Parametric bootstrap confidence intervals for the SIRD-GA MLE

Two bootstrap schemes are supported. method = "gillespie" simulates B independent epidemic trajectories and surveillance records from the fitted parameters via the exact jump-process simulator (sird_ga_simulate_obs()) and re-estimates the model on each replicate — appropriate when extinction risk is low. method = "residual" resamples the standardised one-step-ahead Kalman residuals of the fitted model with replacement, reconstructs bootstrap observation series around the fitted predictive mean, and re-estimates the model on each — appropriate for real-data settings near the epidemic threshold, where re-simulating the jump process can produce early extinction.

Usage

sird_ga_bootstrap(
  fit,
  y,
  N,
  Delta = 1,
  method = c("gillespie", "residual"),
  B = 200L,
  level = 0.95,
  S0 = NULL,
  I0 = NULL,
  R0 = NULL
)

Arguments

fit

Object returned by sird_ga_fit().

y

Original observed series used to produce fit.

N, Delta

Passed through to sird_ga_loglik() / the simulator.

method

"gillespie" or "residual".

B

Number of bootstrap replicates.

level

Confidence level.

S0, I0, R0

Initial head counts, required for method = "gillespie" (derived from fit$theta and N if omitted).

Value

A list with boot_theta (B x p matrix of bootstrap estimates) and ci (matrix of percentile confidence limits, one row per parameter).