baselinenowcast (version 0.2.0)

estimate_uncertainty: Estimate uncertainty parameters

Description

This function ingests a list of point nowcast matrices and a corresponding list of truncated reporting matrices and uses both to estimate a vector of uncertainty parameters from the observations and estimates at each horizon, starting at 0 up until the max delay number of horizons.

Usage

estimate_uncertainty(
  point_nowcast_matrices,
  truncated_reporting_triangles,
  retro_reporting_triangles,
  n = length(point_nowcast_matrices),
  uncertainty_model = fit_by_horizon,
  ref_time_aggregator = identity,
  delay_aggregator = function(x) rowSums(x, na.rm = TRUE),
  validate = TRUE
)

Value

uncertainty_params Vector of length of the number of horizons, with each element representing the estimate of the uncertainty parameter for each horizon. The specific parameter type depends on the chosen error model.

Arguments

point_nowcast_matrices

List of point nowcast matrices where rows represent reference time points and columns represent delays.

truncated_reporting_triangles

List of truncated reporting matrices, containing all observations as of the latest reference time. Elements of list are paired with elements of point_nowcast_matrices.

retro_reporting_triangles

List of n truncated reporting triangle matrices with as many rows as available given the truncation.

n

Integer indicating the number of reporting matrices to use to estimate the uncertainty parameters.

uncertainty_model

Function that ingests a matrix of observations and a matrix of predictions and returns a vector that can be used to apply uncertainty using the same error model. Default is fit_by_horizon with arguments of obs matrix of observations and pred the matrix of predictions that fits each column (horizon) to a negative binomial observation model by default. The user can specify a different fitting model by replacing the fit_model argument in fit_by_horizon.

ref_time_aggregator

Function that operates along the rows (reference times) of the retrospective point nowcast matrix before it has been aggregated across columns (delays). Default is identity which does not aggregate across reference times.

delay_aggregator

Function that operates along the columns (delays) of the retrospective point nowcast matrix after it has been aggregated across reference times. Default is function(x) rowSums(x, na.rm = TRUE).

validate

Logical. If TRUE (default), validates the object. Set to FALSE only when called from functions that already validated.

See Also

Observation error estimation functions fit_by_horizon(), fit_nb()

Examples

# Use example data to create reporting triangle
data_as_of_df <- syn_nssp_df[syn_nssp_df$report_date <= "2026-04-01", ]
rep_tri <- as_reporting_triangle(data = data_as_of_df)

# Create retrospective nowcasts
trunc_rts <- truncate_to_rows(rep_tri, n = 2)
retro_rts <- apply_reporting_structures(trunc_rts)
retro_nowcasts <- estimate_and_apply_delays(retro_rts)

# Estimate dispersion parameters using default negative binomial model
disp_params <- estimate_uncertainty(
  point_nowcast_matrices = retro_nowcasts,
  truncated_reporting_triangles = trunc_rts,
  retro_reporting_triangles = retro_rts
)
disp_params

# Estimate dispersion parameters from rolling sum on the reference times
if (requireNamespace("zoo", quietly = TRUE)) {
  disp_params_agg <- estimate_uncertainty(
    point_nowcast_matrices = retro_nowcasts,
    truncated_reporting_triangles = trunc_rts,
    retro_reporting_triangles = retro_rts,
    ref_time_aggregator = function(x) zoo::rollsum(x, k = 2, align = "right")
  )
  disp_params_agg
}