Question on applying negative binomial distribution

92 Views Asked by At

Let's assume we conduct an experiment which generates two outcomes: success $A$ or failure $B$. The experiment never generates more than $N$ failures. Both $A$ and $B$ have the same probability. After $N+1$ successes the experiment stops. What is the probability to see $m\geq0$ failures before the $(N+1)$-th success?


This looks like a negative binomial distribution, i.e. $P(m)={N+m\choose m}\frac{1}{2^{N+m+1}}$. However, in a negative binomial distribution $m$ is unbounded. Further, in the experiment we ca only see the outcomes $m=0,1,\dots,N$. Hence, $$ P(m=0)+P(m=1)+\dots +P(m=N)=1. $$ However, if we consider the binomial distribution it is certainly not the case that $$ \sum\limits_{m=0}^N{N+m\choose m}\frac{1}{2^{N+m+1}}\color{red}{=}1. $$


So I am wondering if this distribution is appropriate in this setting? How could we justify the application of the negative binomial distribution (if it is possible at all)?

1

There are 1 best solutions below

1
On

With Wimbledon on, it strikes me that it would be easier to model it instead as a truncated binomial distribution (thus probabilities not adding up to $1$) in which you lose a $(2N+1)$ set match

$$P(m\geq 0)=\frac{1}{2^{2N+1}}\sum_{m=o}^N\binom{2N+1}{m}$$


Added at OP's request

To mould it into a negative binomial distribution, given $N+1$ successes, we find the number of trials $T$ (variant #2 of Alternative Formulations in Wikipedia)

$$P(X=T) = \binom{T-1}{N}p^{N+1}(1-p)^{T-N-1}$$

Of course, due to the conditions laid down,
$(N+1) \leq T \leq (2N+1)$,
so again, it is a truncated NB distribution.