A simulated dataset derived from Gostic, Katelyn M., et al. "Practical
considerations for measuring the effective reproductive number, Rt."
PLoS Computational Biology 16.12 (2020): e1008409. The data are simulated
from a stochastic SEIR compartmental model. The original timeseries and Rt
are available in the incidence and true_rt columns and with additional
columns added or modified to increase noise and add a day-of-week effect.
Format
stochastic_sir_rt A data frame with 299 rows and 6 columns:
- reference_date
The date cases were observed.
- true_rt
The known, true Rt of the epidemic system.
- dow
The magnitude of the day-of-week effect added to the log of
incidence.- true_cases
The true number of cases occurring on the
datein the simulated system before observation noise or day-of-week effects.- obs_cases
The observed number of cases on
dateafter the day-of-week reporting effect and observation noise have been applied totrue_cases.- obs_cases_no_dow
The observed number of cases on
dateafter observation noise has been applied totrue_cases. It does not include a day of week effect.
Source
https://github.com/cobeylab/Rt_estimation/tree/d9d8977ba8492ac1a3b8287d2f470b313bfb9f1d https://github.com/CDCgov/cfa-epinow2-pipeline/pull/91 https://github.com/CDCgov/cfa-epinow2-pipeline/pull/17
Details
This synthetic dataset has a number of desirable properties:
The force of infection changes depending on the Rt, allowing for sudden changes in the Rt. This allows for modeling of sudden changes in infection dynamics, which might otherwise be difficult to capture.
The realized Rt is known at each timepoint
The dataset incorporates a simple generation interval and a reporting delay.
Gostic et al. benchmark the performance of a number of Rt estimation frameworks, providing practical guidance on how to use this dataset to evaluate Rt estimates.
In practice, we've found that the amount of observation noise is often undesirably low for testing. Many empirical datasets are much noisier. As a result, models built with these realistically noisy settings in mind can perform poorly on this dataset or fail to converge. To better reflect realistic settings, we manually add observation noise and a day-of-week reporting effect.