EpiEstim model for fable

Estimates the current reproduction number (Rt) using EpiEstim and simulates future incidence using projections. Works with any regular aggregation period and integrates into the fable workflow via model().

Usage

EPIESTIM(
  formula,
  mean_si,
  std_si,
  rt_window = 14L,
  n_sim = 100L,
  R_fix_within = TRUE
)

Arguments

formula

Response variable, e.g. observation. Exogenous regressors are not supported.

mean_si

Mean serial interval in days.

std_si

Standard deviation of the serial interval in days.

rt_window

Sliding window width in days for Rt estimation (dt_out in EpiEstim). Controls how many recent days inform the current Rt. Smaller values track recent trends more closely; larger values give a smoother estimate. Defaults to 14 days (~2 weeks).

n_sim

Number of simulation paths for the forecast distribution.

R_fix_within

If TRUE, Rt is held constant within each simulated path (recommended for short horizons).

Value

A model definition for use inside model().

Details

The aggregation period is detected automatically from the tsibble index. Only the most recent data (the rt_window estimation period plus the serial interval period) is passed to EpiEstim's expectation-maximisation (EM) algorithm, which reconstructs daily incidence from the aggregated counts before estimating Rt on rolling rt_window-day windows. The most recent Rt estimate is then used to project forward via stochastic simulation. Forecasts are returned as sample distributions (one per horizon period) drawn from n_sim epidemic paths.