Home/Soccer Analytics/GSAA Shot-Type Decomposition

GSAA Shot-Type Decomposition

Goalkeeper AnalysisLevel 3 — Advanced

What It Is

Splitting Goals Saved Above Average (GSAA) into 7 distinct shot-type bins, each probing a different goalkeeper skill. Standard GSAA lumps all shots together, masking a goalkeeper's specific strengths and weaknesses. By decomposing into headers, close-range 1v1s, long-range 1v1s, long-range foot shots to corners, close-range central foot shots, angled central foot shots, and shots after crosses, analysts can identify exactly which shot-stopping sub-skill a goalkeeper excels or struggles at. Four shot categories are excluded from the model: own goals (accidental), penalties (separate skill, insufficient sample), deflections (too noisy even for post-shot xG models), and forced-out-of-position shots (should penalize cross-claiming, not shot-stopping).

Correct Execution

(1) Exclude: own goals, penalties, deflections (model can't accurately value them — velocity/direction change not captured), forced-out-of-position shots (fast cutbacks/rebounds where GK physically can't be positioned). (2) Bin remaining shots into 7 types:

  • Headers: probes set position height + gaze adjustment (switching focus from cross flight to header direction) + reflexes
  • Close-range 1v1: probes rushing speed + premeditated barrier shape (spread, don't react — no time)
  • Long-range 1v1: probes backward speed + decision-making (rush out vs. hold position) + reaction trust
  • Long-range foot shots to corner: probes power-step footwork (step toward ball, push off, dive)
  • Close-range central foot shots: probes collapse/negative-step dive (sweep standing leg away, drop with gravity — no time for power step)
  • Angled central foot shots: probes foot saves + lateral split positioning
  • Shots after crosses: probes repositioning speed + awareness + footwork reset after initial cross movement

(3) Compute GSAA per bin using post-shot xG. (4) Build a 7-axis radar dedicated entirely to shot-stopping.

Progression Levels

Diagnostic Tree

Coaching Cues

  • "Overall GSAA is a B+. But on headers? It's a D."
  • "The model says he cost us half a goal on that deflection. The model is wrong."
  • "Joe Hart's problem isn't diving. It's his power step to the left."
  • "Shot-stopping is way too important for a third of a radar. It deserves the whole radar."

Common Errors

  1. Not excluding deflections: Post-shot xG models handle deflections poorly — they penalize goalkeepers for shots that changed direction unpredictably.
  2. Not excluding forced-out-of-position shots: A fast cutback creates an unsaveable situation from a shot-stopping perspective. This should penalize cross-claiming, not shot-stopping.
  3. Using overall GSAA for recruitment: A goalkeeper with +5 GSAA overall might be +10 in 1v1 and -5 in headers. If your team concedes mostly headers, this keeper is a bad fit.

Edges

🔑 Hidden Causal Lever

A GK's GSAA Can Swing 2+ Goals Per Season Just By Changing the Team's Defensive System

GSAA (Goals Saved Above Average) is confounded by the team's shot concession profile. A GK who is elite at saving 1v1s but average at everything else will show different GSAA depending on whether their team concedes 20% or 50% of xG from 1v1s. By decomposing GSAA into shot-type components, you can predict how a GK's GSAA would change under a different defensive system — and the swing can be 2+ goals per season, which is often the difference between relegation and safety.

What most people do
Treat GSAA as a stable individual metric, independent of the team's defensive style.
What the best do
Decompose GSAA by shot type. Compute counterfactual GSAA: "If this GK faced Team B's shot concession profile instead of Team A's, their GSAA would change by X." Use this for recruitment: a GK who looks average in one system may be elite in yours.
Why it's an edge: A GK transfer from a team with a different defensive system will show a GSAA change that has nothing to do with their ability — only the system changed. Clubs that don't account for this make systematic evaluation errors.
How to exploit: Before signing a GK, recompute their GSAA under your team's shot concession profile. If the counterfactual GSAA is significantly better (or worse) than their current GSAA, the system difference — not the GK — explains the gap.
Max Odenheim & John Harrison, LAFC, StatsBomb Conference, 2021-11-04. Shot-type GSAA decomposition with R-squared = 0.7 profile repeatability.

Sources

  • Max Odenheim & John Harrison, LAFC, StatsBomb Conference 2021, YouTube, 2021-11-04 — presented 7-bin GSAA decomposition with 4 exclusion categories; mapped each bin to specific GK techniques; Alisson 1v1 mastery; Joe Hart power-step weakness; Ederson long-range 1v1 decision flaw