Notes

Score Model

Ridge regression coefficient using all independent variables as predictors, and the score as the outcome. This corresponds 1:1 with score, e.g. a coefficient of +1 means a predicted score 1 higher than average. Ridge regression is used instead of standard linear regression because of high multicollinearity.

Win Model

Logistic regression coefficient using all independent variables as predictors, and whether the game was won or not as the outcome. Higher values indicate greater likelihood of winning when present in the game. For continuous variables (e.g. number of players), this number is per unit. model.

Differential

The difference between the logistic regression coefficient for this item using this Spirit, as compared to using this item in general. A negative number means that this Spirit underperforms using this item, and a positive number means that it overperforms.

Model (Diff ≥10)

Same as Win Model, but instead counts all games of difficulty 10 or higher, for all time.

Layouts

Diagrams of the layouts are on the main page. Data collected before v.2.7.1 (May 2, 2025) use the end game state, even if the island layout changed due to powers like Cast Down Into the Briny Deep. Data collected from v.2.7.1 onward uses the layout as it was at setup.