adfa, p. 1, 2011.
© Springer-Verlag Berlin Heidelberg 2011
Relating Student Performance to Action Outcomes and
Context in a Choice-Rich Learning Environment
James R. Segedy, John S. Kinnebrew, Gautam Biswas
Vanderbilt University, Nashville TN 37235, USA
{james.r.segedy, john.s.kinnebrew, gautam.biswas}@vanderbilt.edu
Abstract. This paper presents results from a recent classroom study using Bet-
ty’s Brain, a choice-rich learning environment in which students learn about a
scientific domain (e.g., mammal thermoregulation) as they teach a virtual agent
named Betty. The learning and teaching task combines reading and understand-
ing a set of hypertext resources with constructing a causal map that accurately
models the science phenomena. The open-ended nature of this task requires
students to combine planning, targeted reading, teaching, monitoring their
teaching, and making revisions, which presents significant challenges for mid-
dle school students. This paper examines students’ learning activity traces and
compares learning behaviors of students who achieved success with those who
struggled to complete their causal maps. This analysis focuses on students’ ac-
tions leading to changes in their causal maps. We specifically examine which
actions led students to make correct versus incorrect changes to their causal
map. The results of this analysis suggest future directions in the design and tim-
ing of feedback and support for similarly complex, choice-rich learning tasks.
Keywords: Metacognition, Monitoring, Learning Activity Traces, Sequence
Analysis, Learning Environment
1 Introduction
Betty’s Brain is a learning-by-teaching environment where students teach a virtual
agent, named Betty, about science topics by reading a set of hypertext resources and
constructing a causal map (Figure 1) to model the relevant scientific phenomena [1].
Once taught, Betty (the Teachable Agent) can use her map to answer causal questions
(e.g., if cold temperatures increase, what happens to an animal’s blood vessel con-
striction?) and explain those answers by reasoning through chains of links [1]. The
student’s goal is to teach Betty a causal map that matches a hidden, expert model of
the domain using information from the resources. To gauge their progress towards
this goal, students can make Betty take quizzes, which are sets of questions created
and graded by a virtual mentor agent named Mr. Davis, who compares Betty’s an-
swers with those generated by the expert model. Thus, when Betty is unable to answer
quiz questions correctly, the (human) students can use that information to discover
Betty’s (and their own) misunderstandings and correct them.