Students' Strategies for Reasoning About Complex Systems Using Aggregate Data Sources

• In Event: Restructuring Concepts and Tools Through a Complexity Perspective

Abstract

Objective
Research on complex systems reasoning has made considerable progress over the past 20 years (Yoon, Goh, & Park, 2018). Much of this work supports learning about complex systems through the complexity restructuration (Wilensky, here) by focusing on embodied and mechanistic reasoning at the agent level, and then extrapolating to the aggregate level (Danish, 2014; Hmelo-Silver, Marathe, & Liu, 2007; Wilensky & Reisman, 2006). However, investigators often do not have access to the behaviors and interactions that make up a system. Instead they need to infer agent-level behavior from recorded quantitative aggregates constructed through observation and measurement (Grimm et al., 2005). Thus the current study approaches complex systems reasoning from the other direction, exploring the question: How do learners reason about the agent-level of complex systems when interacting with aggregate-level quantitative data?

Theoretical Framework
Connecting agent-level behaviors to aggregate data requires learners to attend to features of data that are typically ignored, such as random noise (Wilkerson-Jerde & Wilensky, 2015). Thus, to explore what Grotzer and colleagues (2017) call “the relationship between pattern and mechanism” (p. 64) means students are likely to engage with graphs in ways that would otherwise be considered unproductive. In this way, working with aggregate data using an emergence-based perspective (Wilensky & Resnick, 1999) represents an agent-based restructuration not of graphs as representations in themselves, but rather in what features of graphs as a representation are valuable for reasoning.

Methods and Data Sources
This study draws from data collected during three week-long enactments with nine 7th-grade public school classrooms in a metropolitan region of California. Students explored climate and natural resources, trophic cascades, and geology by analyzing large, multivariate datasets obtained from state and national agencies. Students used the Common Online Data Analysis Platform (Finzer & Damelin, 2015) to visualize and manipulate data. They were asked to construct explanations that connected lower-level mechanisms (e.g., predator-prey relationships, local or regional causal links, or categorical differences in geology) to the aggregate patterns. Data sources include students’ written work, classroom video, and synchronized screen and iSight video captured from select focal groups. Beginning with focal group data, we (1) tagged instances where students explicitly connected behaviors or agent features to the aggregate data they analyzed, and (2) for each instance, thematically coded for what strategies they used to identify such connections.

Results and Significance
We identified a number of previously unidentified strategies that learners employed to connect macro-level patterns with suspected micro-level mechanisms, usually through isolating or otherwise manipulating subsets of data. These included: divide-and-conquer; clustering; time segmenting (see Figure MW); and zooming. Our findings (Figure 1) reveal how learners reason about aggregate features of complex systems such as equilibrium, initial conditions and perturbations on systems, and system robustness and resilience—features still underrepresented in the literature (Yoon et al., 2018).

Authors

• Michelle Hoda Wilkerson, University of California - Berkeley

• Vasiliki Laina, University of California - Berkeley