Hockey GSAx: Goals Saved Above Expectation and Model Limits
Define GSAx as xGA minus goals allowed, work through an example, and separate the metric from save percentage and GSAA.
What GSAx means
Goals Saved Above Expected (GSAx) compares expected goals against a goalie with the goals he actually allowed. This article uses one explicit sign convention: GSAx = xGA − GA, where xGA is the model’s expected goals for shots faced by that goalie and GA is the goals allowed on exactly the same set of shots. A positive value means fewer goals allowed than the selected model expected.
This is not a standard box-score total; it is derived from a shot-quality model. A model assigns goal probabilities to individual shots and sums them. Different models may use different inputs and produce different xGA values. Check which events are included and which sign the table uses: labels or sign conventions can vary between interfaces.
A numerical example
If a goalie faced shots worth 20.0 expected goals and allowed 16 goals on precisely that same sample, GSAx = 20.0 − 16 = +4.0. With 16 xGA and 20 actual goals allowed, the result would be −4.0. This compares an outcome with one model’s expectation; it does not mean four goals were certainly prevented or predict the next game.
The shot set matters. Do not assign a team’s total xGA to one goalie if a relief goalie faced some of those shots. Empty-net goals should not be included in the expectation for a goalie who was not facing them; if the source reports them separately or includes them under a special rule, say so explicitly. The event sample, periods, and goalie must match between xGA and GA.
Do not confuse it with save percentage or GSAA
Save percentage divides saves by shots on goal but does not account for how dangerous those shots were. GSAx tries to adjust for shot quality, yet it depends on the model and event coding. Our guide to goalie saves versus shots on goal explains why one percentage does not describe the whole performance.
GSAx is also different from Goals Saved Above Average (GSAA), which generally compares a goalie’s results with a league-average baseline rather than expected goals from a specific xG model. The labels sound alike, but the baselines differ. Check the source formula instead of inferring the definition from the abbreviation.
Using it around a hockey market
A high cumulative GSAx may prompt a closer look at a goalie’s performance, but first check sample size, minutes, manpower state, and the confirmed starter. Before a game-total or goalie-saves market, review the starting goalie and the match line: an unknown or changed starter changes the comparison itself.
A model does not fully describe screens, deflections, rebounds, goalie movement, or defensive actions. A small sample can swing after a few goals, and figures from different models are not directly interchangeable. Do not turn GSAx into a win probability or promise; it is a retrospective estimate relative to a defined model.
Before drawing a conclusion, state the formula, sign convention, sample, goalie identity, and empty-net treatment.