Abstract

Should they have passed? A football question with 36 defensible answers

When a player shoots instead of passing, can the data tell us it was the wrong call?

It uses StatsBomb's open data from the 2023 Women's World Cup, co-hosted by New Zealand and Australia, with five matches in Perth. Every shot across all 64 matches comes with a freeze frame of where the visible players were standing. That is enough to ask what else the shooter had: which teammates were closer to goal, how well placed, and how likely the pass was to arrive.

Before you can answer "should they have passed?" you have to settle four things, and they are football questions rather than statistical ones:

• Which teammates count as a genuine alternative, anyone further forward or only those in the box.
• Whether a pass is assumed to arrive, or long balls into traffic treated as riskier.
• Whether the defenders around a teammate count when judging their chance, or just distance and angle.
• How much better the pass has to be before the shot is a mistake.

None has one right answer, so rather than pick a set I ran the analysis under every combination, thirty-six in total. Depending which you go with, shooting was the right call anywhere between 31.7% and 46.3% of the time.

Which of the four mattered most was surprising. The biggest swing came from the first, one line of code easily forgotten. The smallest from the third, the most technical.

It is all in R. I will cover the shot model, the harness that runs the specifications and reports which choices mattered, and a validation gate that refuses to write a figure the analysis did not produce. Two predictions I made went slightly the wrong way. They are included in the talk, as the argument for building the harness rather than reasoning about it.