Constructing Causal Diagrams to Learn Deliberation

Article type: Research Article

Authors: Easterday, Matthew W. | Aleven, Vincent | Scheines, Richard | Carver, Sharon M.

Affiliations: Human-Computer Interaction Institute, Carnegie Mellon University, Pittsburgh, PA, USA. E-mail: [email protected]

Abstract: Policy problems like "What should we do about global warming?" are ill-defined in large part because we do not agree on a system to represent them the way we agree Algebra problems should be represented by equations. As a first step toward building a policy deliberation tutor, we investigated: (a) whether causal diagrams help students learn to evaluate policy options, (b) whether constructing diagrams promotes learning and (c) what difficulties students have constructing and interpreting causal diagrams. The first experiment tested whether providing information as text, text plus a correct diagram, or text plus a diagramming tool helped undergraduates predict the effects of policy options. A second, think-aloud study identified expert and novice errors on the same task. Results showed that constructing and receiving diagrams had different effects on performance and transfer. Students given a correct diagram on a posttest made more correct policy inferences than those given text or a diagramming tool. On a transfer test presented as text only, students who had practiced constructing diagrams made the most correct inferences, even though they did not construct diagrams during the transfer test. Qualitative results showed that background knowledge sometimes interfered with diagram interpretation but was also used normatively to augment inferences from the diagram. Taken together, the results suggest that: causal diagrams are a good representation system for a deliberation tutor, tutoring should include diagram construction, and a deliberation tutor must monitor the student's initial beliefs and how they change in response to evidence, perhaps by representing both the evidence provided and the student's synthesized causal model.

Keywords: Ill-defined domains, causal diagrams, external representations, deliberation

Journal: International Journal of Artificial Intelligence in Education, vol. 19, no. 4, pp. 425-445, 2009