Confusion about relationship between geometric and negative binomial RV's

154 Views Asked by At

(Geo for geometric random variable, NB for negative binomial)

Let $X_1,\ldots,X_n\sim\text{Geo}(1/n)$ be i.i.d and let $X=\sum_{i=1}^nX_i$, I saw on wikipedia that $X\sim\text{NB}(n,1/n)$. However, when taking expectations, I have on the one hand that $$\mathbb{E}X=\mathbb{E}\sum_{i=1}^nX_i=\sum_{i=1}^n\mathbb{E}X_i=\sum_{i=1}^n\frac{1}{1/n}=n^2$$ and on the other hand I have that $$\mathbb{E}X=\frac{n\cdot1/n}{1-1/n}=\frac{1}{1-1/n}$$ using the formula for expectation of a negative binomial variable (also from wikipedia).

Clearly there's a mistake somewhere but I'm drawing a blank and would appreciate if anyone could explain whats going wrong.

1

There are 1 best solutions below

2
On BEST ANSWER

There are two common variations of the geometric distribution. The first counts the total number of trials until the first success (where the probability of success is $p$). The second counts the number of failures until the first success. The difference is that $0$ is in the support of the second distribution, and its expected value is $(1-p)/p$.

The negative binomial distribution $\text{NB}(r, p)$ counts the number of successes until $r$ failures have occurred. So a $\text{Geo}(p)$ random variable (in the second sense) is the same as a $\text{NB}(1, 1-p)$ random variable (by flipping the meaning of "success" and "failure). So a sum of $n$ $\text{Geo}(1/n)$ random variables (of the second kind) is a $\text{NB}(n, 1-1/n)$ random variable.