Quantitative evaluation of gender differences, cognitive development differences and software effectiveness for an elementary mathematics intelligent tutoring system
This dissertation presents results of a formative evaluation of AnimalWatch, an intelligent tutoring software system for 9–12 year-olds, which teaches whole numbers and fractions by adapting the difficulty of problems to students' performance. Results come from a macro-analysis of 350 students using AnimalWatch in two different schools during three years. Data from these studies were integrated to analyze the system's overall effectiveness, as well as gender and cognitive development differences in interactions with the system, particularly in relation to the help component. In general, it was found that students reduced their mistakes as they progressed in the tutoring session, and improved their attitudes towards mathematics after using the system. However, a rigorous study of the internal components of AnimalWatch showed that the system can be further improved to maximize its positive impact. Students saw too many easy problems, and this may explain the fact that they hardly reached the last topics in the system. In addition, students benefited differently from alternative kinds of feedback provided, depending on the level of abstraction within the help component and the amount of help. An analysis of help effectiveness for students of different gender and cognitive development showed that girls were more sensitive to amount and structure of help than to level of abstraction in the provided help. On the other hand, boys of low cognitive development were affected by level of abstraction in the help. While boys were being selective about the kinds of help they were willing to see, girls would obediently go through each of the hints provided.
0280: Mathematics education
0800: Artificial intelligence