The same principle, applied to an ML model
Fictional individual: age = 45, education = Bachelors, hours_per_week = 50. Click a coalition to see what the model receives for it, or isolate a variable to see its SHAP value. The number in each square is the model's prediction.
How to read a coalition: a coalition is the set of variables the model knows for this individual. The model still needs a value for the missing variables: it takes them from the individuals of the reference dataset, one row at a time, and averages its predictions. On those variables, it knows no more than it would about an "average" individual.
What the model receives
Isolate one variable's contribution
Marginal contribution across each pair of coalitions
SHAP value (weighted average)