3
$\begingroup$

This question already has an answer here:

As explained in amoeba's beautiful answer here one can use a singular value decomposition of the data matrix, $\mathbf{X} = \mathbf{USV}^\top$, to do a principal component analysis, if it is assumed that the data matrix $\mathbf{X}$ is centered. The principal directions will then be given as the columns of $\mathbf{V}$.

Now, I have come across some statistical analysis where the data matrix was centered and standardized for the data to have unit variance before being used in a singular value decomposition. The analysis then proceeded in using the columns of $\mathbf{V}$ as principal directions.

Can any of you think of a reason to do this standardization step before running the singular value decomposition? Is the result of such a procedure in terms of what PCA should tell you even meaningful?

$\endgroup$

marked as duplicate by whuber Sep 15 '16 at 16:59

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ This is indeed a duplicate but +1 for a clear and well-formulated question. $\endgroup$ – amoeba Sep 15 '16 at 17:02
3
$\begingroup$

In the context of PCA, the reason for making all the variables have the same variance (by standardizing them) is simple: to make them all be treated as equally important. PCA will treat variables with greater variance as more important, since it tries to account for the greatest possible total variance with each successive component.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.