Classroom discourse
There has been a push for educational institutions to engage in student-centered and collaborative instructional approaches in Science, Technology, Engineering, and Mathematics (STEM) classrooms designed to help students construct their own understanding and take ownership of their learning (Freeman et al., 2014; National Research Council, 2015; Olson & Riordan, 2012; Theobald et al., 2020). These approaches have shown improvements in students’ exam scores, overall course performance, pass rates, and retention rates in STEM courses (Freeman et al., 2014; Loes et al., 2017; Nussbaum, 2008; Wilson & Varma-Nelson, 2016). When having an emphasis on viewing science as “a way of thinking rather than a body of knowledge”, studies have also shown positive impacts on students’ reasoning, problem-solving, communication, and critical-thinking skills (Erduran & Jiménez-Aleixandre, 2007; Talanquer & Pollard, 2010; Wilson & Varma-Nelson, 2016).
Aligned with Vygotsky’s collaborative learning theory, the benefits of learning environments that include group work are centered on the role of classroom discourse and positive peer interactions (Vygotsky, 1978). Discourse is defined as “any stretch of language (spoken, written, signed) which hangs together to make sense to some community of people who use that language”(Gee, 2015). In other words, discourse is the language we use to interact with people in a community. In the context of science, discourse involves asking questions, constructing arguments, and building explanations that can be shared with the scientific community to help understand new phenomena (Lemke, 1990; Osborne, 2010).
Specifically in chemistry education, discourse analysis serves as a qualitative tool to explore the role of classroom talk in students’ development of content knowledge, scientific reasoning, construction of arguments, and the establishment of a shared understanding of phenomena (Cole et al., 2014; Driver et al., 2000; Gee & Green, 1998; Kulatunga et al., 2014; Mercer, 2010; Moon et al., 2016, 2017; Repice et al., 2016; Talanquer, 2014; Young & Talanquer, 2013). Research has shown that discourse moves such as engaging in reasoning, expanding students’ thinking, developing good listening skills, and building accountability as they work towards a common understanding promote the development of the desired knowledge and skills. (Criswell, 2012; Michaels & O’Connor, 2015; Michaels et al., 2008; Towns, 1998). Specifically, students have been encouraged to request clarification or justification, challenge the ideas of their peers, and share what they understood from a previous utterance (Criswell, 2012; Michaels & O’Connor, 2015; Moon et al., 2017). By examining group dialogues of chemistry students regarding solution chemistry, Warfa et al., (2014, 2018) determined that sociochemical norms, which constitute the discursive norms of the classroom that regulate what counts as a chemical justification and explanation, altered how students reasoned about the cause(s) of chemical phenomena. It was noted that students’ statements such as confirmatory and clarifying questions, agreeing or rejecting with reasoning, seeking group consensus, and acknowledging each other’s ideas allowed students to reach a collective understanding of chemistry concepts (Warfa et al., 2014, 2018). Similarly in discourse research, Repice et al. (2016), explored the use of language in peer-led small groups in a first-year general chemistry course as students worked through different problem types. Their study demonstrated that students engaged in regulative talk (communicative interactions that involve everyone present in the space) in order to manage group discussion and dynamics between peers and used instructional talk focused on problem solving to develop their understanding of chemistry concepts (Repice et al., 2016). In addition to cognitive benefits, other studies on collaborative learning and scientific classroom discourse have shown positive impacts at a social and personal level (Eren-Sisman et al., 2018; Hamnett et al., 2018; Nichols, 1996). If members of a group recognize themselves as active participants working towards a shared goal, they can experience higher levels of group trust, perceived consensus, and gain satisfaction that improves group discourse (Harney et al., 2017; Repice et al., 2016; Wells & Arauz, 2006).
Despite these numerous possible benefits, studies have shown that collaborative learning environments do not necessarily lead to productive group engagement, which we define as students engaging with each other’s ideas in a manner that expands students’ understanding of relevant concepts and skills and that support collaboration and inclusion of all group members. Some studies have reported interactions such as antagonistic dynamics between peers, unequal member participation, lack of thoughtful arguments, and focusing on task completion with minimal discussion occurring more often in low-performing groups than in high-performing groups (Barron, 2003; Chan, 2001; Ryu & Sandoval, 2015). Even when these conflicts are not identified, other studies have found that groups only spend a small amount of their discussion time engaging with the content (Summers & Volet, 2010). They even suggest that students tend to avoid high-level content processing interactions such as elaborating, interpreting, and reasoning, which are usually students’ best opportunity to learn from each other (Summers & Volet, 2010). Poor group dynamics are a concern as classroom dialogue reflects the social dynamics and norms that exist within a group that shapes the quality of group learning (Gee & Green, 1998; Mercer, 2004). These findings align with views that students need to be taught how to engage in effective group discourse, and suggests the need for additional investigation of the role of social interactions on collaborative learning in STEM (Osborne, 2010). The work presented in this paper addresses these concerns by demonstrating a method for analyzing both the cognitive and social aspects of student discourse within STEM classrooms. Specifically, this paper builds on the work of existing analytical frameworks within scientific discourse analysis and illustrates how to visualize student discourse for further qualitative analysis. The paper begins by reviewing established methods for analyzing and visualizing student discourse and discussing why a new framework is needed to further analyze student discourse in STEM classrooms. This is followed by a description of the design of the Student Interaction Discourse Moves (SIDM) framework and a method for visualizing the applied framework for analysis. The paper will conclude with implications for research using various lenses to show how this framework can be applied to different research contexts.
Theoretical framework
Vygotsky uncovered that as a group of individuals collectively processes information through clarifying and exchanging ideas with each other, each member constructs their own understanding (Vygotsky, 1978). Thus, each member plays a critical role in contributing to the discussion by responding to previous utterances to advance the discussion to a collective understanding. When students are willing to listen and critically examine alternative perspectives, and when members of the group generate new ideas that extend beyond their initial understanding, then “progressive discourse” can be achieved (Bereiter, 1994). Such engagement in collaborative discourse also provides students the opportunity to build reasoning skills or reconstruct incorrect reasoning (Wegerif et al., 1999). Thus, under a social constructivist perspective, learning extends beyond the accumulation of knowledge and describes the progress of speech, thinking, and behavior that allows individuals to effectively participate in social and intellectual activities (Wells, 1999). These processes and behaviors also define productive engagement when students are working together to complete a task. In support of better understanding of how we can promote student learning in a collaborative environment, we sought a methodological approach to investigate the flow as well as the patterns of student interactions through the lens of social discourse.
Review of discourse analysis frameworks
The analytical frameworks developed by Sinclair and Coulthard (1975), Coulthard (1992), Kaartinen and Kumpulainen (2002), Sampson et al. (2011) & Sampson and Clark (2011) each provide insights into student discourse, although none address all the interactions that are part of productive discourse. This led us to develop the SIDM framework to analyze student discourse, taking direction and inspiration from these earlier analytical frameworks.
Sinclair and Coulthard developed and refined a model called the Initiation-Response-Follow-up model (IRF) that investigated the social structure of discourse in a classroom, primarily between students and instructors at the secondary education level (Atkins, 2001; Coulthard, 1992; Sinclair & Coulthard, 1975). This was some of the first patterned discourse work done and was found to be useful for mapping focused, one-on-one conversations that highlighted the utility of being able to code student discourse across a whole conversation. However, we found this framework lacked the ability to be used for free-flowing small group conversations with more than two members and did not readily capture who or what idea a student was responding to in a conversation. This framework also left out the nonverbal interactions that are sometimes present in a conversation; both critiques have also been mentioned by others in more recent literature (Atkins, 2001).
Kaartinen and Kumpulainen (2002) incorporated Sinclair and Coulthard’s (1975) and Coulthard’s (1992) ideas in a framework they created to analyze how college students in a collaborative inquiry setting constructed their understanding of the definition of dissolving. Part of their analytical framework examined the nature of students’ participation in the manner of initiating, continuing, extending, referring back, agreeing or disagreeing, replying, commenting, and concluding while discussing scientific ideas. The other layers of their framework (logical process, nature of explanation, and cognitive strategies) aimed to characterize students’ explanation-building processes (Kaartinen & Kumpulainen, 2002). This study inspired the idea of designing a tiered framework, as well as the types of codes one could use to characterize an utterance. However, the framework itself was focused on the reasoning patterns of student conversations and did not address types of student discourse outside of cognition. Our framework includes cognition codes like reasoning and presenting claims, but we wished to also characterize the broader social interactions such as motivation, information sharing, and questioning.
Sampson, Grooms, and Walker.(2011) developed a framework to capture how students engaged in scientific argumentation using Argument-Driven Inquiry (ADI), when working in small lab groups. This framework was then expanded outside the lab environment to the classroom environment, where Sampson and Clark (2011) examined students’ argumentation patterns by characterizing high school science students’ responses (agree, reject, discuss, and ignore) and the functions of their utterances (information seeking, exposition, opposition, and co-construction). For a closer examination of the nature of students’ oppositions, they described what criteria students used to evaluate the claims presented in their group. In this way, the researchers were able to differentiate between high and low-performing groups (Sampson & Clark, 2011). From this framework, we found inspiration for writing more specific codes, such as the types of questions students ask each other. However, similar to the Kaartinen and Kumpulainen (2002) framework, this framework was narrowly focused on students’ abilities to construct arguments under the ADI instructional model, which left out many interactions we observed that were not directly related to formulating the answer for a specific task. This framework was missing the nuance to characterize the social interactions among students since it only focused on argumentation building.
Overall, these studies demonstrated how multi-level analysis of student talk can inform our understanding of the way in which students’ interactions influences their way of thinking through science problems (Coulthard, 1992; Kaartinen & Kumpulainen, 2002; Sampson et al., 2011; Sampson & Clark, 2011; Sinclair & Coulthard, 1975). However, rather than focusing SIDM solely on the viewpoint of instructor interaction, argumentation construction, or students’ explanation-building process, we aimed to develop a framework to characterize the nature of student engagement across an entire conversation that could be used to investigate a variety of student interactions.
Review of discourse analysis visualization methods
Reports have been published regarding the visualization of students’ discourse patterns, but most use tabulated frequencies of discourse moves to gain these insights, which is not a descriptive representation of the flow of discourse. Recently, within science education, social networking analyses have been used to characterize and visualize changes in student participation and interaction over time (González-Howard, 2019; Ryu & Lombardi, 2015). Although this method can illustrate who partook in the discussion, how often they participated, and who they responded to most often, these maps rely on the frequencies of interactions and lack the nature of the utterances within the discourse (González-Howard, 2019).
More informative methods of illustrating specific student talk patterns include mapping out speaking sequences as either strings of letters or the use of a graphical coding system (Keefer et al., 2000; Ryu & Sandoval, 2015). Specifically, to explore differences in decision-making and scientific argumentation patterns, Ryu and Sandoval (2015) chronologically mapped students’ speaking sequences using the following form: “an argumentation code (speaker)”. For instance, Dan’s explicit agreement to Jack’s claim was written as Claim (Jack) Explicitly Agreed (Dan)—C(J)EA(D). In a similar fashion of sequencing argument structures, Keefer et al. (2000), used a graphical coding system to map the logic of students’ arguments in peer-led student dialogues to identify what characteristics promoted a productive critical discussion. Different shapes were used to identify an argument, a concession, a challenge, or when an answer to a challenge was made, and arrows were used to identify how these elements were linked to each other (Keefer et al., 2000).
Although both studies illustrate differences in students’ discourse patterns in the context of argumentation, they do not capture the way students engaged in conversation prior to an argument nor what sustained an argument. Given that only a portion of student’s scientific discourse is focused on creating arguments, capturing how students talk and respond to each other in discussion beyond the context of argumentation would be informative of what constitutes productive discourse within STEM classrooms.
Context
Data presented in this paper was collected as part of a larger study investigating student engagement in introductory chemistry courses that was conducted at a large Midwestern university in a large enrollment active learning classroom and associated discussion classrooms. Data for the examples used below was collected in an introductory chemistry course designed for students who did not have an advanced chemistry course in high school. The structure of the course included both lecture and discussion components. The lecture component met for 75-min periods twice a week and was led by the course instructor in a large auditorium setting. In lecture, students worked on activities in self-selected groups of 3–4 people for 20–30 min over the course of the class period. These activities were either from a Process Oriented Guided Inquiry Learning (POGIL) workbook (Garoutte & Mahoney, 2015) or questions delivered using a student response system. The discussion portion of the course consisted of multiple sections of 25–30 students, each meeting for a 50-min period once a week, led by graduate teaching assistants.