code: Contrast Code Factor

In bhagwataditya/autonomics: Unified Statistical Modeling of Omics Data

Contrast Code Factor

Description

Contrast Code Factor for General Linear Model

Usage

code(object, ...)

## S3 method for class 'factor'
code(object, codingfun, verbose = TRUE, ...)

## S3 method for class 'data.table'
code(object, codingfun, vars = names(object), verbose = TRUE, ...)

contr.treatment.explicit(n)

code_control(n)

contr.diff(n)

code_diff(n)

code_diff_forward(n)

code_deviation(n)

code_deviation_first(n)

code_helmert(n)

code_helmert_forward(n)

Arguments

object	factor vector
...	used for s3 dispatch
codingfun	factor coding function contr.treatment: intercept = y0, coefi = yi - y0 contr.treatment.explicit: intercept = y0, coefi = yi - y0 code_control: intercept = ymean, coefi = yi - y0 contr.diff: intercept = y0, coefi = yi - y(i-1) code_diff: intercept = ymean, coefi = yi - y(i-1) code_diff_forward: intercept = ymean, coefi = yi - y(i+) code_deviation: intercept = ymean, coefi = yi - ymean (drop last) code_deviation_first: intercept = ymean, coefi = yi - ymean (drop first) code_helmert: intercept = ymean, coefi = yi - mean(y0:(yi-1)) code_helmert_forward: intercept = ymean, coefi = yi - mean(y(i+1):yp)
verbose	TRUE or FALSE
vars	svars
n	character vector

Details

A General Linear Model contains:
* An Intercept Coefficient: expressing some form of sample average
* For each numeric variable: a slope coefficient
* For each k-leveled factor: (k-1) Contrast Coefficients.
The interpretation of (intercept and contrast) coefficients depends on the contrast coding function used. Several contrast coding functions are available in 'stats' and 'codingMatrices' But their (function and coefficient) namings are a bit confusing and unsystematic. Instead, the functions below offer an intuitive interface (to the otherwise powerful stats/codingMatrices packages). The names of these functions reflect the contrast coding used (treatment, backward, sum, or helmert contrasts). They also reflect the intercept interpretation (either first factor's first level or grand mean). They all produce intuitive coefficient names (e.g. 't1-t0' rather than just 't1'). They all have unit scaling (a coefficient of 1 means a backward of 1).

Value

(explicitly coded) factor vector

Examples

# Coding functions
    x <- factor(paste0('t', 0:3))
    xlevels <- levels(x)
    contr.treatment(         xlevels)
    contr.treatment.explicit(xlevels)
    contr.diff(              xlevels)
    code_control(            xlevels)
    code_diff(               xlevels)
    code_diff_forward(       xlevels)
    code_deviation(          xlevels)
    code_deviation_first(    xlevels)
    code_helmert(            xlevels)
    code_helmert_forward(    xlevels)

# Code
    x %<>% code(contr.treatment)
    x %<>% code(contr.treatment.explicit)
    x %<>% code(contr.diff)
    x %<>% code(code_control)
    x %<>% code(code_diff)
    x %<>% code(code_diff_forward)
    x %<>% code(code_deviation)
    x %<>% code(code_deviation_first)
    x %<>% code(code_helmert)
    x %<>% code(code_helmert_forward)

# Model
    file <- system.file('extdata/atkin.metabolon.xlsx', package = 'autonomics')
    object <- read_metabolon(file)
    object %<>% fit_limma(codingfun = contr.treatment) # default
    object %<>% fit_limma(codingfun = contr.treatment.explicit)
    object %<>% fit_limma(codingfun = contr.diff)
    object %<>% fit_limma(codingfun = code_control)
    object %<>% fit_limma(codingfun = code_diff)
    object %<>% fit_limma(codingfun = code_diff_forward)
    object %<>% fit_limma(codingfun = code_deviation)
    object %<>% fit_limma(codingfun = code_deviation_first)
    object %<>% fit_limma(codingfun = code_helmert)
    object %<>% fit_limma(codingfun = code_helmert_forward)

bhagwataditya/autonomics documentation built on Dec. 1, 2024, 12:27 a.m.