I am coming from reading the selected answer to this question. I have a question about the following bit:

It's not hard to show that if the covariance matrix of the original data points $x_i$ was $\Sigma$, the variance of the new data points is just $u^T \Sigma u$.

I have been playing around with projecting two dimensional data for random variables $(x_1,x_2)$ onto the horizontal axis corresponding to $x_1$. As expected, with $u$ being in the direction of the horizontal axis, the result is $u^T \Sigma u = \textrm{var}[x_1]$ since we are only taking into account $x_1$. However, when setting $u$ to be in direction of the identity line, the result is $$u^T \Sigma u = \textrm{var}[x_1] + \textrm{var}[x_1] + 2\textrm{cov}[x_1,x_2].$$ I don't know much about statistics and don't understand how this represents the variance of the data when projected onto the identity line. Is there a more formal proof for why the quoted bit is true? An intuitive explanation of the result would be appreciated as well.

Edit: My question is why $u^T \Sigma u$ is the variance of the new data points as stated in the question linked above.


I'm not sure why are you interested on projecting onto the identity line, but: assuming zero mean variables, projecting onto $x_2=x_1$ is the same as computing the average: $z= (x_1+x_2)/2$.

Now, $E[z^2]=\frac{1}{4}E[(x_1+x_2)^2]=\frac{1}{4}( E[x_1^2] +E[x_2^2]+2 E[x_1 x_2]) $

So $Var(z)=\frac{1}{4}(Var(x_1)+Var(x_2)+2Cov(x_1,x_2))$


Your Answer

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

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