1
$\begingroup$

Suppose that we have $n$ independent Bernoulli random variables, $x_1,\ldots,x_n$ such that each $x_i$ takes value 1 with success probability $p_i$ and value $0$ with failure probability $1-p_i$ ,i.e.,

\begin{align} P(x_1=1) & = p_1,\ P(x_1=0)= 1-p_1 \\ P(x_2=1) & = p_2,\ P(x_2=0) = 1-p_2 \\ & \vdots \\ P(x_n=1) & = p_n,\ P(x_n=0)=1-p_n \end{align}

So $x_1+...+x_n$ has Poisson binomial distribution since it is a sum of independent Bernoulli trials that are not necessarily identically distributed with the following pdf: $$ \mathsf{Pr}\left(K=k\right) = \sum_{A \in F_k} \prod_{i \in A} p_i \prod_{j \in A^c} (1-p_j) $$ where $F_k$ is the set of all subsets of k integers that can be selected from $\{1,\ldots,n\}$. and $A^c$ is the complement of A.

Suppose that I have calculated the following probability $$P(x_1+x_2+...+x_n<k)$$ Now I substitute one of $x_i$'s with another Bernoulli random variable (let say we substitute $x_1$ with $y_1$ with success probability $p'_1$) now I want to calculate $$P(y_1+x_2+...+x_n<k)$$ My question is : Can I use $P(x_1+x_2+...+x_n<k)$ to calculate $P(y_1+x_2+...+x_n<k)$ ?

$\endgroup$
2
$\begingroup$

Write $z = x_2 + \ldots + x_n$. You have $$P(x_1 + z = k) = p_1 P(z = k-1) + (1-p_1) P(z=k)$$ so $$ \eqalign{P(z = 0) &= \frac{P(x_1 + z = 0)}{1-p_1} \cr P(z = 1) &= \frac{P(x_1 + z = 1) - p_1 P(z = 0)}{1 - p_1} = \frac{P(x_1 + z = 1)}{1-p_1} - \frac{p_1 P(x_1 + z = 0)}{(1 - p_1)^2}\cr P(z = 2) &= \frac{P(x_1 + z = 2)}{1-p_1} - \frac{p_1 P(x_1+z=1)}{(1-p_1)^2} + \frac{p_1^2 P(x_1 + z = 0)}{(1-p_1)^3}\cr P(z = k) &= \sum_{j=0}^k \frac{(-p_1)^j P(x_1 + z = k-j)}{(1-p_1)^{j+1}}}$$ and then $$P(y+z = k) = p'_1 P(z=k-1) + (1-p'_1) P(z=k)$$

$\endgroup$

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.