How PAR Works

What is PAR?

PAR (Points Above Replacement) is a single number that measures a player's total contribution per game above a replacement-level player. It combines shooting, volume, rebounding, assists, turnovers, steals, and free throws into one value. Positive means you're helping your team. Negative means a replacement-level player would produce more.

The model starts with play-by-play data from every WNBA game. Each shot is mapped to a court zone. Each rebound, assist, turnover, and steal is tracked to the player responsible. From there, we build per-game profiles, define a replacement baseline, and calibrate the raw numbers against actual game outcomes.

Shot Zones and Expected Points

Every field goal attempt has x/y coordinates from ESPN's shot chart API. We compute distance from the basket centroid (25.2, 2.0 in ESPN's coordinate system) and classify each shot into one of seven zones:

For each zone, we compute the league-wide field goal percentage. A player's expected points (xPts)on a shot equals the league FG% for that zone multiplied by the points attempted (2 or 3). This is the baseline: what a league-average shooter would produce from these exact spots. If a player takes 10 rim shots in a game and the league converts 62% at the rim, her expected output from those shots is 10 × 0.62 × 2 = 12.4 points. What she actually scores above or below that number is her shooting value.

The Eight Components

PAR decomposes a player's per-game contribution into eight components. Each one converts a box-score stat into a point impact estimate. The constants come from NBA analytics literature and are confirmed by possession-level analysis of WNBA play-by-play data (38,668 possessions across 262 games, all within ~10% of the NBA values).

Volume and shooting are two sides of the same coin. Volume captures how many expected points a player generates just by taking shots from certain locations. Shooting captures whether she converts better or worse than league average from those locations. Added together, they equal her actual field goal scoring. The model splits them internally so we can see why a player scores. A high-volume, low-efficiency scorer like Angel Reese has a large volume component and a negative shooting component. A low-volume sniper has the opposite profile. Both can end up with positive total contributions.

Replacement Level

The total contribution from all eight components gives a raw per-game number, but that number isn't meaningful on its own. A player averaging 12 points of contribution per game could be average or could be excellent. It depends on the baseline.

We define replacement level as follows:

1. Filter to regulars: players with 10 or more games played. This removes short-stint callups and injured players with tiny samples.
2. Sort by total contribution (all eight components summed).
3. Take the bottom 25% of those regulars and average their per-game contribution. This is the replacement-level baseline.

Conceptually, this represents a roughly 12th-roster-spot player: someone who would be available if you needed a body, but who contributes minimally. PAR is then the gap between a player's total contribution and this baseline.

Calibration

Raw PAR is box-score math. It tells us what a player does on the stat sheet, but not whether those contributions actually move the score. To answer that, we run an injury-impact regression.

The idea: when a player misses a game, how does the team's margin shift relative to what a baseline team-strength model predicted? We track player availability for every game of the season. For each game, we compute:

1. Base predicted margin from a team-level offensive/defensive rating model (adjusted for opponent and home court).
2. Actual margin (final score difference).
3. Net missing PAR for each game: the sum of raw PAR for all players missing from the away team minus the sum for all players missing from the home team.

Then we fit a single-variable OLS regression:

The resulting parScalefactor tells us how many real margin points each point of raw PAR is worth. We multiply each player's raw PAR by this scale factor to produce scaled PAR, which is the calibrated number. Scaled PAR is what appears in leaderboards and articles. It represents how many actual margin points a player is worth per game above replacement.

The 2026 Leaderboard

Here are the top 15 players in the 2026 WNBA by scaled PAR through August 18 (262 games). The Shooting column is shooting value only (efficiency vs zone average). Rebounds combines offensive (at 1.0 pts/OREB) and defensive (at 0.4 pts/DREB).

The top of the list is dominated by players who contribute across multiple categories. Wilson leads because she scores efficiently andrebounds at an elite level. Reese ranks 9th despite the worst shooting value in the top 15 because her rebounding (+7.88) is more than double anyone else's. Stewart has a negative shooting value (-0.57) and still ranks 3rd. See our Angel Reese analysis for a deep dive into how rebounding value works.

What PAR Doesn't Measure

• Individual defense beyond steals and blocks— Without lineup-level on/off data, we can't measure how a player's defensive presence affects opponent efficiency. Steals and blocks capture some of it, but contest rate, help defense, and rim protection are invisible to the model.
• No minutes weighting— PAR is per-game, not per-minute. A player who plays 35 minutes and one who plays 22 minutes are compared on the same per-game basis. This is a deliberate choice: per-minute stats can inflate low-minute players, and teams care about total contribution per game, not rate stats in a vacuum.
• No assist quality— All assists are valued equally at 0.5 points. An assist on a wide-open corner three and an assist on a contested fadeaway get the same credit. Play-by-play data doesn't capture the difficulty of the resulting shot.
• No context— Garbage time contributions count the same as clutch-time contributions. A steal in a 30-point blowout has the same value as a steal in a 1-point game. Context-weighting would require win-probability data that we don't currently incorporate.
• Season-long only— PAR is cumulative through the season. There is no rolling window or recency weighting. A player who started slowly and finished strong has the same PAR as one who was consistent throughout, assuming the same total production.

Key Constants