RotateLoadings: 'RotateLoadings'

In Displayr/flipDimensionReduction: Algorithms for dimension reduction

RotateLoadings

Description

Rotate the component loadings from a principal components analysis or factor analysis.

Usage

RotateLoadings(
  loadings,
  rotation = "varimax",
  delta = 0,
  kappa = 4,
  covar = FALSE,
  stds = NULL
)

Arguments

loadings	A matrix containing the un-rotated loadings from a PCA or factor analysis.
rotation	A string specifying the kind of rotation to be conducted. The allowed values are: "varimax", "quartimax", "equamax", "oblimin", and "promax".
delta	The parameter supplied to oblimin.
kappa	The exponent supplied to promax.
covar	Whether or not the loadings were generated by analysing a covariance matrix rather than a correlation matrix. This is only relevant if using a promax rotation, as an extra scaling step is required in order to match the outputs from SPSS.
stds	If a covariance matrix was used (covar == TRUE) then the standard deviations of the original input variables must be supplied. Again, this is required for a scaling step in promax.

Details

This function utilizes the rotation options from GPArotation to match SPSS' component rotations as closely as possible. As SPSS uses different rotation algorithms, the match is not exact in all cases, but testing shows that the rotated loadings are effectively the same. Differences will be most noticable for Oblimin rotations, as the GPA package does a much better job at minimizing the oblimin function compared with SPSS, and for larger loading matrices where the solution space is larger.

A Kaiser normalization step is always included, to match what SPSS does.

The promax rotation, which is done algebraicly rather than by optimizing an objective function, uses our own function, flipPromax, as the promax functions from stats and psych do not match the SPSS outputs.

Displayr/flipDimensionReduction documentation built on Aug. 19, 2024, 9:29 p.m.