2

Is it possible to have the following sample correlation matrix for $x$, $y$, $z$?

$\begin{pmatrix} 1 & 0.8 & 0.2 \\ 0.8 & 1 & 0.7\\ 0.2 & 0.7 & 1\end{pmatrix}$

Where a 3 by 3 correlation matrix is $\begin{pmatrix} 1 & p_{xy} & p_{xz} \\ p_{yx} & 1 & p_{yz}\\ p_{zx} & p_{zy} & 1\end{pmatrix}$, and the $p$'s (partial correlations) are unequal, making this different more general than a similar question with equal partial correlations posed elsewhere.

My answer is yes, and my reasoning is: all the diagonal entries have value $1$, the matrix is symmetric and all entries are between $-1$ and $1$

Is my reasoning good enough? Are there other points that I'm missing?

user59036
  • 205
  • Please search our site to read many more answers to your question. Your reasoning is good but it's not complete. – whuber Feb 24 '17 at 00:53
  • 3
    As I understand it, the question "Bound for correlation of three random variables" referred to above deals only with the case when all the off-diagonal entries have the same value $\rho$ whereas here the off-diagonal terms have different values. Maybe there is more here than meets the eye... I vote to re-open at least until the Moderators find a better question and answer to point to as a duplicate. – Dilip Sarwate Feb 24 '17 at 02:37
  • Note, in OP's Q $p_{xy}\neq p_{xz}\neq p_{yz}$ so that the question mentioned does not answer this. This question is more general but also just algebra. For ans, just derive conditions for probabilities from https://www.easycalculation.com/statistics/learn-correlation-matrix.php that is, their definition. – Carl Feb 24 '17 at 05:12
  • @Dilip I apologize if the apparent duplicate does not directly answer the question in your mind. Since this question has been answered many times, would you mind nominating one of the other questions as a duplicate then? – whuber Feb 24 '17 at 14:25
  • 2
    I have voted to close this question as a duplicate of Completing a $3\times 3$ correlation matrix $\ldots$ but actually it is this answer to the question that has the important stuff. It says "The correlation matrix should be positive semidefinite and hence its principal minors should be nonnegative." That is the point that you are missing, and why @whuber said that your reasoning is not complete. – Dilip Sarwate Feb 24 '17 at 16:06

0 Answers0