This study investigates variations in how users exert agency and control over their choice patterns within the game-based ITS, iSTART-2, and how these individual differences relate to performance. Seventy-six college students interacted freely with iSTART-2 for approximately 2 hours. The current work captures and classifies variations in students' behavior patterns using three novel statistical techniques. Random walk analyses, Euclidean distances, and Entropy measures indicated that students who interacted exhibiting more controlled and systematic patterns demonstrated higher quality strategy performance compared to students who interacted with the system in more disordered fashions. These results highlight the potential for dynamical analyses as stealth assessments indicative of students' degree of agency within adaptive learning environments. [This paper was published in: "Proceedings of the 7th International Conference on Educational Data Mining," 2014, pp. 241-44.]