I have four columns, which I have z-scaled and tried PCA and SVD on, hoping that I can obtain one dimension that explains the majority of the variance.

However, with PCA there is approximately 25% variance explained by each of PC 1-4, SVD is not much better. Are there any common methods that work better when PCA and SVD don't, or is it simply not possible to reduce the dimensionality to one column that explains a greater amount of the variance with this data set?

In case you want more info about why I want to do this: the four variables are all related to one another, and I wish to use them as the dependent variable in a structural equation modeling analysis.

  • 4
    $\begingroup$ Please say more about what you wish to accomplish by obtaining "one dimension that explains the majority of the variance." Most modeling procedures should have no problem working with 4 dimensions of predictors. Please provide that information by editing your question, as information in comments is easily overlooked and comments can sometimes be lost. Note that PCA and SVD are essentially equivalent, with singular values the square roots of corresponding PCA eigenvalues. $\endgroup$ – EdM Feb 27 at 18:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.