Bootstrap Inference for SIR Model Parameters

Computes bootstrap standard errors and confidence intervals for SIR model parameters. This is the recommended approach for inference when the Hessian is singular or ill-conditioned, which is common in models with bilinear influence terms (i.e., when fix_receiver = FALSE).

Usage

boot_sir(
  sir_fit,
  R = 200,
  type = c("block", "parametric"),
  seed = NULL,
  trace = FALSE
)

Arguments

sir_fit: A fitted sir object from sir.
R: Integer. Number of bootstrap replicates. Default is 200. Increase to 500-1000 for publication-quality intervals.
type: Character. Bootstrap type: "block" (default) resamples time periods with replacement; "parametric" simulates new outcomes from the fitted model.
seed: Optional integer for reproducibility. Sets the random seed before resampling.
trace: Logical. If TRUE, prints progress every 10 replicates.

Value

An object of class "boot_sir" with components:

coefs: R x n_params matrix of bootstrap coefficient estimates. Rows for failed replicates contain NA.
se: Named numeric vector of bootstrap standard errors (one per parameter).
ci_lo: Lower 2.5% percentile bounds.
ci_hi: Upper 97.5% percentile bounds.
point_est: Point estimates from the original fit.
param_names: Character vector of parameter names.
n_valid: Number of successful bootstrap replicates.
n_total: Total number of replicates attempted.
type: The bootstrap type used.
family: The distribution family.

Details

Two bootstrap strategies are available:

block: Resamples time periods with replacement. Preserves the within-period dependence structure. Best when T is moderately large (T >= 10).
parametric: Simulates new outcome arrays from the fitted model using the estimated parameters and the specified family distribution. Better when T is small but the model is well-specified.

Each replicate refits the full SIR model. Replicates that fail to converge are dropped and reported. Standard errors are column standard deviations of the successful replicates. Confidence intervals use the percentile method.

Examples

if (FALSE) { # \dontrun{
model <- sir(Y, W, X, family = "poisson")

# block bootstrap with 200 replicates
boot_result <- boot_sir(model, R = 200, seed = 42)
print(boot_result)

# use bootstrap CIs with confint
confint(model, boot = boot_result)

# parametric bootstrap
boot_par <- boot_sir(model, R = 200, type = "parametric")
} # }