R/partial.corr.R

In TiagoOlivoto/cursoR: Curso R do grupo Experimentacao da UFSM-SM

  partial.corr = function(data,
                         n = NULL,
                         method = "pearson") 
{
  
  if(is.matrix(data) == TRUE & is.null(n) == TRUE){
    stop("You have a matrix as entire but do not declared the sample size used to estimate the matrix")
  }
  
  if(is.matrix(data) != TRUE & is.null(n) != TRUE){
    stop("You have a dataframe as entire but declared the sample size used. When a dataframe is used, the sample size is automatically caltulated.")
  }
  if(is.null(n)==TRUE){
    n = nrow(data)
  }else{n = n}
   nvar = ncol(data)
   df = n - nvar
  
  cl = match.call()
  if (!is.matrix(data)) {
    data = cor(data, use = "complete", method = "pearson")
  }
    m = as.matrix(data)
    X.resid = -(solve(m))
    diag(X.resid) = 1/(1 - psych::smc(m))
    X.resid = cov2cor(X.resid)
  
results = data.frame(linear = as.vector(t(m)[lower.tri(m,diag=F)]))
results = dplyr::mutate(results,
                        partial = as.vector(t(X.resid)[lower.tri(X.resid,diag=F)]),
                        t = partial/(sqrt(1-partial^2))*sqrt(n-nvar),
                        prob = 2*(1-pt(abs(t), df = df)))
names = colnames(data)
combnam = combn(names,2, paste, collapse = " x ")
rownames(results) = names(sapply(combnam, names))

return(list(linear.mat = data.frame(m),
            partial.mat = data.frame(X.resid),
            results = results,
            call = cl))
  

}