Chasing Deliberation in the Social Science Classroom. A Study of Deliberative Quality in Whole-Class, Small Group, and Pair Discussions

Abstract

While classroom discussion is often cherished as a cornerstone of effective social science teaching, knowledge on when and where discussions tend to turn deliberative is lacking. To make up for this lack of knowledge, the present study examined classroom discussion in three conditions: a whole-class condition, a small group condition, and a pair condition. The study made use of video observations from a classroom intervention to evaluate how each condition affected the deliberative quality of the discussion process. To this end, 585 student utterances were coded for aspects of deliberation using the Stromer-Galley observation manual. The results showed that both the small group and pair conditions promoted key aspects of deliberation compared to the whole-class condition. The small group condition generated the highest levels of contestation and engagement, whereas the pair condition generated the highest level of equality. While these results suggest that the deliberative quality of classroom discussion might sometimes benefit from the teacher’s absence, more research is needed to determine the circumstances in which this is likely to be the case.

Keywords:

• Social science education
• social studies
• classroom deliberation
• classroom discussion

Introduction

This article builds on a theoretical and methodological approach developed in an article by Teglbjærg (2023) entitled “Chasing deliberation in the Social Science classroom. A study of deliberative quality in factual and controversial issue discussions” and uses this approach to analyze how the deliberative quality of a political discussion in a Danish 9^th grade social science classroom varied across three discussion formats: the whole-class, small group, and pair format.

Homing in on the quality of the discussion process

While studies of classroom discussion are common within the field of Social Science Education, they often focus on the educational outcomes of discussion and neglect the quality of the discussion process (e.g., Andersson, 2012, pp. 192–193; Forsberg, 2011; Latimer & Hempson, 2012; Persson et al., 2020). This lack of attention to process quality is problematic for a number of reasons (Teglbjærg, 2023). First, classroom discussions’ ability to foster civic outcomes might depend on the quality of these discussions. Low quality discussions are likely less able to bring about civic outcomes empirically (Schuitema et al., 2017). This conjecture has been buttressed by Hess and McAvoy (2015, pp. 59), who found that high quality discussion was more effective at producing civic outcomes than low quality discussion. Specifically, students, who were exposed to high-quality “best-practice classroom discussion” scored higher on 7 out of 18 civic outcomes after the discussion, whereas students, who were exposed to low-quality classroom discussion scored higher on just 3 of the 18 civic outcomes after discussion. Hess and McAvoy (2015, pp. 68) described “best-practice discussion” as high-quality discussions involving student preparation, a student-centered atmosphere, and activities structured with an eye to making students learn how to discuss with each other (Teglbjærg, 2023). Second, certain educational outcomes might lose much (or all) of their normative appeal if created through coercive means. For example, instances of opinion change and knowledge gain, which results from classroom discussions where some student voices were systematically marginalized or where evidence and arguments were biased, are arguably not desirable and, hence, loose their normative value (e.g., Myers & Mendelberg, 2013; Teglbjærg, 2023). Third, investigating the quality of classroom discussions is important, because deliberative quality can reasonably be thought of as an end in itself (McAvoy & Hess, 2013; Roth, 2003) and a marker of thoughtfulness, equality, openness, inclusiveness, mutual recognition, and fairness in the present moment of discussion (Teglbjærg, 2023).

The few studies, which do in fact focus on the quality of the discussion process, often find classroom discussions to be of low quality (e.g., Crocco et al., 2018; Elf & Kaspersen, 2012 cited in Beck et al., 2014, pp. 364), though this is not always the case (for counterexamples see e.g., Samuelsson, 2016; Tammi & Rajala, 2018; Teglbjærg, 2023). Classroom discussions are often marred by unequal opportunities for student participation, as they tend to be dominated by high-performing students (Howe & Abedin, 2013; Nishiyama et al., 2020), male students (Baxter, 1999; Howe & Abedin, 2013; Michaels et al., 2008; Nishiyama et al., 2020), students from ethnic or cultural majority groups (Howe & Abedin, 2013), and extrovert students (McMillan & Harriger, 2002; Nishiyama et al., 2020). Furthermore, it has been argued that classroom discussions suffer from a lack of overt disagreement with peers (Howe & Abedin, 2013; King, 2009; McMillan & Harriger, 2002; Nishiyama et al., 2020; Parker, 2010; Savin-Williams and Berndt (1990) cited in Avery et al., 2013), a lack of argumentative sophistication (Gronostay, 2016), and a lack of subject matter content and disciplinary knowledge forms (Elf & Kaspersen, 2012 cited in Beck et al., 2014, pp. 364; Solhaug, 2013).

This article attempts to address some of the above shortcomings by investigating to what extent Social Science teachers can promote the deliberative quality of classroom discussions by letting students discuss on their own (in small groups or pairs) rather than in a traditional whole-class setting. An important first step in this regard is to clarify what exactly is meant by deliberative quality (Teglbjærg, 2023).

Conceptualizing deliberative quality through the ideal procedure of deliberation

Before specifying what is meant by deliberative quality, it should be noted that different widely acknowledged conceptions of high-quality classroom discussion exist (Teglbjærg, 2023). Within dialogic teaching, quality discussion is seen as an open-ended collaborative construction of meaning where control over central aspects of classroom talk is shared between the students and the teacher (e.g., Reznitskaya & Glina, 2013). Relatedly, the concept of Exploratory Talk, understands quality discussion as involving explicit and visible reasoning and stipulates that participants should engage critically and constructively with each other’s ideas (Mercer et al., 1999). Within Education for Deliberative Democracy (Samuelsson, 2016; Samuelsson, 2018), quality discussion is defined as deliberation, and deliberation is often viewed as comprising various elements, including for example contestation, inclusion, argumentation, a lack of coercion, and attempts at reaching consensus or understanding (Englund, 2006; Samuelsson, 2016; Tammi & Rajala, 2018). Finally, scholars within the field of radical democratic citizenship education argue that discussion should also focus on the education of emotions and that permissible forms of discussion need to encompass not just reason-giving but also rhetoric, storytelling, and greeting (Ruitenberg, 2009; Samuelsson, 2016; Young, 1996). Radical democratic citizenship education arose as a response to the deliberative approach and was partly motivated by a worry that efforts at implementing deliberative democracy in the classroom would not succeed but end up suppressing value conflicts and excluding students less familiar with traditional forms of argumentation (Ruitenberg, 2009; Young, 2001). The present study defines quality as deliberative quality in order to narrow the study’s empirical scope. It does not view the deliberative perspective as being a priori better than other approaches to quality discussion but contributes to empirical investigations aimed at clarifying in which settings classroom discussion turns deliberative and in which settings it turns into a display of power and exclusion as feared by the opponents of classroom deliberation.

Deliberative quality was defined by use of the ideal procedure of deliberation formulated by Joshua Cohen (1997) and Habermas (1996, pp. 305–306), since this theoretical construct is recognized as the most systematic and canonical theoretical expression of the deliberative ideal (Goodin, 2018). Formulated concisely, the ideal procedure requires deliberative discussions to be (1) argumentative in form, (2) unconstrained by received norms and values, (3) characterized by inclusion and equality, (4) aimed at reaching understanding (rationally motivated consensus), (5) focused on a publicly relevant topic, and (6) free from any kind of coercion (Cohen, 1997; Habermas, 1996, pp. 305–306). Since the ideal procedure requires discussion to revolve around a publicly relevant issue (i.e., about societal content), it might be said to be well-suited as an ideal for classroom discussion in the social science subject (Teglbjærg, 2023; see also Blanck & Lödén, 2017).

In this article, the term “deliberation” is used to refer to the ideal procedure of deliberation outlined above. The term, “discussion” is, by contrast, used to refer to “the act of talking about something with other people and telling them your ideas or opinions” (Cambridge Dictionary’s American Dictionary, 2023). The term “discussion” is, thus, used to refer to a social practice, which may align to the ideal of deliberation to a greater or lesser extent (see also Teglbjærg, 2023).

Can small group and pair formats improve deliberative quality?

When using deliberative spectacles, the fact that researchers often find classroom discussions to be of low quality is troubling and necessitates more research on how social science teachers can support the deliberative quality of the discussions they initiate. One thing, teachers can do, is to let students discuss on their own in small groups or pairs rather than in a traditional whole-class setting. This section reviews the literature on the deliberative advantages and challenges associated with conducting discussions in a whole-class, small group, and pair formats. Based on this review, it formulates three competing hypotheses, each of which favors a distinct format for conducting discussions aimed at approximating the ideals of deliberation.

The whole-class setting is a setting that includes the teacher and involves a large group of participants. From a deliberative perspective, there are at least four arguments in favor of whole-class discussion. First, whole-class discussions arguably take place in a semipublic space (Nishiyama, 2019), which offers opportunities for students to engage with other people as strangers or acquaintances rather than as close friends (Parker, 2010). In the traditional whole-class setting used in many schools, students are likely to be confronted with some peers, who do not share their worldviews, interests, or experiences. These peers might well be characterized as acquaintances or even strangers rather than as friends (Avery et al., 2013). The whole-class setting, therefore, enables interactions that are not based on intimate relationships but rather on the more anonymous citizen duty to discuss matters of public concern with other members of the public (Parker, 2010). The risk of discussions based on intimate relationships is that they become too appreciative, and, as a consequence, not sufficiently critical, contestatory, and unconstrained by existing norms (Curato et al., 2013). Prioritizing discussion among strangers is also in line with Habermas, who has pointed out that in complex, secularized societies, civic solidarity is a thin kind of solidarity among strangers, who agree to resolve their disputes peacefully through communication (Habermas, 1996, pp. 308). The semipublic character of the whole-class setting, moreover, means that classrooms can function as mediating spaces between students’ private spheres and the larger public sphere (Nishiyama, 2019). Second, whole-class discussions comprise larger and, therefore, usually more diverse pools of information and arguments than discussions in small groups or pairs. While students are usually exposed to a larger diversity of viewpoints in their classroom than in their homes (Hess, 2009, pp. 23), this might not be the case if students are only exposed to a subset of views in the class. Adding an additional person to a discussion group, means adding new beliefs, which might diverge from and be opposed to the beliefs of existing group members. Large discussion groups, therefore, often contain more diverse and opposing views than smaller ones (Hess, 2009, pp. 23). If students can be motivated to participate in the whole-class setting, the benefits of information sharing are, therefore, more pronounced than in the small group or pair setting, where less (and usually less diverse) information is available from the outset. Goodin (2008) among others has pointed out that the value of information sharing is central in deliberative democracy and has a long theoretical history dating as far back as Aristotle, who described it as follows: “different people know different things; if they got together and pooled everything that any of them knew, then together they would know more things than any of them knew separately” (Aristotle, 1984 cited in Goodin, 2008). The problem with small group and especially pair discussions, then, is that the epistemic benefit of being many declines when many becomes few. With respect to the quality of the deliberative process, a smaller and less diverse pool of information and arguments also implies that the prospects for disagreement diminish. Though a diverse information pool is by itself no guarantee for disagreement (Manin, 2011), it is at the very least a necessary condition for disagreement and contestation to occur. Participants with homogenous views and arguments do not disagree with each other. Third, the whole-class setting includes the teacher. The teacher is often seen as resource, who can act as a trained facilitator, help steer the discussion in a deliberative direction as well as encourage and scaffold student contributions (Andersson, 2012, pp. 37–38; Hess, 2009, pp. 22; McAvoy & Hess, 2013; Reinhardt, 2015, pp. 31). As Sybille Reinhardt (2015, pp. 31) has pointed out, the teacher’s mode of facilitation should depend on the composition of the class. (1) When faced with a politically heterogenous student group, the teacher should merely moderate the discussion, since students are sufficiently engaged and their views sufficiently diverse for productive discussion to occur more or less spontaneously. (2) When faced with a politically polarized student group, the teacher should focus on enforcing appropriate rules of discussion to avoid that the discussion turns aggressive. (3) When faced with an engaged but politically homogeneous student group, the teacher should actively try to introduce the students to opposing points of view from outside the classroom. Finally, (4) when faced with a disengaged student group, the teacher should try to galvanize the class for example by provoking the students with her own personal views (Reinhardt, 2015, pp. 31; see also Sætra, 2018). Often, however, the teacher does not drive the discussion forward by provoking students but merely poses questions that students feel more or less obliged to answer (Sætra, 2018). Another important role of the teacher not mentioned by Reinhardt is that of scaffolder. The teacher might, for example, scaffold students’ argumentation by asking them to justify, elaborate, or clarify their opinions (e.g., Grossman, 2021), and she might scaffold intersubjective engagement by connecting related contributions made by different students. Based on an empirical study of classroom discussion, McMillan and Harriger (2002) argue that the teacher’s presence is required if classroom discussions are to turn into deliberative encounters. They write that even college students can seldomly be “turned loose” to deliberate and that discussions must be supported by a skilled teacher or moderator to help students balance between speaking and listening (McMillan & Harriger, 2002). Fourth, it is worth mentioning that keeping the class together means that the teacher should only worry about one dimension of equality (equality between participants). Splitting the class up in multiple smaller groups, by contrast, creates an additional layer of inequality, namely inequality between discussion groups. When the class is split up into smaller groupings, the teacher, therefore, needs to promote equality between discussion groups as well as equality between participants within the same discussion group. The four arguments advanced above all point to the first hypothesis: whole-class discussion is best at promoting the deliberative quality of classroom discussions.

The small group setting is a setting that excludes the teacher and involves a number of smaller discussion groups. Groups may vary in size, and Reich, (2007) has argued that optimal group size is between six and twelve students. The literature contains at least three arguments in favor of small-group discussion. First, although the potential for argument sharing and disagreement might be more pronounced in the whole-class setting, this potential is seldomly realized because students often do not have the courage to (or do not want to) speak up in front of the entire class. Instead, what students need is a safe space, where they can articulate their opinions without embarrassing themselves in front of an entire class (Avery et al., 2013; Sætra, 2018). Second, opportunities for making contributions and responding to others decline as the number of participants increases, and discussion groups should, hence, be small enough to allow for the equal and active participation of all students (Reich, 2007). Compared to the whole-class setting, discussion in the small-group setting therefore gives each student more opportunities for active participation and simultaneously makes it harder for passive students to remain passive and unnoticed. The small-group setting is optimal, because small groups are sufficiently big to contain a range of different arguments and views but at the same time small enough to give time and space for all students to talk (Weasel, 2017). Third, the teacher does not always constitute a resource and might very well detract from the deliberative quality of classroom discussions. Teachers might sometimes misuse their power as facilitators in an effort to make students adopt their own view on a matter. They might act as biased facilitators rather than as neutral arbiters of differing viewpoints (Hess, 2004; Sætra, 2018). Teachers might also misuse their power as initiators of classroom discussion (their agenda-setting power). They might for example let students discuss matters that they consider controversial yet prevent students from discussing and contesting matters that they themselves consider settled (e.g., Hess, 2004). However, even in cases where the teacher acts as a fair and skilled facilitator and does not misuse her authority, the mere presence of the teacher unavoidably infuses the classroom discussion with an element of authority and formal power, which might prompt students to act strategically rather than in concord with the ideals of communicative action (which constitute part of the ideal procedure’s fifth criterion). Since the teacher is bestowed with the administrative power to grade students (Nishiyama et al., 2020), students might for example occasionally engage in argumentation to showoff their reasoning skills to the teacher rather than in a sincere attempt to convince other students of the merits of their opinion. Englund (2006) has emphasized the importance of letting students discuss in the absence of the teacher and has even incorporated ‘teacher absence’ into his criteria for deliberative communication in the classroom. The teacher’s superior level of knowledge is, moreover, part of the reason why teacher-led discussions are said to be necessarily less equal than discussions taking place in the absence of the teacher (Englund, 2006; Nishiyama et al., 2020). It is worth noting, however, that the impact of the teacher’s presence on the quality of discussions probably depends on the past experiences and interactions between the teacher and the students. Teachers with a reputation of being authoritative, close-minded, or biased, are probably less likely to spur engaged high-quality student discussions than teachers known for being responsive, fair, and open to new ideas (see e.g., Sætra, 2018). Howe and Abedin (2013) reviewed the empirical evidence for the benefits of small-group and concluded that the small-group format is especially conducive to discussion of opposing views. Most of their evidence, however, came from discussions in science classes rather than from discussions in social science. The three arguments presented above support the second hypothesis: small group discussion is best at promoting the deliberative quality of classroom discussions.

The pair setting (also sometimes referred to as a dyadic setting) is a setting containing just two students. It contains neither the teacher nor a genuine group, since a pair lacks essential group features (Moreland, 2010). There are at least four arguments in favor of pair discussion. First, the pair setting is likely to constitute a safe space, where students feel more comfortable and less vulnerable than in the semipublic whole-class and small-group settings. The creation of a safe space is often thought to be a central precondition for deliberative classroom interactions because students are more likely to contribute constructively when they feel at ease (Samuelsson & Bøyum, 2015). Second, the pair setting comprises only a single social relationship, whereas the group setting comprises a vast multitude of relationships (Moreland, 2010). The pair setting, therefore, offers a unique venue for relationship building, which Sætra (2018) has singled out as a central precondition for controversial issue discussions. Contrary to Parker (2010), who prizes classroom discussion among peers, who are merely acquaintances or even strangers to each other, Sætra (2018) argues that close social relationships and an amiable learning environment are key to promoting the quality of discussions. Third, the pair setting allows and requires each student to speak much more than the small-group and whole-class settings, since each student’s speaking time is inversely proportional to the number of participants in a discussion session (Reich, 2007). Deliberative learning is sometimes assumed to take place through a process of self-persuasion (e.g., Dryzek, 2000, pp. 46–47, 169, 171; Elster, 1997; Isernia et al., 2014), and if this is the case, more speaking time means that discussions will be more effective a promoting deliberative learning among students. One of the deliberative dynamics assumed to underlie processes of self-persuasion is referred to as the civilizing force of hypocrisy. Proponents of this dynamic assume that while students might initially base their publicly expressed arguments on the public good merely to sound convincing (i.e., for strategic reasons), the public good based arguments they express will end up persuading themselves of the merits of views aligned with the public good (Dryzek, 2000, pp. 46–47, 169, 171; Elster, 1997). At least one empirical study has supported the existence of self-persuasion dynamics in deliberative discussions (Isernia et al., 2014). Fourth, while group interactions are often marred by normatively suspect dynamics related to group pressure, conformity, and social influence (e.g., Asch, 1951; Deutsch & Gerard, 1955; Wood, 2000), pairs are not genuine groups and are, hence, less susceptible to suffer from aspects of coercion related to the social influence and conformity dynamics. For example, majority influence, which refers to the group members’ tendency to yield to the views of the majority, cannot take place in a pair setting, simply because majorities cannot form when only two people are present (Moreland, 2010). A weakness of this argument is, however, that the majority does not always need to be physically present to exert influence on individual group members (e.g., Arnesen et al., 2018). Empirical evidence for the positive effects and deliberative quality of pair discussions comes in part from social network studies investigating everyday informal political talk among friends, colleagues, and partners. These studies show that people tend to become more tolerant, knowledgeable, and aware of arguments contradicting their own opinion as a result of exposure to opposing beliefs in dyadic settings (Mutz, 2006, pp. 73–78; see also Andersen & Hopmann, 2018). Yet, they also point out that such exposure increases attitudinal ambivalence and reduces the likelihood of informal and electoral political participation (Hopmann, 2012; Mutz, 2006, pp. 73–78). The four arguments outlined above all support the third hypothesis: the pair setting is best at promoting the deliberative quality of classroom discussions.

The three hypotheses just presented represent conflicting views on whether the whole-class, small group, or pair format is best able to promote the deliberative quality of classroom discussions. This article explored the three conflicting hypotheses empirically by use of data from an intervention study conducted in a Danish 9^th grade social science classroom in the autumn of 2020.

Method

Research design: the single case experimental design (SCED)

The analysis presented in this article draws on data from an intervention study (see also Teglbjærg, 2023). Interventions and experiments are acknowledged ways of gaining empirical insight into the complexities of classroom life (e.g., Klafki, 2016, pp. 130) and are particularly well-suited for analyzing the conditions of deliberative discussion (Esterling, 2018). The intervention design employed here was a quasi-experimental single-case design. In single case experimental designs (SCED) an action is assessed continually while a treatment condition is alternately introduced and withdrawn (Krasny-Pacini & Evans, 2018). If the action tends to occur more frequently when the treatment condition is introduced and decline again when the treatment is withdrawn, there is evidence to suggest that that the treatment condition positively affects the target action (Richards, 2019, pp. 84, 155). Such an inference, however, can only be made if other factors are held approximately constant. When a treatment condition is introduced, other features of the classroom should, hence, be kept as constant as possible. In the single case experiment conducted in the context of the present study, the participants, time, and location, as well as the assigned topic of discussion were all held constant across the different experimental conditions. The assigned topic consisted of two questions pertaining to the 2020 covid-19 related lock-down of Danish society. The questions were “Was the timing for corona lock down right?” and “Should Denmark lock down again?.”

Even though experimental control is vital in classical experiments (Druckman & Kam, 2011), complete control was not achieved and not strived for in the single case experiment conducted as part of the present study. It was important to relinquish some control because the experiment was carried out in the field and intervened in the everyday practices of self-directed agents with their own legitimate agendas, interests, and concerns. Relatedly, the teacher and students, who took part in the study, were not required to stay on the assigned topic of discussion, since such a requirement would have violated the ideal procedure’s definition of deliberation as unconstrained and free to revolve around any publicly relevant matter (Habermas, 1996, pp. 306,313; see also Teglbjærg, 2023). Importantly, giving up complete experimental control implied that the intervention was turned into a quasi-experiment.

The single case experimental design used as part of the present study had the A-B-A-C form. The letters show how the experimental conditions were ordered chronologically (Richards, 2018, pp. 11, 152–158), and all students were exposed to each condition in the indicated sequence. The As denote the baseline conditions in which students were asked to discuss the assigned topic in a whole-class setting and in the presence of the teacher. The letter B denotes the first treatment condition in which students were asked to discuss the assigned topic in small groups without the presence of the teacher. The letter C denotes the second treatment condition in which students were asked to discuss the assigned topic in pairs, also without the teacher’s presence (see also Teglbjærg, 2023). The students were, hence, first asked to discuss the assigned topic in a traditional whole-class setting; then in a small group setting; then in a whole-class setting again; and finally in a pair setting.

Case and data gathering process

The intervention constituted one of the elements of a larger study, which took place at a Danish lower-secondary school in the autumn of 2020. The school was located in an affluent middle-class neighborhood in a small town comprising about five thousand residents. The backdrop of the intervention was, thus, the Danish Social Science school subject, which goes by the name “Samfundsfag,” and it is important to remember this subject disciplinary context when interpreting the results presented in the results section (Teglbjærg, 2023).

The Danish “Samfundsfag” comprises three core elements: thematical planning, actuality, and the social sciences. The social sciences represented in the Danish curriculum include Political Science, Sociology, and Economy (Christensen, 2023) but exclude Religion, Geography, History, and Law (Christensen, 2023; Ledman, 2019). A 9^th grade class (with students being on average 15 years of age) and their Social Science teacher took part in the pre-planned discussion sessions. The teacher was invited to take part because she often conducted classroom discussions with the aim of educating students for their role as democratic citizens and had many considered views on how to make classroom discussions support students’ democratic education. She perceived the ability to engage deliberatively in political discussions as a key citizen competence, and helping students acquire the skills and virtues of deliberation was a key priority of her teaching. In that sense, the intervention supported the teacher’s own teaching goals (Teacher 2, personal communication, 16.06.20). The class consisted of students whom the participating teacher and another teacher at the same school judged to be dependent learners, who lacked motivation for Social Science classes and needed a large degree of support to achieve the subject’s learning goals (Teacher 1 & Teacher 2, personal communication, 06.08.2020). At the time of the intervention, the class was working on topics related to terrorism, globalization, and the Danish welfare system. The discussion topic used for the intervention, “the Danish society-wide covid-19 lockdown,” therefore, provided students with an opportunity to use their recently acquired scholastic knowledge to form and defend their own opinions on a pressing social issue (Teacher 1 & Teacher 2, personal communication, 06.08.2020). The entire intervention, which ended up lasting for a total of 46.78 minutes was attended by the participating Social Science teacher, 18 students, a videographer, and the author.

Before the study was implemented, approval was sought and obtained from the Research Ethics Committee at the University of Southern Denmark (no. 20/39548). At the first stage of the research process, the teacher was interviewed about her ways of teaching and her views on classroom discussion as a teaching method. The knowledge gained through this in-depth interview served to further qualify and inform the ensuing research process including the planning and implementation of the quasi-experiment, which was conducted as a joint effort by the teacher and the author. The interview was carried out on June 16^th, 2020, and was succeeded by a planning meeting, which took place on August 6^th, 2020. At the meeting, the participating, the author, and another teacher, who took part in the larger study, planned how to carry out the intervention in practice. As part of the planning meeting, it was, for example, decided to show students a short, balanced video about the different ways of handling covid-19 in Sweeden and Denmark to prepare them for the ensuing discussions about the Danish lock down.

The planning meeting was succeeded by the intervention itself. The intervention took place on September 24^th, 2020, and included video observations, which were conducted by use of intrusive gear. While this gear might have reduced authenticity (mundane realism) (Raudaskoski, 2010), it might also have increased the extent to which the students were motivated to take the task seriously and were experiencing the experimental situation as involving and important to them (experimental realism)—and in experimental research the latter form of realism is often believed to be the most important of the two (Druckman & Kam, 2011; Teglbjærg, 2023).

Assessing deliberative quality

While the ideal procedure of deliberation is often characterized as a regulative ideal (Bächtiger et al., 2018), turning it into a construct that is accessible to empirical research is not a senseless undertaking (Bächtiger et al., 2018; Habermas, 1996, pp. 362–363; Teglbjærg, 2023). To assess the degree to which the observed classroom discussions approximated the criteria put forward by the ideal procedure, this study made use of the Stromer-Galley coding manual (Stromer-Galley, 2007), which is widely acknowledged as an observation instrument that captures key empirical correlates of deliberation (Bächtiger et al., 2018; Esterling, 2018; Teglbjærg, 2023). Because of its acknowledgement of the key role of interpretation in coding discussion, the Stromer-Galley manual has earned a positive reputation among researchers committed to quantitative as well as qualitative ways of studying empirical aspects of deliberation (Bächtiger et al., 2018). The Stromer-Galley manual is also a flexible coding scheme, which can be modified to suit specific research contexts and aims (Stromer-Galley, 2007). This feature of the manual was exploited by the present study to capture the empirical aspects of the ideal procedure of deliberation as adequately as possible. Concretely, the manual was supplemented by a number of additional, indicators, which managed to capture two aspects of the ideal procedure, which were not adequately captured by the Stromer-Galley manual (see also Teglbjærg, 2023). These indicators were borrowed from Argumentation Theory (Fisher, 2004, pp. 16–19, 24–28) and from the Discourse Quality Index (DQI), which is another recognized observation manual developed to capture empirical facets of deliberation (Steenbergen et al., 2003). For a thorough description of the indicators used and their connection to the ideal procedure of deliberation, see Teglbjærg (2023). Teglbjærg (2023) analyzed the intra-rater reliability of each of the indicators used and found that Kappa scores were statistically significant for all indicators (p < 0.05) and ranged from 0.720 (satisfactory reliability) to 1 (perfect reliability). The indicators are shown in Tables 3 and 4.

Analytical procedure

The analysis consisted of three stages. To acquire an understanding of how the discussion developed and whether the experimental conditions were delivered as scheduled, the video data were first examined qualitatively and descriptively. Then, the data were coded by use of the indicators outlined in the previous subsection. This process, which is illustrated in the results section, was context-sensitive and relied on interpretation. Further technical details on how it was carried out are available from the corresponding author. Finally, the resulting binary scores for the presence or absence of specific aspects of deliberation were compared quantitatively across the experimental conditions to see which conditions promoted which aspects of deliberation.

Results

How did the discussions evolve?

The discussions took place shortly after the first Danish covid-19 lockdown and at a time when infection rates were rising, and the question of lockdown was back on the political agenda. During the discussions, the students touched upon different themes related to the two questions they had been asked to discuss (1) “Was the timing for the Danish covid-19 lock down right?” and (2) “Should Danish society lock down again?.” On several occasions, however, the question shifted either abruptly or gradually from being pragmatic to being more factual. While pragmatic questions are about what we should do together as a political community (e.g., what we should do to prevent the spread of covid-19?), factual questions are about what is, was, or will be the case (e.g., what parts of society will be locked down again this autumn?). One of the first such shifts happened at the beginning of the first whole-class session. After having noticed the students’ initially hesitant reaction to the pragmatic questions, the teacher decided to pose a more factual question, which was also more closely related to the students’ everyday lives.

Quote 1: Yes, because how many of you have, when now, you think, you were tightly tightly tightly constrained from the beginning. You weren’t allowed; you weren’t allowed to go to school; and you had to stay at home. How many of you have actually, like, loosened up a bit? Are there some, who have been given free rein, and some who have been told by mother and father at home to just stay? [Yes] [name of student 1]. (Teacher, Condition A1, Video 0 [0:00:54–0:01:14])¹

While this shift in emphasis was probably an attempt, on the part of the teacher, to make the topic more accessible for the students and to scaffold the discussion, it also meant that the students quickly tuned in on the factual aspects of the topic. The teacher’s question was answered by student 1 as follows.

Quote 2: I believe that now after Denmark has opened up more; then I believe that there are more of us, who have also, like, begun to meet with friends; not where we meet fifty people, but where one perhaps meets in small groups in the evening (Student 1, Condition A1, Video 0 [0:01:15–0:01:26])²

It is worth noting, however, that the teacher was not the only one, who shifted the topic in a more factual direction. On some occasions, students gave a factual answer despite the teacher posing a clearly pragmatic question. This is illustrated by the below sequence of turns.

Quote 3: Do you think, [name of student 15], do you think, one should; should one lock down again? (Teacher, Condition A1, Video 0 [0:03:08–0:03:11])³

Quote 4: Well, I guess that within the near future, we will probably lock down again, because the infection rate has gone up that much (Student 15, Condition A1, Video 0 [0:03:12–0:03:17])⁴

Though the discussions sometimes tended to move from a focus on “what to do” to a focus on “what will happen now,” the shift in focus only occasionally reduced the deliberative potential of the discussion. This was because the students’ factual utterances were often contestable empirical claims (e.g., I think they will close schools down again in the near future), which were frequently defended by arguments and then disputed by other participants. In line with the coding manual used, the coding (illustrated in the next section) treated such disputable empirical claims as opinions about empirical matters rather than as hardnosed factual statements (e.g., Denmark is a constitutional monarchy).

Table 1 provides a very brief overview of what happened in each of the experimental conditions and in each of the discussion sessions within these conditions. As shown in the table, the content of the discussions varied somewhat between the different groups and pairs. The remainder of the analysis, however, focused only on variations in deliberative quality between the conditions.

Table 1. An overview of experimental conditions and major discussion events.

Experimental conditions	Sessions	Events
Whole-class condition (A1). Baseline. Students asked to discuss the two questions about the covid-19 lock down in a plenary session with the teacher present	Plenary 1	The teacher started by posing the two assigned open questions, but the students reacted hesitantly and unresponsively. To make students talk, the teacher then resorted to closed factual questions directed at specific students
Group condition (B). First treatment. Students asked to discuss the two questions about the covid-19 lock down in small groups and without the teacher’s presence	Group 1	Discussion centered on the assigned pragmatic questions about the covid-19 lock down
	Group 2	Discussion was a mixture of publicly relevant talk pertaining to the lock down and social chitchat, but chitchat was more pronounced than publicly relevant talk
	Group 3	Discussion revolved mostly around questions pertaining to the lock down but around half of the questions were empirical (e.g., what will happen now?) rather than pragmatic (e.g., what should we do about rising infection rates?)
	Group 4	Discussion centered on the pragmatic questions relating to the lock down but around half of the questions discussed were factual or empirical
Whole-class condition (A2). Baseline. Students asked to discuss the two questions about the covid-19 lock down in a plenary session with the teacher present	Plenary 2	The teacher started the session by asking whether the students acquired any new perspectives on the matter, but the students again reacted hesitatingly. The teacher then started to ask factual questions about how the students experienced and handled the lock down in their everyday lives
Pair condition (B). Second treatment. Students asked to discuss the two questions about the covid-19 lock down in pairs and without the teacher’s presence	Pair 1	Discussion centered on both factual and pragmatic questions related to covid-19. Chitchat was somewhat pronounced
	Pair 2	Discussion was a mixture of publicly relevant talk about the lock-down and social chitchat, but the publicly relevant talk was most pronounced. Opinions and judgements were frequent
	Pair 3	Discussion centered on publicly relevant pragmatic questions about what to with covid-19, but many of the stated opinions were uttered in an insincere and at times even sarcastic tone
	Pair 4	Discussion contained mostly factual statements related to covid-19. Opinions and judgements were rare. Chitchat was somewhat pronounced
	Pair 5	Discussion revolved around publicly relevant questions pertaining to the lock down. Most of the statements made were about factual matters. Chitchat was rare
	Pair 6	Discussion revolved around publicly relevant questions pertaining to the lock down. Most of the statements made were about factual matters. Chitchat was somewhat pronounced
	Pair 7	Discussion centered on both factual and pragmatic questions related to covid-19. Opinions were frequent. Chitchat was somewhat pronounced
	Pair 8	Discussion mostly characterized by social chitchat
	Pair 9	The two students (students 4 and 5), who should have constituted pair 9 left the classroom after the second whole-class session and, hence, did not participate in the pair discussions

Some of the events occurring during the intervention deviated from the intervention plan. Deviations, judged to be at odds with the experimental condition in which they occurred, were generally excluded from the analysis to safeguard comparability between conditions. On the other hand, deviations, which were not at odds with the experimental condition in which they occurred, were kept for further analysis. Students 4 and 5 left the classroom before the onset of the pair discussions, and student 18 was mumbling and, hence, mostly inaudible in the group and pair conditions. To ensure that the group and pair condition would remain comparable to the other conditions, all utterances made by students 4, 5, and 18 were, therefore, excluded from the final analysis (further details on this process of sample reduction are available in an appendix to the present article). While some utterances were excluded from the final comparative quantitative content analysis, all utterances were first analyzed in depth to retain a holistic understanding of the discussion’s content and flow. It is worth emphasizing, moreover, that while deviating utterances were excluded from the quantitative content analysis post-hoc in an effort to safeguard comparability between conditions, it would have been squarely against the ethos of deliberation to deny students the possibility of acting in a deviant way from the outset (Esterling, 2018). Demanding strict adherence to the intervention plan would not only have contradicted the ideal procedure but would also have violated good practice and commonly accepted ethical standards within field research (Dupret et al., 2016).

After excluding utterances being at odds with the experimental conditions in which they occurred, the total sample of data consisted of 585 student utterances, and Table 2 shows how these utterances were distributed across the experimental conditions. Table 2 also shows the number of sessions in each condition, the duration of each session, and the duration of each condition. The latter equals the duration of each session times the number of sessions in each condition.

Table 2. Descriptive statistics for the final sample.

	Whole-class discussion (A1)	Group discussions (B)	Whole-class discussion (A2)	Pair discussions (C)	Total
Sessions	1	4	1	7	13
Student utterances n (%)	19 (3.2)	176 (30.1)	23 (3.9)	367 (62.7)	585 (100)
Duration of each session^a	00:05:05	00:05:05	00:05:05	00:05:05	00:20:20
Total duration	00:05:05	00:20:20	00:05:05	00:35:35	01:06:05

Note.N = 585 student utterances.^aStudents were allowed to and sometimes did discuss for longer than 5 minutes and 5 seconds but only the first 5 minutes and 5 seconds of each discussion was analyzed to strengthen comparability between conditions. Duration had to be set equal to 5 minutes and 5 seconds because this was the duration of the shortest session.

A look inside the interpretative coding engine

The coding process consisted in giving each of the 585 student utterances a score on each of the indicators of deliberative discussion displayed in Tables 3 and 4. This coding process was sensitive to the context of the verbal interaction and used non-verbal cues (for example gestural cues and the intonation of students’ voices) to interpret the meaning of the verbalized communication. As recommended by Stromer-Galley (2007) and Fisher (2004, pp. 17–18), utterances were interpreted from an involved participant perspective. Consequently, the assigned binary scores (for the presence or absence of a particular indicator of deliberation) were based on a thorough qualitative judgment. It is worth mentioning, moreover, that the indicators were iteratively refined and adjusted to the empirical material before being used to assign the final scores presented in Table 3 and 4. The coding process is further elucidated below through six example quotes, selected for their exemplary character. The first example is from group 3’s discussion and illustrates the presence of argumentation.

Quote 5: But with respect to; wasn’t it; was it Sweden? No, yes, Sweden; then I will say we [in Denmark] have done it better because there aren’t that many, who; many, who have died (Student 14, Condition B, Video 45 [0:05:50–0:05:58])⁵

This utterance is an example of argumentation because student 14’s conclusion “we have done it better” is backed up by a reason “fewer people have died” (Fisher, 2004, pp. 16–19; 24–28). Still, the utterance did not contest prior views and did not engage with the utterances of prior speakers, because student 14’s comment occurred spontaneously and did not respond to anyone. The second example is also from group 2’s discussion and illustrates the contestation of a value judgment. The utterance occurred as a response to a prejudiced comment from student 5, who claimed that things coming from China are generally disgusting and undesirable. To this student 7 responded:

Quote 6: They are the ones, who have made the most successful [vaccines] so far (Student 7, Condition B, Video 0 [0:02:31–0:02:33])⁶

The tone and content of student 7’s response to student 5 indicate that he opposes her prejudiced value judgment, and student 7’s utterance is, hence, an example of a contestation of a deep-rooted belief. The utterance is also a response to student 5 and, therefore, constitutes an instance of engagement. The third example quote is very brief and was taken from group 4’s discussion. The quoted utterance shows what a critical question looks like in practice. The question occurred as a reaction to a comment made by student 16, who expressed the view that people should be tested for covid-19 before being allowed to enter restaurants. To this, student 17 simply replied:

Quote 7: Why? (Student 17, Condition B, Video 1 [0:03:34–0:03:34])⁷

The tone of student 17’s question clearly indicated his skepticism and signaled that he wanted his question to be understood as a demand for justification along the lines of “are there any sound reasons behind this crazy proposal?.” While his question remained genuine (it was neither closed nor rhetorical), it constituted an instance of cross-examination rather than an invitation for open co-inquiry. The fourth example comes from pair 3’s discussion and illustrates metatalk. The utterance occurred as a reaction to a prompt that encouraged students to explore points of agreement and disagreement.

Quote 8: I think we disagree a little about this thing with Sweden [whether Denmark’s decision to lock down was right compared to Sweden’s decision not to lock down] and whether we should be allowed to meet (Student 13, Condition C, Video 1 [0:01:35–0:01:39])⁸

Student 13’s utterance is an example of metatalk, because she does not just dispute a prior statement but verbalizes the dispute and turns it into the momentaneous topic of discussion. Such verbalized disagreement constitutes an instance of metatalk (talk about the talk) and is a sign of reflective engagement with the other’s point of view. The fifth example is from the first whole-class discussion and exemplifies a pure factual statement. The quoted utterance occurred as a response to a factual question by the teacher, who asked students which parties that tend to support big business.

Quote 9: But that is, I guess, many of the right wing parties, for example Venstre [Danish center-right party] […] (Student 8, Condition A1, Video 0 [0:04:42–0:04:44])⁹

Student 8’s answer to the teacher’s question is a plain factual statement rather than an opinion about an empirical or factual matter. The reason for this is that the statement includes very little room for contestation and was not phrased as a personal judgment but as a fact. While personal judgements about factual matters afford participants with opportunities for disagreement and argumentation, the same is not true of factual statements. The occurrence of factual statements, therefore, tend to depress the contestation and argumentation related aspects of deliberation. Though quote 9 does not include an opinion, it was still coded as talk about a publicly relevant matter. The final example is from pair 5’s discussion and illustrates an instance of harmful social talk (and hence coercion). As with the other examples, knowledge about the immediate context of interaction is a precondition for interpreting the utterance. As a reaction to student 2, who argued that “we have to handle covid-19 the right way,” student 6 simply responded:

Quote 10: We have to handle it the right way” (Student 6, Condition C, Video 3 [0:00:20–0:00:21])¹⁰

As she repeated student 2’s utterance, however, student 6 used a sarcastic voice and turned her face toward the camera and away from student 2. So, what happened from an interactional point of view was that student 6 attempted to ridicule student 2 by presenting a caricature of what she just said. Having given examples of how the coding was carried out, the analysis now turns its gaze toward the quantitative results and shows how each aspect of deliberation and coercion differed across the experimental conditions.

Table 3. Aspects of deliberative quality across conditions.

	Whole-class discussion (A1)	Group discussions (B)^i	Whole-class discussion (A2)	Pair discussions (C)^j	All conditions
Argumentation
Arguments n(%)^a	2 (10.5)	45 (25.9)	5 (21.7)	49 (13.8)	101 (17.7)
Original arguments n(%)^a	2 (10.5)	36 (20.7)	5 (21.7)	39 (11.0)	82 (14.3)
Contestation^h
Statements of disagreement n(%)^a	0 (0.0)	28 (16.2)	0 (0.0)	14 (3.9)	42 (7.3)
Critical questions n(%)^a	0 (0.0)	2 (1.1)	0 (0.0)	1 (0.3)	3 (0.5)
Contestations of deep-rooted beliefs (values or needs) n(%)^a	0 (0.0)	2 (1.1)	0 (0.0)	0 (0.0)	2 (0.3)
Equality and inclusion
Speakers n(%)^b	6 (40.0)	15 (100.0)	8 (53.3)	15 (100.0)	15 (100.0)
Share of speaking time among speakers M (SD)^c	16.7 (13.0)	6.7 (5.4)	12.5 (14.0)	6.7 (2.1)	6.7 (2.6)
Addressees n(%)^d	0 (0.0)	15 (100.0)	1 (6.7)	15 (100.0)	15 (100.0)
Share of received responses among addressees M (SD)^e	–	6.7 (4.9)	–	6.7 (2.6)	6.7 (2.3)
Engagement
Responses n(%)^f	18 (94.7)	82 (62.6)	20 (100.0)	147 (54.4)	267 (60.7)
Responses to other students n(%)^f	0 (0.0)	82 (62.6)	0 (0.0)	147 (54.4)	229 (52.0)
Genuine questions or Metatalk n(%)^a^,g	0 (0.0)	9 (5.1)	0 (0.0)	55 (15.1)	64 (11.0)
Rebuttals n(%)^a^,h	0 (0.0)	28 (16.1)	0 (0.0)	12 (3.4)	40 (7.0)
Original rebuttals n(%)^a^,h	0 (0.0)	21 (12.1)	0 (0.0)	10 (2.8)	31 (5.4)
Public relevance
Publicly relevant utterances n(%)^a	19 (100.0)	133 (75.6)	20 (87.0)	273 (75.0)	445 (76.5)
Opinion statements n(%)^a	11 (57.9)	105 (59.7)	11 (47.8)	161 (44.7)	288 (49.8)

Notes. N = 585 student utterances.

^a As percent of total number of all utterances in a particular condition and overall. Reported percentages are valid percentage and were, hence, based on the total number of utterances that were audible enough to enable discrimination between the presence and the absence of the particular aspect of deliberation in question. Consequently, dividing the number of occurrences in a cell by the total number of utterances and multiplying by hundred does not always yield the percentage reported in the parenthesis of that cell.

^b Number and percent of the 15 present and audible students who spoke at least once in each condition and during the entire discussion.

^c The higher the standard deviation (SD) of share of speaking time among speakers, the more unequal the distribution of the shares of speaking time among speakers.

^d Number and percent of the 15 present and audible students who were addressed at least once in each condition and during the entire discussion.

^e The higher the standard deviation (SD) of share of received responses among addressees, the more unequal the distribution of the shares of responses among addressees. The standard deviation of the distribution of received responses among addressees could not be calculated for the whole-class conditions since to compute a distribution of received responses among addressees at least two addressees are needed.

^f As percent of total number of publicly relevant utterances in a particular condition and overall.

^g Genuine questions and critical questions are not mutually exclusive categories.

^h Rebuttals are indicators of both engagement and contestation.

ⁱ The proportions presented in this column were not calculated as the averages of each discussion group’s proportion but as the proportions of the small group condition overall. Technical details and justifications for this procedure can be obtained from the author.

^j The proportions presented in this column were not calculated as the averages of each pair’s proportion but as the proportions of the pair condition overall. Technical details and justifications for this procedure can be obtained from the author.

Empirical aspects of deliberation across conditions

Table 3 illustrates how each empirical element of deliberation evolved from the first phase of the intervention (condition A1) to the last phase of the intervention (condition C). The last column of Table 3 sums up the presence of the different elements across conditions. On the whole, argumentation was present in 17.7% of the analyzed student utterances. Original argumentation (presentation of an argument not previously voiced in the same session) occurred in 14.3% of analyzed utterances. Original argumentation seemed to be most pronounced in the middle phases of the intervention (conditions B and A2) but did not differ systematically across the experimental conditions.

Overall, the students only rarely contested each other’s beliefs. Only 7.3% of student utterances were statements of disagreement (disputations) and both critical questions and contestations of deep-rooted beliefs made up less than 1% of the total number of student utterances. Though contestation was not frequent overall, it varied widely across the experimental conditions. The small group condition was systematically more favorable to all the analyzed forms of contestation than any of the other conditions. While statements of disagreement were completely absent during the first whole-class session, they constituted 16.2% of the utterances voiced during the small group sessions. Importantly, this increase in disagreement was not due to a general time trend, because the proportion of utterances expressing disagreement dropped back to zero again after the reintroduction of the whole-class condition. Though the proportion of utterances displaying disagreement increased again after the onset of the pair discussions, the increase was markedly less pronounced than the increase that followed the introduction of the small group condition. The proportion of utterances displaying disagreement in the small group condition (16.2%) was more than four times as large as the proportion of utterances displaying disagreement in the pair condition (3.9%).

Remarkably, the other elements of contestation, which included critical questioning, contestations of deep-rooted beliefs, rebuttals, and original rebuttals, all displayed the exact same pattern of variation across conditions as the element of disagreement. Though all elements of contestation displayed the same pattern of variation, the variations were less pronounced for the elements of critical questioning and contestations of deep-rooted beliefs than for the other elements of contestation.

Although the small group setting seemed to unequivocally promote aspects of deliberation related to contestation, the same picture did not emerge in the analysis of inclusion and equality. While all students spoke at some point during the discussion (and were hence included), the proportion of included students was much larger during the group and pair discussions (100%) than during the whole-class discussions. This is not surprising since (as mentioned in the hypothesis section) smaller discussion groups give each individual student more time to speak (Reich, 2007). What is more interesting is how unequally speaking time was distributed among the students who spoke. Overall, the most talkative student’s share of total speaking time (11.65%) was almost four times as high as the least talkative student’s share (3.07%) (see figure 1). While the ratio of the most talkative student to the least talkative student is an illustrative and straight-forward indicator of inequality in speaking time, its weakness is that it only captures the distribution of speaking time between two students and disregards the distribution of speaking time among the remaining students participating in the discussion. To determine how unequally speaking time was distributed among all students, one might look at the normalized standard deviation of the distribution of speaking time.¹¹ The means and standard deviations provided in Table 3 can be used to calculate the normalized standard deviations for each phase of the intervention. This yields a value of 0.78 for the first whole-class discussion (A1); a value of 0.81 for the small group discussions (B); a value of 1.12 for the second whole-class discussion (A2); and a value of 0.31 for the pair discussions (C).

Figure 1. Distribution of shares of speaking time among speakers (entire intervention). Note. The height of each bar represents the number of speakers whose share of total speaking time fell within the bar’s width. E.g., One speaker occupied between 10 and 12 percent of total speaking time. Five speakers occupied between 6 and 8 percent of total speaking time.

While inequality in speaking time did, hence, not differ systematically between the whole-class and small group conditions, it was markedly lower in the pair condition than in the other conditions. This is not as self-evident as it might seem, since pair discussions require discussion turns to alternate frequently but do not automatically guarantee that each member of the pair speaks for an equal amount of time. Pair discussions might in principle constitute very unequal encounters, where one person speaks most of the time, while the other occasionally interjects a brief comment to keep the conversation flowing. While the ideal procedure emphasizes that participants should have equal opportunities to contribute to the discussion, it also stipulates that participants should have an equal chance of being heard (Habermas, 1996, pp. 305–306). The study, therefore, also measured inclusion and equality with respect to received responses and not just with respect to speaking time. Table 3 shows that while all students were addressed (i.e., received a response) by peers at some point during the discussion, students were almost solely addressed by other students during the small group and pair discussions and were almost never addressed during the whole-class sessions. Overall, student responses were, moreover, unequally distributed among addressees (see figure 2).

Figure 2. Distribution of shares of received responses among addressees (entire intervention). Note. The height of each bar represents the number of addressees whose share of received responses fell within the bar’s width. E.g., one addressee received between 13 and 14 percent of total responses.

The student, who was responded to least often, received 3.83% of total student responses. The student, who was responded to most often, by contrast, received 13.41% of total student responses. Before analyzing how inequality in received responses varied across the experimental conditions, it is worth noting that it was only possible to compare the small group and pair conditions, since inequality in received responses between addressees could not be calculated for the whole-class conditions (because they included less than two addressees). Inequality in received responses between addressees was approximately twice as large in the small group condition (normalized SD = 0.73) than in the pair condition (normalized SD = 0.39). To sum up, the pair condition seemed to be most favorable to aspects of deliberation related to inclusion and equality.

As shown in Table 3, the analysis captured five different aspects of deliberative engagement. While the row labeled “responses” displays the occurrence of responses to the teacher or to other students, the column labeled “responses to other students” only displays the occurrence of responses directed at students. Overall, the discussions were characterized by a relatively high level of responding, since 60.7% of utterances were responses to previous comments, and 52.0% of utterances were responses to previous comments made by other students. This suggests that students did not just perceive the classroom discussion as an opportunity to present their own beliefs to an attentive audience but made at least some attempts to connect their thoughts with those of other students. Responses to other students, however, varied widely across conditions. Whereas responses to other students were completely absent from the whole-class discussions, they constituted 62.6% of utterances in the small group discussions and 54.4% of utterances in the pair discussions. The absence of student responses in the whole-class discussions reveals that these discussions were not only teacher-controlled but also teacher-centered. The students did not attempt to communicate with each other through the teacher but rather perceived the teacher as the only interlocutor with whom they were supposed to engage. The other aspects of engagement (metatalk and questioning, rebuttals, and original rebuttals) all followed a pattern of variation across conditions identical to the pattern exhibited by “responses to other students.” The whole-class discussions displayed a complete absence of engagement, while the small group and pair discussions were characterized by relatively high levels of engagement between students. Although the presence of rebuttals was most frequent in the small group discussions, the presence of questions and metatalk was most pronounced in the pair discussions. Whereas the pair discussions contained 46 student questions and 9 instances of metatalk, the small group discussions contained 8 student questions and 1 instance of metatalk.

The last aspect of deliberation displayed in Table 3 is the aspect of publicly relevant talk. Overall, 76.5% of student utterances were about a publicly relevant matter but publicly relevant utterances occurred somewhat more often in the whole-class discussions than in the small group and pair discussions, where social chitchat was slightly more pronounced. Conversely, opinion statements (which are a sub-category of publicly relevant utterances) were more frequent in the small group discussions than in whole-class discussions. When considering the deliberative quality of classroom discussions, the presence of opinion statements about either factual, pragmatic or value-related matters are important because opinion statements offer students opportunities for argumentation and contestation. Factual statements, by contrast, invite neither argumentation nor contestation and, consequently, tended to reduce the deliberative quality of the discussion. To sum up, publicly relevant utterances were more frequent in the whole-class discussions than in the small group and pair discussions. However, since many of the publicly relevant utterances found in the whole-class discussions were factual statements, publicly relevant opinion statements occurred more often in the small group discussions (59.7%) than in any of the whole-class discussions. The pair discussions, by contrast, contained the smallest proportion of publicly relevant opinion statements (44.7%) and, so, did not offer the same opportunities for publicly relevant discourse as the small group discussions.

Empirical aspects of coercion across conditions

Since the criteria for deliberation, put forward by the ideal procedure, are to some extent formulated negatively as the absence of coercion (Habermas, 1996, pp. 305–306), it seems sensible to include measures of coercion when analyzing the deliberativeness of classroom discussions. Since coercion can take countless forms, any empirical analysis of coercion is bound to focus only on a subset of these forms. This study focused on two such forms of coercion, namely interruptions of the current speaker and harmful social talk, which captured instances of blame, insults, imperatives, or threats (Table 4). Overall, interruptions of the current speaker occurred seldomly, since only 5.2% of turns were interruptions of the previous turn. Interruptions were, however, markedly more frequent in the small group discussions, where they constituted 7.2% of discussion turns, than in any of the whole-class discussions, where they were completely absent. Of note is also that interruptions occurred more often in the small group discussions than in the pair discussions, where they made up only 4.8% of discussion turns. On the whole, harmful social talk was present in just 1.0% of the analyzed student utterances. Though harmful social talk followed almost the same pattern of variation across conditions as interruptions of the current speaker, it was just as infrequent in the small group discussions as in the pair discussions.

Table 4. Aspects of coercion across conditions.

	Whole-class discussion (A1)	Group discussions (B)	Whole-class discussion (A2)	Pair discussions (C)	All conditions
Interruptions of the current speaker n(%)^a,b	0 (0.0)	12 (7.2)	0 (0.0)	16 (4.8)	28 (5.2)
Harmful social talk (blame, insults, imperatives, or threats) n(%)^c	0 (0.0)	2 (1.1)	0 (0.0)	4 (1.1)	6 (1.0)

Note. ^aExcluding interruptions of speakers engaging in social talk or process talk. ^bAs percent of the total number of turns, rather than utterances, in a particular condition and overall (i.e., excluding the secondary and tertiary utterances of a turn). ^cAs percent of the total number of utterances in a particular condition and overall.

Before moving on to the discussion section, it is worth noting that many of the differences in proportions between the whole-class and small group conditions reported in Tables 3 and 4 were systematic. An aspect of deliberation differs systematically across conditions if it increases when the whole-class condition (A1) is replaced by the small group condition (B), then decreases when the whole-class condition is reintroduced (A2), and finally increases again following the introduction of the pair condition (C). Such a pattern of chance across conditions suggests that the observed differences between teacher-led conditions (A1 and A2) and student-led conditions (B and C) were reliable and not simply due to the progression of time. While the reported differences between the intervention phases could, hence, often be ascribed to the experimental conditions, the size of the differences were often not large when assessed by use of NAP (Non-overlap of All Pairs) (see Parker et al., 2011; Vannest et al., 2016). NAP is a technique designed to estimate the effect sizes of single case experiments, and though the NAP-values of the observed differences were generally not impressive, the values displayed a considerable degree of variation (from the smallest possible NAP-value equaling 0.5 to the largest possible NAP value equaling 1.0). A more detailed account of the effect size calculations is available from the author.

Discussion

On the whole, the results of this study showed that the investigated classroom discussions exhibited a relatively low level of deliberative quality. Many of the empirical aspects of deliberation (including argumentation, contestation, and some forms of engagement) occurred infrequently and inequality was pronounced, though it was less pronounced in some conditions than in others. The only clear deliberative strength of the analyzed discussions was an almost complete lack of harmful social talk. This study found the investigated classroom discussions to suffer from a low level of deliberative quality. While this negative finding is in line with some empirical studies of deliberation (e.g., Crocco et al., 2018; McMillan & Harriger, 2002), it contradicts other classroom deliberation studies (e.g., Samuelsson, 2016; Tammi & Rajala, 2018; Teglbjærg, 2023), which have found pre-arranged classroom discussions to exhibit a more deliberative character.

Interpreting the findings

This study found the whole-class condition to promote publicly relevant talk as well as the absence of coercion measured as interruptions and harmful social talk. Apart from these strengths, however, the whole-class condition seemed to be the condition that was least conducive to deliberative discussion, since the elements of contestation, engagement, and inclusion were all much less pronounced in the whole-class condition than in the small group and pair conditions. The whole-class condition set the stage for a very teacher-controlled discussion, where students often displayed overt signs of disengagement and only spoke when directed to do so by the teacher. Students were also often interrupted by the teacher and were promptly corrected when making statements that the teacher judged to be even slightly inaccurate. As mentioned, the students did not attempt to communicate with each other through the teacher but rather perceived the teacher as the only person with whom they were required to engage. The low levels of contestation, engagement, and inclusion all seemed to be partly attributable to the tightly controlled nature of the discussion and a general lack of student commitment. The controlled nature of the whole-class discussion arguably not only explains its deliberative deficits but also its (very limited) deliberative strengths, namely the absence of interruptions and harmful social talk. The finding might be said to lend empirical support to Englund (2006), who has emphasized that deliberative communication requires opportunities for discussing in the absence of the teacher.

While the whole-class discussion analyzed here went awry, it is important to emphasize, however, that whole-class discussions are by no means predestined to become tightly teacher-controlled encounters unfavorable to deliberative discussion (see e.g., Teglbjærg, 2023 for a counter-example). The deliberative quality of a whole-class discussion depends in no small part on the relationship between the teacher and the class as well as on the prior interactions between the teacher and the students (including prior classroom discussions) (see e.g., Sætra, 2018). One explanation for why the analyzed whole-class discussions failed to develop into a deliberative encounter might, hence, be the existence of a troubled relationship between the teacher and the students. One indication that the relationship was not entirely smooth was the teacher’s perception that the students were demotivated learners, who would need firm teacher moderation during the discussions (Teacher 1 & Teacher 2, personal communication, 06.08.2020). Another possible, and related, explanation is that a majority of the students were opposed to the culture, norms, and authority figures representing the school (see e.g., Audigier, 2002), and that this culture of opposition toward the school tended to silence some students during the whole-class condition, because the whole-class space is arguably filled with school norms and controlled by an authority figure representing the school. Even students not being particularly opposed to school culture might have yielded to the pressure of the majority to avoid being sanctioned by the majority. It is worth noting that if the latter was indeed the case, the whole-class discussion would not have been characterized by a lack of coercion but would rather have been completely saturated with coercion, albeit of a silent kind not easily detectable by observational methods. Signs that a culture of opposition might have existed included (1) students’ occasional displays of sarcasm in the pair condition (especially pair 3), and (2) the fact that some of the students being most engaged in the group discussions were completely silent during the whole-class discussions.

The results showed that the small group condition promoted aspects of deliberation related to contestation, opinion sharing, and engagement measured as the occurrence of original rebuttals and responses to other students. On the downside, the small group condition produced the largest proportion of interruptions of the current speaker. The small group condition often seemed to pave the way for discussions characterized by a certain degree of student involvement and deliberative quality. Different factors might have contributed to the small group condition’s advantage over the other conditions. First, the small group condition arguably constituted a combination of a public space and a safe space (see e.g., Avery et al., 2013). The small group setting constituted a public space, because it offered opportunities for contestation among the students, who are merely acquaintances. Yet, it also constituted a safe space because it was divorced from the teacher and, hence somewhat insulated from the pernicious struggle between school culture and opposition culture that played out in the whole-class setting. Second, the number of students pr. discussion group (4–6 students) was close to the number of students often thought to be optimal for classroom discussions (Reich, 2007; Weasel, 2017): large enough to ensure the presence of opposing views and, hence, contestation; yet small enough to ensure that every student got a decent chance to speak and be heard by others. Last but not least, the small group discussions were not driven forward by the authority-laden questions of the teacher as is often the case in whole-class discussions (see e.g., Sætra, 2018). Instead, they were driven forward by informal norms of communication among the students, who were immediately compelled to take responsibility for the drive of the discussions to avoid awkward moments of silence.

The pair condition was found to promote equality as well as one indicator of engagement (questioning and metatalk). On the downside, however, it produced the smallest proportion of publicly relevant talk and opinion statements. The pair discussions were, thus, characterized by a considerable amount of social chitchat as well as factual statements incapable of opening the door to argumentation and contestation. The pair condition, moreover, generated much less contestation, engagement, and original argumentation than the small group condition, though it also generated fewer interruptions of the current speaker. While the pair condition had the unique advantage of promoting equality in speaking time and received responses, it generally did not foster the same level of deliberative quality as the small group condition. Three dynamics accounted for the above results. First, the students who had dominated the group discussions immediately took on the role as drivers of the pair discussions, which implied that they started questioning the less vocal and less dominant pair members about their covid-19-related opinions and beliefs. While the dominating students thereby gained control of the agenda and the pace of discussion, they also engaged with the less dominating students and offered them a lot of speaking time. This dynamic, then, simultaneously explains the high level of questioning and the high level of equality in the pair conditions. Second, the lower level of publicly relevant talk and correspondingly higher level of chitchat might have occurred because the pair setting is more reminiscent of a private intimate setting, characterized by intense emotions, than the group setting (Moreland, 2010). Contrary to Sætra (2018) but in line with Parker (2010), these results suggest that settings favorable to relationship building might sometimes run the risk of transforming the public space of the classroom into a private space that limits opportunities for publicly relevant talk and contestation. Third, as predicted by the epistemic approaches to deliberation (Goodin, 2008), the argument and information pools present in the pair discussions seemed to be smaller and less diverse than the corresponding pools in the group discussions. The smaller argument pools became apparent because students often complained that they “had nothing left to say” following less than five minutes of pair discussion. While the argument pools might of course simply have been drained at the time the students reached the final stage of the intervention, the small and homogeneous number of arguments, seemed to constrain the amount of contestation occurring during the pair discussions.

Inspirations for teaching practice

This study produced at least one counter-intuitive insight that could potentially inspire the practice of deliberatively inclined social science teachers: the students who seem to be least ready and motivated to engage in independent forms of classroom discussion (such as discussion in small groups) might sometimes be the ones who need these independent forms the most. This may apply in particular when students show contempt for the authority of the teacher, since the teacher will then be an ineffective galvanizer of discussion. When this is the case, delegating the task of driving the discussion forward to the students, might be an effective way of ensuring that the discussion displays at least some degree of deliberativeness.

Limitations of the study

The participating students were not representative of the Danish population of students, since they had similar socio-economic backgrounds and were homogenous with respect to culture and ethnicity. Consequently, teachers cannot use the findings made here as off-the-rack solutions to the local problems of their classrooms. When considering which discussion formats to use, teachers have to take account of the particularities of the teaching situation they face—including the content and purposes of the specific lesson and the specific student group.

Conclusion

This study has shown that the deliberative quality of classroom discussions may sometimes benefit from the absence of the teacher. This might be especially true when students disregard the teacher’s authority or when past interactions between the students and the teacher have been troublesome. Another conclusion to be drawn from this study is that different kinds of “teacher absence” are not always equally beneficial for the deliberative quality of classroom discussions. While the small group and pair settings seemed to differentially impact the deliberative quality of the investigated discussions, more research is needed to shed light on when, why and to what extend this is the case.

Acknowledgements

The author would like to thank the participating teachers for their invaluable help regarding the organization and implementation of the discussion session, and the students for their engagement.

Disclosure statement

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data supporting this study can be requested from the corresponding author.

Additional information

Funding

This work was supported by NordForsk [grant number 87663] and was carried out as part of the NordForsk financed Quality in Nordic Teaching Center of Excellence (QUINT).

Notes

1 Translated from Danish: “Ja, fordi hvor mange af jer har egentlig, når nu, I tænker, I var stramt stramt stramt bundet fra starten. I måtte ikke; I måtte ikke komme i skole, og I skulle blive hjemme. Hvor mange af jer har egentlig, sådan, løsnet lidt op på det? Er der nogen, der har fået lov til at få frie tøjler, og nogen, der stadig har fået at vide af mor og far, der hjemme, at man bliver lige. [Ja] navn på elev 1.” (Teacher, Condition A1, Video 0 [0:00:54 − 0:01:14]).

2 Translated from Danish:” Jeg tror her efter, hvor at Danmark har åbnet mere op, så tror jeg, der er flere af os, der også sådan er begyndt at mødes med venner; ikke hvor vi mødes halvtreds mennesker, men hvor man måske mødes i små grupper om aftenen” (Student 1, Condition A1, Video 0 [0:01:15 − 0:01:26]).

3 Translated from Danish: “Tænker du, [navn på elev 15], tænker du, man skulle; skulle man lukke ned igen?” (Teacher, Condition A1, Video 0 [0:03:08 − 0:03:11]).

4 Translated from Danish:” Altså, jeg tror nok inden for den nærmeste fremtid, vi nok kommer til at lukke ned igen, fordi smittetallet er steget så meget” (Student 15, Condition A1, Video 0 [0:03:12 − 0:03:17]).

5 Translated from Danish:” Men i forhold til; var det ikke; var det Sverige; nej, jo Sverige; så vil jeg sige, vi har gjort det bedre, fordi der ikke er så mange, der; mange, der er døde” (Student 14, Condition B, Video 45 [0:05:50 − 0:05:58]).

6 Translated from Danish:” De er dem, der har lavet de mest succesfulde indtil videre” (Student 7, Condition B, Video 0 [0:02:31 − 0:02:33]).

7 Translated from Danish:” Hvorfor?” (Student 17, Condition B, Video 1 [0:03:34 − 0:03:34]).

8 Translated from Danish: “Jeg tror, vi er lidt uenige med det der med Sverige; også sådan noget med om vi skal mødes” (Student 13, Condition C, Video 1 [0:01:35 − 0:01:39]).

9 Translated from Danish:” Men det er vel mange af de højreorienterede partier, for eksempel Venstre […]” (Student 8, Condition A1, Video 0 [0:04:42 − 0:04:44])

10 Translated from Danish:” Vi bliver nødt til at håndtere det rigtigt” (Student 2, Condition C, Video 3 [0:00:20 − 0:00:21]).

11 The normalized standard deviation, which is also referred to as the Coefficient of Variance, equals the standard deviation (SD) divided by the mean (M) and shows how unequally the shares of speaking time were distributed among all the students who spoke at some point during a particular session (the speakers).

Appendix

Table A1. An overview of deviations from the intervention plan.

Experimental conditions	Sessions	Deviation	Deviation excluded from analysis?
Whole-class condition (A1). Baseline. Students asked to discuss the two questions about the covid-19 lock down in a plenary session with the teacher present	Plenary 1	None	–
Group condition (B). First treatment. Students asked to discuss the two questions about the covid-19 lock down in small groups and without the teacher’s presence	Group 1	None	–
	Group 2	Two instances of pair talk (group split)	Yes, thirteen turns excluded
		Brief communication with author and assistant	Yes, three turns excluded
	Group 3	As opposed to the other groups, group 3 consisted of six students rather than four	No, deviation was judged to be minor
		Brief communication with the teacher	Yes, two turns excluded
	Group 4	Student 18 almost entirely inaudible	Yes, student 18’s turns excluded from all conditions
Whole-class condition (A2). Baseline. Students asked to discuss the two questions about the covid-19 lock down in a plenary session with the teacher present	Plenary 2	None	–
Pair condition (B). Second treatment. Students asked to discuss the two questions about the covid-19 lock down in pairs and without the teacher’s presence	Pair 1	One instance of cross-pair talk	Yes, two turns excluded
		Brief communication with the author	Yes, two turns excluded
	Pair 2	None	–
	Pair 3	Brief communication with the author	Yes, one turn excluded
	Pair 4	One instance of cross-pair talk	Yes, seven turns excluded
		Brief communication with the assistant	Yes, one turn excluded
		Brief communication with the teacher	Yes, three turns excluded
	Pair 5	Pair composed of students which were not in the same group	No, deviation was judged to be minor
		One instance of cross-pair talk	Yes, four turns excluded
	Pair 6	None	–
	Pair 7	Brief communication with the teacher	Yes, two turns excluded
	Pair 8	Student 18 almost entirely inaudible	Yes, student 18’s turns excluded from all conditions
		Lengthy communication with the teacher	Yes, four turns excluded
	Pair 9	Student 4 and 5, who should have constituted pair 9, left the classroom after the second plenary session, and, hence, did not participate in the pair discussions	Yes, student 4’s and 5’s turns excluded from all conditions