JK

gistfile1.txt

pairedBootstrap <- function(yourData, responseName, numberBoot){
  
  #####CREATE USEFUL VARIABLES FOR THE BOOTSTRAP METHOD
  originalSampleSize <- dim(yourData)[1]  #the data sample size
  numberExplanatory <- dim(yourData)[2]-1 #the number of explanatory variables
  
  #explanatory variable names
  explanatoryNames <- setdiff(names(yourData),responseName) 
  
  #explanatory variable formula (for the lm function)
  explanatoryFormula <- explanatoryNames[1]
  for(i in 2:length(explanatoryNames)){
    explanatoryFormula <- paste(explanatoryFormula, "+", explanatoryNames[i])
  }
  explanatoryFormula <- paste(as.character(responseName), "~", explanatoryFormula)
  
  #store coefficient estimates for each bootstrap sample in this matrix
  bootstrapCoefficients <- matrix(0, nrow=numberBoot, ncol=numberExplanatory+1)
  
  ######IMPLEMENT THE RESAMPLING PROCEDURE
  for(i in 1:numberBoot){
    #Create a vector of indices, which will make up our new sample
    samp <- sample(1:originalSampleSize, size=originalSampleSize, replace=TRUE)
    
    #save the new sample into a data frame called newSamp
    newSamp <- yourData[samp,]
    
    #run a linear regression to obtain bootstrap coef. estimates
    newReg <- lm(as.formula(explanatoryFormula), data=newSamp)
    bootstrapCoefficients[i, ] <- coef(newReg)
  }
  return(bootstrapCoefficients)
}

library(Stat2Data)
data(Perch)
head(Perch)
pairs(~Weight+Length+Width,Perch)
m=lm(Weight~Length+Width+Length:Width,Perch)
par(mfrow=c(2,2))
plot(m)
dataPerch=Perch
dataPerch=dataPerch[,-1]
dataPerch$LW=dataPerch$Length*dataPerch$Width
head(dataPerch)
Boot= pairedBootstrap(dataPerch,"Weight",2000)
hist(Boot[,2],breaks=20)
quantile(Boot[,2],probs=c(0.025,0.975))