max_prob_pred_int: Maximum probability for the prediciton interval 'small sample...

In femiguez/predintma: Prediction Intervals for Meta-Analysis

Description

Finds the maximum probability for a prediction interval using optimization

Usage

max_prob_pred_int(
  x,
  method = c("tdist", "conformal", "npar"),
  m.method = c("quantile", "deviation", "jackknife"),
  interval = c(0, 1),
  alpha.penalty = 0,
  scale = FALSE
)

Arguments

x	should be a vector
method	either 'tdist' (assumes normality), 'conformal' (distribution-free), or non-parametric ('npar')
m.method	method used to compute conformal prediction interval: either "quantile", "deviation" or "jackknife"
interval	maximum and minimum values for the optimization search
alpha.penalty	whether to include a penalty for alpha
scale	whether to scale the input vector

Details

Optimize the maximum probability for the prediction of a small sample size using either the t-distribution or conformal methods

The idea is to find the maximum level of probability that will produce a prediction interval which matches the minimum and maximum values in the observed sample. The distance is calculated as abs(x.min - calc.lower.bound) + abs(x.max - calc.upper.bound).

Value

a single value which represents a 'suggestion' for the maximum level of probability given the data

Examples

## Not run: 
set.seed(12345)
x <- rnorm(10)
mp.pdi <- max_prob_pred_int(x) 
## For this particular sample, the maximum is about 76%
data(soyrs)
soyrs.s <- aggregate(lrr ~ Trial_ID, data = soyrs, FUN = mean)
mp.pdi.soy.tdist <- max_prob_pred_int(soyrs.s$lrr)
mp.pdi.soy.conf <- max_prob_pred_int(soyrs.s$lrr, method = "conformal")
print(round(c(mp.pdi.soy.tdist, mp.pdi.soy.conf1),2))

## End(Not run)

femiguez/predintma documentation built on July 5, 2021, 4:16 a.m.