GamePlan Science

What is GamePlan Science?

GamePlan Science is a Bayesian analysis platform for NFL team strengths. We estimate offensive and defensive team abilities using a hierarchical model that shares information across seasons while allowing each team-season to have its own unique strength profile.

Unlike traditional power rankings or Elo systems, our approach captures full posterior distributions of team strength. This means we can answer questions like: “How likely is it that the Chiefs offense is actually top-5?” rather than just giving a single point estimate.

The Model

Our core model uses a Bayesian hierarchical framework with partial pooling. Team strengths are modeled on a logit scale, where each team has an offensive and defensive parameter for every season. These parameters are drawn from league-wide distributions, which allows us to:

• •Estimate strengths even for teams with small sample sizes
• •Quantify uncertainty with credible intervals
• •Track team trajectories across seasons
• •Simulate hypothetical matchups with proper uncertainty propagation

Game scores are modeled as a function of the offensive strength of the scoring team and the defensive strength of the opposing team, plus a random error term. This structure allows us to decompose team performance into its offensive and defensive components.

Data

We use historical NFL play-by-play data spanning from 1999 to the present. The dataset includes every regular season game, with team scores, game locations, and other contextual information. The model is fit using Hamiltonian Monte Carlo sampling to ensure robust posterior estimates.

How to use this site