 # PCA vs Correlation What is the relationship between (first) principal component(s) and the correlation matrix or the average correlation of the data. For example, in an empirical application I observe that the average correlation is almost the same as the ratio of the variance of the first principal component (first eigenvalue) to the total variance (sum of all eigenvalues). Is there a mathematical relationship? Shannon Andrews 2022-07-16 Answered
PCA vs Correlation
What is the relationship between (first) principal component(s) and the correlation matrix or the average correlation of the data. For example, in an empirical application I observe that the average correlation is almost the same as the ratio of the variance of the first principal component (first eigenvalue) to the total variance (sum of all eigenvalues).
Is there a mathematical relationship?
You can still ask an expert for help

• Live experts 24/7
• Questions are typically answered in as fast as 30 minutes
• Personalized clear answers

Solve your problem for the price of one coffee

• Math expert for every subject
• Pay only if we can solve it minotaurafe
Short Answer: The principal components are the eigenvectors of the correlation matrix. Therefore, each principle component (V) multiplied by the correlation matrix (C) will give us the same correlation matrix times the corresponding eigenvalue $\lambda$:
$CV=\lambda V$
Details: Given n-dimensional data ${x}_{i}\in {R}^{n}$, suppose we have m datapoints represented as rows in a matrix X (An $m×n$ matrix). Given that Cor(i,j) is the correlation of 2 dimensions i and j, the correlation matrix is defined as:
$C=\left[\begin{array}{cccc}Cor\left(0,0\right)& Cor\left(0,1\right)& \cdots & Cor\left(0,n-1\right)\\ Cor\left(1,0\right)& Cor\left(1,1\right)& \cdots & Cor\left(1,n-1\right)\\ ⋮& ⋮& \ddots & Cor\left(n-1,n-1\right)\end{array}\right]$
Since the correlation matrix is a square matrix of size $n×n$, there are n possible eigenvectors for this matrix, and these vectors are the principle components of this data. each principle component V is of size $n×1$, and it's corresponding eigenvalue $\lambda$ is a scalar value.