Connecting two interpretations of the negative binomial distribution

34 Views Asked by At

In my probability course, my professor derived the negative binomial distribution by reasoning about the probability that the time of the $k$-th success, $T_k$, takes some value $n$. If $p$ is probability of success, $X_k$ is the outcome of the $k$-th Bernoulli trial and $S_n$ is the number of successes by trial $n$, then

$$ \mathbb{P}\{T_k = n\} = \mathbb{P}\{X_n = 1\} \mathbb{P}\{S_{n-1} = k-1\} = {n - 1 \choose k - 1} p^k (1-p)^{n-k}, \qquad n \geq k $$

and he denotes this

$$ T_k \sim \text{NB}(k, p) $$

This all makes sense.

But when I try to read about the negative binomial distribution on Wikipedia, it was described as the probability of the number of successes until a fixed number of failures, $r$, and is denoted $X \sim \text{NB}(r, p)$.

$$ \mathbb{P}\{X = k\} = {k + r - 1 \choose k} p^k (1 - p)^r $$

If we note that $n = k + r$, this is almost the same:

$$ \mathbb{P}\{X = k\} = {n - 1 \choose k} p^k (1 - p)^{n - k} $$

Anyway, are these the same modeling assumptions? Why are they parameterized differently? Also, is there some connection between interpreting the r.v. as a time vs. a count?