GAME RESULTS
Results through the selected week
Scores show the selected team first. Opponent records include all exported games through the selected week. Completed games only. P = playoff week.
BAYESIAN MASSEY
Every rank has a probability.
Explore the rankings—and how certain they really are.
Loading ranking snapshots…
Team focus
RANK DISTRIBUTION
Where could this team rank?
All ranks shown. Hover or focus a bar for its exact probability.
SEASON TRAJECTORY
How the picture changed
Band: 90% equal-tailed rank interval. P = playoff week.
THE FIELD
Rankings with uncertainty
Teams ordered by mean rank across the full matrix. “Top 10” shows the first 10 teams in that ordering. Select a team to explore it above.
Each cell is the probability of one exact rank. Displayed ranks are not renormalized; the rest of a team’s probability may lie outside this view.
Understanding the method and these probabilities
The Bayesian Massey model estimates team strength from game margins, with hierarchical shrinkage and strength that can change over time. Each NBA snapshot is fit to that season’s completed games through the selected week, using scores from the NBA Stats API (hoopR). Regular-season, play-in and playoff games count, plus the NBA Cup championship game; preseason and All-Star games do not. Weeks are seven-day blocks counted from the first game of the regular season, and playoff weeks are counted the same way from the first play-in or playoff game.
Each source matrix has ranks in rows (rank 1 is best) and teams in columns. A team’s column sums to 100%. Mean rank is the probability-weighted average of all its possible ranks. P(top 10) sums the probabilities of ranks 1 through 10; it is not a power-poll rank, a standings position, or a probability of making the playoffs.
The R export uses draws retained inside a 90% elliptical credible set of team strengths. The dashboard summarizes those exported probabilities, not the full unfiltered posterior. Its 90% rank intervals are equal-tailed marginal intervals calculated from each exported column; they are not simultaneous guarantees across teams. Discrete intervals can contain more than 90% probability.
Rankings are relative to the teams present in each snapshot. Every season starts from scratch with no preseason information, so early-week distributions are wide, and a week with games still to be played uses only the completed games. The matrices alone do not provide sampling diagnostics or establish model calibration.
Read the model specification (background document; the retained R and Stan files describe the implemented export).