To develop socio-scientific thinking skills in biology, students need learning materials nested in the context of real-world problems (Zeidler et al., 2009). This necessarily involves thinking in systems and across scale. Further, we argue, learners need to develop, model, and support explanations of complex biological systems (e.g., modeling their understanding), while thinking across perspectives (Jordan and Sorensen, 2021). The focus of the learning intervention described here is to use models as a means for students to engage in different perspectives. In this study, we defined models as simplified abstractions that aid in thinking. Below are the three research literatures that inform our approach: (1) systems thinking, (2) model-based learning for systems thinking, and (3) novel perspective taking.

1. Systems thinking. Systems thinking is increasingly recognized as central to being scientifically literate citizens (e.g., Gray, 2018). While a systems-thinking approach to STEM education has been popular for decades (Stave and Hopper, 2007; Skaza and Stave, 2009), there remain significant gaps in understanding how systems thinking is best scaffolded in the undergraduate STEM classroom and how to assess and measure students’ understanding of systems. Nonetheless, systems thinking generates scientific habits of mind (Kay and Foster, 1999; Steinkuehler and Duncan, 2008) that are useful frameworks for reasoning and abstracting about a range of complex interactions that underlie contemporary societal biological-related problems (Tabacaru et al., 2009).

Although definitions of systems thinking vary, most definitions highlight that systems thinking is an ability to recognize and understand the relationships between the structure and function of complex systems and the ability to qualify this understanding through graphical or semantic representation, definition, and explanation. In this manuscript, we our view of systems thinking is akin to the soft Systems-thinking Methodology (i.e., SSM) defined by Checkland and Poulter (2020). Here, learners are using a purposeful activity (such as generating FCM models; see below) to grapple with real- world, often social, issues that are complex and require the learner to approach the problem from different angles. We stray from SSM in that we do not follow the learning cycle exactly as described by the authors, but we do include a cycle where information is gathered, leverage points are defined, trade-offs are compared, and decision or action is supported.

Systems thinking has been shown to be challenging for learners because of the non-linear, multi- scalar dynamics with often embedded feedbacks, hidden mechanism, and emergent properties (Hmelo-Silver and Azevedo, 2006; Jacobson and Wilensky, 2006; Verhoeff et al., 2008). In natural systems such complexity includes macroscopic and microscopic phenomena (Penner, 2000; Samon and Levy, 2017) that interact with the living and nonliving elements of the environment (Hmelo-Silver et al., 2007; Shepardson et al., 2007; Eilam, 2012). Learners also struggle with how mechanisms and outcomes are interrelated (Covitt et al., 2009). Learners often put undue focus on unidirectional causal chains that then cause the learner to miss the reciprocal relationships between structures and the associated processes (Grotzer and Basca, 2003; Mohan et al., 2009). This oversimplicity in relational understanding is especially exacerbated when thinking about how smaller system elements interact to result in larger scale outcomes (Grotzer et al., 2017). In addition, learners can ascribe forces within a system to have intent and purpose to produce a certain outcome, leading to oversimplified and inaccurate interpretations of causality (Cuzzolino et al., 2019).

Ben-Zvi Assaraf and Orion (2005) report that learners’ systems thinking can be supported through interventions that feature knowledge integration, but in a case study of four students 4 years later, few ideas remained stable: suggesting a need for greater thinking support throughout the curriculum. Some (e.g., York and Orgill, 2020; Jordan and Sorensen, 2021) have suggested approaches to systems thinking scaffolding and assessment, which involve clearly defining the system processes within the context of the curriculum, whether it be chemistry, life sciences, or in the case of Mahaffy et al. (2018), within interdisciplinary studies (e.g., chemistry and socio-environmental systems). How can students’ internal representations of their systems understanding, therefore be determined, and assessed? We suggest that models as external representations of learners’ system understanding can be useful in not only teaching about systems but also in the assessment of systems thinking.

2. Model-based learning for systems thinking. Based on the idea that internal mental models are constructed over time as new information is obtained (e.g., see Dauer and Long, 2015), we argue that iterative model construction is an effective practice from which learners can integrate both evidence and domain-specific knowledge into visualizations (i.e., concept maps) that can be used to measure degrees of systems thinking through model-based reasoning. Indeed, iterative model development provides a manner for learning about systems in domain specific contexts (Jordan and Sorensen, 2021). Below we discuss the value of models in supporting the cognitive offloading of complex ideas.

Here we define models as simplified abstractions or representations that characterize an idea or phenomenon (e.g., Crawford and Jordan, 2013). Models allow cognition to be distributed by offloading parts of difficult tasks into the physical environment, where thinking can be organized and discussed. Furthermore, because models often include a small number of semantic representations, individuals coming from different backgrounds, once familiar with model terms, can communicate in a standardized space. Finally, models provide opportunities for learners to make their ideas visible and open for discussion, negotiation, revision, and extension; supporting constructive discourse, which is associated with positive learning outcomes (Greeno, 1998; Chi et al., 2001).

Learner mental models, as made visible through cognitive mapping software, can also serve as a kind of boundary object (Star and Griesemer, 1989) by providing the means for bridging ideas across disciplines. In this manner, students taking ideas and perspectives from different disciplines are given a common language for discourse. We, therefore use models as a way for students to help the instructors and each other make their thinking visible. Doing so can enable the multiple and complex parts and processes of the system to be identified and elaborated. We argue that providing learners with a specific modeling approach can enable a common language and classroom artifact that can be assessed. In addition, this common language serves as an effective means for assessment. Finally, modeling can also aid students in taking multiple perspectives.

3. Perspective taking. Cabrera (2009), and more recently Taylor et al. (2020), argue that perspective taking is a fundamental part of systems thinking but is underdeveloped in terms of STEM and life science learning. Perspective taking is the ability to make inferences about and represent others’ psychological states—their emotions, thoughts, goals, and intentions (Stietz et al., 2019). We argue that promoting such perspective-taking pedagogy is essential for developing social and scientific problem-solving skills, both because it will enhance learners’ understanding of complex systems, as well as help to overcome cognitive biases elicited by controversial socio-biological problems.

Like the ambiguity of defining, understanding, and measuring systems thinking, perspective taking is often referred to in the literature using various terms including theory of mind, mentalizing, and cognitive empathy (in contrast to emotional empathy, the capacity to share others’ emotions; see Kahn and Zeidler, 2019 for a review). When learners are exposed to a view that is contrary to their own on an issue that matters to them (e.g., climate change, Genetically Modified Organisms), their moral feelings are violated. This triggers emotions that take over thinking, therefore much of the reasoning that follows serves solely to find arguments that support one’s own view on the issue and arm against deviating arguments (Molden and Higgins, 2012; Greene, 2014). The false consensus effect, or the tendency to assume that others see the world as we do, leads people to systematically underestimate the extent to which others’ perspectives diverge from their own (Bergquist et al., 2019). In this manuscript, we focus on several issues that present likely moral dilemmas for the learner, though, we acknowledge that this might vary and therefore use positionality (i.e., whether a student agrees or disagrees and to what extent) on an issue as a consideration in our study.

We partnered with instructors at a small midwestern liberal arts college for two semesters: including instructors of introductory classes in psychology, economics, and environmental studies from this institution. These data represent a subset of a larger study in progress. The larger study focuses not only on systems thinking and perspective taking, but also is intended to develop tools that allow for rapid conceptual model assessment in large classes. Instructors informed students that course products produced by them would be included in this study and that students had the option of opting out of the study. While the assignments were mandatory, students were able to elect or decline for their materials to be used in the research study and were given the option to withdraw throughout the semester. Student participation involved filling out two surveys and completing three modeling assignments. Research was done with institutional IRB approval (i054387).

Students participated in a survey at the beginning of the semester. This survey sought student agreement with one side or the other of a controversial issue. Three issues were chosen as the focus of this study: one biological (Genetically Modified Crops), one economic (price gouging), and one social (social media use) controversy. These topics were chosen because the authors thought they would likely elicit significant personal investment from students based on prior experience. Throughout the semester, students engaged in the three modeling assignments, each focused on one topic described above. To complete this modeling assignment, students read two contrasting perspectives written as first person narratives by the authors (pro and con). These perspectives were balanced in length and argument strength. Both shared common arguments for each perspective (see Supplementary material). After reading both perspectives, students were asked to make a model using the software, MentalModeler, for each perspective (see Supplementary material for the homework prompts). These models appear as concept maps (again, see Supplementary material) with arrows that have indicated both a direction, weight, which is positive or negative, and strength represented by line thickness. In this way, these models are semi-quantitative and represent drivers, receivers, and ordinary variables within the system. Note: that the direction was used to quantify the type of relationship between the variables, but through an informal inspection of the data saw no reason to include line strength as a variable. Whether the line was positive or negative was used in the micro-motif analysis. Students were given five or six predetermined (by the study authors) components for each model that they were required to put in their models and were told to add up to 10 more. They were told to think about and add as many connections to their models as they saw fit. All students did the assignments in the same order and roughly the same time across classes.

Surveys and modeling works were collected online. Students were told to attach their dotmmp (Mental Modeler File type) files. However, many failed to do so, instead attaching different file types, which resulted in a reduction of our sample size because data analysis required the dotmmp file. In addition, throughout the semester, students were instructed to complete and upload six models: one for and against each of the three topics, but some students were missing models altogether. We received a total of 104 out of a possible 225 complete student responses from the first semester, and 67 out of 132 complete student responses from the second semester. Attrition was greatest at the first timepoint.

All data were extracted through download of the model structure metrics using the dotmmp files and MentalModeler. Model structure metrics included component information, which included each component put into the model and the nature of the arrows linking the components (and arrow direction and strength). These data included: Total Components, Total Connections, Density (number of lines), Number of Driver Components (causal), Number of Receiver Components (effect), and Number of Ordinary Components (neutral).

Once these data were extracted, model component counts were averaged across the pro and con for each side. They were also separated into “my side” and “other side” based on which stance the student took at the beginning of the semester and which side the model was on. For example, if a student was against GMOs, the GMO pro model would be “other side” and the GMO con model would be “myside.”

Models averages of component information were analyzed through a type II, unbalanced design Analysis as Variance (ANOVA) across the three assignments with time, course enrolled, and issue side as a within-subjects factors. In this manner, model effects were tested separately.

Using these data, information on micro-motifs was also generated. The micro motifs reflect the patterns associated with the components (represented by boxes in the modeling software) and the links (represented by arrows in the modeling software) students create between the boxes. Table 1 explains each micro-motif and how they were calculated using the model structure data.

Bidirectionality is the simplest type of linkage with multiple causes and effects being the next level in that students are interlinking constructs and relationships. Next are the indirect effects, which are essential to representing non-linear and often dynamic component relations, and then finally feedback loops represent cyclic type mechanisms that play a role in complex system outcomes (see Supplementary material for model motif images). We define sophistication as the presence of indirect effects, causal chains, and feedback loops. The four categories of motifs (see Table 1) represent increasing levels of sophistication with feedbacks/feedback loops being the highest level of understanding in terms of system representation motifs.

These micro-motif data were analyzed like model component data in that each motif was counted and averaged across students and then ANOVA was used to compare within side and across time.