Back to The Negative Binomial Distribution
Chapter 9Calculator

Estimating the Dispersion Parameter

Calculating the r parameter from data

Estimating the Dispersion Parameter (r)

Once you've determined that a player's data is overdispersed (VMR > 1.3), the next step is to estimate the dispersion parameter r. This value captures the player's inherent boom-or-bust nature and is essential for calculating Negative Binomial probabilities.

The Method of Moments Formula

Dispersion Parameter (r)

r = μ² / (Variance - μ)
Excel: =AVERAGE(range)^2 / (VAR.S(range) - AVERAGE(range))

This formula derives from rearranging the Negative Binomial variance equation:

Variance = μ + (μ² / r)

Solving for r:
Variance - μ = μ² / r
r = μ² / (Variance - μ)

Warning

Important: This formula only works when Variance > Mean (VMR > 1.0). If variance equals or is less than the mean, the denominator becomes zero or negative, and you should use Poisson instead.

Worked Example: Boom-or-Bust Running Back

Let's calculate r for Running Back B from the previous lesson:

Game Log: 2, 0, 0, 3, 0, 0, 2, 0, 3, 0

MetricCalculationValue
Mean (μ)Sum/Count = 10/101.00
VarianceVAR.S calculation1.78
VMR1.78/1.001.78 ✓ Overdispersed

Now calculate r:

r = μ² / (Variance - μ)
r = 1.00² / (1.78 - 1.00)
r = 1.00 / 0.78
r = 1.28

Note

r = 1.28 indicates a moderately boom-or-bust player. This value will be used with μ to calculate probabilities using NEGBINOM.DIST.

Interpreting r Values

r ValuePlayer ProfileTypical Players
r < 1Extremely boom-or-bustBackup specialists, gadget players
r = 1-2Highly volatileTD-dependent WR3s, streaky shooters
r = 2-5Moderately volatileMost backup RBs, role players
r = 5-10Slightly volatileConsistent starters with occasional big games
r > 10Near-PoissonVery consistent performers

The Critical Insight: r Reflects Personality, Not Matchups

Key Insight

The dispersion parameter r reflects a player's inherent boom-or-bust nature. It doesn't change game-to-game. When adjusting for matchups, you adjust μ (the expected count), not r. Think of r as the player's "personality," and μ as their expected output in a given situation.

Example: Matchup Adjustment

Scenario: Your boom-or-bust RB (r = 1.28) faces a weak run defense.

ParameterBase ValueMatchup AdjustedWhat Changed
μ1.001.15 (+15%)✓ Increased for plus matchup
r1.281.28 (unchanged)✗ Personality stays constant

The player is still boom-or-bust (same r), but you expect more TDs on average (higher μ) due to the favorable matchup.

Excel Implementation

Complete Formula Set

// In cells with historical game data in B2:B21 (20 games)

Mean (μ):
=AVERAGE(B2:B21)

Variance:
=VAR.S(B2:B21)

VMR:
=VAR.S(B2:B21)/AVERAGE(B2:B21)

r (Dispersion Parameter):
=AVERAGE(B2:B21)^2 / (VAR.S(B2:B21) - AVERAGE(B2:B21))

// Error handling version (returns "Use Poisson" if not overdispersed):
=IF(VAR.S(B2:B21)>AVERAGE(B2:B21), 
    AVERAGE(B2:B21)^2/(VAR.S(B2:B21)-AVERAGE(B2:B21)), 
    "Use Poisson")

Sample Size Requirements

Warning

Minimum Data Requirement: You need at least 15-20 games of data to reliably estimate r. With smaller samples, the variance estimate is too noisy to trust, and you should default to Poisson or use empirical frequencies (just counting historical outcomes directly).

What to Do with Limited Data

Games AvailableRecommended Approach
1-5 gamesUse Poisson with qualitative adjustment
6-10 gamesUse Poisson, note uncertainty
11-15 gamesCalculate VMR; if clearly >1.5, consider NB
16+ gamesFull VMR and r calculation
30+ gamesHigh confidence in r estimate

Practice Calculation

Let's work through another example:

Streaky NBA Shooter (Three-Pointers Made)

Game log (20 games): 1, 4, 0, 2, 6, 1, 0, 3, 0, 5, 2, 0, 4, 1, 0, 3, 0, 5, 2, 1

Step 1: Calculate Mean

Sum = 1+4+0+2+6+1+0+3+0+5+2+0+4+1+0+3+0+5+2+1 = 40
Mean = 40/20 = 2.00

Step 2: Calculate Variance Using Excel: =VAR.S({1,4,0,2,6,1,0,3,0,5,2,0,4,1,0,3,0,5,2,1})

Variance = 3.68

Step 3: Calculate VMR

VMR = 3.68 / 2.00 = 1.84

→ VMR > 1.3, proceed with Negative Binomial

Step 4: Calculate r

r = μ² / (Variance - μ)
r = 2.00² / (3.68 - 2.00)
r = 4.00 / 1.68
r = 2.38

Interpretation: This shooter has moderate boom-or-bust tendencies (r = 2.38). They're not as volatile as our touchdown RB (r = 1.28), but still overdispersed enough to warrant Negative Binomial modeling.


📝 Exercise

Instructions

A tight end has the following touchdown data over 16 games:

0, 1, 0, 2, 0, 0, 1, 0, 0, 2, 0, 1, 0, 0, 3, 0

Calculate the mean, variance, VMR, and dispersion parameter r.