Nowcast right-truncated surveillance data

The most recent weeks of surveillance data are almost always incomplete because of reporting delays (right truncation). get_ncast uses baselinenowcast to estimate what those counts will look like once all reports arrive.

Usage

get_ncast(df, max_delay = 4, draws = 1000, prop_delay = 0.5, scale_factor = 3)

Arguments

df

An accidda_data object from check_data or get_data. Must have revision history ($history == TRUE); use get_data(revisions = TRUE).

max_delay

Integer. Maximum reporting delay in weeks. Default 4.

draws

Integer. Number of posterior samples. Default 1000.

prop_delay

Numeric 0-1. Proportion of reference times used for delay estimation. Default 0.5.

scale_factor

Numeric. Multiplicative factor on the maximum delay for the estimation window. Default 3.

Value

An accidda_ncast object (a list) with:

data

Corrected time series. The observation column contains the nowcast median for the corrected weeks. Two extra columns, ncast_lower and ncast_upper (95\ used by get_fcast to propagate nowcast uncertainty.

plot

ggplot2 visualisation of the nowcast correction.

Details

With the default max_delay = 4, the last 4 weeks are treated as right-truncated and replaced by nowcast estimates. Everything before that is left untouched.

The function returns three corrected versions of the full series (nowcast median, lower, and upper 95\ get_fcast can propagate nowcast uncertainty into the final forward-looking forecast.

Examples

if (FALSE) { # \dontrun{
df    <- get_data(pathogen = "covid", geo_value = "ca", revisions = TRUE)
ncast <- get_ncast(df)
fcast <- get_fcast(ncast, eval_start_date = "2025-01-01")
} # }