R/estCoeff.R

In dcoffman/tvmediation: Time Varying Mediation Analysis

#' Function to estimate coefficients at time t.
#' 
#' Part of the set of internal functions called within the \code{tvma} function to assist 
#' in the estimation of the time varying mediation effect.
#' 
#' @param newMO.j.est   a list containing mean centered mediators and outcomes
#' 
#' @return \item{coeff.est}{estimated coefficients of the mediation model}
#' 
#' 

estCoeff <- function(newMO.j.est) {

  ### Current method: Ridge Regression
  sym_newMO <- t(newMO.j.est$M)%*%(newMO.j.est$M)
  diag(sym_newMO) <- diag(sym_newMO) + .001
  coeff.est <- solve(sym_newMO)%*%t(newMO.j.est$M)%*%(newMO.j.est$Y)
  
  ### Original way that rendered singular matrices which were non-invertible
  # coeff.est <- solve(t(newMO.j.est$M)%*%(newMO.j.est$M))%*%t(newMO.j.est$M)%*%(newMO.j.est$Y)
  
  ### Generalized Moore-Penrose Inverse method
  # coeff.est <- MASS::ginv(t(newMO.j.est$M) %*% (newMO.j.est$M)) %*% t(newMO.j.est$M) %*% (newMO.j.est$Y)
  
  return(coeff.est)
}