R/multicollinearity.R

In rvif: Collinearity Detection using Redefined Variance Inflation Factor and Graphical Methods

multicollinearity <-
function(y, x, alpha=0.05){
  if (is.matrix(x) == FALSE) x = as.matrix(x)
  n = nrow(x)
  p = ncol(x) 
  #
  reg = lm(y~x+0) # equivalent to lm(y~X[,-1]): same values of beta, var.beta and sigma2
  beta = as.double(reg$coefficients)
  var.beta = as.double(summary(reg)[[4]][,2])^2
  sigma2 = summary(reg)[[6]]^2
  #
  Xqr=qr(x)
  P = qr.R(Xqr)
  beta.o = P%*%beta
  #
  t.teo = qt(1-alpha/2,n-p)
  t.exp = as.double(summary(reg)[[4]][,3])
  c0 = numeric()
  c3 = numeric()
  RVIFs = rvifs(x, ul=FALSE)[[1]]
  if (is.double(RVIFs) == TRUE) {
    Affects = array("No", p)
    Scenario = array("", p)
    for (i in 1:p){
      c0[i] = (beta[i]/(sqrt(sigma2)*t.teo))^2
      c3[i] = ((t.teo/beta.o[i])^2)*var.beta[i]
      #c3[i] = ((beta[i]/beta.o[i])*(t.teo/t.exp[i]))^2
      levels = c(c0[i], c3[i])
      #
      if((c3[i]<RVIFs[i])&(RVIFs[i]<c0[i])){Scenario[i]="a.1"}
      minimum = min(levels)
      if(RVIFs[i]<minimum){Scenario[i]="a.2"}
      maximum = max(levels)
      if(RVIFs[i]>maximum){Scenario[i]="b.1"}
      if((c0[i]<RVIFs[i])&(RVIFs[i]<c3[i])){Scenario[i]="b.2"}
      if (Scenario[i]=="a.2"){Affects[i]="Contradiction"}
      if (Scenario[i]=="b.1"){Affects[i]="Yes"}
    }
    #
    output = data.frame(RVIFs, c0, c3, Scenario, Affects)
    return(output)  
  }
}