R/flury.AIC.R

In tpepler/cpc: Common principal component (CPC) analysis and applications

flury.AIC <- function(covmats.high, covmats.low, nvec, df)
{
  # covmats.high: estimates of the covariance matrices under the "higher" model
  # covmats.low: estimates of the covariance matrices under the "lower" model (usually unrelated/individual covariance matrices)
  # nvec: vector of sample sizes
  # df: degrees of freedom of the higher model, versus unrelated covariance matrices
  
  k <- length(nvec)
  p <- dim(covmats.high)[1]
  aic.total <- 0
  for(i in 1:k){
    aic.total <- aic.total + (nvec[i] - 1) * (sum(diag(solve(covmats.high[, , i]) %*% covmats.low[, , i])) + log(det(covmats.high[, , i])) - sum(diag(solve(covmats.low[, , i]) %*% covmats.low[, , i])) - log(det(covmats.low[, , i])))
  }
  
  npar <- k * (0.5 * p * (p - 1) + p) - (0.5 * p * (p - 1) + p) - df
  
  aic.criterion <- aic.total + 2 * npar
  return(aic.criterion)
}