1
$\begingroup$

Suppose we have two binomial random variables $X_i \sim B(\frac{a}{p_i},p_i)$ and $X_j \sim B(\frac{a}{p_j},p_j)$, where $a$ is a positive integer, and both $\frac{a}{p_i}$ and $\frac{a}{p_j}$ are integers. Assume $p_i > p_j$. In particular, we have $E[X_i] = E[X_j] = a$.

Let $F_{X_i} (x)$ and $F_{X_j} (x)$ denote the CDF's of $X_i$ and $X_j$, respectively. It seems true (with an example in the image below) that $F_{X_i} (x) < F_{X_j} (x)$ for $x < a$ and $F_{X_i} (x) > F_{X_j} (x)$ for $x > a$. Is there a simple way to prove it (or, is this result somewhat established in literature)?

Example

$\endgroup$
3
  • 2
    $\begingroup$ crossposted: math.stackexchange.com/q/3969967/87355 --- please don't do that, in particular not without disclosing it, to avoid wasteful duplication of efforts. $\endgroup$ Jan 2, 2021 at 13:14
  • $\begingroup$ Thanks for your response! One thing I find tricky here is that in the problem set-up $N$ actually changes with $p$ (i.e., $N = \frac{a}{p}$ for some constant $a$). Does the derivative method still apply here or maybe more tricks would be needed? $\endgroup$
    – messi22
    Jan 2, 2021 at 17:39
  • $\begingroup$ @messi22: Yes, you are right, I didn't quite answer the question you asked. $\endgroup$ Jan 2, 2021 at 18:59

1 Answer 1

4
$\begingroup$

This result is known. See e.g. Corollary 4 in this paper or its arXiv version.

$\endgroup$
1
  • $\begingroup$ Thanks very much for your help! $\endgroup$
    – messi22
    Jan 3, 2021 at 14:41

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.