0

I have performed a scaled and centred PCA on a dataset with nine variables. I now want to quantify the contribution of each factor to the PCA.

Using a scree plot, I have identified that only the first three PCs are required.

My question is: if for each of the three PCs, I **multiply each rotation score by the PCs % explained **, and then sum this across all three PCs, will I be able to compare the 'overall' contribution of a variable to the PCA?

I am working in R, E.g.

pca <- prcomp(data, scale=TRUE)
percents <- pca$sdev^2 / sum(pca$sdev^2)
overall <- (abs(pca$rotation[, 1]) * percents[1]) + 
    (abs(pca$rotation[,2]) * percents[2]) + (abs(pca$rotation[,3]) * 
    percents[3])

Would these scores be a valid way to compare a factor's contribution? Do they have a name?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Mike M
  • 1
  • You will find what is generally counts as contribution of a factor/component to a variable and vice versa here https://stats.stackexchange.com/q/143905/3277 – ttnphns Jan 31 '22 at 11:54

0 Answers0