13.6 Cross validation and log score for logistic regression

  • As in Section 11.8, we can use loo crossvalidation to compare models. In this case the log score is:

\[ \text{log score} = \sum_{i=1}^{n^{\text{new}}}y_i\log{p_i^{\text{new}}}+(1-y_i)\log{(1-p_i^{\text{new}})} \]

  • For calibration purposes: If you have no info and simply assign p=0.5 for every outcome, the log score is \(-0.693n^{\text{new}}\).

  • Examples using loo will be given in the next session