ImputationR Documentation

Imputation of missing parts in compositions from a SGB model

Description

Applied to a completely missing composition, the function returns the Aitchison expectation.
Applied to a partially missing composition, it returns the conditional Aitchison expectation, given the observed sub-composition.
Applied to a complete case, it returns the complete case.

Usage

impute.regSGB(obj, dsup, usup)

Arguments

obj

list, output of regSGB.

dsup

data frame with explanatory variables for the incomplete compositions. Missing values not allowed.

usup

compositions corresponding to dsup. On each row, the non-missing parts sum to 1.

Value

data frame with imputed compositions instead of missing or partially missing compositions. Complete cases are also returned.

Examples

## Arctic lake
data(arc)
arcmis <- arc
arc[11:13,]

## Introduce NA alues

arcmis[11,2] <- NA      # "core" observation
arcmis[12,3] <- NA      # outlying clay value
arcmis[13,1:3] <- NA    # totally missing observation
umis <- arcmis[,1:3]
umis <- umis/rowSums(umis,na.rm=TRUE)
umis[11:13,]

d <- data.frame(depth=arc[["depth"]])

## original compositions
arc[11:13,1:3]

## unconditional predicted value                            
MeanA.SGB(oilr[["par"]][1],oilr[["scale"]],oilr[["par"]][4:6] )[11:13,]

## predicted value given the sub-composition (sand,clay) for 11, (sand,silt) for 12        
impute.regSGB(oilr,arcmis,umis)[11:13, ]

impute.regSGB(oilr,arcmis[11:13, ],umis[11:13, ])  # same result.