Confirmation / retraction process

Describes reports that are later retracted (count-cumulative streams that revise downward as well as upward). Attach it to a model() via the confirmation argument; nowcast() then uses the signed-increment Skellam / SkNB likelihood when the data are count-cumulative.

Usage

confirmation_process(retract_delay = lognormal_delay(), p = numeric(0))

Arguments

retract_delay

A delay_process_class (e.g. lognormal_delay(), gamma_delay()) describing the retraction delay g_C (onset of a report to its retraction). Default a geometric-like short lognormal.

p

The confirmation probability (probability that a report is genuine and never retracted), specified the same way as a delay parameter: either a prior_class (estimated with that prior) or a single numeric in (0, 1] (held fixed). Left unset (the default), default_priors() builds a data-informed, strongly-concentrated Beta prior centred at the empirical retraction rate – exactly as the lognormal delay's mu default is data-informed. The strong prior is deliberate: retractions are empirically rare (~1-2% of reports), and with a weak prior the Skellam variance abuses the retraction stream as an overdispersion knob and p collapses. Pass a fixed value (e.g. p = 0.98) to hold it constant, or your own beta_prior() to estimate it under a prior of your choosing.

Value

A confirmation_process_class object.

Details

Default priors. retract_delay inherits the default priors of its delay family (see delay_process); p gets a strongly-concentrated Beta centred at the empirical retraction rate (see default_priors()). At p = 1 the confirmation layer is inert and the model is the ordinary count model.

See also

model(), delay_process, nowcast()

Examples

# A confirmation process with a lognormal retraction delay
confirmation_process(retract_delay = lognormal_delay())
#> <diseasenowcasting::confirmation_process_class>
#>  @ retract_delay: <diseasenowcasting::lognormal_delay_class>
#>  .. @ name               : chr "LogNormal"
#>  .. @ num_id             : int 1
#>  .. @ num_delay_seasons  : int 1
#>  .. @ season_distribution: <diseasenowcasting::prior_class>
#>  .. .. @ name       : chr "StdNormal"
#>  .. .. @ num_id     : int 0
#>  .. .. @ stan_params: num(0) 
#>  .. @ mu                 : num(0) 
#>  .. @ sigma              : num(0) 
#>  @ p            : num(0) 
#>  @ active       : logi TRUE

# Attach to a model for a count-cumulative stream
model(nb_likelihood(), ar1_epidemic(), lognormal_delay(),
      confirmation = confirmation_process())
#> 
#> ── Bayesian Nowcast Model ──────────────────────────────────────────────────────
#> 
#> ── Likelihood 
#> NegBin(mu, phi ~ LogNormal(2.996, 0.500))
#> 
#> ── Epidemic process 
#> AR(1)(phi, sigma | error)
#> 
#> ── Delay process 
#> LogNormal(mu, sigma)
#> 
#> ── Covariate prior 
#> StdNormal()
#> Strata pooling: "independent"
#> ────────────────────────────────────────────────────────────────────────────────