Pith. sign in

REVIEW 3 major objections 3 minor 54 references

Fast, Frequentist Estimation of Epidemic Reproduction Numbers

T0 review · 3 major / 3 minor · reviewed 2026-07-08 · grok-4.5

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

Pith's one-line read ConvRt estimates epidemic reproduction numbers with a fast frequentist two-step method that matches Bayesian accuracy and uncertainty while running far faster and clarifying which Rt narratives the data support.

desk verdict Useful frequentist Rt pipeline on paper; the whole claim stack still sits on a fixed delay kernel the abstract treats as given. read the letter →

arxiv 2607.05887 v1 pith:PMHSMLNZ submitted 2026-07-07 stat.AP stat.ME

Jeremy Goldwasser , Ryan J. Tibshirani , Alyssa Bilinski This is my paper
classification stat.APstat.ME MSC 62P1062G0892D30
keywords effectivereproductionnumberRtestimationdeconvolutionpenalizedlikelihoodsplinesfrequentistmethodsinfectiousdiseasesurveillanceepidemicdynamics
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

The paper proposes ConvRt, a frequentist pipeline for estimating the effective reproduction number Rt from delayed case or hospitalization counts. It first deconvolves observed reports into latent infections using a known delay distribution, then recovers a smooth Rt curve through successive penalized-likelihood spline fits. Across stylized and data-driven simulations the method delivers competitive point estimates and uncertainty quantification at a fraction of the runtime of popular Bayesian estimators. Because smoothness is imposed only on the estimated past and is kept separate from any forward projection, researchers can also check which qualitative stories about rising or falling transmission remain consistent with the data. A sympathetic reader cares because Rt is the central real-time indicator of epidemic trajectory, yet existing Bayesian tools are slow and prior-sensitive; a transparent, fast frequentist alternative could make routine surveillance more practical and easier to interrogate.

What carries the argument

A two-step pipeline: deconvolution of delayed observed counts into latent infections (using a supplied delay distribution), followed by successive penalized-likelihood spline estimations of Rt that model smooth curves without embedding forecast assumptions into the smoothness penalty.

What would settle it

Run the same stylized or data-driven simulations with a deliberately misspecified delay distribution and check whether ConvRt’s point estimates and interval coverage still match or beat the Bayesian competitors the paper claims to rival; systematic degradation under realistic delay error would refute the performance claim.

Watch

Extended reading notes

Core claim

ConvRt is a two-step frequentist procedure that first recovers latent infection incidence by deconvolution and then estimates the effective reproduction number Rt via successive penalized-likelihood spline steps that enforce temporal smoothness. In both stylized and data-driven simulations it achieves favorable point estimation, uncertainty quantification, and runtime relative to popular Bayesian Rt methods. By untangling the smoothness penalty from future projections, the same pipeline also lets researchers assess which qualitative narratives about Rt the observed data actually support.

Load-bearing premise

The delay distribution that maps the moment of transmission to later observed cases or hospitalizations is treated as known and correctly specified; if it is wrong, the deconvolution step and every subsequent Rt estimate are biased.

Editorial extensions

If this is right

  • Public-health teams can produce real-time Rt estimates with calibrated uncertainty intervals in a small fraction of the compute time required by Bayesian methods.
  • Analysts can separately test which qualitative Rt trajectories remain consistent with the data once smoothness is imposed only on the observed past.
  • Retrospective reconstruction of past epidemic waves becomes faster and less dependent on prior choices for the infection process.
  • The same pipeline applies equally to case counts and hospitalization series once the appropriate delay distribution is supplied.
  • Method-comparison frameworks gain a transparent frequentist baseline that isolates deconvolution error from smoothing error.

Reading between the lines

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

  • If the delay distribution is even moderately misspecified, the entire Rt path inherits systematic bias, so validating or jointly estimating that distribution is the highest-leverage next experiment.
  • The successive-spline structure may extend to multi-pathogen or age-stratified settings where hierarchical Bayesian models currently dominate but scale poorly.
  • Untangling smoothness from projection suggests a practical diagnostic: report the envelope of smooth Rt curves that stay inside a fixed likelihood contour of the data rather than a single point estimate.
  • Computational speed makes daily bootstrap or sandwich-based uncertainty feasible at both national and local scales without specialized hardware.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The manuscript proposes ConvRt, a frequentist two-step procedure for estimating the effective reproduction number Rt from delayed epidemic counts (cases or hospitalizations). Step 1 deconvolves latent infections via a fixed delay kernel; Step 2 estimates Rt by successive penalized-likelihood spline fits. The authors claim favorable point estimation, uncertainty quantification, and runtime versus popular Bayesian methods in stylized and data-driven simulations, and that separating smoothness from projections lets users assess which qualitative Rt narratives the data support.

Significance. If the claims hold under realistic delay misspecification and proper UQ calibration, a fast frequentist alternative that avoids prior sensitivity and enables narrative assessment would be a useful addition to the Rt toolkit for both retrospective analysis and real-time surveillance. The separation of smoothness from projections is a genuine methodological contribution if demonstrated with clear, reproducible diagnostics. Strengths to credit if present in the full text: simulation design with baselines, runtime comparisons, and any open code or parameter-free aspects of the spline steps.

major comments (3)
  1. The central claim rests on deconvolution of latent infections with a delay/lag kernel treated as known. The abstract and method framing do not establish how the kernel is obtained, validated, or whether its uncertainty is propagated into Rt intervals. Misspecification would bias the infection curve and all subsequent Rt estimates; intervals that ignore kernel uncertainty will be overconfident relative to Bayesian methods that place priors on delays. Simulation wins under a correctly specified kernel therefore do not automatically transfer to real surveillance data. This is load-bearing for both the performance claims and the narrative-assessment feature.
  2. Uncertainty quantification comparisons versus Bayesian methods are only credible if the frequentist intervals account for (or explicitly exclude and justify excluding) delay-kernel uncertainty and the two-step estimation error. Without that accounting, favorable UQ claims risk being an artifact of not paying for the same sources of uncertainty the Bayesian baselines include.
  3. The free parameters of the procedure—spline smoothness/penalty and the delay distribution—must be specified without circular use of the same series used for evaluation. How penalties are chosen (cross-validation, fixed defaults, etc.) and whether delay parameters are estimated or fixed from external sources needs to be stated and stress-tested; otherwise the “frequentist, prior-free” framing is incomplete.
minor comments (3)
  1. Clarify notation for the delay kernel, the successive penalized-likelihood steps, and the spline basis so that the two-step pipeline is fully reproducible from the text alone.
  2. Report simulation design details (baselines, error metrics, delay-model specification, number of replicates) in tables or figures that allow direct comparison of point error, coverage, and runtime.
  3. State explicitly how real-time (nowcasting) versus retrospective estimation differ in the algorithm and in the evaluation protocol.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for a careful, constructive report that correctly identifies three load-bearing under-specifications in the submitted manuscript: provenance and uncertainty of the delay kernel, the precise scope of the frequentist intervals relative to Bayesian baselines, and the non-circular selection of free parameters. We agree these points must be clarified and stress-tested before the performance, UQ, and narrative-assessment claims can be fully evaluated. In revision we will (i) document external sources for delay kernels and add misspecification sensitivity analyses, (ii) state explicitly which uncertainties enter the intervals and supply matched comparisons that isolate estimation procedure from kernel priors, and (iii) give the exact penalty-selection algorithm together with stress tests. These changes strengthen rather than alter the core contribution—a fast, transparent two-step frequentist pipeline. Detailed point-by-point replies follow.

read point-by-point responses

  1. Referee: The central claim rests on deconvolution of latent infections with a delay/lag kernel treated as known. The abstract and method framing do not establish how the kernel is obtained, validated, or whether its uncertainty is propagated into Rt intervals. Misspecification would bias the infection curve and all subsequent Rt estimates; intervals that ignore kernel uncertainty will be overconfident relative to Bayesian methods that place priors on delays. Simulation wins under a correctly specified kernel therefore do not automatically transfer to real surveillance data. This is load-bearing for both the performance claims and the narrative-assessment feature.

    Authors: We agree the delay kernel is load-bearing and that the submitted text did not adequately document its provenance or the consequences of conditioning on it. ConvRt, like the EpiEstim and EpiNow2 baselines we compare against, takes the delay distribution as an external input estimated from independent contact-tracing, clinical, or serological studies, never from the incidence series under analysis. We will revise Methods to state this explicitly, cite the precise external sources used in every simulation and real-data example, and add a dedicated subsection on kernel misspecification. There we will report controlled experiments in which the analysis kernel is shifted or overdispersed relative to the data-generating kernel, quantifying resulting bias in the infection curve and in Rt. We will also state clearly that the reported intervals condition on the kernel and therefore do not automatically cover kernel uncertainty; when that uncertainty is material, users should run the sensitivity analyses we will supply. Simulation comparisons under a correctly specified kernel remain the fair head-to-head setting used throughout the Bayesian literature as well, but automatic transfer to poorly characterized surveillance delays is limited—an important caveat we will foreground. Once the kernel is fixed, the subsequent spline steps (and therefore the narrative-assessment feature) are unaffected by how the kernel was chosen; any kernel-induced bias of course propagates into the narratives that appear supported, which the new sensitivity results will illustrate. revision: yes

  2. Referee: Uncertainty quantification comparisons versus Bayesian methods are only credible if the frequentist intervals account for (or explicitly exclude and justify excluding) delay-kernel uncertainty and the two-step estimation error. Without that accounting, favorable UQ claims risk being an artifact of not paying for the same sources of uncertainty the Bayesian baselines include.

    Authors: The referee is correct that credible UQ comparisons require transparent accounting of included uncertainties. In the present implementation the frequentist intervals are obtained from the penalized-likelihood Hessian (or a nonparametric bootstrap) after deconvolution; they therefore capture sampling variability of the observed counts and the two-step estimation error conditional on the kernel and the chosen penalty. They do not integrate over kernel uncertainty. We will revise the Uncertainty Quantification section to state this scope explicitly and to contrast it with the Bayesian baselines, which place priors on delays and can produce wider intervals when those priors are diffuse. In the simulations we will continue to report coverage under the correctly specified kernel (the conventional validation setting), and we will add a parallel suite in which the Bayesian methods are also run with a fixed, correctly specified delay (removing the prior contribution) so that the comparison isolates the estimation procedure itself. Where the Bayesian intervals remain better calibrated once kernel uncertainty is removed, we will report that fact. Together with the misspecification studies of the preceding comment, these changes will make the UQ claims precise rather than overstated. revision: yes

  3. Referee: The free parameters of the procedure—spline smoothness/penalty and the delay distribution—must be specified without circular use of the same series used for evaluation. How penalties are chosen (cross-validation, fixed defaults, etc.) and whether delay parameters are estimated or fixed from external sources needs to be stated and stress-tested; otherwise the “frequentist, prior-free” framing is incomplete.

    Authors: We agree that the “prior-free” framing is incomplete without a clear, non-circular account of free-parameter selection. The delay distribution is fixed from external sources (as detailed above) and is never estimated from the evaluation series. The spline penalty is selected by restricted maximum likelihood (REML) or by leave-one-out cross-validation performed solely on the deconvolved infection curve up to the estimation time; neither procedure uses held-out future observations. We will revise Methods to give the precise selection algorithm, default values, numerical tolerances, and software calls, and we will add a sensitivity analysis in which the penalty is deliberately under- or over-smoothed to show how point estimates and qualitative narratives change. We will also report the wall-clock cost of automatic penalty selection so that the runtime claims remain transparent. These additions fully specify the frequentist character of the procedure and allow exact reproduction of the reported fits. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; ConvRt is a self-contained frequentist two-step estimator evaluated in simulations, not a derivation that reduces by construction to its inputs.

full rationale

The paper introduces ConvRt as a frequentist procedure that first deconvolves latent infections from delayed observations and then estimates Rt via successive penalized-likelihood spline steps. Its claims are empirical performance (point estimates, UQ, runtime) versus Bayesian alternatives in stylized and data-driven simulations, plus the practical feature that separating the smoothness penalty from future projections lets users inspect which qualitative Rt narratives the data support. None of these reduce by construction to fitted targets or self-defined quantities: the delay kernel is an explicit external input to the inverse problem (standard, not circular), the spline penalties are transparent regularization choices rather than tautological fits renamed as predictions, and no uniqueness theorem, ansatz, or load-bearing self-citation is invoked to force the estimator. Simulation wins under controlled kernels are external benchmarks, not self-referential identities. Concerns about kernel misspecification or unpropagated delay uncertainty are correctness/assumption risks, not circularity. Per the analyzer rules, with no quotable reduction of a claimed result to its own inputs, the score is 0 and steps remain empty.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

Abstract-only audit. The method rests on standard statistical tools (deconvolution, penalized likelihood, splines) plus domain assumptions about epidemic delay structure and smoothness of Rt. Free parameters (smoothing penalties, any delay-distribution parameters) are implied but not quantified. No new physical entities are invented; ConvRt is a statistical procedure name.

free parameters (2)
  • spline smoothness / penalty parameter(s)
    Penalized-likelihood spline estimation of smooth Rt curves requires one or more penalty or bandwidth parameters; the abstract does not state how they are chosen (CV, fixed, etc.).
  • delay / lag distribution parameters
    Deconvolution of latent infections from delayed case or hospitalization counts requires a delay distribution; if any of its parameters are estimated or tuned rather than taken as known external inputs, they are free parameters load-bearing for Rt.
assumptions (3)
  • domain assumption Observed cases or hospitalizations are delayed noisy observations of latent infections via a known or estimable delay distribution.
    Core premise of any deconvolution-based Rt estimator; stated as the lag problem the method addresses.
  • domain assumption Rt is sufficiently smooth to be well represented by a spline basis under a penalized-likelihood criterion.
    Abstract states use of a spline basis to model smooth curves; validity of recovered Rt depends on this smoothness assumption.
  • standard math Standard frequentist penalized-likelihood and deconvolution theory apply to the successive estimation steps.
    Background statistical machinery the method invokes without re-deriving from first principles.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast, Frequentist Estimation of Epidemic Reproduction Numbers." pith.science (2026). https://pith.science/paper/PMHSMLNZ

@misc{pith2026260705887,
  author       = {Pith},
  title        = {Pith review of: Fast, Frequentist Estimation of Epidemic Reproduction Numbers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PMHSMLNZ}},
  note         = {Machine review of arXiv:2607.05887}
}
abstract

The effective reproduction number $R_t$ is one of the most important indicators of epidemic dynamics. Estimating $R_t$, typically from case reports or hospitalization counts, poses a challenging inverse problem. One key issue is lag: $R_t$ acts at the moment of transmission, while the data it generates surface days later. To handle this delay and infer recent infections in real time, popular methods take a Bayesian approach, which can be slow and sensitive to prior specification. As an alternative, we propose ConvRt, a frequentist method for retrospective and real-time estimation. ConvRt deconvolves latent infections and then estimates $R_t$ with successive penalized-likelihood steps, using a spline basis to model smooth curves. Across both stylized and data-driven simulations, we demonstrate favorable performance in point estimation, uncertainty quantification, and runtime. Moreover, by untangling smoothness from future projections, ConvRt enables researchers to assess which qualitative narratives about $R_t$ the data support.

Figures

Figures reproduced from arXiv: 2607.05887 by the authors.

Figure 1
Figure 1. Comparison of retrospective Rt estimates on benchmark datasets. We organize our comparison around three axes: point accuracy (MAE), uncertainty quantification (CE), and runtime. We begin with EpiNow2, ConvRt’s closest competitor, before turning to the others. Relative to EpiNow2, ConvRt reduces MAE by 20% or more on 5 of 6 datasets, and is effectively tied on the last. In the best case—the periodic rtestim benchmark… view at source ↗
Figure 2
Figure 2. Rt predictions for seasonal influenza. ConvRt and EpiNow2 fit retrospectively on NHSN hospitalization data, with 95% pointwise uncertainty bands. 2023/24 was a longer flu season than 2022/23, so its plotting window is extended. Both ConvRt and EpiNow2 generate similar Rt curves when fit retrospectively at the end of each flu season ( [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Real-time Rt estimates and uncertainty bands for seasonal influenza in 2023/24 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (21 more)
Figure 4
Figure 4. Figure 4: Rt predictions at varying levels of the smoothness hyperparameters λ (overall) and γ (tail). Cross-validation provides a sensible default for λ and γ, but two factors limit its authority in real time. First, at the boundary these cannot be well-estimated from data. Inf…
Figure 5
Figure 5. Figure 5: Residual (left) and studentized (right) conformal nowcast bands for ConvRt on the [PITH_FULL_IMAGE:figures/full_fig_p034_5.png]
Figure 6
Figure 6. Figure 6: Real-time coverage calibration on the flu simulation, scored against the true sim [PITH_FULL_IMAGE:figures/full_fig_p035_6.png]
Figure 7
Figure 7. Figure 7: Posterior-mean Rt from EpiNow2’s Gaussian-process model at successive MCMC snapshots on the retrospective wiggly-simulation fit (4 chains; 1500 warmup and 500 post￾warmup iterations per chain). Each line is one snapshot; color encodes the cumulative number of post-warm…
Figure 8
Figure 8. Figure 8: rtestim benchmark datasets: Rt curves and the cases they generate. An exponential takeoff is partially curbed by a late-stage response that bends but does not break the trajectory. 3. Piecewise linear. Four linear segments separated by jump discontinuities, Rt =  …
Figure 9
Figure 9. Figure 9: Retrospective Rt estimates on the rtestim benchmarks, comparing ConvRt with and without the jump augmentation against EpiNow2, EpiEstim, and estimateR. The jump augmentation dramatically sharpens ConvRt’s tracking at discontinuities while leaving smooth segments largel…
Figure 10
Figure 10. Figure 10: Ground-truth Rt and daily hospitalizations for the flu simulations. 0.6 0.8 1.0 1.2 Oct Nov Dec Jan Rt Rt Estimates, Flu Simulation (Real−Time, Trailing 7 Days) ConvRt EpiNow2 estimateR EpiEstim Ground Truth [PITH_FULL_IMAGE:figures/full_fig_p044_10.png]
Figure 11
Figure 11. Figure 11: Real-time Rt estimates on wiggly influenza benchmark, with ConvRt’s 95% point￾wise confidence bands. φb ≈ 2.4 for 2022/23 and 2.7 for 2023/24. A Poisson observation model leaves 15–20% of days outside its 95% band around the fitted trend; φ = 2.5 leaves about 5%, matc…
Figure 12
Figure 12. Figure 12: Rt estimates on the smooth flu simulation. ConvRt’s 95% pointwise interval shaded. has lower MAE than EpiNow2 on this dataset, and considerably lower CE. The remaining four methods sit well behind [PITH_FULL_IMAGE:figures/full_fig_p045_12.png]
Figure 13
Figure 13. Figure 13: ConvRt predictions by smoothness parameter [PITH_FULL_IMAGE:figures/full_fig_p046_13.png]
Figure 14
Figure 14. Figure 14: Retrospective ConvRt Rt for the 2022/23 and 2023/24 flu seasons, with 90% pointwise (dark) and simultaneous (light) confidence bands. Pointwise intervals cover each timestep marginally at the nominal level; simultaneous bands cover the entire curve jointly. 0.6 0.7 0.…
Figure 15
Figure 15. Figure 15: Real-time flu Rt , ConvRt vs EpiNow2. (a) 2022/23: last-7-day nowcasts, con￾textualized by end-of-season fits. (b,c) 2023/24: weekly real-time vintages for EpiNow2 and ConvRt. 47 [PITH_FULL_IMAGE:figures/full_fig_p047_15.png]
Figure 16
Figure 16. Figure 16: Real-time Rt predictions by hyperparameter values on 2023/24 flu season. Top row shows predictions across γ, for λ tuned by the min-rule and 1se-rule. Bottom row shows all last-7-day nowcasts. sharply, except when the data exhibits a longstanding trend. This causes er…
Figure 17
Figure 17. Figure 17: Forward-validation error curves (Poisson deviance) for taper hyperparameter [PITH_FULL_IMAGE:figures/full_fig_p049_17.png]
Figure 18
Figure 18. Figure 18: Real-time ConvRt nowcasts with a constant-tail constraint (B-spline basis, [PITH_FULL_IMAGE:figures/full_fig_p050_18.png]
Figure 19
Figure 19. Figure 19: COVID-19 Rt estimates from ConvRt and CovidEstim on four major U.S.counties. rored by the infection counts in [PITH_FULL_IMAGE:figures/full_fig_p051_19.png]
Figure 20
Figure 20. Figure 20: Cases and seroprevalence-based infection counts for pre-Omicron COVID-19 in [PITH_FULL_IMAGE:figures/full_fig_p052_20.png]
Figure 21
Figure 21. Figure 21: California Rt curve, CovidEstim and ConvRt. 52 [PITH_FULL_IMAGE:figures/full_fig_p052_21.png]
Figure 22
Figure 22. Figure 22: ConvRt’s predictions of COVID-19 Rt in Los Angeles under various delay distri￾butions. 0.7 0.8 0.9 1.0 1.1 1.2 1.3 Oct Nov Dec Jan Feb Rt Daily Weekly Truth (a) Retrospective 0.9 1.0 1.1 Dec 17 Dec 24 Dec 31 Jan 07 Jan 14 Jan 21 Jan 28 γ 10−2 100 102 104 Retrospective…
Figure 23
Figure 23. Figure 23: ConvRt predictions given data at a weekly resolution, in retrospect and in real [PITH_FULL_IMAGE:figures/full_fig_p053_23.png]
Figure 24
Figure 24. Figure 24: ConvRt predictions on NHSN flu data, modeled as a cubic spline and trend filter. [PITH_FULL_IMAGE:figures/full_fig_p055_24.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages

  1. [1]

    Thompson, Katharine Sherratt, Hamish P

    Sam Abbott, Joel Hellewell, Robin N. Thompson, Katharine Sherratt, Hamish P. Gibbs, Nikos I. Bosse, James D. Munday, Sophie Meakin, Emma L. Doughty, June Young Chun, Yung-Wai Desmond Chan, Flavio Finger, Paul Campbell, Akira Endo, Carl A. B. Pearson, Amy Gimma, Tim Russell, Stefan Flasche, Adam J. Kucharski, Rosalind M. Eggo, and Sebastian Funk. Estimatin...

    work page

  2. [2]

    doi: 10.12688/wellcomeopenres.16006.2. 20

    work page doi:10.12688/wellcomeopenres.16006.2

  3. [3]

    Anderson and Robert M

    Roy M. Anderson and Robert M. May.Infectious Diseases of Humans: Dynamics and Control. Oxford University Press, Oxford, 1991

    work page 1991

  4. [4]

    Angelopoulos, Emmanuel J

    Anastasios N. Angelopoulos, Emmanuel J. Candès, and Ryan J. Tibshirani. Conformal PID control for time series prediction. InAdvances in Neural Information Processing Systems, volume 36, 2023

    work page 2023

  5. [5]

    A simple model for COVID-19.Infectious Disease Modelling, 5:309–315, 2020

    Julien Arino and Stéphanie Portet. A simple model for COVID-19.Infectious Disease Modelling, 5:309–315, 2020. doi: 10.1016/j.idm.2020.04.002

    work page doi:10.1016/j.idm.2020.04.002 2020

  6. [6]

    Estimates of the reproduction number for seasonal, pandemic, and zoonotic influenza: a systematic review of the literature.BMC Infectious Diseases, 14(1):480, 2014

    Matthew Biggerstaff, Simon Cauchemez, Carrie Reed, Manoj Gambhir, and Lyn Finelli. Estimates of the reproduction number for seasonal, pandemic, and zoonotic influenza: a systematic review of the literature.BMC Infectious Diseases, 14(1):480, 2014. doi: 10.1186/1471-2334-14-480

    work page doi:10.1186/1471-2334-14-480 2014

  7. [7]

    Bouman, Anthony Hauser, Simon L

    Judith A. Bouman, Anthony Hauser, Simon L. Grimm, Martin Wohlfender, Samir Bhatt, Elizaveta Semenova, Andrew Gelman, Christian L. Althaus, Julien Riou, and Virginia E. Pitzer. Bayesian workflow for time-varying transmission in stratified com- partmental infectious disease transmission models.PLOS Computational Biology, 20 (4):e1011575, 2024. doi: 10.1371/...

    work page doi:10.1371/journal.pcbi.1011575 2024

  8. [8]

    Donnelly, Carrie Reed, Azra C

    Simon Cauchemez, Christl A. Donnelly, Carrie Reed, Azra C. Ghani, Christophe Fraser, Charlotte K. Kent, Lyn Finelli, and Neil M. Ferguson. Household transmission of 2009 pandemic influenza A (H1N1) virus in the United States.New England Journal of Medicine, 361(27):2619–2627, 2009. doi: 10.1056/NEJMoa0905498

    work page doi:10.1056/nejmoa0905498 2009

Show all 54 references
  1. [9]

    Current epidemic trends (based onR t) for states, 2026

    Centers for Disease Control and Prevention, Center for Forecasting and Outbreak Analytics. Current epidemic trends (based onR t) for states, 2026. URLhttps: //www.cdc.gov/cfa-modeling-and-forecasting/rt-estimates/index.html. Ac- cessed April 13, 2026

    2026

  2. [10]

    Morris, Melissa S

    Louis Yat Hin Chan, Sinead E. Morris, Melissa S. Stockwell, Natalie M. Bowman, Ed- win Asturias, Suchitra Rao, Karen Lutrick, Katherine D. Ellingson, Huong Q. Nguyen, Yvonne Maldonado, Son H. McLaren, Ellen Sano, Jessica E. Biddle, Sarah E. Smith- Jeffcoat, Matthew Biggerstaff...

    2025 doi

  3. [11]

    Melanie H Chitwood, Marcus Russi, Kenneth Gunasekera, Joshua Havumaki, Fayette Klaassen, Virginia E Pitzer, Joshua A Salomon, Nicole A Swartwood, Joshua L Warren, Daniel M Weinberger, Ted Cohen, and Nicolas A Menzies. Reconstructing the course of the COVID-19 epidemic over 202...

    2020 doi

  4. [12]

    Kristie E. N. Clarke, Jefferson M. Jones, Yangyang Deng, Elise Nycz, Adam Lee, Ronaldo Iachan, Adi V. Gundlapalli, Aron J. Hall, and Adam MacNeil. Seropreva- lence of infection-induced SARS-CoV-2 antibodies — United States, september 2021– february 2022.MMWR. Morbidity and Mor...

    2021 doi

  5. [13]

    Ferguson, Christophe Fraser, and Simon Cauchemez

    Anne Cori, Neil M. Ferguson, Christophe Fraser, and Simon Cauchemez. A new frame- work and software to estimate time-varying reproduction numbers during epidemics. American Journal of Epidemiology, 178(9):1505–1512, Nov 2013. doi: 10.1093/aje/ kwt133

    2013 doi

  6. [14]

    Covidestim methodology update – summer 2021.https://blog

    Covidestim Team. Covidestim methodology update – summer 2021.https://blog. covidestim.org/posts/summer-2021-changes/, August 2021. Accessed: 2026-04-24

    2021

  7. [15]

    Cowling, Vicky J

    Benjamin J. Cowling, Vicky J. Fang, Steven Riley, J. S. Malik Peiris, and Gabriel M. Leung. Estimation of the serial interval of influenza.Epidemiology, 20(3):344–347, 2009. doi: 10.1097/EDE.0b013e31819d1092

    2009 doi

  8. [16]

    Gao, Daniela Witten, and Jacob Bien

    Ameer Dharamshi, Anna Neufeld, Keshav Motwani, Lucy L. Gao, Daniela Witten, and Jacob Bien. Generalized data thinning using sufficient statistics.Journal of the American Statistical Association, 120(549):511–523, 2025. doi: 10.1080/01621459.2024. 2353948

    2025 doi

  9. [17]

    Diekmann, J

    O. Diekmann, J. A. P. Heesterbeek, and J. A. J. Metz. On the definition and the computation of the basic reproduction ratioR 0 in models for infectious diseases in heterogeneous populations.Journal of Mathematical Biology, 28(4):365–382, 1990. doi: 10.1007/BF00178324

    1990 doi

  10. [18]

    Tibshirani

    Tiffany Ding, Isaac Gibbs, and Ryan J. Tibshirani. Calibrated multi-level quantile forecasting, 2026. URLhttps://arxiv.org/abs/2512.23671

    2026

  11. [19]

    Dublin and Alfred J

    Louis I. Dublin and Alfred J. Lotka. On the true rate of natural increase: As exemplified by the population of the United States, 1920.Journal of the American Statistical Association, 20(151):305–339, 1925. doi: 10.1080/01621459.1925.10503498

    1920 doi

  12. [20]

    Figgins and Trevor Bedford

    Marlin D. Figgins and Trevor Bedford. Inferring variant-specific effective reproduction numbers from combined case and sequencing data.eLife, 14:RP104802, 2025. doi: 10.7554/eLife.104802

    2025 doi

  13. [21]

    Estimating individual and household reproduction numbers in an emerging epidemic.PLOS One, 2(8):1–12, 08 2007

    Christophe Fraser. Estimating individual and household reproduction numbers in an emerging epidemic.PLOS One, 2(8):1–12, 08 2007. doi: 10.1371/journal.pone.0000758. URLhttps://doi.org/10.1371/journal.pone.0000758

    2007 doi

  14. [22]

    Estimating the generation interval for coronavirus disease (COVID-19) based on symptom onset data, March 2020.Eurosurveillance, 25(17): 2000257, 2020

    Tapiwa Ganyani, Cécile Kremer, Dongxuan Chen, Andrea Torneri, Christel Faes, Jacco Wallinga, and Niel Hens. Estimating the generation interval for coronavirus disease (COVID-19) based on symptom onset data, March 2020.Eurosurveillance, 25(17): 2000257, 2020. doi: 10.2807/1560-...

    2020 doi

  15. [23]

    Hu, Alyssa Bilinski, Daniel J

    Jeremy Goldwasser, Addison J. Hu, Alyssa Bilinski, Daniel J. McDonald, and Ryan J. Tibshirani. Estimating time-varying epidemic severity rates with adaptive deconvolu- tion.arXiv preprint arXiv:2510.16180, 2025

    2025

  16. [24]

    Tibshirani, and Alyssa Bilinski

    Jeremy Goldwasser, Ryan J. Tibshirani, and Alyssa Bilinski. On the equivalence of instantaneous and mechanistic reproduction numbers, 2026. URLhttps://arxiv. org/abs/2606.15397

    2026

  17. [25]

    Gostic, Lauren McGough, Edward B

    Katelyn M. Gostic, Lauren McGough, Edward B. Baskerville, Sam Abbott, Keya Joshi, Christine Tedijanto, Rebecca Kahn, Rene Niehus, James A. Hay, Pablo M. De Salazar, Joel Hellewell, Sophie Meakin, James D. Munday, Nikos I. Bosse, Katharine Sherratt, RobinN.Thompson, LauraF.Whit...

    2020 doi

  18. [26]

    Althaus, Niel Hens, and Christel Faes

    Oswaldo Gressani, Jacco Wallinga, Christian L. Althaus, Niel Hens, and Christel Faes. EpiLPS: A fast and flexible Bayesian tool for estimation of the time-varying reproduction number.PLOS Computational Biology, 18(10):e1010618, 2022. doi: 10.1371/journal.pcbi.1010618

    2022 doi

  19. [27]

    Hart, Elizabeth Miller, Nick J

    William S. Hart, Elizabeth Miller, Nick J. Andrews, Pauline Waight, Philip K. Maini, Sebastian Funk, and Robin N. Thompson. Generation time of the alpha and delta SARS-CoV-2 variants: an epidemiological analysis.The Lancet Infectious Diseases, 22 (5):603–610, 2022. doi: 10.101...

    2022 doi

  20. [28]

    Lau, Peng Wu, Xilong Deng, Jian Wang, Xinxin Hao, Yiu Chung Lau, Jessica Y

    Xi He, Eric H.Y. Lau, Peng Wu, Xilong Deng, Jian Wang, Xinxin Hao, Yiu Chung Lau, Jessica Y. Wong, Yujuan Guan, Xinghua Tan, et al. Temporal dynamics in viral shedding and transmissibility of COVID-19.Nature Medicine, 26(5):672–675, 2020. doi: 10.1038/s41591-020-0869-5

    2020 doi

  21. [29]

    J. A. P. Heesterbeek. A brief history ofR 0 and a recipe for its calculation.Acta Biotheoretica, 50(3):189–204, 2002. doi: 10.1023/A:1016599411804

    2002 doi

  22. [30]

    Tibshirani

    Maria Jahja, Andrew Chin, and Ryan J. Tibshirani. Real-time estimation of covid-19 infections: Deconvolution and sensor fusion.Statistical Science, 37(2):207–228, May

  23. [31]

    doi: 10.1214/22-STS856

    doi

  24. [32]

    William Ogilvy Kermack and A. G. McKendrick. A contribution to the mathematical theory of epidemics.Proceedings of the Royal Society of London. Series A, 115(772): 700–721, 1927. doi: 10.1098/rspa.1927.0118

    1927 doi

  25. [33]

    Kucharski, Timothy W

    Adam J. Kucharski, Timothy W. Russell, Charlie Diamond, Yang Liu, John Edmunds, Sebastian Funk, and Rosalind M. Eggo. Early dynamics of transmission and control of COVID-19: a mathematical modelling study.The Lancet Infectious Diseases, 20(5): 553–558, 2020. doi: 10.1016/S1473...

    2020 doi

  26. [34]

    Data fission: Splitting a single data point.Journal of the American Statistical Association, 120(549): 135–146, 2025

    James Leiner, Boyan Duan, Larry Wasserman, and Aaditya Ramdas. Data fission: Splitting a single data point.Journal of the American Statistical Association, 120(549): 135–146, 2025. doi: 10.1080/01621459.2023.2270748

    2025 doi

  27. [35]

    Reconstructing COVID-19 incidences from positive RT-PCR tests by deconvolution.BMC Infectious Diseases, 23:679, 2023

    Mengtian Li, Jiachen Li, Ke Wang, and Lei M Li. Reconstructing COVID-19 incidences from positive RT-PCR tests by deconvolution.BMC Infectious Diseases, 23:679, 2023. doi: 10.1186/s12879-023-08667-1

    2023 doi

  28. [36]

    Leung, Eric H.Y

    Qun Li, Xuhua Guan, Peng Wu, Xiaoye Wang, Lei Zhou, Yeqing Tong, Ruiqi Ren, Kathy S.M. Leung, Eric H.Y. Lau, Jessica Y. Wong, et al. Early transmission dynamics in Wuhan, China, of novel coronavirus–infected pneumonia.New England Journal of Medicine, 382(13):1199–1207, 2020. d...

    2020 doi

  29. [37]

    McDonald

    Jiaping Liu, Zhenglun Cai, Paul Gustafson, and Daniel J. McDonald. Time-varying reproduction number estimation with trend filtering.PLOS Computational Biology, 20 (8):e1012324, August 2024

    2024

  30. [38]

    Lopez, Sarah Kada, Pragati V

    Velma K. Lopez, Sarah Kada, Pragati V. Prasad, Taylor Chin, Betsy L. Cadwell, Jes- sica M. Healy, Rachel B. Slayton, Matthew Biggerstaff, and Michael A. Johansson. COVID-19 mitigation behaviors and policies limited SARS-CoV-2 transmission in the United States from September 20...

    2020 doi

  31. [39]

    The analysis of equilibrium in malaria.Tropical Diseases Bulletin, 49(9):813–829, 1952

    George Macdonald. The analysis of equilibrium in malaria.Tropical Diseases Bulletin, 49(9):813–829, 1952

    1952

  32. [40]

    McGough, Michael A

    Sarah F. McGough, Michael A. Johansson, Marc Lipsitch, and Nicolas A. Men- zies. Nowcasting by Bayesian smoothing: A flexible, generalizable model for real- time epidemic tracking.PLOS Computational Biology, 16(4):e1007735, 2020. doi: 10.1371/journal.pcbi.1007735

    2020 doi

  33. [41]

    Real-timeestimationoftheepidemic reproduction number: Scoping review of the applications and challenges.PLOS Digital Health, 1(6):e0000052, 2022

    RebeccaK.Nash, PierreNouvellet, andAnneCori. Real-timeestimationoftheepidemic reproduction number: Scoping review of the applications and challenges.PLOS Digital Health, 1(6):e0000052, 2022. doi: 10.1371/journal.pdig.0000052

    2022 doi

  34. [42]

    J. Y. Noh, J. Y. Song, S. Y. Hwang, W. S. Choi, J. Y. Heo, H. J. Cheong, and W. J. Kim. Viral load dynamics in adult patients with A(H1N1)pdm09 influenza.Epidemiology and Infection, 142(4):753–758, 2014. doi: 10.1017/S0950268813001817

    2014 doi

  35. [43]

    Chaves, Pam Daily Kirley, Rachel Ember, Erin Arber, Emily B

    Carrie Reed, Sandra S. Chaves, Pam Daily Kirley, Rachel Ember, Erin Arber, Emily B. Hancock, Joan Baumbach, Nancy M. Bennett, Ken Gershman, James Meek, Kimberly Yousey-Hindes, Patricia Ryan, Malini DeSilva, Monica M. Farley, Carla Barber, Shikha Garg, and Alicia M. Fry. Estima...

    2015

  36. [44]

    Danielle Iuliano, Neha N

    Heather Reese, A. Danielle Iuliano, Neha N. Patel, Shikha Garg, Lindsay Kim, Ben- jamin J. Silk, Aron J. Hall, Alicia Fry, and Carrie Reed. Estimated incidence of 24 coronavirus disease 2019 (COVID-19) illness and hospitalization — United States, February–September 2020.Clinic...

    2019 doi

  37. [45]

    Gourgoulianis, and Erasmia Rouka

    Eleni Rousogianni, Garyfallia Perlepe, Stylianos Boutlas, Georgia Rapti, Evdoxia Gouta, Eleni Mpaltopoulou, Giorgos Mpaltopoulos, Dimitrios Papagiannis, Aristome- nis Exadaktylos, Konstantinos I. Gourgoulianis, and Erasmia Rouka. Clinical features and outcomes of viral respira...

    2023 doi

  38. [46]

    estimater: an r package to estimate and monitor the effective reproductive number.BMC Bioinformatics, 24 (1):310, 2023

    Jérémie Scire, Jana S Huisman, Ana Grosu, Daniel C Angst, Adrian Lison, Jinzhou Li, Marloes H Maathuis, Sebastian Bonhoeffer, and Tanja Stadler. estimater: an r package to estimate and monitor the effective reproductive number.BMC Bioinformatics, 24 (1):310, 2023. doi: 10.1186...

    2023 doi

  39. [47]

    Different epidemic curves for severe acute respiratory syndromerevealsimilarimpactsofcontrolmeasures.American Journal of Epidemiology, 160(6):509–516, 09 2004

    Jacco Wallinga and Peter Teunis. Different epidemic curves for severe acute respiratory syndromerevealsimilarimpactsofcontrolmeasures.American Journal of Epidemiology, 160(6):509–516, 09 2004. ISSN 0002-9262. doi: 10.1093/aje/kwh255. URLhttps: //doi.org/10.1093/aje/kwh255. A O...

    2004 doi

  40. [48]

    The observed countsyt are conditionally independent and Poisson with meanΛt = Z⊤ t θ

  41. [49]

    The delay distributions are known; and

  42. [50]

    non_parametric_si

    The smoothing parameterλis held fixed. We perform inference with a standard approach: the sandwich variance for a penalized M-estimator (or quasi-likelihood). Without loss of generality, our notation assumes no day- of-week effects or under-reporting. 29 B.1 Confidence interva...

    2000

  43. [51]

    EpiEstim sliding window

    = 0to match EpiEstim’s indexing convention).Rt is estimated on a 7-day sliding window (t_start = 2:(n-6),t_end = t_start + 6); the gamma prior onR t has mean 1 and SD 5, the package defaults. Per the caveat in the main text, EpiEstim is fit directly to observed reports rather ...

  44. [52]

    Piecewise exponential 3

    Piecewise constant 2. Piecewise exponential 3. Piecewise linear 4. Periodic 0.5 1.0 1.5 2.0 2.5 True Rt 0 150 300 0 150 300 0 150 300 0 150 300 101 102 103 104 105 106 Time (days) Daily count True daily infections Observed reported cases Figure 8: rtestim benchmark datasets:Rt...

  45. [53]

    Periodic

    Piecewise linear 4. Periodic

  46. [54]

    smooth” influenza simulation combines two modifications of its “wiggly

    Piecewise constant 2. Piecewise exponential 50 150 250 50 150 250 0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5 2.0 2.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0 2.5 Time (days) Rt Truth ConvRt ConvRt (jumps) EpiNow2 EpiEstim estimateR Figure 9: RetrospectiveR t estimates on the rtestim benchmarks, compa...

Pith tools

Reviewed July 8, 2026 · model on record in the stance chip above.