Hi,
I’m using the IPCA for some of my analysis and the Kurtosis values are displaying negative values.
Are negative values considered low values, or is it just the magnitude that matters?
In one of my analysis I’m even getting rising negative kurtosis values.
e.g. -0.3125531 -0.6767299 -0.8193821 for PC1, 2, & 3 respectively.
This also results in explained variance of 22% (PC1), 26% (PC2), 16% (PC3)
hi @strkiky,
We calculate the kurtosis on the loading vector to focus on the hyper gaussian ones (and then reorder the associated components based on this). A negative kurtosis would indicate that is not heavy tailed (thus you may be fine just using PCA?)
Maybe run the code for your data to visualise the distribution of your loading vectors:
data(liver.toxicity)
# IPCA on a microarray dataset
ipca.res <- ipca(liver.toxicity$gene, ncomp = 3, mode="deflation")
ipca.res
hist(ipca.res$loadings$X[,1])
hist(ipca.res$loadings$X[,2])
hist(ipca.res$loadings$X[,3])
Kim-Anh
Hi Kim,
I tried out the histograms, you are right, the ones with the negative kurtosis values still have a normal gaussian distribution whereas the ones with positive kurtosis values are hyper gaussian.
Many thanks,
Best regards,