0
$\begingroup$

Consider two matrices of random variables, X and W, of same dimension. Now, consider the product X'W. I've been told that this matrix gives us the covariances between the elements of X and W. However, this is not immediately apparent to me. You do get sums of products, but these aren't exactly covariances.

$\endgroup$
6
  • 1
    $\begingroup$ Usually, the covariance matrix refers to the expectation of the outer product of two vectors of random variables, i.e. the $n \times n$ real valued matrix $\mathbb{E}[\bf{X} \bf{X}^t]$ where $\bf{X}$ $= [X_1,...,X_n]$ is an $n \times 1$ vector of random variables. $\endgroup$ – zoidberg Dec 5 '18 at 2:56
  • $\begingroup$ What you are referring to sounds like a $\textit{random}$ sample covariance matrix. See Wishart matrix $\endgroup$ – zoidberg Dec 5 '18 at 2:59
  • $\begingroup$ Thanks, but does the matrix that I have mentioned give some sort of indication about covariances? Intuitively, I think it does, but am not convinced. $\endgroup$ – Student Dec 5 '18 at 3:06
  • $\begingroup$ It does if the columns of X and W are i.i.d. random variables. Then the $ab$ entry of $X^tW$ is $\sum_i X_{ai} W_{bi}$, which can be thought of as a sample covariance between the random variables $X_a$ and $W_b$ where $X_a$ is distributed like the entries in column $a$ of $X$ and $W_b$ is distributed like the entries in column $b$ of $W$. $\endgroup$ – zoidberg Dec 5 '18 at 3:38
  • $\begingroup$ Not exactly the sample covariances right? Need to subtract sample averages. $\endgroup$ – Student Dec 5 '18 at 8:57

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.