residuals

The difference between each observed y value and the predicted value (y - predicted). Residuals sum to approximately zero for a correctly fitted model.

Returns a defensive copy — modifications to the returned array do not affect this result.