Learn R Programming
apricom (version 1.0.0)

compdist: Comparison Distribution Descriptives

Description

calculate a comparison distribution and associated descriptive statisics.

Usage

compdist(a, b, model, output, lambda1, lambda2, trim, strat1, strat2)

Arguments

a
a vector containing the sum of square errors or -2 * log likelihood estimates derived using modelling strategy 1.
b
a vector containing the sum of square errors or -2 * log likelihood estimates derived using modelling strategy 2.
model
type of regression model. Either "linear" or "logistic".
output
logical. If output is TRUE the function will return two graphical representations of the comparison distribution.
lambda1
a vector or matrix containing the estimated shrinkage factors derived using strategy 1
lambda2
a vector or matrix containing the estimated shrinkage factors derived using strategy 2
trim
logical. If trim is TRUE a "trimmed" comparison distribution will be returned, along with a victory rate and median precision ratio derived using the trimmed distribution. The trimmed distribution only contains precision ratios within a range of plus or minus two times the interquartile range around the median precision ratio.
strat1
a list containing the strategy name and strategy-specific parameter values.
strat2
a list containing the strategy name and strategy-specific parameter values.

Details

This function works within compare to calculate the comparison distribution of strategy 2 against strategy 1. If trim == TRUE, a truncated distribution consisting of +/- 2 * interquartile range around the median precision ratio.

The victory rate for linear regression is the left tail probability with a cut-off of 1. The victory rate for logistic regression instead taken at a cut-off of 0. For more details, refer to the accompanying PDF document.

If output == TRUE the function will plot a histogram of the distribution and a scatterplot of the precision (SSE or -2 log likelihood) of strategy 2 against strategy 1. Points that lie below the red or blue line represent trials in which strategy 2 outperformed strategy 1 (the victory rate).