What measures the fraction of primary-target cases with a predicted score lower than the predicted score of secondary-target cases?

The measure that evaluates the fraction of primary-target cases with a predicted score lower than that of secondary-target cases is known as Discordance. In the context of predictive modeling, particularly in binary classification tasks, discordance indicates situations where a positive case (primary-target) is incorrectly assigned a lower risk score than a negative case (secondary-target).

When assessing the performance of models, understanding discordance helps in identifying how well the model differentiates between groups. A high level of discordance can indicate that the model is not performing effectively. Therefore, it's crucial for model validation to ensure that primary-target cases are scoring higher than secondary-target cases, reinforcing the importance of achieving lower discordance rates for better model performance.

In contrast, concordance refers to scenarios where the primary-target cases have a higher predicted score than secondary-target cases, and classification error pertains more generally to the overall accuracy of classification decisions, which doesn't specifically measure the score comparison between target cases. Predictive validity assesses how well the model's predictions capture real-world outcomes but does not specifically focus on the score comparisons central to discordance.