robCompositions (version 2.4.1)

lmCoDaX: Classical and robust regression of non-compositional (real) response on compositional and non-compositional predictors

Description

Delivers appropriate inference for regression of y on a compositional matrix X or and compositional and non-compositional combined predictors.

Usage

lmCoDaX(
  y,
  X,
  external = NULL,
  method = "robust",
  pivot_norm = "orthonormal",
  max_refinement_steps = 200
)

Value

An object of class ‘lts’ or ‘lm’ and two summary objects.

Arguments

y

The response which should be non-compositional

X

The compositional and/or non-compositional predictors as a matrix, data.frame or numeric vector

external

Specify the columns name of the external variables. The name has to be introduced as follows: external = c("variable_name"). Multiple selection is supported for the external variable. Factor variables are automatically detected.

method

If robust, LTS-regression is applied, while with method equals “classical”, the conventional least squares regression is applied.

pivot_norm

if FALSE then the normalizing constant is not used, if TRUE sqrt((D-i)/(D-i+1)) is used (default). The user can also specify a self-defined constant.

max_refinement_steps

(for the fast-S algorithm): maximal number of refinement steps for the fully iterated best candidates.

Author

Peter Filzmoser, Roman Wiedemeier, Matthias Templ

Details

Compositional explanatory variables should not be directly used in a linear regression model because any inference statistic can become misleading. While various approaches for this problem were proposed, here an approach based on the pivot coordinates is used. Further these compositional explanatory variables can be supplemented with external non-compositional data and factor variables.

References

Filzmoser, P., Hron, K., Thompsonc, K. (2012) Linear regression with compositional explanatory variables. Journal of Applied Statistics, 39, 1115-1128.

See Also

lm

Examples

## How the total household expenditures in EU Member
## States depend on relative contributions of 
## single household expenditures:
data(expendituresEU)
y <- as.numeric(apply(expendituresEU,1,sum))
lmCoDaX(y, expendituresEU, method="classical")

## How the relative content of sand of the agricultural
## and grazing land soils in Germany depend on
## relative contributions of the main chemical trace elements,
## their different soil types and the Annual mean temperature:
data("gemas")
gemas$COUNTRY <- as.factor(gemas$COUNTRY)
gemas_GER <- dplyr::filter(gemas, gemas$COUNTRY == 'POL')
ssc <- cenLR(gemas_GER[, c("sand", "silt", "clay")])$x.clr
y <- ssc$sand
X <- dplyr::select(gemas_GER, c(MeanTemp, soilclass, Al:Zr))
X$soilclass <- factor(X$soilclass)
lmCoDaX(y, X, external = c('MeanTemp', 'soilclass'),
method='classical', pivot_norm = 'orthonormal')
lmCoDaX(y, X, external = c('MeanTemp', 'soilclass'),
method='robust', pivot_norm = 'orthonormal')