Back to The Negative Binomial Distribution
Chapter 9

What Is Negative Binomial?

Adding the dispersion parameter (r)

What Is a Negative Binomial Distribution?

The Negative Binomial distribution is a discrete probability distribution—like Poisson, it models counts (0, 1, 2, 3...), not continuous values. But unlike Poisson, which has only one parameter (λ), Negative Binomial has two parameters that give it more flexibility.

The Two Parameters

Negative Binomial Parameters

μ (mu) = Mean (expected occurrences)\nr = Dispersion parameter (controls spread)
Excel: μ = AVERAGE(range), r = calculated from variance

Parameter 1: μ (mu) — The Mean

This is the expected number of occurrences, the same as λ in Poisson. If a player averages 0.85 TDs per game, then μ = 0.85.

Parameter 2: r — The Dispersion Parameter

This is what makes Negative Binomial more flexible than Poisson. The r parameter controls how spread out the distribution is around the mean.

Key Insight

Think of r as a "consistency dial":

  • High r (10+): Player is consistent, outcomes cluster tightly around the mean → Poisson is fine
  • Low r (0.5-2): Player is boom-or-bust, outcomes are spread wide → Negative Binomial shines

The Variance-Mean Relationship

The key mathematical relationship in Negative Binomial is:

Negative Binomial Variance

Variance = μ + (μ² / r)
Excel: =B2+(B2^2/C2) where B2=mean, C2=r

Notice what happens as r changes:

As r approaches...Variance approaches...Interpretation
∞ (infinity)μ (the mean)Distribution becomes Poisson
0 (zero)∞ (infinity)Extreme overdispersion, very boom-or-bust
1-10 (typical)Between μ and highNormal sports props range

Visual Comparison: How r Affects Shape

When we hold the mean constant at μ = 1 and vary only r:

r ValueDistribution Behavior
r = 0.5Very spread out, high probability at 0 and 3+
r = 1.0Moderately spread
r = 2.0Somewhat clustered
r = 5.0More clustered, approaching Poisson
r → ∞Converges to Poisson (variance = mean)

Note

For sports props, typical r values range from 0.5 to 10. Lower r means more boom-or-bust behavior, which is common for:

  • Backup players with inconsistent opportunity
  • Touchdown-dependent receivers
  • Streaky shooters in basketball
  • Power play specialists in hockey

Poisson vs. Negative Binomial: Side-by-Side

Let's compare the two distributions with identical means:

Setup: μ = 1.0 (both distributions expect 1 occurrence on average)

Outcome (k)Poisson (VMR=1.0)Neg. Binomial (r=1, VMR=2.0)
P(0)36.8%50.0%
P(1)36.8%25.0%
P(2)18.4%12.5%
P(3)6.1%6.3%
P(4+)1.9%6.2%

Warning

Notice the pattern:

  • At k=0: Negative Binomial is 13+ percentage points higher (fatter left tail)
  • At k=1: Negative Binomial is 12 percentage points lower (thinner middle)
  • At k=3+: Negative Binomial is higher (fatter right tail)

This is the signature of overdispersion: more mass at extremes, less in the middle.

Why the "Fatter Tails" Matter for Betting

When you use Poisson for a boom-or-bust player:

  1. You underestimate P(0): Missing value on "Under 0.5" or "No TD" props
  2. You overestimate P(1): Potentially betting -EV on "exactly 1" props
  3. You underestimate P(3+): Missing long-shot value at +800 to +1500 odds

The Negative Binomial corrects these systematic errors by properly accounting for the player's volatile nature.

Key Takeaway

Key Insight

The Negative Binomial distribution adds one parameter (r) to Poisson's single parameter (μ). This extra flexibility allows it to model boom-or-bust players whose outcomes are more variable than their average suggests. The lower the r value, the more "boom-or-bust" the player's profile.

Negative Binomial Calculator

Try the interactive calculator for this concept

Open Tool

📝 Exercise

Instructions

A player has μ = 0.80 touchdowns per game. Using the variance formula Variance = μ + (μ²/r), calculate the variance for different values of r.

If μ = 0.80 and r = 2.0, what is the variance?