Skip to contents

Model outliers via an extreme value distribution

Source: R/model_sampling.R
outliers_estimate.Rd

This option models outliers in sampled concentrations using a generalized extreme value distribution.

Usage

outliers_estimate(
  gev_prior_mu = 0.0025,
  gev_prior_sigma = 0.005,
  gev_prior_xi = 2,
  modeldata = modeldata_init()
)

Arguments

gev_prior_mu

Prior (location) of the GEV distribution.

gev_prior_sigma

Prior (scale) of the GEV distribution.

gev_prior_xi

Prior (shape) of the GEV distribution.

modeldata

A modeldata object to which the above model specifications should be added. Default is an empty model given by modeldata_init(). Can also be an already partly specified model returned by other EpiSewer modeling functions.

Value

A modeldata object containing data and specifications of the model to be fitted. Can be passed on to other EpiSewer modeling functions to add further data and model specifications.

The modeldata object also includes information about parameter initialization (init), meta data (.metainfo), and checks to be performed before model fitting (.checks).

Details

EpiSewer can automatically identify independent spikes in the concentration time series and model them as outliers to reduce the impact on transmission dynamic estimates. Moreover, when using this modeling option, a summary of potential outlier observations will be produced after model fitting (see summary$outliers).

Naturally, the modeling of outliers works better retrospectively than in real-time. EpiSewer will produce a warning if it thinks that the most recent observations contain one or more outliers. In that case, it might make sense to manually remove the outlier observation or to wait until more data is available.

Outliers are modeled as independent spikes in the load on a given sampling day. The spikes follow a generalized extreme value distribution (GEV) that is scaled by the average load per case. The default prior for the GEV is chosen such that distribution of spikes is extremely right-tailed, and the 95% quantile of spikes is below the load equivalent of 1 case, i.e. has minimal impact on estimated infection dynamics. Note that because of the properties of the GEV, the mean posterior estimates of loads and concentrations will be theoretically infinite if this modeling option is used. The median remains be well-behaved.

See also

Other outlier models: outliers_none()