In recent years, political scientists have become increasingly interested in the measurement of numerous social psychological constructs (e.g., personality traits, moral foundations, dark triad, values) at both the mass and elite levels. Measuring these traits for political elites, however, is a difficult if not impossible enterprise when using traditional survey-based measurement approaches. In this paper, we help to solve this challenge by leveraging a new data collection effort. Using over 3,000 survey-based responses for 60 psychopolitical constructs as well as open-ended freeform text, we use recent advances in machine learning to train models for automatic recognition of these constructs from language alone. To demonstrate the power of this approach, we then use these models to predict various traits of political elites from electoral debates and social media posts. The results provide scholars with a powerful new tool to measure and study the psychological traits of political leaders around the world.