baselinenowcast (version 0.2.0)

estimate_uncertainty_retro: Estimate uncertainty parameters using retrospective nowcasts

Description

Estimates uncertainty parameters for nowcasting by creating a series of retrospective datasets from the input reporting triangle, generating point nowcasts for those datasets, and calibrating uncertainty parameters based on retrospective nowcast performance.

This function chains the retrospective nowcasting workflow:

1. truncate_to_rows() - Create retrospective snapshots

2. apply_reporting_structures() - Generate retrospective reporting triangles

3. estimate_and_apply_delays() - Generate point nowcasts

4. estimate_uncertainty() - Estimate uncertainty parameters

For full probabilistic nowcasts (uncertainty estimation + sampling), use estimate_and_apply_uncertainty().

For more control over individual steps (e.g., custom matrix preparation, alternative aggregation), use the low-level functions directly.

Usage

estimate_uncertainty_retro(
  reporting_triangle,
  n_history_delay,
  n_retrospective_nowcasts,
  structure = get_reporting_structure(reporting_triangle),
  delay_pmf = NULL,
  validate = TRUE,
  ...
)

Value

A numeric vector of uncertainty parameters with length equal to one less than the number of columns in the reporting triangle, with each element representing the estimate of the uncertainty parameter for each horizon. Returns NULL if insufficient data is available for estimation.

Arguments

reporting_triangle

A reporting_triangle object with rows representing reference times and columns representing delays. Can be a reporting matrix or incomplete reporting matrix. Can also be a ragged reporting triangle, where multiple columns are reported for the same row (e.g., weekly reporting of daily data).

n_history_delay

Integer indicating the number of reference times (observations) to be used in the estimate of the reporting delay, always starting from the most recent reporting delay.

n_retrospective_nowcasts

Integer indicating the number of retrospective nowcast times to use for uncertainty estimation.

structure

Integer or vector specifying the reporting structure. If integer, divides columns evenly by that integer (with last possibly truncated). If vector, the sum must not be greater than or equal to the number of columns. Default is 1 (standard triangular structure).

delay_pmf

Vector or list of vectors of delays assumed to be indexed starting at the first delay column in each of the matrices in retro_reporting_triangles. If a list, must be of the same length as retro_reporting_triangles, with elements aligning. Default is NULL.

validate

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

...

Additional arguments passed to estimate_uncertainty().

See Also

High-level workflow wrapper functions allocate_reference_times(), estimate_and_apply_delay(), estimate_and_apply_delays(), estimate_and_apply_uncertainty()

Examples

# Create a reporting triangle from syn_nssp_df
data_as_of <- syn_nssp_df[syn_nssp_df$report_date <= "2026-04-01", ]
rep_tri <- as_reporting_triangle(data_as_of) |>
  truncate_to_delay(max_delay = 25)

uncertainty_params <- estimate_uncertainty_retro(
  rep_tri,
  n_history_delay = 30,
  n_retrospective_nowcasts = 10
)
uncertainty_params