Basic idea

  1. If an explanatory variable has a large influence on prediction for a particular instance, then its corresponding CP profile must present large fluctuations.

  2. If an explanatory variable has little influence on prediction for a particular instance, then its corresponding CP profile must present bare fluctuations and be close to the original prediction of the model.

  • The sum of differences between the profile and the prediction (across all possible values of the explanatory variable) should be close to zero.