{"id":"cde4d236-c526-4a96-a440-f0efeddb0467","arxiv_id":"1908.05517","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey paper that assembles known reproduction number formulas, vaccination thresholds, and inference methods for epidemic models on random networks.","lead":"This paper is a survey of mathematical models for how infectious diseases spread on social networks, and of the statistical methods used to fit those models to data. It is a useful orientation for researchers who want the main formulas and open problems in one place.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the Section 5 reproduction-number formulas are standard, and the size-biased heuristic is the correct basis for them; the paper's explicit non-rigor is a presentation choice, not a defect.","rationale":"The reader's weakest assumption correctly locates the load-bearing premise in the size-biased degree heuristic of Section 5. I agree this is where the formulas rest, but the heuristic is valid in the standard large sparse configuration-model regime that the paper explicitly targets (Section 2 fixes a finite mean degree). The paper does not claim full rigour, and the results are standard textbook material. I found no internal inconsistency, no algebra error in the central formulas, and no unsupported claim that changes the survey's usefulness. The labelled opinion in Section 6.3 is presented as opinion and is not evidence. The only caveats are the unproved status of Results 5.2 and 5.3, the asymptotic interpretation of R0 = \\infty for preferential attachment, and the generally heuristic tone; these are appropriate for a survey and were flagged by the author. Therefore the reader's ACCEPT verdict stands unchanged.","tokens_in":14115,"tokens_out":15459,"duration_ms":156799,"concrete_test":"Independently re-derive Result 5.2 for the configuration model using a two-type branching process with generating functions for D and \\tilde D; compute the spectral radius of the mean offspring matrix and verify that it equals p(E(\\tilde D - 1)) for Reed-Frost and [\\beta/(\\beta+\\gamma)] E(\\tilde D - 1) for the Markovian epidemic, thereby confirming that the size-biased heuristic reproduces the stated threshold.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The only structural weakness is the paper's own Section 5 disclaimer ('We will not prove this result'): Result 5.2 and Result 5.3 are presented heuristically rather than proved. However, the size-biased degree heuristic is the standard and correct basis for these formulas on sparse, locally tree-like random graphs. Early infected individuals are reached via a randomly selected edge, so their degree is distributed as \\tilde p_k = k p_k / \\mu_D; with \\tilde D - 1 susceptible neighbours, the mean number of infections is p E(\\tilde D - 1) for Reed-Frost and [\\beta/(\\beta+\\gamma)] E(\\tilde D - 1) for the Markovian model. For the configuration model, a two-type branching process (index case with mean p \\mu_D, later infectives with mean p E(\\tilde D - 1)) confirms that the major-outbreak threshold is governed by the latter expression. Erdős-Rényi is covered as the Poisson limit. For preferential attachment, the infinite second moment should be read in the limiting sense, and the stated conclusion R0 = \\infty agrees with the known zero-threshold behaviour of scale-free networks. For uniform vaccination, the residual susceptible degree satisfies K | D ~ Bin(D, 1-v), giving E(\\tilde K - 1) = (1-v) E(\\tilde D - 1), so Result 5.3 is also correct. The omitted proofs are a survey convention and do not undermine the central claims.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":14389,"tokens_out":21790,"duration_ms":178415,"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).","major_comments":[],"minor_comments":[{"comment":"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":"Section 5 (Results 5.1-5.3; the paragraph \"We will not prove this result\")"},{"comment":"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":"Section 5 (Result 5.2)"},{"comment":"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":"Section 5 (Result 5.1)"},{"comment":"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":"Section 5 (Definition 5.1)"},{"comment":"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.","section":"Section 3 (Remark numbering)"},{"comment":"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).","section":"Throughout (typos)"},{"comment":"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]).","section":"References"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is a single-author survey; its self-citations (e.g., Britton and O'Neill 2002; Ball, Britton and Sirl 2013; Leung et al. 2018; Giardina et al. 2017; Spricer and Britton 2015; and the author's co-authored textbook [13]) are all original sources for the topics covered and are appropriate rather than excessive. The paper is a natural fit for a journal that publishes survey papers on stochastic epidemic modeling or statistical network analysis; there is no novelty-disclosure issue because the paper claims no new results. My minor_revision recommendation is driven by the proof-pointer requests and copy-editing needs, not by any substantive correctness concern."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","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.\n\nThe 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.\n\nCitations look fine. Self-citations point to original derivations, not to this survey, so there is no circularity problem.\n\nWho 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.\n\nRecommendation: 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.","headline":"Competent, honest survey of network epidemic models and inference; no new results, but a useful orientation with correct standard formulas.","tokens_in":14914,"tokens_out":2118,"would_cite":true,"duration_ms":19685,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D30","05C80"],"pacs":[],"model":"deepseek-v4-flash","headline":"For three common random networks, R0 is determined by the size-biased degree distribution and the transmission probability.","keywords":["random networks","epidemic models","SIR epidemic","Reed-Frost model","reproduction number","configuration model","preferential attachment","statistical inference"],"falsifier":"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.","tokens_in":1539,"feed_emoji":"🦠","tokens_out":2523,"duration_ms":68402,"temperature":0.7,"pith_summary":"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.","feed_headline":"R0 for three network models comes down to one formula","feed_subtitle":"Degree variance plus vaccination coverage decides whether an outbreak takes off or dies out.","key_machinery":"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).","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"Defines the Erdős-Rényi random graph, whose asymptotically Poisson degree distribution is the baseline case for the R0 formula.","marker":"[14]"},{"why":"Defines the configuration model with arbitrary degree distribution, the network class for which the size-biased degree computation is made.","marker":"[25]"},{"why":"Defines preferential attachment, the network model whose power-law degree distribution makes R0 infinite.","marker":"[5]"},{"why":"Supplies the acquaintance vaccination strategy presented as a targeted alternative to uniform vaccination.","marker":"[12]"},{"why":"Provides the household pseudo-likelihood used for final-size outbreak data on known household networks.","marker":"[23]"},{"why":"Supplies Bayesian MCMC inference that treats the unknown network as a latent variable.","marker":"[10]"},{"why":"Provides the phylogenetic-tree method for inferring epidemic contact structure from virus sequences.","marker":"[21]"},{"why":"Reports that adding incidence data to phylogenies can worsen precision, which the paper discusses as a counterintuitive finding to revisit.","marker":"[22]"},{"why":"Describes how to correct the household pseudo-likelihood for dependencies using the proper stochastic epidemic model and its central limit theorem.","marker":"[4]"}],"fun_headline_variants":["In scale-free nets, infinite variance makes R0 infinite","One formula for R0: size-biased degree and vaccine","Scale-free networks: infinite R0, so vaccination can't stop outbreaks","Three networks, one R0 formula: degree variance decides"],"cache_read_input_tokens":17024,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["In scale-free nets, infinite variance makes R0 infinite","One formula for R0: size-biased degree and vaccine","Scale-free networks: infinite R0, so vaccination can't stop outbreaks","Three networks, one R0 formula: degree variance decides"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001363,"raw_usage":{"total_tokens":5437,"prompt_tokens":760,"completion_tokens":4677,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":376,"completion_tokens_details":{"reasoning_tokens":4606}},"tokens_in":376,"tokens_out":4677,"duration_ms":32101,"temperature":1.0,"reasoning_tokens":4606,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:10:56.644409+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":"and R´ enyi, A","cited_arxiv_id":null,"evidence_quote":"Defines the Erdős-Rényi random graph, whose asymptotically Poisson degree distribution is the baseline case for the R0 formula."},{"cited_title":"and Reed, B","cited_arxiv_id":null,"evidence_quote":"Defines the configuration model with arbitrary degree distribution, the network class for which the size-biased degree computation is made."},{"cited_title":"and Albert, R","cited_arxiv_id":null,"evidence_quote":"Defines preferential attachment, the network model whose power-law degree distribution makes R0 infinite."},{"cited_title":"and Ben-Avraham, D","cited_arxiv_id":null,"evidence_quote":"Supplies the acquaintance vaccination strategy presented as a targeted alternative to uniform vaccination."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the household pseudo-likelihood used for final-size outbreak data on known household networks."},{"cited_title":"and O’Neill, P","cited_arxiv_id":null,"evidence_quote":"Supplies Bayesian MCMC inference that treats the unknown network as a latent variable."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the phylogenetic-tree method for inferring epidemic contact structure from virus sequences."},{"cited_title":"and Fraser, C","cited_arxiv_id":null,"evidence_quote":"Reports that adding incidence data to phylogenies can worsen precision, which the paper discusses as a counterintuitive finding to revisit."},{"cited_title":"G., Mollison D","cited_arxiv_id":null,"evidence_quote":"Describes how to correct the household pseudo-likelihood for dependencies using the proper stochastic epidemic model and its central limit theorem."}],"review_version":1}