Predictive performance — bayes_R2.stan

Calculate Bayesian-\(R^2\), LOO-\(R^2\), and leave-one-out predictive metrics for a regression model. Currently this is only calculated for IPD regression models. Any models with both IPD and AgD (e.g. ML-NMR) will ignore the AgD part for this calculation.

Usage

# S3 method for class 'stan_nma'
bayes_R2(object, ..., probs = c(0.025, 0.5, 0.975), summary = TRUE)

# S3 method for class 'stan_nma'
loo_R2(object, ..., probs = c(0.025, 0.5, 0.975), summary = TRUE)

# S3 method for class 'stan_nma'
loo_predictive_metric(x, metric, ...)

Arguments

object, x: A stan_nma object.
...: Not used.
probs: Numeric vector of quantiles of interest to present in computed summary, default c(0.025, 0.25, 0.5, 0.75, 0.975)
summary: Logical, calculate posterior summaries? Default TRUE.
metric: The type of predictive metric to be calculated. See loo::loo_predictive_metric(). The default is root mean squared error ("rmse") for continuous or poisson outcomes, and true positive prediction rate ("acc") for binary outcomes.

Value

For bayes_R2() and loo_R2(): a nma_summary object if summary = TRUE, otherwise a 3D MCMC array of samples of the Bayesian or LOO \(R^2\). For loo_predictive_metric(): a list with the estimate and standard error.