1
$\begingroup$

I would like to know which is the best method to do the following: - generate a random variable $Y$ over $Z_2$ (possibly following a Bernoulli distribution) such that $Y$ has correlation $\rho$ with a random variable $X$ following the uniform distribution over $Z_2^k$?

I.e., If $x \in X$ is uniformed sampled over $Z_2^{k}$, then I want to generate $y \in Y$ over $Z_2$ such that $\mathrm{Corr}(X,Y)$ is a fixed, given $\rho$? If possible, I'd like $Y$ to follow a Bernoulli distribution.

As standard, $Z_2=\{0,1\}$ and $Z_2^{k}=\{(x_1,..., x_i,,..., x_k) | x_i \in \{0,1\}\}$.

$\endgroup$
7
  • $\begingroup$ You need to put the math between dollar signs to get the Latex to display - e.g. $x$ produces $x$ $\endgroup$
    – Silverfish
    Jun 27, 2016 at 15:56
  • $\begingroup$ Please say in the question what is $Z$, $Z_2^k$. $\endgroup$
    – ttnphns
    Jun 27, 2016 at 17:43
  • $\begingroup$ How do you define correlation between a random vector and a random scalar? $\endgroup$
    – Yair Daon
    Jun 28, 2016 at 2:01
  • $\begingroup$ $Z_2$=${0,1}$, $Z_2^k$=$\{0,1\} \times ... \times \{0,1\}$ (k times). As for correlation between a variable over $Z_2$ and one over $Z_2^k$, I think it is in fact cross-correlation, to be correct. $\endgroup$
    – icb123
    Jun 28, 2016 at 13:55
  • $\begingroup$ So $\rho$ is a vector, nor a number. $\endgroup$
    – Yair Daon
    Jun 28, 2016 at 18:26

1 Answer 1

1
$\begingroup$

I will provide solutions to two different interpretations of your question. I don't claim that they are the only or the best solutions.

Every place in which independent Bernoulli is mentioned below, it is to be independent of everything else (all other Bernoullis).

Interpretation A: $X$ has a Bernoulli distribution having success probability $0.5$.

Solution:

Step 1. Generate $X$ per Bernoulli distribution having success probability $0.5$.

Step 2.

If $\rho \ge 0$, then for each sample of $X$, set $Y = X$ with probability $\rho$ (i.e., per independent Bernoulli having success probability $\rho$), otherwise set $Y$ = independent Bernoulli having success probability $0.5$.

If $\rho \lt 0$, then for each sample of $X$, set $Y = 1-X$ with probability $-\rho$ (i.e., per independent Bernoulli having success probability $-\rho$), otherwise set $Y$ = independent Bernoulli having success probability $0.5$.

Result: $Y$ has a Bernoulli distribution having success probability $0.5$, and $\text{Corr}(X,Y) = \rho$.

Interpretation B: $X$ has a Bernoulli distribution having success probability $b$, and $Y$ must also have a Bernoulli distribution having success probability $b$, with $\text{Corr}(X,Y) = \rho$.

Solution is provided for $\rho \ge 0$ (note that solution for $\rho \lt 0$ and $b = 0.5$ is provided under Interpretation A; otherwise things are problematic).

Step 1. Generate $X$ per Bernoulli distribution having success probability $b$.

Step 2. For each sample of $X$, set $Y = X$ with probability $\rho$ (i.e., per independent Bernoulli having success probability $\rho$), otherwise set $Y$ = Independent Bernoulli having success probability $b$.

Result: $Y$ has a Bernoulli distribution having success probability $b$, and $\text{Corr}(X,Y) = \rho$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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