MCA in R, including many important variables

mca.R

require(FactoMineR)
# load data tea
data(tea)
# select these columns
newtea = tea[, c("Tea", "How", "how", "sugar", "where", "always")]
mca1 = MCA(newtea, graph = FALSE, ncp = 5)

# coordinates of the categories of the variables along the dimensions
y_kl <- mca1$var$coord

# eigenvalues
lambda_l <- mca1$eig$eigenvalue[1:5]

# Q =nbr of active variables
Q <- ncol(newtea)

# nb categories per variables
cats = apply(newtea, 2, function(x) nlevels(as.factor(x)))

# nb of modalities
K <- sum(cats)

# get_freq(newtea)
get_freq <- function(df){
  df_table <- do.call(rbind, lapply(df, function(x)as.data.frame(table(x))))
  out <- df_table$Freq/nrow(df)
  names(out) <- df_table$x
  return(out)
}

# frequency by modality
f_k <- get_freq(newtea)

# weight by modality
p_k <- f_k / Q

# contribution rel to coud
ctr_k <- (1-f_k)/(K-Q)

# contribtion rel to dimension
ctr_kl <- sweep(y_kl^2,1,p_k,FUN="*")
ctr_kl <- sweep(ctr_kl, 2, lambda_l, FUN = "/")

# compare with:
mca1$var$contrib

# conclusion: mca1$var$contrib = ctr_kl * 100