pscore (version 0.4.0)

mahalanobisComposite: Score Data Using the Mahalanobis Distance

Description

Create a composite using the Mahalanobis Distance

Usage

mahalanobisComposite(object, ncomponents, pca)

Arguments

object

An object of class CompositeReady

ncomponents

the number of components to use from the principal component analysis. If missing, defaults to the number of columns in the data.

pca

An optional PCA object from princomp to use. If not passed, will be calculated from the data.

Value

An S4 object of class MahalanobisScores.

See Also

Other composite: factorComposite(), sumComposite()

Examples

# NOT RUN {
# this example creates distances for the built in mtcars data
# see ?mtcars for more details
# The distances are calculated from the "best" in the dataset
# First we create an appropriate CompositeData class object
# higher mpg & hp are better and lower wt & qsec are better
d <- CompositeData(mtcars[, c("mpg", "hp", "wt", "qsec")],
  thresholds = list(one = with(mtcars, c(
    mpg = max(mpg),
    hp = max(hp),
    wt = min(wt),
    qsec = min(qsec)))
  ),
  higherisbetter = c(TRUE, TRUE, FALSE, FALSE))

# create the distance scores
# and the composite
# covariance matrix will be calculated from the data
# and data will be standardized to unit variance by default
dres <- prepareComposite(d)

# see a density plot of the distance scores
dres@distanceDensity
# regular summary of distance scores
summary(dres@distances)

# examine covariance matrix
round(dres@covmat,2)

# now we can create the composite based on mahalanobis distances
# from our defined thresholds
mcomp <- mahalanobisComposite(dres, 1)

# view a histogram of the composite scores
mcomp@scoreHistogram

# summarize the composite scores
summary(mcomp@scores)

# check the screeplot and loadings
mcomp@screePlot
mcomp@loadingGraph
# examine the loadings as a table
mcomp@loadingTable

# one component is adequate to explain these data
# to be safe can pick first two and re-run model

# use only first two components
mcomp2 <- mahalanobisComposite(dres, ncomponents = 2)

# view a histogram of the updated composite scores
mcomp2@scoreHistogram

# summarize the composite scores
summary(mcomp2@scores)

# compare using all versus two components
plot(mcomp@scores, mcomp2@scores)

# cleanup
rm(d, dres, mcomp, mcomp2)
# }