AIC.evmOpt: Information Criteria

View source: R/AIC.evm.R

AIC.evmOptR Documentation

Information Criteria

Description

Compute AIC and (approximate) DIC for evmOpt objects

Usage

## S3 method for class 'evmOpt'
AIC(object, penalized = FALSE, nsamp = 1000, DIC, WAIC, ..., k = 2)

Arguments

object

fit model object

penalized

whether to use the penalized log-likelihood

nsamp

Number of approximate Gaussian sample to use in computing DIC. Defaults to nsamp=1e3. Only used when the object has class 'evmOpt'.

DIC

Logical. Whether to compute DIC. Defaults to DIC = TRUE. Only applicable to objects of class 'evmSim'.

WAIC

Logical. Whether to compute WAIC. Defaults to WAIC = TRUE. Only applicable to objects of class 'evmSim'.

...

other arguments currently ignored

k

numeric, the penalty per parameter to be used; the default k = 2 is the classical AIC.

Details

If the object has class 'evmOpt', nsamp random draws are made from the Gaussian distribution with mean and covariance inferred from the model object. The result will be an approximate DIC. Note that AIC should not be trusted if priors are not flat. For example, if you use a regularizing prior on xi, say xi ~ N(0, 0.25), AIC can be misleading and DIC should be preferred. If the object has class 'evmSim', the actual posterior draws are used in the computation. Also note that sometimes the optimizer returns an approximatae covariance that is not postive-semidefinite, in which case the DIC will be reported as NA.

Value

The AIC and DIC

See Also

AIC


texmex documentation built on June 22, 2024, 12:26 p.m.