standardize-package: standardize: Tools for Standardizing Variables for Regression...
In standardize: Tools for Standardizing Variables for Regression in R

Description Details Author(s)

The standardize package provides tools for standardizing variables prior to regression (i.e. placing all of the variables to be used in a regression on similar scales). When all of the predictors in a regression are on a similar scale, it makes the interpretation of their effect sizes more comparable. In the case of gaussian regression, placing the response on unit scale also eases interpretation. Standardizing regression variables also has computational benefits in the case of mixed effects regressions, and makes determining reasonable priors in Bayesian regressions simpler. To view the package vignette, call vignette("using-standardize", package = "standardize"). To see the version history, call standardize.news().

The named_contr_sum function gives named sum contrasts to unordered factors, and allows the absolute value of the non-zero cells in contrast matrix to be specified through its scale argument. The scaled_contr_poly function gives orthogonal polynomial contrasts to ordered factors, and allows the standard deviation of the columns in the contrast matrix to be specified through its scale argument. The scale_by function allows numeric variables to be scaled conditioning on factors, such that the numeric variable has the same mean and standard deviation within each level of a factor (or the interaction of several factors), with the standard deviation specified through its scale argument.

The standardize function creates a standardized object whose elements can be used in regression fitting functions, ensuring that all of the predictors are on the same scale. This is done by passing the function's scale argument to named_contr_sum for all unordered factors (and also any predictor with only two unique values regardless of its original class), to scaled_contr_poly for all ordered factors, and to scale_by for numeric variables which contain calls to the function. For numeric predictors not contained in a scale_by call, scale is called, ensuring that the result has standard deviation equal to the scale argument to standardize. Gaussian responses are always placed on unit scale, using scale (or scale_by if the function was used on the left hand side of the regression formula). Offsets for gaussian models are divided by the standard deviation of the raw response (within-factor-level if scale_by is used on the response).