2
$\begingroup$

What are some examples of random variables X, A, B such that X is independent to A, and to B, but not to A and B jointly, i.e., X is not independent to (A,B). In other words, $X \perp A$ and $X \perp B$ but not $X \perp A, B$

I got curious while reading http://en.wikipedia.org/wiki/Conditional_independence

It is enough to find examples such that $X \perp A$ and $X \perp B$ but not $X \perp A \mid B$.

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ Welcome to MathOverflow! Unfortunately, this question isn't appropriate for this web site. Please read the FAQ. $\endgroup$
    – Mark Meckes
    Commented Sep 27, 2012 at 14:06

2 Answers 2

Reset to default
4
$\begingroup$

Let $A$ and $B$ be independent random variables with $P(A=1)=P(A=-1)=P(B=1)=P(B=-1)=1/2$. Let $X = AB$. Then any two of $A$, $B$ and $X$ are independent, but the three are not. It's easy to generalize this to generate a sequence of $n$ random variables, any $n-1$ of which are independent, but where they aren't all independent.

Improve this answer
$\endgroup$
3
$\begingroup$

Let $A$ and $B$ be indipendent and uniformly distributed in $S^1$ (for example you can take $\Omega = S^1 \times S^1$ with the uniform probability and $A$ and $B$ as the two projection); then it is clear that $X=A+B$ is indipendent of $A$, it is indipendent of $B$, but it isn't indipendent of $(A,B)$.

In this reasoning it's crucial that the uniform law on $S^1$ is traslation invariant.

Improve this answer
$\endgroup$
3
  • 3
    $\begingroup$ ...where $S^1 = \mathbf{R}/\mathbf{Z}$ and not $\lbrace z\in\mathbf{C}: |z|=1\rbrace$. (In the latter case $X=A+B$ is not what you mean.) $\endgroup$
    – Sean Eberhard
    Commented Sep 27, 2012 at 15:22
  • $\begingroup$ Sure... By the way, also this method could be generalized to $n$ variables. It seems to be true also for countable many variables... Think of this example: take $\{ X_i \}_{i \in \mathbb{N}}$ i.i.d. variables which take values in $\mathbb{Z} / 2 \mathbb{Z}$ uniformly. Then consider $Y_n = \sum_{ i \leq n} X_i$ (where the sum is also in $\mathbb{Z} / 2 \mathbb{Z}$), and then take $Z$ as the limit of the $Y_i$s along some non-principal ultrafilter. Then $Z$ is indipendent of any $A$ proper subset of $ \{ X_i \}_{i \in \mathbb{N}}$ but clearly not indipendent from all of them. $\endgroup$
    – Simo_the_Wolf
    Commented Sep 27, 2012 at 18:17
  • $\begingroup$ I've always taken a set of rvs to be independent iff every finite subset is independent. Also, I think that by the Kolmogorov 0-1 law, $Z$ isn't going to be measurable (since if $Z$ is measurable, then by symmetry $P(Z=1) = P(Z=0) = 1/2$). $\endgroup$
    – Alexander Pruss
    Commented Mar 19, 2013 at 15:36

Not the answer you're looking for? Browse other questions tagged .