Modeling Science Inquiry in an Interactive Simulation Task

• In Event: Cognitive Models for Assessment

Abstract

In this presentation, I will describe a cognitive model of science inquiry developed in ACT-R cognitive architecture (Anderson et al., 2004) and discuss its applications to assessment research. As a theory of human cognition, ACT-R has been used to develop fine-grained process models for various tasks ranging from math problem solving to driving (Salvucci, 2006) and science discovery (Schunn & Anderson, 1989). In educational research, one of the successful applications of ACT-R is the cognitive tutors for mathematics that have been widely used in thousands of schools across country (Anderson, Corbett, Koedinger, & Pelletier, 1995).
In ACT-R, intelligent behaviors occur through interactions between declarative knowledge and procedural knowledge. One of the task domains where these interactions are critical is science inquiry. Science inquiry involves not only declarative knowledge of science facts and principles and but also procedural knowledge of knowing how to use the facts to carry out scientific investigations (NAEP, 2009). In particular, these interactions can play a significant role in interactive, simulation-based inquiry tasks in which students can design their own experiments and interpret the results to test a hypothesis. Students’ inquiry skills and strategies can influence their content-specific knowledge, and conversely, their knowledge can influence choice of strategies in designing experiments and interpreting the results.
In an effort to demonstrate how ACT-R can be used to model the interaction between declarative knowledge and procedural skills in science inquiry, an ACT-R model was developed for a simulation-based inquiry task in physics domain. In Keehner et al. (2013), adult participants were presented with a cannon-firing simulation interface and asked to evaluate the hypothesis that as the angle of the cannon launch increases, the distance it travels increases. In each simulation trial, participants could set launch variables and observe the resulting travel distance. Analysis of behavioral data indicated that participants who made a correct conclusion differed from those who did not in various process measures including the number of simulation trials, sampling range of the angle variable, variable manipulation strategies, and evaluation of the simulation results in conflict with the given hypothesis.
These group differences were captured by developing two ACT-R models that share the same general inquiry procedures and varying key parameters associated with the models’ declarative and procedural mechanisms. Based on memory blending mechanism (Lebiere, Gonzalez, & Martin, 2007), the two models differ how they weigh the simulation results that are in conflict with the hypothesis. The utility-based procedural mechanism was used to model the group differences in variable manipulation strategies (Chen & Klahr, 1999) and sampling range. The model offers a computational account for how the combination of declarative and procedural knowledge can lead individuals to reach different conclusions. Overall, the model results were consistent with the participants both in the outcome measure and various process measures. Ongoing efforts are underway to test whether the model can be generalized to inquiry tasks in different content domains (e.g., chemistry). Further application of the model to assessment of science inquiry practices in student population will be discussed.

Author

• Jung Aa Moon, Educational Testing Service