R/regDiag.R

In tengku-hanis/niceFunction: These are random nice functions that I found

#' Regression diagnostic
#'
#' \code{regDiag} is for screening of outliers and influential cases.
#'
#' @param model  lm or glm model
#'
#' @return Outlier : leverage, Studentized deleted residual (SDR)
#' @return Influential case : DFFits, DFBetas, cook's distance (cook.d)
#' @return also noted that we use dfbetas (scaled/standardized) NOT dfbeta function
#' @return TRUE : outlier and influential case present,
#' @return FALSE : no outlier and influential case
#'
#' @return Cut-off values:
#' @return leverage > 0.2
#' @return SDR >= 3 or <= -3
#' @return DFFits > 1
#' @return DFBetas > 1
#' @return cook.d > 0.5
#' @export
#'
#' @importFrom stats cooks.distance dfbetas dffits hatvalues rstudent
#' @importFrom knitr kable
#'
#' @examples
#' mod <- lm(Sepal.Length ~ Species + Sepal.Width, data = iris)
#' regDiag(mod)
regDiag <-  function(model){
  lev <-  hatvalues(model)
  s.dr <-  rstudent(model)
  dfit <-  dffits(model)
  dbeta <-  dfbetas(model)
  coo <-  cooks.distance(model)

  leverage <-  lev >0.2
  SDR <-  s.dr >=3 | s.dr <=-3
  DFFits <-  dfit >1
  DFBetas <-  dbeta >1
  cook.d <-  coo >0.5

  outlier <-  list(leverage = table(leverage), SDR = table(SDR))
  influence <-  list(DFFits = table(DFFits), DFBetas = table(DFBetas), cook.d = table(cook.d))
  result <-  list(outlier = outlier, influence = influence)
  print(knitr::kable(result, "simple"))
}