blmeco (version 1.4)

loo.cv: Bayesian leave-one-out cross-validation

Description

Bayesian leave-one-out cross-validation based on the log pointwise predictive density

Usage

loo.cv(mod, nsim = 100, bias.corr = FALSE)

Arguments

mod

an object obtained by the functions lm or glm

nsim

number of Monte Carlo simulations used to describe the posterior distributions. Computing time is large!

bias.corr

The leave-one-out cross-validation underestimates predictive fit because each prediction is conditioned n-1 data points. For large n this bias is negligible. For small n, a bias correction is recommended.

Value

LOO.CV

leave-one-out cross-validation estimate of out-of-sample predictive fit, (log pointwise predictive density)

bias.corrected.LOO.CV

bias corrected leave-one-out cross-validation estimate of out-of-sample predictive fit, (log pointwise predictive density)

minus2times_lppd

-2*LOO.CV, transformed LOO.CV to scale of deviance

est.peff

estimate for the number of effective parameters

Details

For details see Gelman et al. (2014) p 175

References

Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A and Rubin DB (2014) Bayesian Data Analysis, Third edn. CRC Press.

See Also

ocv

Examples

# NOT RUN {
x <- runif(20)
y <- 2+0.5*x+rnorm(20, 0, 1)
mod <- lm(y~x)
loo.cv(mod, bias.corr=TRUE)  # increase nsim!!
  
# }
# NOT RUN {
  
# }