Sign in to save

Bookmark this page so you can find it later.

Sign in to save

Bookmark this page so you can find it later.

The negative binomial distribution models repeated independent Bernoulli trials until a fixed number of successes occurs. This cheat sheet helps students recognize the setup, choose the correct parameterization, and compute probabilities accurately. It is especially useful in college statistics courses because the distribution appears in reliability, quality control, biology, and repeated sampling problems.

The two most common forms count either the total number of trials needed to get rr successes or the number of failures before the rrth success. The probability mass function uses a binomial coefficient to count possible trial sequences and a power of pp for successes. The mean and variance depend strongly on the success probability pp, so interpreting the parameters is as important as substituting into formulas.

Key Facts

  • If XX is the total number of trials needed to get rr successes with success probability pp, then XNegBin(r,p)X \sim \operatorname{NegBin}(r,p) has support x=r,r+1,r+2,x=r,r+1,r+2,\ldots.
  • For the trials-counting form, the PMF is P(X=x)=(x1r1)pr(1p)xrP(X=x)=\binom{x-1}{r-1}p^r(1-p)^{x-r} for x=r,r+1,r+2,x=r,r+1,r+2,\ldots.
  • If YY is the number of failures before the rrth success, then Y=XrY=X-r and the support is y=0,1,2,y=0,1,2,\ldots.
  • For the failures-counting form, the PMF is P(Y=y)=(y+r1r1)pr(1p)yP(Y=y)=\binom{y+r-1}{r-1}p^r(1-p)^y for y=0,1,2,y=0,1,2,\ldots.
  • For XX equal to total trials, the mean is E[X]=rpE[X]=\frac{r}{p} and the variance is Var(X)=r(1p)p2\operatorname{Var}(X)=\frac{r(1-p)}{p^2}.
  • For YY equal to failures before rr successes, the mean is E[Y]=r(1p)pE[Y]=\frac{r(1-p)}{p} and the variance is Var(Y)=r(1p)p2\operatorname{Var}(Y)=\frac{r(1-p)}{p^2}.
  • When r=1r=1, the negative binomial distribution reduces to the geometric distribution, with P(X=x)=(1p)x1pP(X=x)=(1-p)^{x-1}p for x=1,2,3,x=1,2,3,\ldots.
  • The trials must be independent, each trial must have the same success probability pp, and the process must stop when the rrth success occurs.

Vocabulary

Bernoulli trial
A Bernoulli trial is a random trial with exactly two outcomes, usually called success and failure, where the success probability is pp.
Success probability
The success probability pp is the fixed probability that one independent trial results in a success.
Negative binomial distribution
A negative binomial distribution gives probabilities for the number of trials or failures needed to obtain a fixed number rr of successes.
Support
The support is the set of possible values a random variable can take, such as x=r,r+1,r+2,x=r,r+1,r+2,\ldots for total trials.
Probability mass function
A probability mass function assigns a probability P(X=x)P(X=x) to each possible value of a discrete random variable.
Parameterization
A parameterization is the chosen way to define the random variable, such as counting total trials XX or failures YY before the rrth success.

Common Mistakes to Avoid

  • Using the wrong support, such as allowing X<rX<r, is wrong because at least rr trials are required to get rr successes.
  • Confusing total trials with failures is wrong because XX and YY differ by rr, so Y=XrY=X-r and their means are different.
  • Using (xr)\binom{x}{r} instead of (x1r1)\binom{x-1}{r-1} in the trials-counting PMF is wrong because the last trial must be the rrth success.
  • Swapping pp and 1p1-p is wrong because the exponent on pp counts successes and the exponent on 1p1-p counts failures.
  • Applying the negative binomial model when trials are not independent or pp changes is wrong because the standard PMF assumes identical independent Bernoulli trials.

Practice Questions

  1. 1 Let XX be the total number of trials needed to get r=3r=3 successes when p=0.40p=0.40. Find P(X=7)P(X=7).
  2. 2 Let YY be the number of failures before the r=5r=5th success when p=0.25p=0.25. Find E[Y]E[Y] and Var(Y)\operatorname{Var}(Y).
  3. 3 A basketball player makes each free throw independently with probability p=0.80p=0.80. What is the probability that the player makes the 44th successful free throw on the 66th attempt?
  4. 4 Explain why the last trial must be a success in the PMF P(X=x)=(x1r1)pr(1p)xrP(X=x)=\binom{x-1}{r-1}p^r(1-p)^{x-r} when XX counts total trials until the rrth success.

Understanding Negative Binomial Distribution Reference

The combination term comes from the fact that the last trial is not flexible. If a process ends on the eighth trial with its third success, trial eight must be a success. Among the first seven trials, there must be exactly two successes and five failures.

The combination counts every possible placement of those two earlier successes. Each valid sequence has the same probability because it contains the same number of successes and failures. This is why the distribution has a combination term, a success probability raised once for each required success, and a failure probability raised once for each earlier failure.

A common source of errors is mixing up what the random variable counts. In one version, a result of eight means eight attempts were needed. In the other version, a result of five means five unsuccessful attempts occurred before the final required success.

These describe the same event in different units. The difference is the fixed number of successes being waited for.

Before using a table, calculator, or software command, write a short sentence defining the count. This step prevents off by one errors, especially when the required number of successes is one.

The model can be understood as adding several waiting times. Waiting for three successes means waiting for the first success, then the next one, then the next one. Each piece behaves like a geometric waiting time when the conditions remain unchanged.

This explains why lower success probabilities produce longer and more variable waits. It also shows why rare successes create a long right tail.

A few runs may take far more trials than expected. The mean is useful for planning, but it does not mean most runs finish exactly near that value.

Check the assumptions against the real process before applying the model. Repeated manufacturing tests may be close to independent if each item is made under stable conditions. Repeated shots by a player may not be independent because fatigue, practice, or changing defense affects later attempts.

Sampling without replacement from a small group changes the chance of success after each draw. In these cases, a negative binomial calculation can look precise while describing the wrong situation.

For probability questions involving at least or more than a stated number, it is often easier to use a cumulative probability and subtract from one. Pay close attention to whether the endpoint includes the final success, since that wording determines which count is being used.