define_transform: Transform parameter estimates before conducting statistical...

In maxwestphal/SEPM: Statistical Evaluation of Prediction Models

Transform parameter estimates before conducting statistical inference

Description

Define a transform which shall be used for a statistical test, see infer.

Usage

define_transform(transform, link, inv, se.t)

Arguments

transform	character, name of the transform.
link	function, which maps the raw estimate to the transformed estimate.
inv	function, inverse of link: inv = link^{-1}: y -> x
se.t	function, standard error of the transformed quantitiy (delta method). se.t needs to have formal input arguments se, n, estimate (even if the result does not depend on them).

Details

If transform corresponds to a known SEPM.transform object, this will object be returned and all other input arguments are ignored. define_transform() will print a list of known transforms.

The use of a transform (other than "none") may help with making the test statistic(s) more "gaussian". In addition, it can prevent confidence bounds outside the parameter space. This a realistic issue in binomial models, e.g. when working with proportions like accuracy or classification error (ce). In this case the implemented transforms "logit" and "asin.sqrt" enforce confidence intervals in [0,1].

The output of the function se.t can be computed analytically with help of the Delta method, see Delta Method (Wikipedia).

Value

A SEPM.transform which may be passed to infer.

maxwestphal/SEPM documentation built on April 12, 2024, 12:09 a.m.