Poisson Distribution for Count Props: Touchdowns, Strikeouts, Home Runs | DMP Learn
Definition
The Poisson Distribution for Count Props in sports betting how to model discrete, countable events like touchdowns, home runs, and strikeouts using the Poisson distribution.
Think of it this way
Poisson is like counting lightning strikes in a storm. You know the average rate, and from that single number you can calculate the probability of seeing exactly 0, 1, 2, or 3 strikes in any given hour.
The Poisson Distribution for Count Props
What You'll Learn
The Poisson distribution models rare, discrete events — things you count in whole numbers like touchdowns, home runs, and strikeouts. When the normal distribution doesn't fit because outcomes are small integers rather than continuous values, Poisson steps in. This lesson teaches you when to use it, how to calculate probabilities with it, and where it breaks down.
Why You Need a Different Tool for Count Stats
The normal distribution assumes outcomes spread symmetrically around an average and can take any value (including decimals). That works for passing yards or assist totals where a player might put up 7.5-ish assists and individual games cluster smoothly around the mean.
But some stats don't work that way. A player either scores 0, 1, 2, or 3 touchdowns — never 1.7. Home runs are 0, 1, or rarely 2. These are discrete counts of relatively rare events, and they produce skewed distributions where zero is the most common outcome.
The Poisson distribution was designed for exactly this situation: modeling the probability of a specific count occurring when the average rate is known.
When Does Poisson Apply?
The Poisson distribution works well when three conditions are met:
- The outcome is a count of events (0, 1, 2, 3...)
- Events are relatively independent (one occurrence doesn't dramatically change the probability of the next)
- The average rate (lambda) is known or estimable
Good Poisson candidates:
- Pitcher strikeouts (especially 4-8 K range)
- Anytime touchdown scoring (expected TDs per game)
- Home runs per game
- NHL goals per player
- Soccer goals per player
When Poisson starts to struggle:
- When a player's variance is much higher than their mean (we'll cover this as "overdispersion" in the VMR lesson)
- When the count is large enough that the normal distribution works just as well (a pitcher averaging 10+ Ks is roughly normal)
- When events aren't independent (a big inning in baseball where one HR changes the pitcher's approach)
The Core Concept: Lambda (λ)
Lambda is the average rate of occurrence — the one number that drives the entire Poisson distribution. If a running back scores an average of 0.6 touchdowns per game, his lambda is 0.6.
Once you know lambda, you can calculate the probability of any specific count:
Poisson Probability Formula:
P(X = k) = (λ^k × e^(-λ)) / k!
Where:
- λ = average rate (lambda)
- k = the specific count you're calculating probability for
- e = 2.71828 (Euler's number)
- k! = factorial of k (e.g., 3! = 3 × 2 × 1 = 6)
Don't worry if the formula looks intimidating — spreadsheets handle this instantly.
Step-by-Step: Anytime Touchdown Scorer
Scenario: A running back's touchdown line is "Anytime TD Scorer: Yes -130 / No +110." You estimate his expected touchdowns this game at 0.72 based on red zone carries, goal-line usage, and matchup.
Step 1: Set lambda.
λ = 0.72 (his expected TD rate for this game)
Step 2: Calculate P(0 touchdowns).
P(X = 0) = (0.72^0 × e^(-0.72)) / 0!
P(X = 0) = (1 × 0.4868) / 1
P(X = 0) = 48.7%
Step 3: Calculate P(at least 1 touchdown).
P(X ≥ 1) = 1 - P(X = 0) = 1 - 0.487 = 51.3%
Step 4: Compare to the sportsbook's implied probability.
"Yes -130" implies about 56.5% probability (after removing vig). Your model says 51.3%. This bet is actually -EV — the book is overcharging relative to the true probability. Pass.
Step 5: What if lambda were higher?
If the same player had λ = 0.95 (heavy goal-line role, weak run defense):
P(X = 0) = e^(-0.95) = 38.7%
P(X ≥ 1) = 61.3%
Now 61.3% vs the book's 56.5% implied — that's a potential edge.
Calculating Exact Count Probabilities
Poisson lets you calculate the probability of any specific count. Here's a table for a pitcher with λ = 6.5 expected strikeouts:
| Strikeouts | Probability |
|---|---|
| 0 | 0.2% |
| 1 | 1.0% |
| 2 | 3.2% |
| 3 | 6.9% |
| 4 | 11.2% |
| 5 | 14.5% |
| 6 | 15.8% |
| 7 | 14.6% |
| 8 | 11.9% |
| 9 | 8.6% |
| 10+ | 12.2% |
For an Over 6.5 strikeouts line:
P(Over 6.5) = P(7) + P(8) + P(9) + P(10+) = 14.6% + 11.9% + 8.6% + 12.2% = 47.3%
P(Under 6.5) = 52.7%
If the sportsbook prices Over 6.5 Ks at +105 (implied 48.8%), and your Poisson model says 47.3%, the under is the better side.
How to Estimate Lambda
Lambda should reflect the expected rate for this specific game, not just the season average. Adjust for:
For pitcher strikeouts:
- Season K/9 rate as baseline
- Opponent's team strikeout rate (K%)
- Home/away splits
- Recent form (last 4-5 starts)
- Weather conditions (wind, temperature affect ball movement)
For touchdown props:
- Red zone opportunities per game
- Goal-line carry share or target share
- Opponent's red zone defense ranking
- Game total (higher-scoring game scripts = more TD chances)
- Recent usage trends
For home runs:
- Season HR rate per plate appearance
- Ballpark factor
- Pitcher's HR/FB rate
- Weather (temperature, wind direction)
- Platoon advantage (lefty vs righty)
Practical shortcut: If you don't have time to build a full model, use the player's rate over their last 10-15 games in similar contexts as a quick lambda estimate. It won't be perfect, but it's far better than using the raw season average without adjustment.
Spreadsheet Formulas
In Excel or Google Sheets, you don't need to calculate Poisson by hand:
Exact probability of k events:
=POISSON.DIST(k, lambda, FALSE)
Cumulative probability (k or fewer events):
=POISSON.DIST(k, lambda, TRUE)
P(Over k):
=1 - POISSON.DIST(k, lambda, TRUE)
Example: P(Over 6.5 strikeouts) with λ = 6.5:
=1 - POISSON.DIST(6, 6.5, TRUE) = 47.3%
(Use 6, not 6.5, because Poisson only takes integers and "over 6.5" means "7 or more," which equals 1 minus "6 or fewer.")
Where Poisson Breaks Down
The Poisson distribution assumes that the variance equals the mean. This is called equidispersion. When it holds, Poisson works cleanly. When it doesn't, you get misleading probabilities.
Overdispersion is when the variance is significantly larger than the mean. This happens with "boom-or-bust" players — guys who either go off for 3+ of something or put up zero, with less middle ground than Poisson predicts.
How to check: calculate the Variance-to-Mean Ratio (VMR) from the player's game log.
- VMR ≈ 1.0 → Poisson fits well
- VMR > 1.3 → The player is overdispersed. Poisson will underestimate both the probability of zero and the probability of big games. The Negative Binomial distribution handles this better. (Covered in the VMR lesson.)
Example: Two players both average 0.8 touchdowns per game.
- Player A: Scores in 55% of games, rarely more than 1. VMR ≈ 1.05. Poisson fits.
- Player B: Scores in 40% of games, but has multi-TD games when he does. VMR ≈ 1.6. Poisson overstates his anytime TD probability because it doesn't account for the clumping.
Common Mistakes
Applying Poisson to continuous stats. Passing yards, points, rebounds — these aren't counts of rare events. Use the normal distribution instead.
Using season averages without context adjustment. A running back's TD lambda should reflect this week's game script, opponent, and usage — not his 17-game season average.
Ignoring overdispersion. If your Poisson model consistently overestimates "anytime" probabilities for boom-or-bust players, the player is likely overdispersed. Check the VMR before trusting Poisson for volatile players.
Forgetting that Poisson is for whole numbers. When calculating P(Over 6.5), use P(X > 6) = 1 - P(X ≤ 6). Don't try to plug 6.5 into the Poisson formula.
How DMP Handles This
DMP's prop analysis accounts for the discrete nature of count-based stats. When you see fair probabilities for strikeout props, HR props, or TD scorer markets on DMP, the underlying calculations reflect the appropriate distribution for the stat type — not a one-size-fits-all approach.
The platform's 14-signal scoring pipeline evaluates each prop's statistical properties as part of identifying +EV candidates, including whether the stat's variance structure supports reliable modeling.
Key Takeaways
- Poisson models discrete, countable events (TDs, HRs, Ks) where the normal distribution doesn't apply.
- Lambda (λ) is the single input: the expected rate for this specific game.
- P(at least 1) = 1 - e^(-λ) is the quick formula for anytime scorer markets.
- Always adjust lambda for matchup context rather than using raw season rates.
- Check the Variance-to-Mean Ratio — if VMR > 1.3, the player is overdispersed and Poisson will mislead you.
- Use POISSON.DIST in spreadsheets to avoid manual calculation.
Next lesson: Which Distribution Should You Use? →
How DMP uses this
DMP's prop analysis accounts for the discrete nature of count-based stats. The platform's 14-signal scoring pipeline evaluates each prop's statistical properties, including whether the stat's variance structure supports reliable modeling.
Common mistake
Applying Poisson to continuous stats like passing yards or points. Poisson is for discrete counts of rare events (TDs, HRs, Ks). Also, ignoring overdispersion — if VMR > 1.3, Poisson will mislead you on boom-or-bust players.
After this lesson
You can calculate probabilities for any count-based prop using the Poisson distribution, and you know when to check for overdispersion using the VMR.