Jul 22 LV 99 WSH 100 FinalJul 22 CONN 88 IND 123 FinalJul 22 DAL 101 POR 97 FinalJul 28 CONN 84 WSH 92 FinalJul 28 TOR 93 MIN 100 FinalJul 28 IND 105 SEA 95 FinalJul 28 NY 113 LA 109 FinalJul 28 POR 83 LV 98 FinalJul 29 GS 61% PHX 39% Win probJul 29 ATL 50% DAL 50% Win probJul 30 NY 40% LV 60% Win probJul 30 CONN 37% CHI 63% Win probJul 30 MIN 73% TOR 27% Win probJul 31 IND 67% POR 33% Win probJul 31 SEA 28% ATL 72% Win probJul 31 DAL 56% WSH 44% Win prob
dubmetrics

Reference

Glossary

Plain-English definitions for the stats across the site. If a number on a player card or team page is unfamiliar, it is explained here.

Team ratings

How good a team is, adjusted for who it played and how fast it plays.

Net rating Points better than average per 100 possessions
Offensive rating minus defensive rating. A +6 net rating means the team outscores an average opponent by about six points per 100 possessions. Opponent-adjusted and pace-adjusted, so it is not just a record in disguise.
Offensive / defensive rating Points scored / allowed per 100 possessions
Scoring efficiency, stripped of pace. We fit them with a ridge regression that adjusts every team for strength of schedule and home court, so a gaudy offense against weak defenses gets discounted. A note on signs: in the underlying math a defensive rating is points allowed relative to average, so a negative number is actually good defense — you prevent points. To keep the site readable, and to match the convention most stats sites use, we flip the sign on display: a positive defensive rating always means a better-than-average defense, just like offense and net.
Pace Possessions per game
How fast a team plays. Pace inflates or deflates raw box-score totals, which is why we rate efficiency per 100 possessions rather than per game.

Four Factors

Dean Oliver’s four drivers of winning, in order of weight. See the team table on the Four Factors page.

Effective FG% (eFG%) Field-goal % that credits the extra value of a three
A made three is worth 1.5 made twos, so eFG% = (FGM + 0.5×3PM) / FGA. The single biggest factor in scoring efficiency.
Turnover % (TOV%) Turnovers per possession
TOV / (FGA + 0.44×FTA + TOV). On offense, lower is better. On defense, a higher number is good — it means you force opponents into giveaways.
Offensive rebound % (ORB%) Share of available offensive boards grabbed
OREB / (OREB + opponent DREB). On defense we show the offensive boards you allow, so lower is better there.
Free-throw rate (FT Rate) Free-throw attempts per field-goal attempt
FTA / FGA — how reliably a team gets to the line. On defense it is how often you send opponents there (lower is better).

Player metrics

Per-player rates, normalized for pace and minutes, then ranked against the league.

Impact (RAPM) How much a player swings the score, adjusted for teammates and opponents
Impact answers one question: how much does a player help her team win, beyond the box score? It measures how the score moves while she is on the floor, then adjusts for who she played with and against, so a player on a stacked roster does not get all the credit and a star carrying a weak lineup is not punished for it. That is why the leaderboard can look different from the scoring leaders. A passer or defender who never fills the stat sheet can still swing games. We split it three ways: offensive Impact (the points she adds), defensive Impact (the points she prevents, sign flipped so higher is always better), and net (the two combined). Technically it is a ridge regression over play-by-play, regularized because single-season on/off numbers are noisy on the WNBA’s short schedule. We pool several seasons and pull thin samples toward league average, which is why a player’s Impact trend reads as a smooth trajectory instead of bouncing around.
True shooting % (TS%) Scoring efficiency across twos, threes, and free throws
PTS / (2×(FGA + 0.44×FTA)). The most complete one-number shooting measure because it folds in free throws and the three-point bonus.
Usage % (USG%) Share of a team’s plays a player finishes
How often a player ends a possession with a shot, free throws, or a turnover while on the floor. High usage with high efficiency is the mark of a primary option.
Per-100 rates Box stats scaled to 100 possessions on the floor
Counting stats (points, rebounds, assists, steals, blocks, turnovers) divided by the player’s own possessions on court. This lets a high-minute starter and a bench player be compared on the same footing.
Percentile rank Where a player sits versus qualified peers
The bars on a player card rank each metric against all qualified players that season (a minutes floor filters out tiny samples). 90th percentile means better than 90% of the league on that stat.

Predictions

What the model says will happen — and how we check whether it was right.

How a game is simulated Team-level Monte Carlo, not an assembled lineup
A game is simulated at the team level, not by building a lineup. Each team carries an opponent-adjusted offensive rating, defensive rating, and pace, fit from every team-game this season and last. For a given matchup we work out each side’s expected points per 100 possessions and the expected pace, then play the game about ten thousand times: every run draws both teams’ efficiency from a distribution (plus a shared game pace), converts to points, and tallies the result. Win probability is just the share of runs a team wins; the projected score is the average. Because the team ratings are a minutes-weighted picture of who has been on the floor, they implicitly reflect the usual rotation — so today’s simulation already assumes a roughly normal lineup. The flip side: the game simulation is only as current as the latest box scores and does not yet adjust for tonight’s specific injuries or scratches (that lineup detail currently affects individual player projections, not the team win probability).
Win probability Model chance a team wins a specific game
From a Monte Carlo simulation: we draw each team’s scoring from its ratings thousands of times and count how often each side wins. A 65% favorite should win about 65% of the time — not every time.
Projected score Expected final score, not a guarantee
The average outcome of those simulations, converted to points at the expected pace. Real games scatter widely around it; it is a center of gravity, not a prediction of the exact final.
Player projection Expected points / rebounds / assists next game
Recency-weighted recent production adjusted for the upcoming opponent’s pace and defense. A model estimate for one game, where variance is large.
Calibration Do stated probabilities match reality?
A calibrated model that says 70% is right about 70% of the time. We track this on the Track record and Methodology pages — being calibrated matters more than being confident.