Shots on goal is a definition problem
Shots on goal is the oldest quantitative habit in hockey. It is on every scoreboard, in every intermission report, and underneath two of the sport’s most cited rates — shooting percentage and save percentage — as the denominator that makes them possible. It is also a stat whose definition is far narrower than its name suggests, applied by human scorers whose standards vary by building, and blind to the single most important thing about a shot. Before arguing about who won the shot count, it is worth knowing what the shot count agreed to count.
The definition is exclusionary
A shot on goal is a puck directed at the net that either enters it or is stopped by the goaltender. That is the whole rule, and it excludes three large categories. Blocked shots are not shots on goal; they never reached the goaltender. Missed shots are not shots on goal; they were wide or high. And a puck that beats the goaltender cleanly and strikes the post or crossbar is not a shot on goal either, because it was neither saved nor scored. The best scoring chance of a period can be recorded as nothing at all, while a soft backhand from the half wall that the goalie catches with his glove without moving is a full unit.
The crossbar problem is not a curiosity
It is tempting to dismiss the post case as rare. It is not rare enough to ignore, and it is systematically biased. Shots that hit iron are, on average, better shots than shots that get saved — they were aimed at the parts of the net that beat goaltenders. Excluding them removes disproportionately good attempts from both the numerator of a shooter’s volume and the denominator of a goaltender’s save rate. The goaltender who is beaten five times off the post in a month is rewarded by the stat for having been beaten. That is not a rounding error. That is the definition pointing the wrong direction.
The exclusion also distorts how a game reads in real time. A team can spend a period generating chances that clang off iron, sail an inch wide, and get blocked by a collapsing box, and walk to the intermission trailing the shot count while having thoroughly outplayed the opponent. Nothing about that period was ambiguous to anyone watching. The scoreboard stat simply had no slot to record it in, and the postgame conversation then treats the shot count as the objective record of a period it failed to describe.
Scorers, and the buildings they work in
Shots on goal are recorded by off-ice officials, one crew per arena, and their standards drift. Researchers have documented persistent building-level effects for decades: some rinks credit shots at rates several percent above league norm, some below, and the gaps are stable enough that analysts adjust for them before publishing rate stats. The mechanism is mundane. Deciding whether a deflected puck was a shot, whether a wraparound that trickled into a pad was directed at the net, or whether a point drive that grazed a stick was a shot or a blocked attempt requires judgment, and judgment made by different crews produces different totals. Venue-adjusted numbers exist for exactly this reason, and the raw scoreboard figure is not one of them.
Every shot is worth one
The larger issue is the one hockey analytics spent fifteen years fixing. A shot from the slot with a screen and a shot from the boards at a bad angle are the same event to this stat. League-wide, all-situations shooting percentage sits in the neighborhood of nine to ten percent, but that average conceals a range from well under five percent on low-danger perimeter attempts to well over twenty on the highest-danger chances. A team that outshoots an opponent 38–24 while taking its shots from the perimeter has not necessarily created more. It has directed more pucks toward a goaltender who was happy to see them.
What the shot-attempt family fixed, and what it did not
Corsi and Fenwick were built on precisely this complaint, though from the other side. Their insight was that the exclusions — blocks and misses — carry information, and that counting all attempts produces a bigger, more stable sample that predicts future goals better than goals themselves do over short runs. That was correct and it changed the sport. But shot-attempt metrics accepted the same quality-blindness in exchange for volume, which is why the next generation of public work moved to location models and expected goals. Each generation traded one flaw for another. None of them removed the scorer.
It is worth being precise about what survived the transition. Shot volume still predicts goals, because attempts are the raw material of scoring and no team scores from possessions it does not shoot on. What did not survive is the idea that the shot count settles who deserved to win. Volume tells you about territory and puck management. Danger tells you about the chances. A model that has both beats either one alone, and the scoreboard number has only the first.
The player-level version is a role stat
Shots per game at the individual level measures opportunity and assignment at least as much as shooting instinct. A defenseman quarterbacking a first power-play unit accumulates attempts from the blue line by job description. A winger parked at the net front deflects pucks and generates few recorded shots of his own while being central to the chances created. A high-minute forward on a possession-dominant line will out-shoot a more talented player buried in defensive-zone starts. Reading a shot total as an aggression rating gets all three of those backwards.
Goaltending inherits the same distortion in reverse. Save percentage divides saves by shots on goal, so a goaltender who faces forty perimeter attempts is measured on the same scale as one who faces twenty-five from the slot, and the first will post the better number for having had the easier night. That is the whole reason goals-saved-above-expected models exist. They are not a refinement of save percentage so much as an admission that its denominator was never comparable across goaltenders in the first place.
How to read it on StatLine
StatLine’s NHL player tables carry shots alongside games, goals, assists, and ice time, which is the right pairing: the shot column is only interpretable next to the minutes and the role that produced it. Convert to a per-60 basis before comparing players, check the ice-time column before crediting a volume shooter, and treat single-game shot differentials the way you would treat a single-game possession edge — as weak evidence that regresses hard. Over a season, sustained shot volume is meaningful. Over a night, it frequently describes a team that lost.
The honest read
Shots on goal is not a broken stat so much as a badly named one. It counts a specific, narrow class of events — pucks the goaltender handled or failed to — and excludes blocks, misses, and the shots that came closest to going in. It is recorded by people, and where those people work affects the total. It weights a tap-in attempt and a harmless perimeter drive identically, and it sits underneath the save and shooting percentages that get used to judge careers. Use it as a rough proxy for territorial play, insist on per-60 rates and venue adjustment when precision matters, and remember that the puck off the crossbar, the best shot of the night, went into the record book as nothing.