Loading...
Traditional stats like goals and assists only tell part of the story. Advanced analytics reveal which teams actually control play, generate quality chances, and are due for a hot or cold streak. Here's your starting point.
TL;DR
Four stats cover most of what you need: Corsi and Fenwick measure possession through shot attempts, expected goals (xG) measures shot quality, and PDO measures luck. A team that dominates the first three and has a normal PDO is legitimately good. A team that wins despite poor numbers in those areas is living on borrowed time.
Hockey is a low-scoring sport with a lot of randomness. A lucky bounce, a hot goalie, or a single deflection can decide a game. Traditional box score stats like goals, assists, and plus-minus are heavily influenced by that randomness, which makes them unreliable over small samples.
Advanced stats solve this by measuring the process instead of the outcome. They ask: which team is generating more chances? Which team is getting better looks? Is this winning streak sustainable, or is it built on an unsustainably high shooting percentage? The answers are better predictors of future results than the standings alone.
Corsi counts all shot attempts at 5-on-5: shots on goal, missed shots, and blocked shots. If your team has 60 of the 100 total shot attempts in a game, your Corsi is 60%.
Named after Jim Corsi, a former NHL goalie coach who tracked shot attempts.
Fenwick is Corsi minus blocked shots. It counts shots on goal plus missed shots at 5-on-5. The idea is that blocked shots can be a strategy (some teams block more by design), so removing them gives a cleaner possession signal.
Named after Matt Fenwick, a hockey blogger who proposed the adjustment.
Both stats answer the same core question: which team is spending more time on offense? Teams that consistently out-attempt their opponents tend to outscore them over time. Corsi and Fenwick don't care about shot quality, though. That's where expected goals come in.
Not all shots are created equal. A one-timer from the slot is worth way more than a point shot through traffic. Expected goals (xG) captures this by assigning a scoring probability to every shot based on location, shot type, whether there was traffic in front, and the game situation.
Add up all those probabilities and you get a team's total xG for a game. If a team generated 3.2 xG but only scored 1 goal, they got unlucky. If they generated 1.1 xG and scored 4, they got very lucky. Over time, actual goals tend to converge toward expected goals.
xG is widely considered the most predictive single stat in hockey because it measures both quantity and quality of chances.
PDO is the simplest luck indicator in hockey. It adds a team's shooting percentage (how often their shots go in) to their save percentage (how often their goalie stops shots). The league average is always 100 (or 1.000, depending on the scale).
A team with a PDO of 103 is getting great shooting and goaltending at the same time. That combination is extremely hard to sustain. Over the course of a season, almost every team regresses toward 100. A high PDO team that also dominates Corsi and xG is the real deal. A high PDO team with bad underlying numbers is a house of cards.
PDO quick reference
PuckCast's prediction model doesn't rely on any single stat. It combines 154 features spanning possession, shot quality, goaltending, special teams, rest, and travel into an ensemble model trained on 16 NHL seasons.
Corsi and Fenwick differentials feed into the possession picture. Rolling xG windows at 3, 5, and 10 games capture both current form and underlying quality. PDO signals help the model identify teams that are due for regression. The result is a win probability for every game that reflects what's actually happening on the ice, not just what the scoreboard says.
In the model
The four stats most analysts start with are Corsi (all shot attempts), Fenwick (unblocked shot attempts), expected goals (xG, which measures shot quality), and PDO (shooting percentage plus save percentage). Together they tell you whether a team controls play, generates quality chances, and is benefiting from luck.
Winning with poor underlying numbers usually means a team is riding hot goaltending or an unsustainable shooting percentage. Over a full season, teams that dominate possession and shot quality tend to hold their records. Teams that don't tend to regress. Advanced stats help you see that correction coming before it shows up in the standings.
At 5-on-5, anything above 52% is strong and above 54% is elite. The league average is 50% by definition since every shot attempt for one team is a shot attempt against another. Teams that consistently sit above 52% are controlling play and spending more time in the offensive zone.
PuckCast's ensemble model uses 154 features built from advanced stats like rolling Corsi differentials, expected goals at multiple game windows, high-danger chance rates, PDO regression signals, and goalie quality metrics. The model weighs these underlying indicators alongside traditional stats to produce win probabilities for every NHL game.