Epidemic models on social networks -- with inference

Tom Britton

Epidemic models on social networks -- with inference

T0 review · 0 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read For three common random networks, R0 is determined by the size-biased degree distribution and the transmission probability.

desk verdict Competent, honest survey of network epidemic models and inference; no new results, but a useful orientation with correct standard formulas. read the letter →

arxiv 1908.05517 v1 pith:PU2KQLIR submitted 2019-08-15 q-bio.PE math.PRstat.ME

Tom Britton This is my paper · ORCID

classification q-bio.PEmath.PRstat.ME MSC 92D3005C80

keywords randomnetworksepidemicmodelsSIRReed-Frostmodelreproductionnumberconfigurationpreferentialattachmentstatisticalinference

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This survey paper assembles the standard toolkit for modelling an infectious disease outbreak on a social network and for estimating the model from data. Its central concrete results are explicit formulas for the basic reproduction number R0 on three widely used random network models—Erdős-Rényi, configuration, and preferential attachment—in both the Reed-Frost and Markovian SIR formulations. It also states that uniformly vaccinating a fraction v scales R0 down by the factor 1-v, so the critical vaccination coverage is 1-1/R0. The rest of the paper inventories what can be inferred about transmission and network parameters under different data scenarios and lists open problems.

What carries the argument

The size-biased degree distribution is the central object: if the community degree distribution is p_k with mean \mu_D, then an infected individual during the early outbreak has degree distribution \tilde p_k = k p_k / \mu_D. The expected number of susceptible neighbours of an early infective is E(\tilde D - 1), and multiplying by the per-neighbour transmission probability p (Reed-Frost) or \$\beta$/(\$\beta$+\gamma) (Markovian) gives R0. This same quantity also determines the uniform vaccination reduction R0(1-v).

What would settle it

Simulate a Markovian SIR outbreak on a configuration-model network with a known degree distribution, transmission rate $\beta$, and recovery rate gamma; if the average number of secondary infections caused by early infectives does not approach ($\beta$/($\beta$+gamma)) E(\tilde D - 1) as the network grows, the size-biased heuristic fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that in a large Erdős-Rényi, configuration, or preferential attachment network with degree distribution D, the early-stage reproduction number is governed by the size-biased degree distribution \tilde D, where an individual is k times more likely to be infected if it has degree k. For the Reed-Frost epidemic, R0 = p(E(\tilde D - 1)) = p(\sum_k $k^{2}$ p_k / \mu_D - 1); for the Markovian SIR epidemic, R0 = (\$\beta$/(\$\beta$+\gamma)) E(\tilde D - 1), with the same size-biased term. Uniform vaccination of a fraction v reduces R0 to R0(1-v). Because the preferential attachment degree distribution has infinite variance, its R0 is infinite.

Load-bearing premise

The load-bearing premise is that an individual with k contacts is k times as likely to be infected as someone with one contact, so early infectives have degree distribution proportional to k p_k, and all non-infector neighbours are still susceptible in the early stage.

Editorial extensions

If this is right

R0 for these network epidemics can be computed from the mean and variance of the degree distribution alone, without simulating the epidemic.
Uniform vaccination coverage v reduces R0 by the factor 1-v, so the critical coverage needed for herd immunity is 1-1/R0.
On preferential attachment networks, where the degree distribution has infinite variance, R0 is infinite and no finite uniform vaccination coverage can bring it below 1, so targeted vaccination of high-degree individuals is necessary.
When only the final infected fraction is observed and the network model is known, a transmission parameter can be estimated by matching the predicted final size to the observed fraction.
When temporal or viral sequence data are available, Bayesian and likelihood-based methods can resolve some of the ambiguity about who infected whom, although several inferential problems remain open.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same size-biased-degree formula suggests that early epidemic growth may depend on the degree distribution only through its mean and variance, a simplification that could be tested on networks with different local structures such as clustering.
If behavioural changes make high-degree individuals less likely to be infected, the size-biased assumption breaks down and R0 would be lower than the formula predicts; this gives a testable distinction between structural and behavioural drivers of spread.
Combining contact-tracing data with virus sequence data could resolve the edge-level likelihood ambiguity the paper identifies, since sequences can indicate the direction and timing of transmission on a known network.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This is a survey, and it should be read as one. No new theorem, estimator, or dataset. What it does well is give a compact, accurate map of the field: random graph models, SIR on networks, reproduction numbers, vaccination strategies, and the main inference settings with pointers to the literature. The formulas in Section 5 are standard and correctly stated. The size-biased degree heuristic is the right basis for them, and the conclusion that early infectives see degree distribution ~p_k = k p_k / mu_D is correct for the models considered. The uniform vaccination reduction R_v = R_0(1-v) is also standard and correct. The author is explicit that proofs are omitted; that is a presentation choice typical for a survey, and the reader is directed to the original papers.

The soft spots are proportionate. Section 5 is the load-bearing part, and the 'without 100% rigor' disclaimer does a lot of work. If a reader wants to understand why the formulas are true, the heuristic is concise but not self-contained; a more detailed derivation or more specific references at that point would help. The inference sections are overviews, not manuals; likelihoods are sketched and then the reader is sent elsewhere. There are typos and informal phrases, but nothing that misleads. In Section 6.3 the author states a strong opinion about sequence data plus incidence data; it is clearly labeled as an opinion, so it does not contaminate the rest.

Citations look fine. Self-citations point to original derivations, not to this survey, so there is no circularity problem.

Who it is for: people entering network epidemic modeling, or researchers in adjacent fields who want a quick orientation before diving into the primary literature. It has that value. I would not build new research on it without going to the cited papers, but that is true of any survey.

Recommendation: it deserves a serious referee. I would accept it for a journal that publishes surveys, with light revision to expand the Section 5 proof sketch or at least provide a roadmap to the rigorous results, and a cleanup of typos. It is not a desk reject.

Referee Report

0 major / 7 minor

Summary. This paper is a survey of stochastic SIR-type epidemic models on random social networks and of statistical methods for fitting such models to data. Section 2 defines a menu of network models: Erdos-Renyi graphs, the configuration model, preferential attachment, the Watts-Strogatz small-world model, and extensions (weighted, directed, bipartite, dynamic, and exponential random graph models). Section 3 defines the discrete-time Reed-Frost model and the continuous-time Markovian SIR model on a network, optionally supplemented by global "random contacts". Section 4 discusses vaccination and other control measures, contrasting uniform, acquaintance, and degree-based strategies. Section 5 states the paper's main quantitative claims: the early-stage reproduction numbers R0^(RF) = p E[D(D-1)]/E[D] and R0^(M) = (beta/(beta+gamma)) E[D(D-1)]/E[D] for the Erdos-Renyi, configuration, and preferential attachment networks (with additive random-contact terms), the threshold criterion R0 <= 1, and the uniform-vaccination reduction R_v^(U) = R0(1-v) with critical coverage 1 - 1/R0. Section 6 reviews inference for known networks (including a pseudo-likelihood for household data and a temporal-data likelihood), for known network models with aggregate observations, and for unknown networks using virus sequence data; Section 7 closes with open problems.

Significance. Assuming the central formulas are correct -- and they are standard results, correctly assembled -- the survey fulfills its purpose well. Result 5.2 matches the accepted branching-process thresholds for locally tree-like configuration-type networks, with the Erdos-Renyi case as the Poisson limit and the preferential-attachment case reflecting the known vanishing threshold for power-law networks; Result 5.3's vaccination reduction and critical coverage are likewise standard. Strengths I want to credit explicitly: the paper is honest about its own level of rigor ("without 100% rigor", "We will not prove this result"), carefully distinguishes exact likelihoods from pseudo-likelihoods in Section 6.1, flags modeling assumptions such as the one-dominant-strain simplification in Section 6.3, and lists a genuinely useful set of open problems (standard errors for the temporal-data MLE, credible combination of sequence and incidence data, inference under unknown networks).

minor comments (7)

[Section 5 (Results 5.1-5.3; the paragraph "We will not prove this result")] The paper explicitly states "We will not prove this result" and that results are given "without 100% rigor"; these disclaimers are honest and appropriate for a survey, but Results 5.1-5.3 are the quantitative core of the paper, so the survey should point to the specific literature where these facts are proved: the branching-process derivations for configuration-type networks (e.g., the network chapter of Diekmann, Heesterbeek and Britton [13]) and the threshold behavior for preferential-attachment networks. Adding these pointers is a local fix that lets readers verify the claims without re-deriving them, and it does not affect my verdict on correctness.
[Section 5 (Result 5.2)] Result 5.2 groups the Erdos-Renyi, configuration, and preferential attachment models as if one derivation covered all three; the size-biased formula is rigorously the threshold parameter for sparse locally tree-like configuration-type networks, with the Erdos-Renyi case as the Poisson limit, whereas for preferential attachment the size-biased heuristic is not a proof and the stated R0 = infinity is the known threshold behavior of power-law networks; one sentence delimiting the domain of validity of the heuristic would prevent misapplication.
[Section 5 (Result 5.1)] The statement that tau_n -> 0 in probability if and only if R0 <= 1 is formally true but misleading, because for R0 > 1 the final fraction does not converge in probability at all: it converges in distribution to a mixture of 0 (minor outbreak) and a positive limit. I suggest phrasing the threshold as "a major outbreak occurs with positive probability if and only if R0 > 1", or explicitly marking the statement as informal in the result itself.
[Section 5 (Definition 5.1)] Because the index case has mean offspring p E[D] while later infectives have mean p E[D(D-1)]/E[D], and it is the latter that determines the threshold, Definition 5.1's reference to "typical infected individuals" should be clarified with a sentence stating that R0 refers to early non-index infectives; this prevents the common confusion between the index-case mean and the threshold parameter.
[Section 3 (Remark numbering)] There are two Remarks both numbered 3.2, one following Definition 3.2 and one following Definition 3.3; renumber the second remark so that citations to specific remarks are unambiguous.
[Throughout (typos)] The manuscript needs a full copy-edit: examples include "proporties" (Abstract and Section 1), "probaility" and "studs" (Section 2), "revoers", "efter", and "callibrated" (Section 3), "inte next section" (Section 4), "dived by" (Section 6.1), "paramaters" (Section 6.4), "extentions" (Sections 3 and 7), and "indivuduals" and "wames" (Section 7).
[References] Reference [33] lists the first author of the small-world model as "Watts, S.C."; the correct initials are D.J., and the reference formatting is inconsistent across entries (for example, the journal-name conventions differ between [5] and [6]).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the survey's reproduction-number results rest on standard model definitions and a size-biased heuristic, not on fitted inputs or self-citations that force the conclusions.

full rationale

This is a survey paper, not an original derivation with fitted parameters. The central formulas in Result 5.2 are presented as known properties of the defined network and epidemic models, and the paper explicitly states 'We will not prove this result, but give a quick heuristic explanation of the first equality.' The heuristic is the standard size-biased degree argument: during the early stage, an infected individual's degree is distributed as p_k tilde = k p_k / mu_D, and each such individual has tilde(D)-1 susceptible neighbours, giving R0 = p E(tilde(D)-1) for Reed-Frost and R0 = (beta/(beta+gamma)) E(tilde(D)-1) for the Markovian model. This is not circular because the degree distribution and transmission parameters are model inputs, not outputs of any fitting procedure, and the size-biased reasoning is an independent probabilistic argument rather than an assumption equivalent to the result. Result 5.3, R_v^(U) = R0(1-v) for uniform vaccination, follows from the thinning of susceptible degrees under random vaccination and is again a standard consequence of the model definition, not a renamed input. The self-citations in the paper (e.g., Britton and O'Neill 2002; Ball, Britton and Sirl 2013) are references to the original sources of surveyed results and are not used as unverified premises that force the paper's conclusions. The paper's own disclaimer about lack of full rigor is a presentation choice typical of surveys and does not indicate circularity. No equation or claim reduces by construction to its own inputs, and no fitted quantity is later relabeled as a prediction. The analysis is therefore self-contained for the purposes of the circularity pass.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper is a survey, so it introduces no fitted parameters and no invented entities. The axioms are the standard modelling assumptions of the field: sparse static networks, finite mean degree, constant infectivity, and the size-biased degree heuristic used without proof in Section 5. These are domain assumptions rather than ad hoc inventions.

assumptions (5)

domain assumption The network is sparse with finite mean degree: E(D)=mu with 0<mu<infinity as n grows.
Section 2 states the focus is on sparse graphs where the number of edges is O(n), so each individual has mean degree mu fixed as n grows.
domain assumption The network is static and unweighted for the main results.
Section 2 says static networks may be sufficient for short-term outbreaks, and the R0 results in Section 5 are stated for static undirected networks.
domain assumption Infectious individuals have constant infectivity, with infectious period either fixed or exponentially distributed.
Section 3 defines the Reed-Frost and Markovian SIR models under this simplifying assumption, and the reproduction number formulas in Section 5 use p for Reed-Frost and beta/(beta+gamma) for the Markovian model.
domain assumption Early infectives have size-biased degree: P(degree=k) proportional to k p_k, and all neighbours except the infector are susceptible.
Section 5 uses this heuristic to derive R0 without proof; it is the weakest load-bearing premise behind the central formulas.
domain assumption Initially all individuals except a random index case are susceptible.
Definitions 3.1 and 3.2 assume one randomly chosen index case and the rest susceptible, and the inference sections discuss corrections for prior immunity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Epidemic models on social networks -- with inference." pith.science (2026). https://pith.science/paper/PU2KQLIR

@misc{pith2026190805517,
  author       = {Pith},
  title        = {Pith review of: Epidemic models on social networks -- with inference},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PU2KQLIR}},
  note         = {Machine review of arXiv:1908.05517}
}

read the original abstract

Consider stochastic models for the spread of an infection in a structured community, where this structured community is itself described by a random network model. Some common network models and transmission models are defined and large population proporties of them are presented. Focus is then shifted to statistical methodology: what can be estimated and how, depending on the underlying network, transmission model and the available data? This survey paper discusses several different scenarios, also giving references to publications where more details can be found.

Figures

Figures reproduced from arXiv: 1908.05517 by the authors.

**Figure 2.** Illustration of the configuration model for a very small network. a) The degree [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Illustration of the final outcome in a small random social network. Red nodes [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Illustration of an transmission tree (A) and the corresponding virus genealogy [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

[1]

L., Longini I

Addy C. L., Longini I. M. and Haber M. (1991). A generalized stochastic model for the analysis of infectious disease ﬁnal size data. Biometrics 47, 961-974

work page 1991
[2]

K., Borgan Ø, Gill R

Andersen P. K., Borgan Ø, Gill R. D., Keiding N. (1993). Statistical models based on counting processes. New York: Springer

work page 1993
[3]

and Sirl, D

Ball, F., Britton,T. and Sirl, D. (2013) A network with tunable clustering, degree correlation and degree distribution, and an epidemic thereon. J. Math. Biol., 66:979- 1019

work page 2013
[4]

G., Mollison D

Ball F. G., Mollison D. and Scalia-Tomba G. (1997). Epidemics with two levels of mixing. Ann. Appl. Prob. 7, 46-89

work page 1997
[5]

and Albert, R

Barab´ asi, A.-L.. and Albert, R. (1999). Emergence of Scaling in Random Networks. Science, 286, 509-512

work page 1999
[6]

and Vespignani, A

Barrat, A., Barth´ elemy, M., Pastor-Satorras, R. and Vespignani, A. (2004). The architecture of complex weighted networks. PNAS. 101, 3747-3752

work page 2004
[7]

(2001).Random Graphs, Academic Press, 2nd ed

Bollob´ as, B. (2001).Random Graphs, Academic Press, 2nd ed

work page 2001
[8]

and Riordan, O

Bollob´ as, B., Janson, S. and Riordan, O. (2007). The phase transition in inhomoge- neous random graphs. Rand. Str. Alg. 31, 3-122

work page 2007

Show all 33 references

[9]

and Turova, T

Britton, T., Lindholm, T. and Turova, T. (2011). A dynamic network in a dynamic population: asymptotic properties. J. Appl. Prob. 48: 1163-1178

2011
[10]

and O’Neill, P

Britton, T. and O’Neill, P. D. (2002). Bayesian inference for stochastic epidemics in populations with random social structure. Scand. J. Stat., 29, 375-390

2002
[11]

Cauchemez, S., Valleron, A.-J., Bo¨ elle, P.-Y., Flahault, A., and Ferguson, N.M. (2008). Estimating the impact of school closure on inﬂuenza transmission from Sen- tinel data. Nature. 452, 750–754

2008
[12]

and Ben-Avraham, D

Cohen, R., Havlin, S. and Ben-Avraham, D. (2003). Eﬃcient immunization strategies for computer networks and populations. Phys. Rev. Lett. 91: 247901

2003
[13]

and Britton, T

Diekmann O., Heesterbeek, J.A.P. and Britton, T. (2013). Mathematical tools for understanding infectious disease dynamics . Princeton UP

2013
[14]

and R´ enyi, A

Erd˝ os, P. and R´ enyi, A. (1959). On random graphs,Publ. Math. 6, 290-297

1959
[15]

Felsenstein, J. (2003). Inferring phylogenies, 2nd Ed . Oxford UP. 18

2003
[16]

Giardina, F., Romero-Severson E.O., Albert J., Britton T., and Leitner T.K. (2017). Inference of transmission network structure from HIV phylogenetic trees.PLoS Comp Biol. 13:e1005316

2017
[17]

and Struchiner, C.J

Halloran, M.E., Longini, I.M. and Struchiner, C.J. (2010). Design and analysis of vaccine studies. Springer

2010
[18]

and Wallinga, J

Held, L., Hens, N., O’Neill, P.D. and Wallinga, J. (Eds). (2019). Handbook of infec- tious disease data analysis . CRC Press

2019
[19]

I.M., Watson, C.H., Edmunds, W.J

Henao-Restrepo, A.M., Camacho, A., Longini. I.M., Watson, C.H., Edmunds, W.J. et al. (2017). Eﬃcacy and eﬀectiveness of an rVSV-vectored vaccine in preventing Ebola virus disease: ﬁnal results from the Guinea ring vaccination, open-label, cluster- randomised trial (Ebola C ¸ a...

2017
[20]

and Britton, T

Leung, K., Ball, F., Sirl, D. and Britton, T. (2018). Individual preventive social distancing during an epidemic may have negative population-level outcomes. Journal Royal Society: Interface , 15:20180296

2018
[21]

Leventhal, G.E., Kouyos, R., Stadler, T., Von Wyl, V., Yerly, S., B¨ oni, J., et al. (2012). Inferring epidemic contact structure from phylogenetic trees. PLoS Comp Biol. 8:e1002413

2012
[22]

and Fraser, C

Li, L.M., Grassly, N.C. and Fraser, C. (2017). Quantifying transmission heterogeneity using both pathogen phylogenies and incidence time series. Mol. Biol. Evol. , 34: 2982–2995

2017
[23]

Longini I. M. and Koopman J. S. (1982). Household and community transmission parameters from ﬁnal distributions of infections in households. Biometrics, 38: 115- 126

1982
[24]

and Halloran, M.E

Longini, I.M., Nizam, A., Xu, S., Ungchusak, K., Hanshaoworakul, W., Cummings, D.A.T. and Halloran, M.E. (2005). Containing pandemic inﬂuenza at the source. Science. 309, 1083-1087

2005
[25]

and Reed, B

Molloy, M. and Reed, B. (1998). The size of the giant component of a random graphs with a given degree sequence, Comb. Prob. Comp. 7, 295-305

1998
[26]

and Strogatz, S.H

Newman, M.E.J., Watts, D.J. and Strogatz, S.H. (2002). Random graph models of social networks. PNAS. 99, 2566-2572

2002
[27]

and Reittu, H

Norros, I. and Reittu, H. (2006). On a conditionally Poissonian graph process. Adv. Appl. Prob., 38, 59-75

2006
[28]

and Snijders, T.A.B

Nowicki, K. and Snijders, T.A.B. (2001). Estimation and prediction for stochastic blockstructures. J. Am. Stat. Ass. , 96, 1077-1087

2001
[29]

and Vespignani, A

Poletto, C., Gomes, M.F.C., Piontti, A.P., Rossi, L., Bioglio, L., Chao, D.L., Longini, I.M., Halloran, M.E., Colizza, V. and Vespignani, A. (2014). Assessing the impact of travel restrictions on international spread of the 2014 West African Ebola epidemic. Eurosurveillance, 1...

2014
[30]

and Halloran, M.E

Potter, G.E., Handcock, M.S., Longini, I.M. and Halloran, M.E. (2011). Estimat- ing within-household contact networks from egocentric data. Ann. Appl. Stat . 5: 1816–1838

2011
[31]

and Handcock, M.S

Snijders, T.A.B., Pattison, P.E., Robins, G.L. and Handcock, M.S. (2006). New speciﬁcations for exponential random graph models. Soc. Method., 36: 99-153

2006
[32]

and Britton, T

Spricer K. and Britton, T. (2015): The conﬁguration model for partially directed graphs. J. Stat. Phys. 161, 965-985

2015
[33]

and Strogatz, S.H

Watts, S.C. and Strogatz, S.H. (1998). Collective dynamics of ‘small-world’ networks, Nature 393, 440–442. 20

1998

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

L., Longini I

Addy C. L., Longini I. M. and Haber M. (1991). A generalized stochastic model for the analysis of infectious disease ﬁnal size data. Biometrics 47, 961-974

work page 1991

[2] [2]

K., Borgan Ø, Gill R

Andersen P. K., Borgan Ø, Gill R. D., Keiding N. (1993). Statistical models based on counting processes. New York: Springer

work page 1993

[3] [3]

and Sirl, D

Ball, F., Britton,T. and Sirl, D. (2013) A network with tunable clustering, degree correlation and degree distribution, and an epidemic thereon. J. Math. Biol., 66:979- 1019

work page 2013

[4] [4]

G., Mollison D

Ball F. G., Mollison D. and Scalia-Tomba G. (1997). Epidemics with two levels of mixing. Ann. Appl. Prob. 7, 46-89

work page 1997

[5] [5]

and Albert, R

Barab´ asi, A.-L.. and Albert, R. (1999). Emergence of Scaling in Random Networks. Science, 286, 509-512

work page 1999

[6] [6]

and Vespignani, A

Barrat, A., Barth´ elemy, M., Pastor-Satorras, R. and Vespignani, A. (2004). The architecture of complex weighted networks. PNAS. 101, 3747-3752

work page 2004

[7] [7]

(2001).Random Graphs, Academic Press, 2nd ed

Bollob´ as, B. (2001).Random Graphs, Academic Press, 2nd ed

work page 2001

[8] [8]

and Riordan, O

Bollob´ as, B., Janson, S. and Riordan, O. (2007). The phase transition in inhomoge- neous random graphs. Rand. Str. Alg. 31, 3-122

work page 2007