Registrations for a scientific conference with a deadline accumulate slowly at first and then surge as the deadline approaches, so a linear extrapolation of early data badly underestimates the final attendance. Alfi, Parisi and Pietronero (Nature Physics 3, 746, 2007; companion paper in Central European Journal of Physics 7, 483, 2009) proposed that the pressure to register is inversely proportional to the remaining time, giving a cumulative curve N(t)=Cln⁡(T∗/(T∗−t))N(t)=C\ln(T^*/(T^*-t)), a separate model for fee payments (an irreversible commitment) that multiplies this rate by an exponential postponement factor, and a rule of thumb that the final total is roughly three times a linear extrapolation of the early data. On the basis of only two physics conferences they suggested that the dynamics may be universal. The open problem is to determine the quantitative law of human response to deadlines and its domain of validity: whether the inverse-time law (or one of the saturating forms later proposed by Flandrin and by Fenner, Levene and Loizou) survives systematic multi-event validation; whether simple prediction rules such as the threefold-multiplier heuristic are accurate; whether the payment model with a characteristic postponement time holds beyond the original credit-card data; under what conditions the pool of potential registrants is effectively unlimited; and whether a microscopic, agent-based mechanism can derive the aggregate curves. Later work has enlarged the data and produced competing functional forms and partial mechanistic models, but none of these validations exists, so the problem remains open.

Background

The quantitative study of human dynamics treats the timing of human actions — emails, library loans, purchases, form submissions — as a stochastic point process and searches for statistical regularities. A landmark observation of Barabási (Nature 435, 207, 2005) was that such timing is typically bursty: bursts of rapid activity separated by long idleness, with heavy-tailed waiting times. He explained this with a priority queuing model, in which an agent keeps a list of tasks with random priorities and executes the highest-priority task with some probability (otherwise a random one), so that low-priority tasks wait a long time; this queuing framework became the standard microscopic model family of the field.

A natural testbed for deadline-driven dynamics is conference registration. The Statphys conferences (the International Conference on Statistical Physics, triennial) and the EP2DS conferences (International Conference on Electronic Properties of Two-Dimensional Systems) are ordinary scientific meetings: a registration website opens months ahead, registrations accumulate, and a registration/abstract deadline T∗T^* produces a last-minute rush. Organizers need the final attendance number early — for venue, housing and catering planning — but early registrations grow only roughly linearly, and a linear extrapolation to the deadline grossly underestimates the total.

Alfi, Parisi and Pietronero analyzed the registration and fee-payment records of Statphys 23 (Genova, July 2007) and EP2DS 17 and proposed, first, a registration model in which the pressure to register is inversely proportional to the time left before the deadline: the per-capita registration rate is p(t)=C/(T∗−t)p(t)=C/(T^*-t), so that the cumulative number of registrations grows as

N(t)=Cln⁡T∗T∗−t,N(t)=C\ln\frac{T^*}{T^*-t},

with the constant CC fixed by the total number of participants and the logarithmic divergence at the deadline regularized by the one-day discreteness of registrations. Second, for payments — an irreversible commitment of money, which they call strong engagement — the rush toward the deadline is even sharper, and they multiply the rate by an exponential postponement factor exp⁡(−(T∗−t)/τ)\exp(-(T^*-t)/\tau) with a characteristic postponement time τ\tau (reported as about 19.5 days for the Statphys 23 credit-card payments), introduced by analogy with a Boltzmann factor. Third, they offered a rule of thumb: the final number of registrants is roughly three times a linear extrapolation of the initial data. Fourth, the good fit without any saturation term implies that the pool N′N' of people who might consider registering is much larger than the final total NtotN_{\text{tot}}, so that the factor (N′−N(t))(N'-N(t)) is effectively constant. Finally, the similarity of the two conferences' rescaled curves suggested a universal behavior, and they pointed to Barabási-type microscopic models as the possible mechanistic origin.

Later work modified this picture without settling it. Flandrin (Advances in Complex Systems 13, 439, 2010) revisited the question of the model's universality with a larger dataset of electronic conference submissions and proposed a new model that better describes the complete evolution of the process, including its saturation near the total, and supports running prediction of the total number of submissions. Fenner, Levene and Loizou (Central European Journal of Physics 11, 2013) fitted six years of weekly registration counts of a conference with an early-bird deadline (a discounted-fee deadline preceding the final one) with a bi-logistic function — the sum of two logistic growth curves, an S-shaped saturing form — reflecting two distinct surges. On the mechanistic side, Jo (Journal of the Korean Physical Society 79, 407, 2021) studied a deadline-concerning priority queuing model in which a task's priority is adjusted by the inverse of the remaining time to its deadline, and Zhang and Ma (Scientific Reports 14, 2024) gave real-world evidence that temporal discounting — the behavioral tendency to devalue rewards that arrive later — predicts procrastination. Deadline-driven activity acceleration is also documented in financial closing auctions, social-media posting around events, and project schedules, but these adjacent domains have not been connected back to a validated law for registration and payment dynamics.

Problem Statement

Determine the quantitative law of human response to a deadline and its domain of validity, replacing the two-conference evidence base of Alfi, Parisi and Pietronero with systematic validation across diverse events, deadline structures and commitment types:

(1) Functional form and universality. Does the inverse-time registration law — per-capita rate p(t)=C/(T∗−t)p(t)=C/(T^*-t) and cumulative N(t)=Cln⁡(T∗/(T∗−t))N(t)=C\ln(T^*/(T^*-t)) up to the discreteness-regularized singularity at the deadline T∗T^* — hold for events other than the two original physics conferences (for example, other disciplines, event types, sizes, regions, and deadline structures), or do the dynamics instead follow the saturating forms reported in later work (Flandrin's submission model; the bi-logistic form appropriate to early-bird fee structures)? Is there a universal scaling after rescaling by the deadline and the total, and on which structural features (single versus multiple deadlines, fee incentives, irreversible versus revocable commitments) does the functional form depend?

(2) Prediction rule. Is there a validated rule for predicting the final number of registrants from early registration data — in particular, does the rule that the final total is approximately three times a linear extrapolation of the initial linear stretch hold for other meetings with quantified accuracy — and what is the operational definition of the extrapolation window that makes the rule reproducible, with that window and the accuracy criterion fixed before the data are evaluated (or calibrated on data held out from the evaluation set)?

(3) Payment dynamics. Does the payment model — inverse-time pressure multiplied by the reported exponential postponement factor exp⁡(−(T∗−t)/τ)\exp(-(T^*-t)/\tau) — describe irreversible monetary commitments in datasets beyond the original credit-card subset (other conferences, other payment channels, larger samples), and is the characteristic postponement time τ\tau a stable, transferable parameter?

(4) Pool saturation. Under what conditions is the pool N′N' of potential registrants much larger than the final total NtotN_{\text{tot}}, the regime in which the saturation factor (N′−N(t))(N'-N(t)) is effectively constant and absorbable into the rate constant, and how must the model be generalized when this fails, for example for small or capacity-limited events?

(5) Mechanism. Can a microscopic model — an agent-based or priority-queue model in the spirit of Barabási's human-dynamics framework, or a decision-theoretic mechanism such as temporal discounting — derive the aggregate registration and payment curves from individual behavior rather than merely fit them, and does it remain consistent with the validated phenomenology?

These questions jointly serve one objective, a validated, predictive quantitative description of how people react to a deadline. As framed by the original papers, the core open question is whether the inverse-time registration model, the strong-engagement payment model with its deadline postponement, and the early-data prediction rule hold across multiple independent events and deadline structures beyond the two original conferences: whether early registration data permit accurate prediction of the final number of registrants, and whether the models' features are universal, as against the alternative functional forms reported in later work (Flandrin's saturating submission model; the bi-logistic form appropriate to early-bird fee structures). Comparison against the alternative forms and appropriate statistical scrutiny are part of such a validation, and no particular statistical protocol is required by the question; the mechanistic account of question (5) is a separate explanatory direction that need not be settled for the empirical claims. The validation covers at least three independent events beyond the two original conferences and spans more than one deadline structure and commitment type, jointly addressing sub-questions (1) to (4) — functional form and universality, the prediction rule, payment dynamics, and pool saturation — with the event sample fixed before any model comparison; within (1), a refutation of the inverse-time law settles the functional-form question only when it also determines which candidate form the evidence supports. The empirical evidence base otherwise remains open: the datasets range over events with a single- or multiple-deadline structure and, where the payment model is tested, an irreversible monetary-commitment series, with no particular dataset required by the question.

References

  1. V. Alfi, G. Parisi, L. Pietronero, 'Conference registration: how people react to a deadline', Nature Physics 3(11), 746 (2007), DOI 10.1038/nphys761, https://doi.org/10.1038/nphys761
  2. V. Alfi, A. Gabrielli, L. Pietronero, 'How people react to a deadline: time distribution of conference registrations and fee payments', Central European Journal of Physics 7(3), 483-489 (2009), DOI 10.2478/s11534-009-0059-z, https://doi.org/10.2478/s11534-009-0059-z
  3. Albert-László Barabási, 'The origin of bursts and heavy tails in human dynamics', Nature 435(7039), 207-211 (2005), DOI 10.1038/nature03459, https://doi.org/10.1038/nature03459
  4. P. Flandrin, 'An empirical model for electronic submissions to conferences', Advances in Complex Systems 13(3), 439-449 (2010), DOI 10.1142/S0219525910002554, https://doi.org/10.1142/S0219525910002554
  5. T. Fenner, M. Levene, G. Loizou, 'A bi-logistic growth model for conference registration with an early bird deadline', Central European Journal of Physics 11(7) (2013), DOI 10.2478/s11534-013-0275-4, https://doi.org/10.2478/s11534-013-0275-4
  6. H.-H. Jo, 'Numerical study on the deadline-concerning priority queuing model', Journal of the Korean Physical Society 79(4), 407-411 (2021), DOI 10.1007/s40042-021-00219-7, arXiv:2105.14718, https://arxiv.org/abs/2105.14718
  7. L. Chen, H. Zheng, W. Wei, H. Pan, W. Nie, Z. Hao, Z. Zhao, 'Exploring Temporal Asymmetry in Human Behavior in Social Media Platforms', Entropy 28(1), 27 (2025), DOI 10.3390/e28010027, https://doi.org/10.3390/e28010027
  8. P. Y. Zhang, W. J. Ma, 'Temporal discounting predicts procrastination in the real world', Scientific Reports 14, 14642 (2024), DOI 10.1038/s41598-024-65110-4, https://doi.org/10.1038/s41598-024-65110-4
  9. D. Terasawa, A. Fukuda, K. Fukushima, 'Correlation between exam score and quiz submission day in an introductory physics course', Physics Education 61(2), 025026 (2026), DOI 10.1088/1361-6552/ae4935, https://doi.org/10.1088/1361-6552/ae4935
  10. M. Salek, D. Challet, I. Muni Toke, 'Equity auction dynamics: latent liquidity models with activity acceleration', Quantitative Finance 24(10), 1381-1398 (2024), DOI 10.1080/14697688.2024.2367680, arXiv:2401.06724, https://arxiv.org/abs/2401.06724
  11. A. Vazquez, C. Marasinou, G. Kalogridis, C. Ellinas, 'Activity delay patterns in project networks', Physica A: Statistical Mechanics and its Applications 637, 129570 (2024), DOI 10.1016/j.physa.2024.129570, arXiv:2304.10304, https://arxiv.org/abs/2304.10304
  12. T. Fujiyama, C. Matsui, A. Takemura, 'A Power-Law Growth and Decay Model with Autocorrelation for Posting Data to Social Networking Services', PLOS ONE 11(8), e0160592 (2016), DOI 10.1371/journal.pone.0160592, https://doi.org/10.1371/journal.pone.0160592
  13. R. Shibue, A. Takemura, 'An Extended Power Model for Point Processes in a Finite Interval and its Application to Bid Arrival Processes of Online Auctions', Japanese Journal of Applied Statistics 44(3), 89-100 (2015), DOI 10.5023/JAPPSTAT.44.89, https://doi.org/10.5023/JAPPSTAT.44.89
  14. C. Castellano, S. Fortunato, V. Loreto, 'Statistical physics of social dynamics', Reviews of Modern Physics 81(2), 591-646 (2009), DOI 10.1103/RevModPhys.81.591, arXiv:0710.3256, https://arxiv.org/abs/0710.3256
  15. M. Karsai, H.-H. Jo, K. Kaski, 'Bursty Human Dynamics', Springer (2018), DOI 10.1007/978-3-319-68540-3, arXiv:1803.02580, https://arxiv.org/abs/1803.02580