0
$\begingroup$

I am self learning the discrete random variable distributions, and I think I have some problems understanding the difference between these two distributions.

For what I understand at this stage, I know in binomial distribution, we are curious about the probability of doing n trials with r times of success, and in negative binomial distributions, we are curious about the probability of a "SPECIFIC order of success" in the total of n trials, like the second success in the total of n trials that kind of stuff. (I am not sure if my understanding about this is correct, but this is what comes to my mind after reading Wikipedia).

I have a practice question in my textbook that states the following:

[1]

My friend and I are arguing whether this is a binomial distribution or a negative binomial distribution. In my opinion, I think this is a binomial distribution, since we only care about like the best four out of seven, but not the 4th win in the seven.

Can someone help me understand about this? Thanks.

$\endgroup$
1
$\begingroup$

The usual way of conceptualizing the difference between the two models is to ask whether a fixed number of trials need to be observed, or whether a fixed number of successes need to be observed.

For example: if I have a die that, when rolled, shows 6 with probability $p$, then I can ask about the probability of observing $k$ sixes in $n$ rolls. That's a binomial model, and $k$ is a realization of the underlying random variable. I could also ask about the probability that it takes $n$ rolls to observe exactly $k$ sixes. That's a negative binomial model, and here, $n$ is a realization of the underlying random variable.

For your practice question, you can use either model.

Here's how you'd do it using a binomial model. Your choice is between $n = 5$ or $n = 7$. Let $X$ be the random number of wins that $A$ achieves. Then in the first case, we need $X \ge 3$ (best of 3), and in the second case, we need $X \ge 4$ (best of 4). So we want to compare the probabilities of these two events: $$\Pr[X \ge 3 \mid n = 5], \quad \text{vs.} \quad \Pr[X \ge 4 \mid n = 7].$$ In both cases, $p = 0.4$. Using the binomial model, $$\Pr[X \ge 3 \mid n = 5] = \binom{5}{3} (0.4)^3 (0.6)^2 + \binom{5}{4} (0.4)^4 (0.6)^1 + \binom{5}{5} (0.4)^5 (0.6)^0,$$ and $$\Pr[X \ge 4 \mid n = 7] = \binom{7}{4} (0.4)^4 (0.6)^3 + \binom{7}{5} (0.4)^5 (0.6)^2 + \binom{7}{6} (0.4)^6 (0.6)^1 + \binom{7}{7} (0.4)^7 (0.6)^0.$$ After computing these, the probability that is larger tells you which choice of $n$ is preferable for team $A$.

Now, here's how you'd use the negative binomial model. Rather than having two cases based on the total games played, you now have two cases based on the number of wins $w$ achieved by A. Let $Y$ be the random number of games played. In the first case, what is the probability that it takes $Y \le 5$ games for $A$ to win exactly $w = 3$ games? In the second, what is the probability that it takes $Y \le 7$ games for $A$ to win exactly $w = 4$ games? In other words, now the comparison is $$\Pr[Y \le 5 \mid w = 3] \quad \text{vs.} \quad \Pr[Y \le 7 \mid w = 4].$$ The first probability is $$\Pr[Y \le 5 \mid w = 3] = \binom{2}{2} (0.4)^3 (0.6)^0 + \binom{3}{2} (0.4)^3 (0.6)^1 + \binom{4}{2} (0.4)^3 (0.6)^2.$$ The second is $$\Pr[Y \le 7 \mid w = 4] = \binom{3}{3} (0.4)^4 (0.6)^0 + \binom{4}{3} (0.4)^4 (0.6)^1 + \binom{5}{3} (0.4)^4 (0.6)^2 + \binom{6}{3} (0.4)^4 (0.6)^3.$$ You should get the same results as with the binomial model; i.e., $$\Pr[X \ge 3 \mid n = 5] = \Pr[Y \le 5 \mid w = 3], \\ \Pr[X \ge 4 \mid n = 7] = \Pr[Y \le 7 \mid w = 4].$$ Therefore, both of you are right and both of you are wrong: which model you use depends on your choice of stopping criterion. In the binomial model, the game play stops after a fixed number of games are played. In the negative binomial model, the game play stops after a fixed number of wins are observed. The latter may be more commonly the case in a playoff situation (since typically, once a team wins a decisive number of games, there is no point to continuing to play), but the binomial model is mathematically equivalent.

| cite | improve this answer | |
$\endgroup$

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.