shipunov (version 1.7)

MDSv: MDS: dimension importance ("explained variance" surrogate)

Description

Calculates R-squared coefficients of the linear relationships between each of derived variables and original data

Usage

MDSv(scores)

Arguments

scores

Data frame or matrix with values (e.g., result of 'isoMDS()'

Value

Numeric vector, one values per column of scores

Details

MDSv() converts each of the derived variables and original data into distance matrices, and then uses lm() to calculate adjusted R-squared coefficients. These coefficients may be used to understand the "importance" of each new dimension. They work for any dimension reduction techique including multidimensional scaling.

Examples

# NOT RUN {
iris.dist <- dist(iris[, -5], method="manhattan")

library(MASS)
iris.m <- isoMDS(iris.dist + 1e-9) # isoMDS() does not like zero distances
(vv <- MDSv(iris.m$points)) # MDS explained variance surrogate
xlab <- paste0("Dim 1 (", round(vv[1], 2), "%)")
ylab <- paste0("Dim 2 (", round(vv[2], 2), "%)")
plot(iris.m$points, col=as.numeric(iris$Species), xlab=xlab, ylab=ylab)

iris.cmd <- cmdscale(iris.dist)
MDSv(iris.cmd)

iris.p <- prcomp(iris[, -5], scale=TRUE)
MDSv(iris.p$x)
100*summary(iris.p)$importance[2, ] # compare with MDSv() results

# }