Does it make sense to average principal components?

Hi,

I have ran a PCA on several participants data sets and would like to get a list of averaged principal components for my entire data set. Hence I am wondering, does it make sense to average PC's together and create means for the coefficients?

Thanks!

Answer 1

[prabhat kumar sharma]

Hi @lil brain,

I understand you are facing issue with the principal component analysis.

I would suggest not to calculate the average of the principal components (PCs) themselves or the mean of the coefficients when performing PCA on multiple participants' datasets. The PCs represent orthogonal directions capturing the maximum amount of variance in each individual dataset and averaging them would not yield meaningful results.

Instead, consider calculating the mean or median of the original data points to obtain an average representation of the data across all participants.

For more information about PCA , you can refer to the following documentation link:

https://www.mathworks.com/help/stats/pca.html

Note : PCA is sensitive to the relative scaling of the original variables. If the scales of the variables in your different datasets are not the same, you'll need to standardize them before combining them and running PCA.

Regards,

Prabhat Sharma