International Journal of Educational Research 90 (2018) 234–247

Contents lists available at ScienceDirect

International Journal of Educational Research

journal homepage: www.elsevier.com/locate/ijedures

Instructional discourse and argumentative writing

T
⁎
Joshua A. Morrisa, , Brian W. Millerb, Richard C. Andersonc, Kim Thi Nguyen-Jahield,
Tzu-Jung Line, Theresa Scottf, Jie Zhangg, Jingjing Suni, Shufeng Mah
a
Institute for the Science of Teaching & Learning, Arizona State University, 1000 S. Forest Mall, Tempe, AZ 85281, United States
b
Department of Elementary Education, Towson University, 8000 York Road, Towson, MD 21252, United States
c
Center for the Study of Reading, University of Illinois at Urbana-Champaign, 51 Gerty Dr., Champaign, IL 61820, United States
d
College of Engineering, University of Illinois at Urbana-Champaign, Champaign, IL, United States
e
Department of Educational Studies, Ohio State University, 1945 North High Street, Columbus, OH 43210, United States
f
Greene Street Friends School, Philadelphia, PA, United States
g
College of Education, University of Houston, United States
h
Institute of Education, Tsinghua University, Beijing, China
i
Department of Teaching and Learning, University of Montana, 32 Campus Drive, Missoula, MT 59812

A R T IC LE I N F O ABS TRA CT

Keywords: Fifth-grade students from two urban school districts completed an integrated unit on wolves.
Written argumentation Classes received either direct instruction (DI) or collaborative group work (CG). Analysis of
Classroom discourse reasoning in classroom talk indicated that CG students more often used connective and con-
Connectives trastive words and the performative verb phrases I think and I know than DI students. Analysis of
Audience
written arguments about a controversial question raised by the unit indicated that, compared to
DI students, CG students included more logical connectives, contrastives, and performative verbs,
produced fewer unelaborated arguments, more frequently asked rhetorical questions, and more
often considered both sides of the policy issue. The study provides fresh insight into how in-
structional frameworks can affect how students view themselves as writers in relation to a pro-
spective audience.

1. Introduction

Preparing students for productive civic engagement requires teaching them about the balance of the common good with in-
dividual liberty and how to think critically about issues of policy. Policy issues are too ill-structured for people to strictly apply the
rules of formal logic. The policy controversies for which educators are preparing their students involve reasoning from plausibility
instead of accepting or rejecting arguments based on undeniable truth (Kock, 2003). Incorporating the needs of the many into one’s
decision making involves affective as well as logical considerations. The question becomes, how do we best prepare our students to
engage constructively in the informal reasoning necessary to make sensible decisions about the sometimes emotional issues that
impact their community?
In answer to this question, this paper addresses ways in which classroom discourse impacts students’ reasoning about authentic
policy questions that involve multiple stakeholders, some of whom hold positions antithetical to one another. We examine the
reasoning of students taught in two common instructional frameworks, teacher-centered direct instruction and peer-led collaborative

⁎
Corresponding author.
E-mail addresses: [email protected] (J.A. Morris), [email protected] (B.W. Miller), [email protected] (R.C. Anderson),
[email protected] (K.T. Nguyen-Jahiel), [email protected] (T.-J. Lin), [email protected] (T. Scott), [email protected] (J. Zhang),
[email protected] (J. Sun), [email protected] (S. Ma).

https://doi.org/10.1016/j.ijer.2018.03.001
Received 16 May 2017; Received in revised form 26 February 2018; Accepted 1 March 2018
Available online 02 May 2018
0883-0355/ © 2018 Elsevier Ltd. All rights reserved.
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

groups, with the goal of identifying how the classroom talk within the two frameworks influences student reasoning. Our approach is
to identify differences in talk between the instructional conditions that could influence students’ individual reasoning and then
analyze variation in the students’ written argumentation attributable to differences in classroom talk.

2. Theoretical framework

This study is situated within a sociocultural view of learning according to which learners become attuned to the means of
communication within their learning community. The attunement involves positioning themselves in relation to the “constraints and
affordances” of the learning context (p. 255, Greeno, 2016; Lave & Wenger, 1991). By attending to the norms of learning contexts,
students discover how to use language to maneuver within a problem space while adhering to any locally organized rules of en-
gagement. Students’ language is dynamically scaffolded by their peers and teachers (Rogoff, 1990). Over time, students come to
represent these discourse processes, and their role in the process, as generalized frames, or schemas, and these schemas can influence
communication and meaning making between students and across learning episodes (Green & Harker, 1982). Schemas (Reznitskaya &
Anderson, 2002) or frames (Goffman, 1974) represent the commonalities among communication events and, “are both the tools that
facilitate the co-construction of knowledge and the means that are internalized to aid future independent problem solving activity”
(Palinscar, 1998, p. 353). The development of such schemes is scaffolded by the implicit and explicit ways in which the student is
invited by teachers and fellow students to engage in the activities of the classroom (Fernandez, Wegerif, Mercer, & Rojas-Drummond,
2001).
In this study we characterize the discourse processes within two instructional frameworks based on a systematic sampling of
lesson videos recorded during a six week intervention in which students were charged with making a decision about a controversial
public policy issue. Then, following an analysis of the students’ independent reasoning in a post-intervention essay, we made con-
nections between features of the classroom talk and the nuances of the student reasoning revealed in the essay.

2.1. Written argumentation

As our interest is in comparing the written arguments of students who experienced different instructional conditions, we first
present perspectives on the cognitive processes in constructing a written text. Writing is a process in which an author creates a text for
an intended audience (Flower & Hayes, 1981; Bereiter & Scardamalia, 1982). A critical element is the recognition that the audience
contextualizes the text in a parallel process. The success of the text largely depends on the degree to which mutual understandings are
exploited and potential confusions are anticipated (Nystrand, 1983).
Young writers struggle with balancing their expressive goals and the needs of the audience (Knudson, 1992; Scardamalia,
Bereiter, & Steinbach, 1984). Novices’ written productions suggest an additive conception where thoughts are simply translated to
writing in the order in which they occur to the author (Flower & Hayes, 1981). Bereiter, Scardamalia, and colleagues (1984, 1988)
refer to this process as knowledge-telling and explain that it, “Consists of reducing writing assignments to topics, then telling what one
knows about the topic” (Scardamalia, Bereiter, & Steinbach, 1984, p. 174). In contrast, knowledge-transforming (Scardamalia et al.,
1984) is a more sophisticated mode of writing that relates expressive means to perceived audience characteristics.
The goal of educators, then, is shifting novice writers from knowledge-telling to knowledge-transforming, but this has proven a
very difficult task in the case of written argumentation (Knudson, 1992). Argumentative writing is a genre acutely concerned with the
intended audience, as the audience determines the sorts of reasons and types of rhetoric that will be effective (Walton, 1998). Because
successful argumentation is so dependent on managing the problem space between the content of arguments and the audience, and
since the metacognitive oscillation between the goals of delivering content and being understood is difficult for students, it is not
surprising that argumentative writing is a difficult genre for students.
Studies have shown improvement in argumentative writing through prompting about the structure and purpose of argumentative
writing (Ferretti, Macarthur, & Dowdy, 2000; Knudson, 1992; Scardamalia et al., 1984). Similarly, argumentation can be improved
by establishing an authentic audience of peers and teachers as critics during the revision process (Midgette, Haria, & MacArthur,
2008; Wong, Butler, Ficzere, & Kuperis, 1996). Bearing in mind the purposes of argumentation and keeping in mind the audience
while writing have both been shown to improve argumentative writing.
In the current study we will explore how two common instructional frameworks, collaborative learning and direct instruction,
contribute to students’ argumentative writing, in particular whether texts created by the students include indicators of the students’
awareness of their audience.

2.2. Direct instruction

Teacher-led direct instruction is a straightforward approach for teaching students about a policy controversy. By controlling the
flow of information, teachers can methodically introduce difficult concepts and scaffold students’ understanding with examples and
differentiated explanations (Duffy et al., 1987; Pressley, 2006). Using questioning techniques, teachers can draw out misconceptions
and build any background knowledge they perceive their students are missing (Edwards & Mercer, 1987). By making themselves the
focal point of the lesson, teachers can alter their instruction on the fly in order to address misunderstandings or redirect non-
productive talk (Alvermann, O’Brien, & Dillon, 1990; Wells, 1999).
However, talk during teacher-centered instruction can be one-sided with little opportunity for students to make lengthy con-
tributions to the discussion or engage in extended reasoning themselves (Cazden, 1988; Mehan, 1979). Direct instruction usually has

235
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

an initiation-response-evaluation (IRE) (Mehan, 1979) pattern, during which speaking turns follow the predictable sequence of the
teacher asking a question, the student responding, and the teacher evaluating the student answer or giving feedback. Although the
basic structure of an IRE can, in the hands of a skillful teacher, be effectively exploited to encourage explicitness or elaboration (Boyd
& Rubin, 2006; O’Connor & Michaels, 1993; Wells, 1993), in its most common form the IRE yields inauthentic questions that can
negatively affect students’ engagement (Dillon, 1985). Tharp and Gallimore (1988) showed that during recitation teachers ask a
preponderance of yes/no questions and only give specific feedback when answers are incorrect. The teacher holds interpretive
authority and may present feedback in a way that suggests the interpretation is non-negotiable.
Direct instruction is highly teacher-centric with control of the turntaking, control of topic, and control of interpretation all in the
hands of the teacher. After studying teacher-student dynamics in over a hundred classrooms, Nystrand et al. (1997) concluded that
most classroom discourse was monologic in nature, meaning that there was an assumption of “coincidence of speaker’s meaning and
listener’s interpretation,” with the goal being to pass on factual information (p. 384, Wells & Arauz, 2006). Wells and Arauz (2006)
explain that this sort of discourse has its place in the classroom but is problematic in that it is “by nature authoritative, not open to
question or alternative perspectives (p. 385).”
A classroom that is largely monologic can produce passive learners who are not expected to question or give alternative or
personal interpretations; rather, they are expected to reproduce previously presented information. Because answers involve re-
plication more than generation or synthesis, a student can be brief and rely unthinkingly on the teacher to fill in any missing element.
Displays of knowledge can become perfunctory and serve no real dialogic purpose (Nystrand, Gamoran, Kachur, & Prendergast,
1997). There is mounting evidence that such discourse not only affects knowledge construction, but can have a lasting impact on how
students think about knowledge, learning, and the nature of intelligence (Dweck, 2006; Johnston, 2004).
Direct instruction has been employed in several types of instructional programs to improve student reasoning. Often instruction is
focused on student mastery of component parts of an idealized reasoning or decision-making process (Beyer, 1985; Kuhn & Udell,
2003). Other studies have presented students with a normative model of reasoning to help them perceive the requisite parts of a good
argument and to organize their own thinking. Two of the more common approaches to improving students’ reasoning are through
establishing goals (Ferretti, Lewis, & Andrews-Weckerly, 2009; Nussbaum & Kardash, 2005) or providing scripts or other organi-
zational tools (Munneke, Amelscoort, & Andriessen, 2003; Stegmann, Weinberger, & Fischer, 2007). In this study, direct instruction
teachers did not utilize any of these approaches. There was no explicit instruction in the components of a good argument, and there
was no systematic use of prompts to remind students to employ desirable components of arguments.
In this study, direct instruction teachers led their students through a unit on a controversial public policy issue. Unlike most
curriculum materials for middle grade students, the unit was organized in terms of arguments on both sides of the controversy. Direct
instruction was expected to be a useful pedagogy for making sure students understood the sides of the controversy and imparting
knowledge relevant for reasoning about the issue. But, the one-sided nature of the discourse during direct instruction may have
undermined students’ ability to engage in an authentic reasoning process themselves, due to the reduced necessity to consider the
knowledge and potential disagreement of their addressee, who was usually, and most importantly, the teacher.

2.3. Collaborative group work

In small-group collaborative discussions the students control turntaking and hold interpretive authority. A ‘shared forum’ (Wells &
Arauz, 2006) enables students to see how their words can shape the actions, attitudes and ideas of their peers. Unlike the more
circumscribed role students play in teacher-led instruction, collaborating students can contribute ideas that are not necessarily known
or accepted by the other students in the group. Collaborating students’ use of language is not tightly constrained by teacher directives.
They are engaged in the process of crafting what they say while managing how they say it.
In contrast to direct instruction, peer-led group work provides ample opportunity for students to personally and collaboratively
explore their reasoning about a controversy. Previous research establishes that collaborative discussions can lead to improved rea-
soning in writing (Dong, Anderson, Lin, & Wu, 2009; Reznitskaya et al., 2001), interviews (Kuhn, Shaw, & Felton, 1997; Kuhn &
Udell, 2003), and on critical reasoning tasks (Mercer, Dawes, Wegerif, & Sams, 2004; Wegerif, Mercer & Dawes, 1999). Particularly
relevant to the current study’s design are findings from the Collaborative Reasoning project (Reznitskaya & Anderson, 2002) that
demonstrated that even without any explicit instruction about argument form or function, collaborative group work has a positive
impact on reasoning (Reznitskaya et al., 2001).
This distinction between student-managed and teacher-led instruction is usefully characterized through a consideration of the
concept of an addressee in language production. In direct instruction the addressee is the teacher, a known authority who requests
answers only to what has been asked. In collaboration, the addressees are fellow students with different opinions and agendas.
Because each student in a group represents a separate addressee for other students, discourse choices are shaped by students’ sen-
sitivity to what is being said and who is saying it. We hypothesize that varied experiences with different addressees improves
students’ sensitivity to an audience, which is an integral part of argumentative writing (Bereiter & Scardamalia, 1982; Harris,
Graham, Brindle, & Sandmel, 2009).

2.4. Current study goals and hypotheses

In the current study, fifth-graders worked through a unit on wolf management while participating in direct instruction or col-
laborative group work. In direct instruction, students read arguments on both sides of the unit’s major policy issue – whether a town
should be allowed to kill a pack of wolves. They were led over the arguments by their teacher, but had restricted opportunities to

236
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

actively argue about the issue among themselves. In collaborative group work, students also read the arguments on both sides of the
major policy issue, but unlike the direct instruction students, they actively argued about the policy issue and related subsidiary issues.
Our goal was not to champion one instructional framework over another. Instead, we assume that each instructional frame would
benefit students differently. The criterion assessment was a writing task where students were asked to justify a position on the central
policy issue in the curriculum unit. Since an argument schema will be elaborated differently depending on the characteristic features
of the instructional discourse in which students were involved, nuanced differences in the persuasive writing can be theoretically
linked to the instructional discourse during direct instruction and collaborative discussion.
We have specific hypotheses about how the instructional discourse would affect student writing. First, because students in col-
laborative groups initiate and sustain arguments we hypothesized that students from the collaborative groups would show a greater
sensitivity to audience through consideration of counterarguments and through rhetorical moves to engage the audience, including
the rhetorical engagement of the audience through questions. Second, we predicted that students in direct instruction condition
would include a greater variety of reasons as a result of their being lead through the material by a teacher who could emphasize all
the key concepts.
Based on previous research demonstrating the positive effects of collaboration on reasoning (e.g. Reznitskaya et al., 2001) the
forecast was that the decision letters of CG students would be more likely to contain all of the elements of a complete argument –
position, argument, counterargument, and rebuttal – than the decision letters of DI students. However, unlike previous CR studies,
which involved business-as-usual controls, in this study DI students experienced arguments on all sides of the wolf controversy, which
may have enabled them to provide complete arguments. One possibility is that students in collaborative groups may write more
complete arguments because their experience with the unit and arguing with their classmates over the policy issue, may lead to an
argument schema with better articulated slots for counterarguments and rebuttals as they had embodied experience with individuals
disagreeing with them and responding to those disagreements. Alternatively, teachers may successfully characterize the dialectical
complexity of the policy issue ensuring that students understand the multiple sides, leading to more students in the direct instruction
condition to provide complete arguments.

3. Method

3.1. Participants

This study examines data from the first year of a two-year study, during which twelve fifth-grade teachers from seven schools in
two districts implemented the Wolf Reintroduction and Management Unit in their classrooms. The study design was quasi-experimental
with classrooms matched on demographics randomly assigned to one of two conditions: Direct Instruction (DI) or Collaborative
Group Work (CG). Six classrooms were from three schools in a Midwestern industrial city. These schools serve predominantly low-
income African-American families. The percentage of students eligible for free or reduced-price lunch ranged from 80% to 98%. Six
classrooms were from four schools in a district in a major metropolitan area in the Midwest serving predominantly low-income
Hispanic-American families. The percentage of students in these schools eligible for subsidized lunch ranged from 80% to 86%.
Overall, 45.7% of the participants were Hispanic American, 41.5% African American, 9.4% European American, and 3.5% Other.
At the beginning of the study, students received a reading comprehension assessment (MacGinitie, MacGinitie, Maria & Dreyer,
2000), and a task that required rapid naming of pictures of common objects (hereafter, RAN) that served as a measure of basic oral
English fluency. For six students, missing pretest reading comprehension scores were imputed from the RAN measure and other
student characteristics. There were no pre-instructional condition differences in reading comprehension (t(235) = 1.02, p = .30) or
RAN (t(235) = 0.12, p = .90).
There were 294 students who were enrolled at some point in either a DI or a CG classroom. Due to high student mobility in the
participating schools, some students left before writing the essay that provided the outcome measures. There were also students who
were not included in the analysis because they came into classrooms over halfway through the unit. There were a few students who
did not agree to participate and some students who did not return a parental consent form. At final count the essays of 237 students
were available for analysis.

3.2. Implementation of direct instruction and collaborative group work

Teachers participated in a two-day workshop pertaining to their assigned instructional condition. The first day of the workshop
was dedicated to the specifics of the instructional method. The second day focused on the particulars of the curriculum. Teachers
remained segregated for the second day in order to best explain aspects of the research design that were specific to each instructional
condition (e.g. jigsaw design in CG).
The curriculum was the Wolf Reintroduction and Management Unit. All the classes worked through the unit for six weeks. Students
were asked to role-play being officials at the Wolf Management Agency. The agency had been contacted by an imaginary town,
Winona, asking for permission to hire professional hunters to kill a pack of wolves that many citizens regarded as a problem. The
students were told that at the end of the unit they would be asked to write the town with their decision about whether or not, as
officials at the Wolf Management Agency, they would allow Winona to hire hunters to kill the wolves.
The unit was divided into three sections that introduced the students to Winona’s ecosystem, how Winona’s businesses compete
and rely on each other, and the rudiments of how a public policy is made. Each of these three sections was comprised of an
information booklet and an activity booklet. The information booklet contained readings specific to the topic and the activity booklet

237
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

contained a variety of tasks designed to reinforce concepts covered in the information booklets as well as challenge the students with
extension questions. Many of the key concepts of the unit (e.g. interdependence, relationship between producers and consumers)
were covered in each sub-section. For example, the concept of interdependence was covered in all three sub-sections: Food web
(Ecosystem), Production web (Economy), Effects of a policy for the common good (Public Policy).
Students in the DI condition completed all three pairs of information and activity booklets. Students in the CG condition were split
into heterogeneous groups, each a cross section of the class in terms of reading level, talkativeness, ethnicity, gender, and attitude
towards learning. Each CG group was charged with becoming ‘experts’ in one of the unit’s three sections. Each CG classroom had at
least one ecosystem expert group, economy expert group and public policy expert group. Each of these expert groups read the
information booklet and completed the activity booklet on its topic. Then, each group prepared a poster on its topic and made a
presentation to the whole class. While CG groups did not read through each information booklet, they were exposed to all the key
concepts and relationships through the poster presentations. To assure that the main concepts and relationships were presented by
each group, and time was used efficiently, production and presentation of the posters was designed as a structured task. The assembly
of the poster, and which students would do which tasks, was left to the group.
Before the Wolf Unit, students in the CG condition had three literature discussions in the Collaborative Reasoning format (see
Waggoner, Chinn, Yi, & Anderson, 1995). The students read a story and discussed a big question that related to an ethical and
practical dilemma the protagonist was facing. Collaborative Reasoning involves open participation, meaning that students speak
freely without being nominated by the teacher. Teachers were encouraged to follow a “coach-on-the-sideline” approach to scaffolding
collaborative interaction such that the students were responsible for regulating their discussions and the teachers would only in-
tervene briefly when necessary. Students in DI also read these stories and participated in whole-class discussions about the big
question.
In their professional development, both CG and DI teachers were instructed in ways to frame the discussions they would have in
their classes (peer-led group discussions and whole-class discussions, respectively). Teachers were encouraged to set guidelines
around taking positions on the big question, advancing reasons to justify their positions, and supporting reasons with evidence and
other backing.
Students in the CG condition also held two Collaborative Reasoning discussions during the Wolf Unit about whether Winona
should be allowed to hire hunters to kill the wolves. The first discussion was held at the beginning of the unit when the students were
first presented with Winona’s request. The second discussion was held near the end of the unit after the students had spent a number
of weeks gathering evidence and shaping their opinions. Teachers in DI also coordinated whole-class discussions about the central
question before and after completing the wolf unit. These discussions were guided by teachers but were not scripted in anyway.
Student-to-student talk was allowed; however it was exceedingly rare.
An RA was a participant observer in each classroom, recording video as well as supporting the implementation of the unit. The RA
was in daily contact with the teacher through conferences, phone, and email. The RA assisted the teacher several ways, including
providing reminders about supplies needed for activities, answering questions a teacher had about the unit or procedures for the
study, highlighting key concepts for instructional focus, facilitating group work (CG condition), providing one-on-one help (DI and
CG condition).

3.3. Features of classroom discourse

Seventy-two, 4-min excerpts from videos of lessons during the Wolf Unit were transcribed and analyzed, six excerpts from each
class, one excerpt from each week of the unit, usually drawn from lessons on Tuesday. Within a lesson, the starting point for the 4 min
excerpt was chosen at random with two constraints. The first 4 min and final 4 min of lessons were not sampled. This reduced the
likelihood of sampling non-instructional activities (e.g. filing into class, passing out and collecting materials). Within each class three
of the sampled excerpts came from the first half of lessons and three from the second half as determined by a randomized, coun-
terbalanced schedule.
The goals of the discourse analysis were to document the amount of student and teacher talk, and to examine the talk for markers
of connectedness, relational thinking, and interactional positioning. A word search was performed for the common conjunctions and,
but and the causal connectives because, so, and then. Previous research has shown conjunctions are an indicator of elaborated and
connected argumentative talk (Anderson, Chinn, Chang, Waggoner, & Yi, 1997; Mercer, Wegerif, & Dawes, 1999). As a low-inference
index of interactional positioning, a search was made for the verb phrases I think and I know, which are often utilized in making
arguments and conversational alignments with other discussants. Other performative verbs and verbs of propositional attitude,
including believe, claim, conclude, and doubt, were not present in the transcribed segments.
To increase validity, terms were screened in context to ensure that counted words were related to the Wolf Unit and that the uses
were indicative of relational thinking. Terms were removed that were simple repetitions after a maze or false start; in a speech
fragment with no propositional content; read verbatim from the text; not related to the unit (e.g. talk about after-school activities);
not expressing relational thinking.

3.4. Wolf decision letter

At the end of the unit, students individually wrote an essay – a ‘decision letter’ – about whether Winona should be allowed to hire
professional hunters to kill the wolves. The writing prompt was presented by an RA following a protocol, so that the task would be
uniformly administered. Students were given 45 min to write. The handwritten letters were keyed into electronic files with spelling

238
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Fig. 1. Use of connectives in classroom discourse.

corrected to enable computer-assisted coding. The letters were then coded for the quality of students’ reasoning in order to investigate
how instructional conditions impacted the students’ written arguments.

4. Results

4.1. Classroom discourse during direct instruction and collaborative groups

To gauge the stability and consistency of the measures, four correlations were computed for each connective and performative
verb: one between the first three excerpts and the last three, one between the odd excerpts and the even excerpts, and the same first/
last and odd/even correlations separately by instructional condition. These correlations were all above 0.80, suggesting stability in
the measures and a substantial degree of consistency across excerpts.
Fig. 1 shows that students employed almost all of the connectives used during CG lessons, while teachers supplied more than half
of the connectives during DI lessons. To account for differences among segments in total words spoken, and differences among
classrooms in average pretest reading comprehension, these variables were included in the regression model. Total words was
incorporated as an offset variable, as recommended by Agresti (2007), to control for the fact that counts of connectives are related to
total words. This is a more conservative treatment than entering total words as a covariate. Condition was dummy-coded with CG = 1
and DI = 0. The analysis indicated that the students in CG condition used more connectives than students in the DI condition
(β = 1.52 SE = 0.22, χ2(1) = 45.23, p < .001), teachers in the DI condition employed more connectives than teachers in the CG
condition (β = −1.07, SE = 0.26, χ2(1) = 17.40, p < .001) and including uses by both students and teachers, CG students and
teachers together employed more connectives (β = 1.85, SE = 0.48, χ2(1) = 15.12, p < .01). Models were run assuming a negative
binomial distribution for count data with over-dispersion, which in every case fit better than either Poisson or zero-inflated Poisson
models.
Fig. 2 provides a further break down of the six conjunctions indicative of connected talk. These counts do not conform to a Poisson
distribution because of over-dispersion, so we again assumed a negative binomial distribution. Total words spoken during segments,
and average classroom reading comprehension were included in the models, although these factors were not significant in every
analysis. As before, condition was dummy-coded with CG = 1 and DI = 0. Students in the CG condition significantly more often used
the connectives because (β = 1.11, SE = 0.36, χ2(1) = 9.26, p < .001), and (β = 1.31, SE = 0.24, χ2(1) = 28.12, p < .0001), but
(β = 1.60, SE = 0.31, χ2 (1) = 25.58, p < .001), so (β = 2.11, SE = 0.58, χ2(1) = 13.32, p < .001), and then (β = 1.66,
SE = 0.47, χ2(1) = 12.34, p < .01). The analysis of performative verb phrases showed that CG students used the phrases I think

Fig. 2. Use of connectives and performative verbs in classroom discourse.

239
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Fig. 3. Use of logical connectives and performative verbs in student decision letters.

significantly more than students in DI (β = 2.04, SE = 0.51, χ2(1) = 12.56, p < .01). There was no significant difference between
conditions in the use of I know.
We evaluated whether there was an increase in connectives indicative of relational thinking from the beginning to the end of the
Wolf Unit. A model of use of connectives was constructed in which lesson order was entered as an ordinal variable, along with other
class-level variables such as mean comprehension scores, mean RAN scores, and instructional condition. Controlling for other factors,
order was not a significant predictor (B = 0.03, SE = 0.05, X2(1) = 0.42, p > 0.10) (Fig. 3).

4.2. Analysis of connectives and performative verbs in wolf decision letters

The first analysis entailed a search of students’ wolf decision letters for the connectives and, because, so, and if, the most commonly
used contrastives but and instead, and the performative verb phrases I think and I know. CG students made significantly more use of
connectives (β = 0.20, SE = 0.08, χ2(1) = 6.65, p < .01) than DI students (Fig. 3). The same comparison was rerun removing
counts of and in order to obtain a better indication of use of logical connectives. CG students outperformed students in DI in use of
logical connectives (β = 0.28, SE = 0.07, χ2(1) = 12.66, p < .001). Although it did show up in classroom talk, the adverb then was
not included in this analysis because it was never used in students’ writing. The most commonly used contrastive was but while a few
students used instead. Students in CG were more likely to use these contrastives in their writing (β = 0.42, SE = 0.17, χ2(1) = 5.39,
p < .05), and more likely to use the performative verb phrases (β = 0.70, SE = 0.18, χ2(1) = 15.03, p < .001) than DI students.
These results parallel the findings from the analysis of classroom talk.

4.3. Coding argument elements in wolf decision letters

The next analysis of the wolf decision letters involved aspects of argumentation. Like previous work (Golder & Coirier, 1994;
Reznistkaya et al., 2001), we started with codes that would capture a minimal argument structure (position, reason) and built on this
in an attempt to capture writing that indicated increased complexity of argumentation (elaboration of reasons, counterarguments).
We chose not to use Toulmin’s model because, in our experience, it can be difficult to use to characterize long stretches of written
discourse. Coding proceeded in two steps. First, the letters were divided into communication units (Loban, 1966), defined as a main
clause and its subordinate clauses. We chose C-units as opposed to the similar T-unit (Crooks, 1990), because the rules of C-unit
division are less syntactically driven and allow for rule-based division of colloquial and substandard writing. The essays were divided
into C-units by three raters working independently with satisfactory reliability.
The argument analysis involved four categories: statement of position, support for that position, consideration of a counter
position, and response to a counter position. Table 1 provides illustrations of these categories. Reasons in support of a stated position
were coded as elaborated or unelaborated. Elaborations were propositions that included information increasing the relevance or
further justifying an explicitly stated reason. Elaborations were explanations that provided causal chains, and/or details for why a
circumstance envisioned in a reason might come about, or evidence that pointed to a fact or authority to increase the credibility of the
reason. Unelaborated reasons were discrete premises that were in support of the position, but were not further developed by the
student. Some writers also supported their arguments with rhetorical questions, which were separately tabulated.
Counterarguments and rebuttals to counterarguments were coded. Propositions that contained a reason that ran counter to the
stated position were coded as counterarguments. Context of use was evaluated to further ensure that the student was considering a
counterargument as opposed to entertaining two equally held positions. Some letters expressed uncertainty about a position and

240
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Table 1
Most common elements found in student essays.
Statement of position
I think they should hire professional hunters to kill wolves…
My decision is no, I am not going to give professional hunters permission …
Support for position
Elaborated reason
Wolves are bad for the ecosystem. Wolves are bad because they eat elk and the elk disappear
Wolves should not be hunted. The wolf biologist said that the wolf is important to the ecosystem.
Unelaborated reason
Wolves should be hunted. They eat elk.
Rhetorical questions
How would you feel if you were a wolf and someone tried to kill you?
Consideration of reasons on both sides
Counter reason
…shouldn’t kill the wolves…but I know that they keep attacking other animals
Consideration of counterposition
…the wolves should be killed…I know that some people do not want the wolves killed but they're dangerous
Response to counterposition
Concession
They think the wolves will attack the dogs, and that's true.
Rebuttal
They are worried about the livestock but wolves don't kill that many sheep

provided reasons on both sides; these letters were holistically coded as both positions, and the individual reasons were coded as
consideration of both positions. Some students explicitly named the opposing position but did not give any reasons in support of that
position; these were coded as consideration of a counterposition.
Counterarguments were also categorized in terms of whether they expressed an elaborated or unelaborated explanation of the
counterargument. Likewise, rebuttals were coded as elaborated or unelaborated. Finally, instead of rebutting the counterargument, a
few students were coded as making a concession to the counterargument.
The coding scheme was put through a trial run on decision letters produced by fourth graders who had received the Wolf Unit but
were not participating in the current study. A few modifications were made to fine-tune a low-inference protocol for determining
when children were considering a counterargument and rebutting a counterargument. The 237 essays from students in the current
study were then coded by two graduate students blind to condition. After coding 98 essays, the raters reviewed the coding scheme but
determined that no major changes were needed. The only alteration was a stricter definition of what counted as a hypothetical
argument. Previously coded essays were re-coded to reflect the slight modification. Then, the remaining 139 essays were coded.
Interrater reliability across all codes was satisfactory (Cohen’s κ=0.89).
The average length of essays was M = 15.5 C-units (SD = 7.3). Most children took a position on the issue (97%) and supported
their position with at least one unelaborated reason (97%). Fewer students supported their positions with elaborated reasons (81%).
Around 35% of the students considered a counterargument, and 27% attempted to rebut the counterargument, while four students
conceded to the proffered counterargument. Table 2 provides further descriptive statistics on how students in the two instructional
conditions supported their arguments.
Letters were evaluated in terms of whether they contained a complete argument and elaborated arguments. An essay was defined as
presenting a complete argument if it included a position statement, support for the position in the form of a reason, consideration of a

Table 2
Descriptive statistics for elements of arguments and relative frequency of binary features.
CG DI
Counts Mean SD Mean SD

Unelaborated Reasons 1.42 1.58 2.66 2.12

Elaborated Reasons 1.95 1.51 1.88 1.46
Rhetorical Questions 0.78 1.28 0.27 0.81
Counter reason
Unelaborated 0.56 0.42 0.61 0.66
Elaborated 0.32 0.86 0.34 0.67
Both positions 0.02 0.25 0.03 0.24
Counter position 0.11 0.38 0.12 0.43

Relative frequencies CG DI

For killing wolves 9% 11%

Against killing wolves 82% 81%
Total argument 29% 21%
Considered both sides 46% 31%

241
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Table 3
Generalized hierarchical linear models predicting argument elements.
Elements of Argument

Unelaborated Reasons Elaborated Reasons Rhetorical Questions Counterarguments Rebuttals

Effect Est. SE Est. SE Est. SE Est. SE Est. SE

Intercept −1.90 0.31 −1.25 0.27 −3.88 −0.68 −3.31 0.53 −3.58 0.56
RAN 0.31 0.30 −0.19 0.01 −0.11 0.02 −0.01 0.55 −0.01 0.59
Compr. *
−0.02 −0.01 **
0.02 −0.01 −0.01 0.02 0.01 0.01 0.02 0.01
Condition *
−0.50 −0.16 0.05 0.17 **
1.07 0.30 0.38 0.31 0.38 0.26

−2LnLike 814.78 1176.57 401.89 537.87 499.53

AIC 826.78 1188.57 413.89 549.87 509.53

* p < .05.
** p < .01.

counterargument, and a rebuttal to that counterargument. The letters of 59 (25%) students met these criteria. Elaborated arguments
were the total of elaborated reasons used in any of the three parts of the argument. There were 191 students who displayed at least
one elaborated argument element, and these essays that were more argumentatively rich, averaging M = 5.3 (SD = 4.4) elaborated
reasons per essay.

4.4. Statistical models predicting argument elements in wolf decision letters

Because students were assigned to instructional conditions by classrooms, hierarchical linear modeling was used to accommodate
the nested data structure. Students were nested within 12 classrooms. This is a satisfactory although minimal number of groups for
HLM (Snijders & Bosker, 2009). We also ran every analysis using multiple regression analysis. The HLM model fit better in every case,
although the parameter estimates were virtually identical. We report the results of the HLMs because they fit the data better.
Because the outcome variables were over-dispersed count data, all models were fit assuming a negative binomial distribution that
was more appropriate than a Poisson distribution (Barron, 1992). For every outcome variable, models were fit assuming each of the
two distributions allowing us to perform a log likelihood ratio test to ascertain which distribution was more appropriate. A negative
binomial distribution improved overall model fit except where noted.
To test the hypothesis that the DI and CG conditions would produce unique types of argument support, different types of argument
support were regressed on instructional condition, controlling for letter length, student characteristics, reading comprehension, and
RAN time. Of the student characteristics, age, ethnicity, and gender, only age was significant in some models, but in most cases these
characteristics were not significant or led to poorer model fits. In the final models, then, the measures of argument were regressed on
reading comprehension, RAN time, and instructional condition. Due to missing data the models are based on 228 of the essays.
Parameter estimates are provided in Table 3.
Models were fit with and without control for letter length. When models were fit without length, the only change was that the
reading comprehension measure accounted for more variance. Importantly, whether letter length was in a model did not have any
influence which would change the findings reported below; that is, no null instructional condition effects were made significant, and
no significant instructional condition effects were made null. For each of the hierarchical linear models a random intercept was tested
as well as a random slope. In all models, fitting a random slope led to a failure to converge because the Hessian matrix was not
positive definite. Models failed to converge even after rescaling the predictor variables.
There was no difference between the CG (=1) and DI (=0) conditions in number of elaborated reasons (β = 0.05, SE = 0.17,
p > .10). Following earlier research (Reznitskaya et al., 2001), we were interested in counterarguments and rebuttals. But again,
instructional condition proved to be a nonsignificant predictor of number of counterarguments (β = 0.38, SE = 0.31, p > .10) and
rebuttals (β = 0.38, SE = 0.26, p > .10). Thus, the essays of students in the DI and CG conditions contained similar numbers of
elaborated reasons, counterarguments, and rebuttals.
Because Murphy, Wilkinson, Soter, Hennessey, and Alexander (2009) concluded that dialogical methods are “more potent for
students of below average ability” (p. 760), instructional condition by reading level interactions were examined to ascertain whether
CG had a disproportionate effect on low readers. The interaction was not significant for elaborated arguments (p > .05) or un-
elaborated arguments (p > .05) indicating that DI and CG had comparable effects on students of different reading levels.
Turning to more subtle variations in argument production, we looked at the rate of unelaborated arguments and found that
students in the CG condition produced significantly fewer unelaborated arguments than students in the DI condition (β = −0.50,
SE = −0.16, p < .05). CG students were significantly more likely (β = 1.07, SE = 0.30, p < .01) than DI students to employ
rhetorical devices meant to draw the reader in (e.g. What would you do if you were a wolf?). These results are in line with our
hypothesis that the two instructional conditions would support their arguments in distinctive ways, DI with a greater number of
unelaborated arguments and CG with questions addressed to the reader.
Two aggregate codes were created to capture additional features of the student’s writing. Some students appeared to be aware of
the dialectical nature of the policy issue, but their essays did not adhere to a strict position, reason, counterargument, rebuttal format.

242
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Table 4
Types of reasons used in essays. Proportion of students who used at least one of the reasons in the category.
CG DI

Killing wolves is hypocritical 0.37 0.39

If we kill wolves then you have to kill other animals that kill to live
They only attack you if you're provoking thema
They're related to dogs and we don't kill dogs
We stole where they're living in the first placea
We kill and eat livestock too, but we don't get in trouble

Concern for the community's safety and economic welfare 0.32 0.49
Removing wolves will negatively affect businesses in towna
Tourism will go down, businesses that depend on tourism will go downa
Government compensates ranchers for dead livestocka
The town will be overrun with wolf prey
If we don't do something now it will get worse
Wolves are eating livestocka
Keeping wolves will negatively affect businesses in towna
People in town are worrieda

Wolves are keystone species 0.54 0.49

They are endangered/were once nearly extincta
Wolves help balancea
Wolves will cause deer and elk to decreasea
Wolves kill elk. Elk Population down. Plant diversity upa
Wolves leave their prey which helps population of scavengersa

Natural Law 0.51 0.36

Animals can't speak for themselvesa
It's cruel to killa
They are beautiful animalsa
They deserve to live
How would you like it if you were killed?
They are like people (e.g. have feelings, emotions, needs)
Puppies won't have mothers
It's their nature to eat/need meata

Misunderstanding 0.34 0.34

They don't kill that many animalsa
A wolf has never killed a humana
They eat animals not people
They keep to themselves/not aggressivea
Sightings in town may not have been wolvesa
People who want the wolves killed are misinformeda

a
Facts or opinions of stakeholders that were explicitly mentioned in the unit.

Some students made a rebuttal against an implied counterargument or defended both sides of the policy issue. There were too few of
these instances to model separately, but we combined these essays with essays of students who established a position and considered
a counterargument as well as essays of students who entertained both positions. A binary code was created to represent individuals
who were clearly aware of the different sides of the issue (1 = at least one counterargument, rebuttal to implied counterargument, or
consideration of reasons on each side of issue, 0 = none of the three). A multilevel logistic regression indicated that CG students were
more likely to recognize the different sides of the issue than DI students (β = 0.81, SE = 0.37, p < .05).
There was no significant difference between instructional conditions on the composite measure of a complete argument
(β = 0.42, SE = 0.32, p > .10) indicating that DI and CG students were equally likely to support a claim with at least one reason,
consider an argument counter to their own position, and ultimately rebut the counterargument. The results of these two logistic
regression analyses shows that CG students were more likely to address both sides of the issue; however, among those who took a
definite position, CG and DI students were equally likely to produce a complete argument.
The types of reasons that students advanced to support their arguments was investigated by reading the letters, keeping a running
list of all reasons used, and then categorizing the reasons according to the underlying theme. Assigning similar reasons to categories
was a low-inference endeavor as most of the reasons had little variability in word choice or syntax. Reasons were separated into
categories by one coder, and a second coder blind to condition labeled reasons using provided categories. Percentage agreement on
the types of reasons for a quarter of the essays was above 90%. The reasons and categories are presented in Table 4, along with the
proportion of students in each instructional condition who incorporated at least one reason from each category. The number of
distinct reasons advanced by CG students (M = 3.2) and DI students (M = 3.3) was almost identical. Furthermore, CG and DI students
were equally likely to use concepts from the unit (β = 0.01, SE = 0.15, p > .10).
Students in the two conditions varied in the types of reasons they preferred. DI students preferred reasons that pertained to the economic
well-being of the businesses in Winona, and the safety of the town citizens. Whereas CG students preferred natural law arguments about the
rights of animals. Reasons in the other three categories were employed equally often by students in the two instructional conditions.

243
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Fig. 4. Number of elaborated reasons in student writing as a function of the frequency of because and so in classroom discourse.

To conclude this section, CG students and DI students were equally likely to write essays that contained all of the basic elements of
an argument. However, CG students were more likely to link propositions using because, so, and if, made more use of performative
verb phrases, wrote essays with fewer unelaborated arguments, and were more likely to incorporate a rhetorical question. CG
students more often made arguments based on the natural rights of animals whereas DI students more often based their arguments on
the economic well-being and safety of citizens in the community.

4.5. Relationship between spoken and written discourse

Three of the connectives found in classroom discourse (because, so, and but) have likely corollaries in student writing. Because and
so should be related to students’ elaboration of the reasons they use in their arguments whereas but should be related to the number of
counterarguments. Figs. 4 and 5 show that the data are consistent with the idea that use of certain connectives in classroom discourse
is related to use of connectives in the writing task. To further test this relationship, classrooms were divided into high, medium and
low in terms of the rate that certain connectives were used in the class. Because and so were classified as causal connectives and
classrooms were divided into three categories of use: High use (more than one per excerpt), medium use (around one per excerpt),
low use (less than one per excerpt). A generalized linear mixed model with categories of causal use dummy coded (0 = low,
1 = medium, 2 = high) documented that classrooms with high (B = 0.48, SE = 0.17, p < .05) and medium (B = 0.57, SE = 0.17,
p < .05) use of connectives were associated with greater number of elaborated reasons in the wolf decision letter than classrooms

Fig. 5. Number of counterarguments in student writing as a function of frequency of but in classroom discourse.

244
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

with low usage. Similarly, students from classrooms with high use of but (B = 1.55, SE = 0.51, p < .01) presented a greater number
of counterarguments in the wolf decision letter than classes with less frequent use. Each of the models controlled for instructional
condition, reading comprehension, and RAN with model building and testing similar to methods described in previous analyses. No
interactions were significant.

5. Discussion

A major finding of this study is that students talking with one another during collaborative group work were several times more
likely to use the words but, because, then, so, and and than students speaking during direct instruction, after screening to insure that
the terms indexed relational thinking. The increase in use of connective words is one indication that CG students were expressing
arguments with more explicit relationships between argument elements. Another major finding is that students engaging in colla-
borative group work were several times more likely to use the phrase I think than students receiving direct instruction. This statement
represents student positioning in relation to ideas and to other students.
Differences in classroom discourse during collaborative group work and direct instruction were reflected in the decision letter that
students wrote at the end of the Wolf Unit. One of the main differences was that CG students made much more use of logical
connectives, contrastives, and performative verb phrases than DI students. Another difference was that the essays of CG students
contained significantly fewer unelaborated arguments than the essays of DI students, even after controlling for letter length. While
some DI students produced both elaborated and unelaborated arguments, others just supported their position with a few brief
unelaborated reasons.
CG students had a distinctive way of supporting their positions that can be traced to their experience during the Wolf Unit. These
students wrote a greater number of rhetorical questions, often in the form that Anderson et al. (2001) called a placing stratagem (e.g.
What would you do if [SCENARIO]?). More CG students articulated reasons on both side of the policy issue. The analysis of types of
reasons indicated that, although students from the two conditions averaged a similar number of distinct reasons per essay, more
students in CG produced natural law arguments focusing on the inherent rights of the wolves, whereas more DI students wrote about
economic and public safety issues involving wolves.
From the standpoint that “cognition is embedded in language” (Wegerif & Mercer, 1997, p. 50), differences in the classroom
discourse during direct instruction and collaborative group work would be expected to lead to different understandings of the Wolf
Unit. In DI, the teacher was the chief audience for students when they spoke and the primary authority on whether contributions were
acceptable. The students had limited individual presence within DI discourse in that they had little or no say about the topic of
discussion, hardly any control over how they contributed to the discourse, and negligible interpretive authority. During direct
instruction students were typically in the position of completing thoughts initiated by the teacher. As exhibited in the counts of
performative verb phrases, students in DI were not indexing themselves when contributing to classroom dialogue. In collaborative
group work, students could choose the topic to address as well as when and how they wanted to contribute. Small groups of students,
working independently from the teacher most of the time, were charged with becoming experts in a knowledge domain. Thus,
knowledge and authority were dispersed among the students rather than centralized.
We theorize that the essential differences between the two conditions are the students’ understanding of self as agent and others as
audience. A child who participated in CG will have embodied concepts of taking a position, giving reasons, and entertaining coun-
terarguments, as these were actions undertaken many times. A child involved in CG will have a view of an audience as people with
different viewpoints with the potential to be persuaded. Both a stronger sense of agency and a fuller concept of audience lead to
consequential elaborations in a student’s argument schema. Because of the characteristics of the discourse they experienced, students
from the two conditions had differently elaborated argument schemas that showed up in predictable ways in the wolf decision letter.
Although there were marked differences in the decision letters of students who experienced the two instructional approaches,
their letters were equivalent in two important respects. DI and CG students were equally likely to write decision letters that contained
the core elements of an argument and they included equal amounts of support from the unit. This is in contrast to other studies (Dong
et al., 2009; Kim, Anderson, Miller, Jeong, & Swim, 2011; Reznitskaya et al., 2001) that have invoked argument schema theory to
explain why students in collaborative reasoning groups provided more reasons, counterarguments, and rebuttals in their written
argumentation than comparable children in control groups. Both conditions had extensive experience with a complex socio-scientific
controversy including exposure to arguments, reasons, counterarguments, and rebuttals on both sides. Evidently, this experience was
sufficient to allow students in both conditions to replicate structured arguments in their decision letters, despite differences in
classroom discourse.
Another way to conceptualize these findings is to use the theoretical framework of Bereiter and Scardamalia (1987). When writing
decision letters, DI students, like most young writers, presumably engaged primarily in knowledge telling, supporting their argument
with any reason they could think of, elaborated or unelaborated. Without a well-developed concept of audience as people to be
persuaded, the DI students did not have a sense that they needed to provide more than unelaborated and unconnected reasons since
elaboration and relational thinking are necessarily tied to who is being persuaded. Having participated in classroom dialogue where
their contributions were often minimal and almost always delivered to the teacher, someone who could elaborate or make con-
nections for them, DI students provided less connected, unelaborated reasons; they did not know what were relevant elaboration or
connections or even that elaboration and connection were desirable.
During the Wolf Unit, CG students had two Collaborative Reasoning discussions [plus three beforehand] where they took sides on
a controversial issue and supported and defended their positions with reasons, were sometimes challenged with counterarguments,
and tried to respond with rebuttals. Beyond these planned opportunities for argument, students talked, and sometimes argued, with

245
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

one another every day while reading refutational texts, completing activities as a group, and sharing the responsibility for achieving
the day’s task.
Students working in collaborative groups experienced the multiple viewpoints expressed in the curriculum, but they also ex-
perienced the multiple viewpoints of their classmates. They heard arguments from others and, importantly, they presented their own
arguments, formulated challenges to their classmates, and responded to the challenges of others. This sort of embodied experience
with argument provided CG students with a better understanding of how to position themselves in an argument and how that position
relates to an audience. This explanation is born out in the content analysis of the wolf decision letters. Students who had participated
in CG were likely to use natural law arguments which emphasized the plight of a helpless animal. While these arguments are based
less on information in the Wolf Unit, they have more rhetorical force with a group of classmates.
Like students in DI classrooms, students in CG classrooms were novice writers and their writing also conforms to the knowledge-
telling model. Because writing an argument was not explicitly taught in either condition and the prompt was administered via
protocol by a research assistant, both DI and CG students no doubt probed their memories with similar “topical cues extracted from
the task assignment” (Bereiter, Burtis, & Scardamalia, 1988, p. 262). The difference came in what CG students possessed in the way of
“structural cues of the intended genre” (p. 262). CG students had an argument schema that included embodied concepts of proponent
and respondent abstracted from dozens of experiences in arguing over all manner of things from procedural matters to open-ended
questions posed in the Wolf Unit. Evidence for more fully formed concepts of self in relation to audience was manifest in CG students’
decision letters through their significantly greater use of logical connectives, performative verbs, and rhetorical questions.
The current study goes beyond previous research in several ways. Students in both conditions were exposed to structured ar-
guments surrounding the complex policy issue of wolf management. Students in the DI condition were explicitly shown different sides
of the argument. Students in the CG condition, in addition to also being shown arguments on both sides, actively took up and argued
the questions themselves. Analysis of the wolf decision letter made it possible to isolate how the experience of arguing leads to
differences in students’ sense of self and sense of audience. Going beyond most previous research, systematically sampling the
instructional discourse made it possible to make a link between the experience students had during classroom activities and the kind
of writing they produced in the wolf decision letter. Thus, the study provides significant new evidence about how contrasting
instructional environments can affect the sensitivity of students to themselves in relation to a prospective audience as they create a
piece of reflective writing. However, it should be cautioned that this study enrolled mostly ethnic minority children from poor homes
and, of course, it remains to be seen whether the findings would generalize to children from more privileged homes.
The goal of educating people who can flexibly match their communication to the demands of a task and an audience is highlighted
in the Common Core State Standards Initiative (2010), which describe college-ready high-school graduates as able to,
…adapt their communication in relation to audience, task, purpose, and discipline… They appreciate nuances, such as how the
composition of an audience should affect tone when speaking and how the connotations of words affect meaning. (p. 7)
Collaborative group work enables students to experience first-hand why aligning their talk and writing with an audience is an
essential part of communication.

References

Agresti, A. (2007). An introduction to categorical data analysis. New York: Wiley.

Alvermann, D. E., O’Brien, D. G., & Dillon, D. R. (1990). What teachers do when they say they’re having discussions of content area reading assignments: A qualitative
analysis. Reading Research Quarterly, 25, 296–322.
Anderson, R. C., Chinn, C., Chang, J., Waggoner, J., & Yi, H. (1997). On the logical integrity of children’s arguments. Cognition and Instruction, 15, 135–167.
Anderson, R. C., Nguyen-Jahiel, K., McNurlen, B., Archodidou, A., Kim, S., Reznitskaya, A., ... Gilbert, L. (2001). The snowball phenomenon: Spread of ways of talking
and ways of thinking across groups of children. Cognition and Instruction, 19, 1–46.
Barron, D. N. (1992). The analysis of count data: Overdispersion and autocorrelation. Sociological Methodology, 22, 179–220.
Bereiter, C., & Scardamalia, M. (1982). From conversation to composition: The role of instruction in a developmental process. In R. Glaser (Vol. Ed.), Advances in
instructional psychology: 2. Hillsdale, NJ: Lawrence Erlbaum Associates.
Bereiter, C., & Scardamalia, M. (1987). The psychology of written composition. Hillsdale, NJ: Erlbaum.
Bereiter, C., Burtis, P. J., & Scardamalia, M. (1988). Cognitive operations in constructing main points in written composition. Journal of Memory and Language, 27,
261–278.
Beyer, B. K. (1985). Critical thinking: What is it? Social Education, 49, 270–276.
Boyd, M., & Rubin, D. (2006). How contingent questioning promotes extended student talk: A function of display questions? Journal of Literacy Research, 38(2),
141–169.
Cazden, C. (1988). Classroom discourse: The language of teaching and learning. Portsmouth, NH: Heinemann.
Common Core State Standards Initiative (2010). Common Core State Standards for English Language arts and literacy in history/social studies, science, and technical subjects.
Retrieved from http://www.corestandards.org/assets/CCSSI_ELA%20Standards.pdf.
Crooks, G. (1990). The utterance and other basic units for second language discourse analysis. Applied Linguistics, 11(2), 183–199.
Dillon, J. T. (1985). Using questions to foil discussion. Teaching and Teacher Education, 1, 109–121.
Dong, T., Anderson, R. C., Lin, T., & Wu, X. (2009). Concurrent student-managed discussions in a large class. International Journal of Educational Research, 48, 352–367.
Duffy, G. G., Roehler, L. R., Sivan, E., Rackliffe, G., Book, C., Meloth, M. S., ... Bassiri, D. (1987). Effects of explaining the reasoning associated with using strategies.
Reading Research Quarterly, 22(3), 347–368.
Dweck, C. (2006). Mindset: The new psychology of success. New York: Random House.
Edwards, D., & Mercer, N. (1987). Common knowledge: The development of understanding in the classroom. London: Methuen/Routledge.
Fernandez, M., Wegerif, R., Mercer, N., & Rojas-Drummond, S. (2001). Reconceptualizing “Scaffolding” and the zone of proximal development in the context of
symmetrical collaborative learning. Journal of Classroom Interaction, 36(2), 40–54.
Ferretti, R. P., MacArthur, C. A., & Dowdy, N. S. (2000). The effects of an elaborated goal on the persuasive writing of students with learning disabilities and their
normally achieving peers. Journal of Educational Psychology, 92(4), 694–702.
Ferretti, R. P., Lewis, W. E., & Andrews-Weckerly, S. (2009). Do goals affect the structure of students’ argumentative writing strategies? Journal of Educational
Psychology, 101(3), 577–589.

246
J.A. Morris et al. International Journal of Educational Research 90 (2018) 234–247

Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32(4), 365–387.
Goffman, E. (1974). Frame analysis: An essay on the organization of experience. New York, NY: Harper & Row.
Golder, C., & Coirier, P. (1994). Argumentative text writing: Developmental trends. Discourse Processes, 18, 187–210.
Green, J. L., & Harker, J. O. (1982). Gaining access to learning: Conversational, social, and cognitive demands of group participation. In L. C. Wilkinson (Ed.).
Communicating in classrooms. New York: Academic Press.
Greeno, J. G. (2016). Classroom talk sequences and learning. In L. Resnick, C. Asterhan, & S. Clarke (Eds.). Socializing intelligence through academic talk and dialogue.
Washington DC: AERA.
Harris, K. R., Graham, S., Brindle, M., & Sandmel, K. (2009). Metacognition and children’s writing. In D. Hacker, J. Dunlosky, & A. Graesser (Eds.). Handbook of
metacognition in education (pp. 131–153). Mahwah, NJ: Erlbaum.
Johnston, P. (2004). Choice words: How our language affects children’s learning. Portland, ME: Stenhouse.
Kim, I.-H., Anderson, R. C., Miller, B., Jeong, J., & Swim, T. (2011). Influence of cultural norms and collaborative discussions on children's reflective essays. Discourse
Processes, 48, 501–528.
Knudson, R. E. (1992). Analysis of argumentative writing and two grade levels. The Journal of Educational Research, 85(3), 169–179.
Kock, C. (2003). Multidimensionality and non-deductiveness in deliberative argumentation. In F. H. van Eemeren, J. A. Blair, C. A. Willard, & A. F. S. Henkemans
(Eds.). Anyone who has a view: Theoretical contributions to the study of argumentation (pp. 319–334). New York: Guilford.
Kuhn, D., & Udell, W. (2003). The development of argument skills. Child Development, 74, 1245–1260.
Kuhn, D., Shaw, V., & Felton, M. (1997). Effects of dyadic interaction on argumentative reasoning. Cognition and Instruction, 15, 287–315.
Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge: Cambridge University Press.
Loban, W. (1966). Language ability: Grades seven, eight, and nine. Washington, DC: Government Printing Office.
MacGinities, W., MacGinities, R., Maria, K., & Dreyer, L. (2000). Gates-MacGinities reading tests. Rolling Meadows: The Riverside Publishing Company2–28.
Mehan, H. (1979). Learning lessons. Cambridge, MA: Harvard University Press.
Mercer, N., Wegerif, R., & Dawes, L. (1999). Children’s talk and the development of reasoning in the classroom. British Educational Research Journal, 25, 95–111.
Mercer, N., Dawes, L., Wegerif, R., & Sams, C. (2004). Reasoning as a scientist: Ways of helping children to use language to learn science. British Educational Research
Journal, 30(3), 359–377.
Midgette, E., Haria, P., & MacArthur, C. (2008). The effects of content and audience awareness goals for revision on the persuasive essays of fifth- and eight-grade
students. Reading and Writing, 21, 131–151.
Munneke, L., van Amelsvoort, M., & Andriessen, J. (2003). The role of diagrams in collaborative argumentation-based learning. International Journal of Educational
Research, 39, 113–131.
Murphy, P. K., Wilkinson, I. A. G., Soter, A. O., Hennessey, M. N., & Alexander, J. F. (2009). Examining the effects of classroom discussion on students’ high-level
comprehension of text: A meta-analysis. Journal of Educational Psychology, 101, 740–764.
Nussbaum, E. M., & Kardash, C. M. (2005). The effects of goal instructions and text on the generation of counterarguments during writing. Journal of Educational
Psychology, 97(2), 157–169.
Nystrand, M., Gamoran, A., Kachur, R., & Prendergast, C. (1997). Opening dialogue: Understanding the dynamics of language and learning in the English classroom. Teachers
College Press: New York.
Nystrand, M. (1983). The role of context in written communication. Nottingham Linguistic Circular, 12, 55–65.
O’Connor, M. C., & Michaels, S. (1993). Aligning academic task and participation status through revoicing: Analysis of a classroom discourse strategy. Anthropology and
Education Quarterly, 24(4), 318–335.
Palinscar, A. S. (1998). Social constructivist perspectives on teaching and learning. Annual Review of Psychology, 49, 345–375.
Pressley, M. (2006). Reading instruction that works: The case for balanced teaching. New York, NY: Guilford.
Reznitskaya, A., & Anderson, R. C. (2002). The argument schema and learning to reason. In C. C. Block, & M. Pressley (Eds.). Comprehension instruction (pp. 319–334).
New York: Guilford.
Reznitskaya, A., Anderson, R. C., McNurlen, B., Nguyen-Jahiel, K., Archodidou, A., & Kim, S. (2001). Influence of oral discussion on written argument. Discourse
Processes, 32, 155–175.
Rogoff, B. (1990). Apprenticeship in thinking: Cognitive development in social context. New York: Oxford University Press.
Scardamalia, M., Bereiter, C., & Steinbach, R. (1984). Teachability of reflective processes in written composition. Cognitive Science, 8, 173–190.
Snijders, T. A. B., & Bosker, R. J. (2009). Multilevel analysis: An introduction to basic and advanced multilevel modeling. London: Sage Publishers.
Stegmann, K., Weinberger, A., & Fischer, F. (2007). Facilitating argumentative knowledge construction with computer-supported collaboration scripts. International
Journal of Computer-Supported Collaborative Learning, 2(4), 421–447.
Tharp, R., & Gallimore, R. (1988). Rousing minds to life. New York: Cambridge University Press.
Waggoner, M. A., Chinn, C. A., Yi, H., & Anderson, R. C. (1995). Collaborative reasoning about stories. Language Arts, 72, 582–589.
Walton, D. N. (1998). The new dialectic. Toronto: University of Toronto Press.
Wegerif, R., & Mercer, N. (1997). A dialogical framework for investigating talk. In R. Wegerif, & P. Scrimshaw (Eds.). Computers and talk in the primary classroom (pp.
49–65). Clevedon: Multilingual Matters.
Wells, G., & Arauz, R. M. (2006). Dialogue in the classroom. Journal of the Learning Sciences, 15(3), 379–428.
Wells, G. (1993). Reevaluating the IRF sequence: A proposal for the articulation of theories of activity and discourse for the analysis of teaching and learning in the
classroom. Linguistics and Education, 5(1), 1–37.
Wells, G. (1999). Dialogic inquiry: Towards a sociocultural practice and theory of education. New York: Cambridge University Press.
Wong, B. Y. L., Butler, D., Ficzere, S., & Kuperis, S. (1996). Teaching low achievers and students with learning disabilities to plan, write, and revise opinion essays.
Journal of Learning Disabilities, 29(2), 197–212.

247