one \(p\) may no longer capture the relevant structure. The Baltimore household extension illustrates this: a basic internal model was retained as a comparison, while outside introduction and heterogeneous infectiousness changed the account of the whole study.^[4]

Abstract Reasoning¶

To analyze a proposed Reed–Frost instance, specify \((S_0,I_0)\), the generation interval, \(p\), and the population boundary. At each step compute \(q^{I_t}\), draw \(I_{t+1}\) from the corresponding binomial law, update \(S\) and the removed class, and stop internal spread at \(I=0\). Running the chain many times distinguishes a probability of an outcome from the outcome of one realization. Alternatively, computing the state distribution can characterize final size and intermediate states without substituting a deterministic average path.^[1]

The discriminating counterfactual is to change only the number of current infectious people. Under Reed–Frost, one and two infectious people expose a susceptible to risks \(p\) and \(1-(1-p)^2\), respectively. Under Greenwood's rule, both positive infectious counts use the same risk \(p\). If a proposed use neither changes risk with \(I\) in the Reed–Frost way nor contains an identifiable classic component, it belongs to a different chain-binomial or epidemic model.^[1]

Knowledge Transfer¶

The model's literal transfer is within epidemic modeling: the same formal pair-escape and binomial-generation rule can organize a classified household chain or a simulated ship outbreak after the carrier, interval, initial state, and parameter interpretation are changed.^[2]^[3] Transfer does not make a common-cold observation equivalent to a measles voyage. It transfers a conditional rule whose assumptions must be tested or declared in each setting.

An indexed random state with a conditional transition also instantiates broader stochastic and Markov structures. Those broader structures travel far beyond epidemics; the specifically infectious–susceptible pair risk, one-generation removal, and epidemic extinction interpretation do not thereby become a cross-domain Prime. Elsewhere, an analogous branching or stopping pattern would need its own literal mechanism and evidence before receiving the Reed–Frost name.^[1]

Examples¶

Canonical: five-person family common-cold chains¶

Heasman and Reid analyzed common-cold outbreak episodes in families of five.^[2] They used onset information to classify cases into generations, tested a first-stage binomial fit for single-primary episodes, and compared observed complete chain frequencies with Reed–Frost expected chains.^[2] Their Table V distinguishes 664 classified single-primary episodes from 92 double-primary episodes.^[2] These are episode counts, not counts of independent families.^[2] Approximate generation assignment, variation among families, and fitting assumptions bound what the comparison shows: a classic model applied to whole classified chains, not direct observation of identical independent contacts.^[2]

Mapped back: finite state → five-person family episode with identified primary case count and remaining susceptible members; turnover → onset-based generation classification under a one-generation model; common pair chance and escape power → classic Reed–Frost assumptions used in expected chains; binomial update → first-stage comparison and complete-chain frequencies; stopping → an episode's terminal generation, without denying later new household introductions.^[2]^[1]

Applied: measles-voyage simulation¶

Paterson and colleagues reconstructed historical measles outbreaks aboard ships and used a named Reed–Frost stochastic model to simulate how many subsequent generations might occur before arrival.^[3] Their simulation inputs included initial cases, estimated susceptibles, an outbreak reproduction parameter, and ship population.^[3] They used binomial random draws and ran 1,000 simulations for each outbreak.^[3] The modeled voyage is a prospective trajectory question conditioned on inferred starting values; incomplete logs and estimated immunity prevent treating its outputs as a complete record of every historical transmission.^[3]

Mapped back: finite state → bounded ship population with estimated susceptible and index-case counts; turnover → modeled successive infectious generations; common pair chance and escape power → the paper's homogeneous-mixing Reed–Frost assumption, with the symbolic rule independently supplied by the formal model definition; binomial update → stochastic case counts in repeated simulations; stopping → modeled chain duration compared with voyage length.^[3]^[1]

Structural Tensions¶

T1: tractable homogeneous exposure versus observed heterogeneity. A single pair chance and independent opportunities make the classic chain-binomial transition calculable.^[1] Real household analysis can require differences among infectious people or exposure from outside the household.^[4] Keeping the classic assumptions makes a sharper baseline but can misdescribe the whole observation; adding those features improves fit while changing the model being claimed. Diagnostic: Is the Reed–Frost rule the whole fitted model, or only an identifiable internal component?

T2: stochastic scenario versus historical reconstruction. A ship simulation can ask how outbreak generations might extend under estimated immunity and starting cases. The historical log can still be incomplete.^[3] Treating every modeled generation as observed makes the simulation seem more certain than its inputs; refusing to model at all loses the scenario comparison the data can support. Diagnostic: Which counts came from the record, which were inferred, and which exist only inside the model?

T3: comparable generation classes versus approximate onset classification. Complete-chain comparisons require cases assigned to discrete generations. In the common-cold study, that assignment is a convention applied to symptom onsets, while natural transmission need not occur in exact synchronous rounds.^[2] The convention permits a model comparison at the cost of possible classification error; dropping the classification removes the chain comparison. Diagnostic: Would a reasonable change in generation assignment change the reported fit?

Structural–Framed Character¶

The Reed–Frost model is strongly structural within mathematical epidemiology, with a domain-bound interpretation. Evaluative weight: it is a way to compute and test a conditional epidemic model, not a recommendation that every outbreak be modeled this way. Human-practice dependence: analysts choose population boundaries, time steps, parameter estimates, and case classifications, but the mathematical transition has determinate consequences once those choices are fixed. Institutional origin: the model is a named scientific construction, not a rule created by a jurisdiction or organization. Vocabulary travel: binomial draws and stochastic transitions travel broadly, while infectious–susceptible pair exposure and removal give this name its epidemiological content. Import versus recognition: an analyst can impose the rule to simulate a scenario or recognize and test it in a fitted chain; neither act proves the world obeys the assumptions.^[1]^[2]^[3] Its character: a domain-specific formal model whose mathematically portable stochastic-transition skeleton is broader than its named epidemic pair-contact mechanism.

Structural Core vs. Domain Accent¶

The skeletal relation is a finite random state updated by a conditional transition and repeated until a stopping state. That skeleton belongs to broader stochastic-process and Markov-process families; it does not distinguish Reed–Frost from Greenwood or many other chains. The identity-bearing mechanism is more specific: independent equal infectious–susceptible pair chances compound as an escape power, generate a binomial number of new cases, and expire after one infectious generation.^[1]

The domain accent is therefore constitutive rather than a removable set of examples. A queue, price series, or random walk may be a stochastic process, but it has no infectious–susceptible contact interpretation merely because it has random steps. The named Reed–Frost model has demonstrated literal reach across epidemic settings, not across unrelated domains. Its wider parent structure should be credited to actual broader catalog entries, while any proposed cross-domain analogue remains an analogy or a separately defined model.^[1]

A stochastic process describes a collection of random states over time. The Reed–Frost state \((S_t,I_t)\) is indexed by generation and governed by a conditional next-state law. Its next-step distribution is determined by the present counts and fixed \(p\), so the model has a Markov reading.^[1] It advances in discrete steps, making it a discrete-time Markov chain in the stated finite-state setting.

The current catalog's Epidemic Models on Lattices (Epidemic models on lattices) is a nearby domain concept with a different carrier: its definition makes a lattice, local neighborhood exposure, and spatial updating constitutive. Classic Reed–Frost instead composes homogeneous pair chances across a bounded susceptible population.^[1] The two are therefore distinct catalog identities. Greenwood's model is an even closer formal neighbor, distinguished by an infection chance that does not rise with positive \(I\) in the Reed–Frost power form.^[1]

Neighborhood in Abstraction Space¶

Reed–Frost Model sits in a sparse region of the domain-specific corpus (95^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2835 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-31

Not to Be Confused With¶

Greenwood's chain-binomial model: it also has discrete generations and binomial new-case counts, but its positive infection chance is constant for any \(I>0\). The \(I\)-dependent escape power is the discriminating Reed–Frost condition.^[1]
A deterministic mean recursion: it follows a representative expected increment instead of drawing the next number of cases. It cannot by itself display early extinction or the distribution over chains that the stochastic model specifies.^[1]
A whole heterogeneous or externally seeded household model: a classic internal Reed–Frost kernel may survive inside it, but importing cases or varying infectiousness changes the whole model's transition.^[4]
A spatial lattice epidemic model: the catalog's Epidemic models on lattices requires a lattice and neighborhood-dependent updates. A local rule that happens to use an escape product needs its own declared population and adjacency assumptions. The classic Reed–Frost rule uses independent common pair chances instead.^[1]
The frozen source's logarithmic reproduction formula: \(\log(1/(1-p))\) is not a reproduction count derived from the pair chance \(p\) alone.^[1] Under the formal source's population convention, the stated basic reproduction number is \(np\).^[1] This entry does not repeat the frozen formula as fact.

References¶

[1] reedfrost project, “Model descriptions”, CDCgov repository owner, sections “General assumptions and corollaries,” “Reed-Frost,” and “Greenwood.” Formal model definition and comparison; not empirical evidence that homogeneous independent contacts occur in nature. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28 ↩29 ↩30 ↩31 ↩32 ↩33 ↩34 ↩35 ↩36 ↩37 ↩38 ↩39 ↩40

[2] M. A. Heasman and D. D. Reid, “Theory and Observation in Family Epidemics of the Common Cold,” British Journal of Preventive & Social Medicine 15, no. 1 (1961): 12–16, https://doi.org/10.1136/jech.15.1.12. See printed pp. 12–15, especially first-stage Table I, generation classification, and complete-chain Table V. Scanned original available at https://pmc.ncbi.nlm.nih.gov/articles/PMC1058809/. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l

[3] Beverley J. Paterson, Martyn D. Kirk, A. Scott Cameron, Catherine D’Este, and David N. Durrheim, “Historical Data and Modern Methods Reveal Insights in Measles Epidemiology: A Retrospective Closed Cohort Study,” BMJ Open 3, no. 1 (2013): e002033, https://doi.org/10.1136/bmjopen-2012-002033. See “Reed-Frost stochastic model,” Methods, and “Stochastic measles outbreak modelling,” Results. The article contains a later corrected numerical \(R_0\) value; no numeric \(R_0\) finding is used here. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k

[4] Christophe Fraser et al., “Influenza Transmission in Households During the 1918 Pandemic,” American Journal of Epidemiology 174, no. 5 (2011): 505–514, https://doi.org/10.1093/aje/kwr122. See “Basic Reed-Frost model” and “An extended model.” Used here as a bounded internal-component and extension comparison, not a second whole classic example. registry ↩a ↩b ↩c ↩d ↩e ↩f