lmer_pi function

Prediction intervals for future observations based on linear random effects models (DEPRECATED)

This function is deprecated. Please use lmer_pi_unstruc(), lmer_pi_futvec() or lmer_pi_futmat().

lmer_pi(
  model,
  newdat = NULL,
  m = NULL,
  alternative = "both",
  alpha = 0.05,
  nboot = 10000,
  lambda_min = 0.01,
  lambda_max = 10,
  traceplot = TRUE,
  n_bisec = 30
)

Arguments

• model: a random effects model of class "lmerMod"
• newdat: a data.frame with the same column names as the historical data on which the model depends
• m: number of future observations
• alternative: either "both", "upper" or "lower". alternative specifies if a prediction interval or an upper or a lower prediction limit should be computed
• alpha: defines the level of confidence (1-alpha)
• nboot: number of bootstraps
• lambda_min: lower start value for bisection
• lambda_max: upper start value for bisection
• traceplot: if TRUE: plot for visualization of the bisection process
• n_bisec: maximal number of bisection steps

Returns

If newdat is specified: A data.frame that contains the future data, the historical mean (hist_mean), the calibrated coefficient (quant_calib), the prediction standard error (pred_se), the prediction interval (lower and upper) and a statement if the prediction interval covers the future observation (cover).

If m is specified: A data.frame that contains the number of future observations (m) the historical mean (hist_mean), the calibrated coefficient (quant_calib), the prediction standard error (pred_se) and the prediction interval (lower and upper).

If alternative is set to "lower": Lower prediction limits are computed instead of a prediction interval.

If alternative is set to "upper": Upper prediction limits are computed instead of a prediction interval.

If traceplot=TRUE, a graphical overview about the bisection process is given.

Details

This function returns a bootstrap calibrated prediction interval

with $\hat{y}$ as the predicted future observation, $y$ as the observed future observations, $\sqrt{\hat{var}(\hat{y} - y)}$

as the prediction standard error and $q$ as the bootstrap calibrated coefficient that approximates a quantile of the multivariate t-distribution.

Please note that this function relies on linear random effects models that are fitted with lmer() from the lme4 package. Random effects have to be specified as (1|random_effect).

Examples

# This function is deprecated.
# Please use lmer_pi_unstruc() if you want exactly the same functionality.
# Please use lmer_pi_futmat() or lmer_pi_futvec() if you want to take care
# of the future experimental design