maskAggregators: Multi-mask/weight based aggregators

Description

Methods to aggregate the same vector with different masks or frequency weights. Useful for fast bootstrapping or split-half scoring. A single aggregate value of x is computed for each column of the mask or weight matrix.

Usage

mediansByMask(x, mask)
meansByMask(x, mask)
sdsByMask(x, mask)
mediansByWeight(x, weights)
meansByWeight(x, weights)
sdsByWeight(x, weights)

Value

a vector with each value representing an aggregate of the same single input vector but with different masks or frequency weights applied.

Arguments

x: A vector to aggregate over with different masks or weights.
mask: Logical matrix where each column represents a separate vector of masks to aggregate x with. Only values marked TRUE are included in the aggregation.
weights: Integer matrix where each column represents frequency weights to weight the aggregation by.

Author

Sercan Kahveci

Examples

Run this code


# Demonstration of mediansByMask()
x<-1:6
mask<-rbind(c(TRUE,FALSE,FALSE),
            c(TRUE,FALSE,FALSE),
            c(FALSE,TRUE,FALSE),
            c(FALSE,TRUE,FALSE),
            c(FALSE,FALSE,TRUE),
            c(FALSE,FALSE,TRUE))
mediansByMask(x,mask)

# Compute split-halves for a single 
# participant, stratified by stimulus
data(foodAAT)
currdata<-foodAAT[foodAAT$subjectid==3,]
currdata$stratfactor<-
  interaction(currdata$is_pull,
              currdata$is_target,
              currdata$stimid)
currdata<-currdata[order(currdata$stratfactor),]
groupsizes<-
  rle(as.character(currdata$stratfactor))$lengths
mysplits<-
  stratifiedItersplits(splits=1000,
                       groupsizes=groupsizes)

# Median for half 1
mediansByMask(currdata$RT,mysplits==1)
 
#How to use meansByMask()
meansByMask(x,mask)
sd(meansByMask(currdata$RT,mysplits==1))

# How to use sdsByMask() to compute
# mask-based D-scores
meansByMask(currdata$RT,mysplits==1) / 
  sdsByMask(currdata$RT,mysplits==1)

# Compute the bootstrapped 
# standard error of a median
weights<-
  bootstrapWeights(size=nrow(currdata),
                   times=1000)
bootmeds<-mediansByWeight(currdata$RT,weights)
sd(bootmeds) # bootstrapped standard error

# Compute the bootstrapped 
# standard error of a mean
bootmeans<-meansByWeight(currdata$RT,weights)
sd(bootmeans) # bootstrapped standard error
# exact standard error for comparison
sd(currdata$RT)/sqrt(length(currdata$RT)) 

# Use sdsByWeight to compute bootstrapped D-scores
bootsds<-sdsByWeight(currdata$RT,weights)
# bootstrapped standard error of D-score
sd(bootmeans/bootsds)