Negative Binomial Distribution
A probability distribution that models count data with more variability than Poisson allows. It adds an overdispersion parameter to handle stats where variance exceeds the mean.
Like modeling daily customer complaints: some days have none, but bad days cluster together more than pure randomness would predict.
Why it matters
Some player stats are "streakier" than Poisson predicts — a player might have many 0-block games and occasional 4-block games, and negative binomial captures that extra variance.
How DMP uses this
DMP evaluates negative binomial as an alternative to Poisson for stats where the variance-to-mean ratio is high, ensuring P(over) calculations account for real-world streakiness.
Common mistake
Defaulting to Poisson when the data is overdispersed — this underestimates the probability of both very low and very high outcomes.