Predictive Validity of Psychological Tests from a Statistical Learning Perspective
Predictive validity is usually established by finding a relationship between test score X in the first measurement and a criterion test score Y on the second measurement (rXY). This measure is obtained within a specific sample and does not give a good indication on how accurate test scores of X will forecast the criterion Y for new individuals. If the goal of a test is to predict future performance, a different coefficient of predictive validity which focuses on measuring prediction accuracy rather than a measure of association. In this project we propose a new definition of predictive validity using a statistical learning perspective, that is, being the out of sample predictive accuracy of a prediction rule of test items. In establishing predictive validity, there are two aspects that should be considered:
- the way we construct a prediction rule from a set of items of the test, and
- the out-of-sample predictive accuracy of this prediction rule.
We will closely study the influence of known concepts such as reliability, on predictive validity given our new definition and their impact on applied situations such as in high stakes testing.
Prof. dr. M. De Rooij, Dr. E. Dusseldorp
Leiden University / Parnassia Groep