NHL Analytics Blog · Aug 28, 2026 · 8 min read
What expected goals measure and miss, why the confirmed goalie moves an NHL line, shot share, special teams, puck line versus moneyline and overtime grading.
By Sportslyx · Published
Hockey is the highest-variance major sport, and the reason is arithmetic: goals are rare. A team can outshoot its opponent two to one and lose to a hot goaltender and a deflection, and a month of results can say very little about which team is better. Analyzing a hockey game with numbers therefore means separating the process that persists — shot volume, shot quality, special teams — from the results that do not, and giving the goaltender the weight the market gives him. This guide covers expected goals, shot share, goaltending, rest, special teams, the puck line and how overtime is graded, then explains how Sportslyx reads an NHL game.
As everywhere on Sportslyx, the win probability on a hockey game page is market-derived: the moneyline pair is de-vigged into a fair probability, and the analysis explains that number rather than replacing it.
Expected goals (xG) assigns every unblocked shot a probability of becoming a goal based on where it was taken, the shot type, whether it followed a rebound or a rush, and similar public details, then sums those probabilities. A team with 2.8 xG created chances that would usually produce close to three goals. Its value is that it describes chance quality rather than the bounce: a game in which one side generated 3.5 xG and lost 2-1 was a game that side mostly controlled. Over a season, xG for and against predicts future goal difference better than actual goal difference does, which is why an analyst reads the xG table before the standings.
Three limits are worth stating. Public xG models see shot location and type but not the goaltender’s sightline, a screen, or a pass across the slot that moved him, so two shots with the same xG can differ enormously in real danger. A single game’s xG is a description of that game, not a verdict — finishing talent is real but small, and one night’s shooting percentage is nearly all noise. And xG says nothing about who stops the shots, which is where hockey analysis has to go next.
Before xG, analysts used shot-attempt share — the percentage of all attempts, or all unblocked attempts, a team takes while on the ice at even strength — and it remains useful because it is measured on dozens of events a game rather than a handful of goals. A team that consistently takes 54% of attempts is controlling play; one at 46% is being controlled, whatever its record. Shot share and xG share together answer the first question about a team: does it drive play, or does it depend on its goaltender and its shooting luck?
Save percentage is simple and blunt: saves divided by shots faced. A league-average goaltender stops a little over ninety percent of shots, and the difference between a .920 goalie and a .900 goalie on thirty shots is 0.6 goals per game — enormous in a sport decided by one goal more often than any other. It is also noisy: save percentage takes a season or more to stabilize, and a backup’s .930 over ten starts is mostly luck. Goals saved above expected corrects for shot quality by comparing goals allowed with the xG of the shots faced, which separates a goalie behind a bad defense from a bad goalie.
The market prices an NHL game against an expected goaltender, and the starter is often not confirmed until the morning skate or an hour before puck drop. When the expected starter is replaced by his backup, the fair probability moves by several points in one step. Suppose a home side opens at -135 with the visitors at +115: implied 57.4% and 46.5%, summing to 103.9%, de-vigged to 55.3% home and 44.7% away. The backup is announced and the line moves to -105 home and -115 away — fair probabilities of 48.9% and 51.1%. A six-point swing from one name, with nothing about the skaters changed.
Hockey’s schedule is dense, and the second game of a back-to-back is the clearest situational disadvantage in the sport: the team is tired, and it usually starts its backup goaltender, which folds directly into the point above. The effect is larger when the opponent is rested and when the second game follows travel. None of this is decisive, and the market knows all of it, but it explains a good deal of why a weaker team is sometimes a favorite. Rest days for both sides are part of the Sportslyx NHL package.
A league-average power play converts roughly one chance in five, and a meaningful share of all goals are scored on it, so a matchup between a strong power play and a weak penalty kill is a real difference in expected goals — provided penalties are actually called. Penalty differential, how often a team draws penalties minus how often it takes them, is the underrated half of the picture. Special-teams percentages are noisy over a month, so read them alongside power-play shot rates.
Two markets, two different questions.
| Market | What it asks | What to know |
|---|---|---|
| Moneyline | Which team wins, including overtime and the shootout | The direct expression of the matchup, and where goaltender news lands first |
| Puck line (±1.5) | Does the favorite win by two or more goals | Cannot be covered in overtime or a shootout, where the margin is one by rule; empty-net goals late in regulation make two-goal wins more common than the flow of play suggests |
| Total | Combined goals over or under a number | Set around the two goaltenders and both teams’ shot rates; a shootout adds exactly one goal to the winner’s total |
A moneyline favorite at -200 may be a plus price at -1.5 because so many one-goal games end in overtime. The empty-net effect runs the other way: a team protecting a one-goal lead in the last two minutes often faces an opponent that has pulled its goalie, and a goal into the open net turns a one-goal win into a two-goal cover. Reading the two markets together tells you what shape of game the market expects.
Regular-season NHL games tied after sixty minutes go to a five-minute three-on-three overtime and then a shootout; playoff games play full overtime periods until someone scores. The main market — the moneyline — settles on the final result including overtime and the shootout, and a shootout win is recorded as a one-goal win in the official score. On Sportslyx a moneyline read is graded the same way: the final result, overtime and shootout included, matching how the market settles the game. A postponed game is void. Because so many games are decided after regulation by something close to a coin flip, a month of hockey reads is thin evidence.
Every NHL game on the board follows the same pipeline. The moneyline pair is de-vigged into a fair win probability — one number, shown on the Overview, the Matches board and inside the breakdown. A per-game package is assembled from the public feeds: the current moneyline, puck line and total and their movement, each club’s goaltenders with games played, save percentage and goals-against average, goals and shots for and against, a season-level expected-goals share for each team taken from a public model (a season aggregate, not shot-by-shot xG for the game), recent form, head-to-head history and rest days. The package is scored with a data-quality tier — rich, partial or weak — and passed to the language model with the probability. The model is asked to name the likely starter for each side, to treat goaltending as the primary lever, and to say what the market appears to be pricing and what it may be underweighting.
The company board publishes one read per game — the market favorite on the moneyline — and grades it after the final result; the model’s job is the explanation, not the estimate.
A moneyline read on Sportslyx settles on the final result including overtime and the shootout; a postponed game is void. Results aggregate over one calendar month in UTC on the pick tracker, with hit rate excluding pushes and ROI at a flat stake per read. None of this is advice, and no outcome is certain — it is a method that can be checked. The NHL predictions board shows tonight’s games with their market-derived probabilities and goaltending context, and the NHL statistics hub holds the shot, special-teams and goaltending numbers this guide refers to.
Because the market priced the game against an expected starter, and the gap between a starter and his backup on thirty shots is a measurable fraction of a goal in a sport decided by one goal more often than any other. When the backup is confirmed the moneyline moves several points, and the Sportslyx probability, which is derived from that line, moves with it.
Not on its own. A single game’s xG describes the quality of chances each side created; it does not see screens, sightlines or the goaltender, and one night’s finishing is mostly noise. Its value is over many games, where xG for and against predicts future goal difference better than actual goals do.
A moneyline read settles on the final result including overtime and the shootout, matching how the market settles the game; a shootout win counts as a one-goal win in the official score. A postponed game is void. Results are published by calendar month with hit rate excluding pushes and ROI at a flat stake.
Take this guide to the NHL board: the 7-day Pro trial opens the AI breakdowns, the market-derived probabilities, the daily company board, Combo Builder, Studio and the pick tracker.
7-day trial · $0 today · card required · cancel anytime before the trial ends
Sportslyx produces statistical estimates and analytical breakdowns for information and entertainment. Nothing on this page is betting advice, and no outcome is certain. 18+ only. Please play responsibly — see our Responsible Play page.
