- Email: [email protected]
What is the state of the art in self-, co- and socially shared regulation in CSCL?
Computers in Human Behavior xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh
What is the state of the art in self-, co- and socially shared regulation in CSCL? q Philip H. Winne ⇑ Simon Fraser University, Canada
a r t i c l e
i n f o
Article history: Available online xxxx
a b s t r a c t Articles in this special issue on regulation of learning in computer-supported collaborative learning apply tools across the spectrum of qualitative and quantitative methods to investigate self-, co- and socially shared regulation of learning. As well, a careful consideration of each of these constructs is provided. I briefly review these contributions to identify unique and forward-looking approaches to research in this vibrant area of research. A particular opportunity is recommended for future research regarding the use of process mining, sequence mining, social network analysis and an as-yet to be invented amalgam of these methods in constructing intelligent software agents that could guide participants in CSCL to assemble an optimum mix of self-, co- and socially shared regulation of learning. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction This special issue includes five diverse contributions, each addressing from a different perspective the topic of how learners regulate learning in a group. The collection illustrates a variety of methods in contexts where software supports and constrains what information learners share and how they share it. Amidst an extensive background of research on the more general topic of computer supported collaborative learning (CSCL), these five shine light on matters that I recommend be researched more thoroughly. To set a stage for those recommendations about researching regulated learning and CSCL, I first encapsulate each contribution and select some features they exhibit.
2. Comments on reports in this special issue Miller and Hadwin (2015) tackle with helpful clarity the central matter of defining forms of regulated activity. They begin with an expansive assertion: Activity in groups is regulated when there is ‘‘. . . intentional, goal directed metacognitive activity in which learners and groups take strategic control of their actions (behavior), thinking (cognitive), and beliefs (motivation, and emotions) q Author note: This work was supported by grants to Philip H. Winne from the Canada Research Chairs Program and the Social Sciences and Humanities Research Council of Canada SRG 410-2011-0727. ⇑ Address: Faculty of Education, Simon Fraser University, Burnaby, British Columbia V3H 4R2, Canada. Tel.: +1 778 782 4858. E-mail address: [email protected]
in the context of dynamic social interactions’’ (p. XX). I foreground two features of their claim. First, consider Miller and Hadwin’s definition of socially shared regulated learning (SSRL) as occurring when a ‘‘group collectively regulates their thinking, behaviour, motivation, emotions in the joint task’’ (p. XX, Table 1). For Miller and Hadwin, a task involves a goal, a plan for achieving that goal, skills for working cooperatively and collaboratively, and standards for metacognitively monitoring any or all of these. This begs a question: How does a group transition from a collection of individuals to acting as a collective? Consider three possibilities. The first possibility is that, before the group is formed, each member of that future group has nearly identical knowledge, motivational stances and emotional connections to a task. In other words, each of the group’s members has previously achieved mastery of content, and there are clear and widely shared sociocultural norms and values. In authentic instructional settings and research contexts, the former is improbable. The latter is often assumed but rarely corroborated by data about that particular group and seems improbable when groups are purposively formed to introduce diversity of members’ views about the task. A second path for a group to reach the status of a collective is when one or fewer than all members of a group exercise co-regulation of learning (CoRL). Miller and Hadwin describe CoRL as when ‘‘Individual(s) temporarily guide, prompt, nudge and support each other‘s self-regulation of thinking, behaviour, and beliefs in the joint task’’ (p. XX, Table 1). In this case, at least some members of a group migrate from a prior state of not belonging to the collective to an initial and perhaps steady state of SSRL. The key parameter that may shape shifts across self-regulated
http://dx.doi.org/10.1016/j.chb.2015.05.007 0747-5632/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
2
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
learning (SRL), CoRL and SSRL is new information that is introduced into the group context by members or via resources that members consult. An third and important alternative to this second path to forging a collective has become possible as a result of recent research on quasi-intelligent software agents, often realized as an avatar. In this case, new information is introduced to the group not because any member of the group is intentional, strategic, goal-directed or metacognitively attentive. Rather, one or more members of the group recognizes the tactical or strategic value of information introduced by a software agent. What is noteworthy about this case is it affords rigorous experimental control when investigating standards group members use in monitoring information introduced into group work. When new information shifts members’ work from individual to shared, this is CoRL triggered by the agent. When the group transforms from cooperative forms of work, where ‘‘partners split the work, solve sub-tasks individually and then assemble the partial results into the final output’’ (Dillenbourg, 1999, p. 8), to collaboration, where ‘‘partners do the work ‘together’’’ (Dillenbourg , 1999, p. 8), SSRL is in play. I note another central feature of Miller and Hadwin’s definition of SSRL. It is that metacognition is integral in all three forms of regulated learning. The scope of this claim needs mapping. Metacognition in regulated learning (see Winne, 2011) arises within an individual. It could be a ‘‘collective act’’ under two strict conditions. One is that group members hold identical standards for metacognitive monitoring. The other is overlap of group members’ options for exercising metacognitive control; i.e., not every member must have exactly the same set of skills from which to choose in applying metacognitive control to the task or to managing the group, but some members must have some of the same skills. As I previously noted, the likelihood that group members share standards and skills is generally low. To the extent this is true, and given the fact that instances of CoRL and SSRL arise, it suggests researchers should use a lens of metacognition to illuminate the evolution of CoRL and SSRL over time as a group works. Elsewhere, colleagues and I described challenges in measuring features of regulated learning (e.g., Winne, 2010; Winne, Zhou, & Egan, 2011). Software systems that provide a medium for collaborators’ work can record comprehensive trace data needed to address some of these challenges. Mapping how metacognition is manifested by members of groups and how groups and their members oscillate across SRL, CoRL and SSRL should be a prime focus in future research. The sorts of systems and tools Miller and Hadwin describe will play key roles in these endeavors, as will approaches applied in the study by Lajoie, Lee, Poitras et al.’s study (2015). Lajoie and her colleagues (2015) examined the sensitive and challenging task of transmitting bad news of kinds that physicians sometimes must convey to patients. They explored how online collaboration with veteran physicians and medical students unfolded as the latter sought to learn this delicate craft using a synchronous conferencing system. Chat windows were the medium for exchanging comments between students and the facilitator using a problem-based learning protocol when discussing about video episodes showing how a physician communicated bad news. Transcripts of turns (successive contributions to the chat) formed the corpus analyzed after turns were coded to reflect metacognitive activities, co-regulation and socio-emotional interactions. The researchers sought to identify patterns that ‘‘represented sensible sequences of metacognitive activities’’ as a way to open a window onto ‘‘metacognitive strategies that contribute to learning’’ (p. XX). Lajoie and her colleagues’ view of co-regulation requires ‘‘purposeful mediation of planning, monitoring, evaluating or changing specific beliefs and strategies for motivation, cognition or behavior’’ through verbally contributions to the task at hand (p. XX).
They note that co-regulatory activities can be productive in two ways: moving the group toward shared goals, or turning the group’s focus away from unproductive work and toward a more gainful approach to reaching objectives. Over two separate sessions, discourse turns were coded as metacognitive if they represented moves for orienting, planning, executing, monitoring, evaluating and elaborating aspects of communicating bad news. Co-regulatory contributions were coded as activating or confirming when they facilitated collaboration, or as slowing or changing if turns inhibited co-regulation. Socio-emotional interactions were coded in several sub-categories nesting under the general division of positive vs. negative socio-emotional interactions. Lajoie’s team applied state-of-the-art data mining algorithms to coded events in the corpus to identify patterns across reciprocating turns that representing co-regulation as ‘‘a complex construct that consists of both cognitive and metacognitive activities . . . in which multiple group members contribute to the task at hand’’ (p. XX). In a first stage, they mined codes for sequences in a way that spanned levels of granularity. This afforded stronger representations of context for transitions across three pairings: co-regulatory to metacognitive events, co-regulatory to socio-emotional events, and metacognitive to socio-emotional events. In stage 2 of the analysis, a heat map was constructed to visualize these relationships and aid interpretation. Within the corpus, discourse changed across sessions and showed patterns within sessions. Metacognition increased over time. The first half of sessions evidenced more orientation and planning while the second half showed more fluid shifts with a greater emphasis on monitoring, evaluating and elaborating. Co-regulation and socio-emotional events were similar in quantity but changed pattern across sessions. Cohesion increased and correlated with fewer change-related co-regulatory moves. Cultivating social presence elevated community. Heat maps of patterns showed that activating a new but related topic or completely changing topics led to co-regulatory events. In these instances, ‘‘group members ease[d] cognitive demands by sharing metacognitive demands’’ (p. XX). Lajoie and colleagues’ analyses demonstrate how advanced quantitative methods can be used to compare discourse to statistically expected likelihoods of occurrence; and how co-regulation manifests as a contextualized social exchange involving a mix of cognitively-, metacognitively- and socio-emotionally-referenced information. These sophisticated interpretations were made possible by the advanced quantitative methods and modern information visualization this team used. In real life, as well as in research settings, not all group work is productive. Why? Rogat and Adams-Wiggins (2015) approached this question by probing fine-grained features of information that group members introduce into the group’s shared environment. They distinguished two forms of CoRL. Each draws differently on a view that group members have limited resources for attending to information. Directive CoRL is characterized when standards for metacognitive monitoring lead to superficial monitoring of whether products meet goals. In directive CoRL, a leader strives to sustain elevated social status and diminish others’ contributions to the group’s task. Social exchange in directive CoRL disrespects some group members. The consequence is an erosion of group cohesion. In contrast, when CoRL is facilitative, the leader emphasizes standards for metacognitive monitoring that focus on understanding. In the social plane, the leader promotes an inclusive atmosphere for information exchange. This promotes group cohesion and elevates group productivity. To investigate these relationships, Rogat and Adams-Wiggins analyzed discourse in two small groups of grade 7 students who worked on inquiry-based science tasks. The reseachers purposively
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
selected groups to contrast facilitative versus directive CoRL. Expected correlations between the forms of CoRL and attention to information were observed. Rogat and Adams-Wiggins’ study demonstrates the value of having ‘‘all the data’’ when investigating in fine-grained detail how a group’s discourse evolves over time. Their work is helpful in pointing out need to investigate not only ‘‘what works’’ but ‘‘what interferes with work.’’ Attending to the latter can set a stage for interventions that can help groups avoid downward spirals that undermine productivity and reduce group cohesion. In this regard, future research might focus on developing algorithms to detect features of discourse that mark counterproductive directive CoRL in text exchanges or, when software is more advanced, on-the-fly transcriptions of participants’ speech. More general algorithms driving software agents that can intervene to steer groups toward facilitative CoRL and model productive SSRL are natural extensions. Rogat and Adams-Wiggins findings, though clearly limited by context and group size, plant a seed for these next stages of research on CSCL. Malmberg, Järvelä, Järvenoja, and Panadero (2015) studied how 30 groups of students in a teacher education program navigated challenges in group work across six weekly meetings as they developed a group essay on assigned topics. These researchers focused on events that represented SSRL. A premise underlying their study was that learners must be vigilant in monitoring how information is introduced and how it relates to task, to self and to social qualities of work. Interestingly, they also speculated that focusing cognitive resources on anticipating potential challenges in the group may impede progress. Expressed in theoretical terms of my own, members must be clear about and share standards for metacognitively monitoring information related to these three spheres of work, and they must avoid cognitive overload that interferes with tracking and exercising productive SSRL. Malmberg et al. examined structure in discourse using a sophisticated quantitative method called process mining. They sought to identify patterns in temporally evolving data and markers of a pattern’s onset and offset. Process mining also supports comparing patterns and markers to a standard. This opens a window onto statistical tests of how well group’s process corresponds to a theoretical model. To avoid an excess of variance in their data, as well as to scaffold collaboration, Malmberg et al. invited each group to collectively complete two standard online forms, ‘‘OurPlanner’’ and ‘‘OurEvaluator.’’ The former prompted group members to plan how they would collaborate and how they would avoid challenges. The latter shaped a group’s reflections on challenges encountered and how collaboration played out in those situations. These interventions shaped data a priori. Findings emerging from the discourse data were shaped a posteriori to fit a grid that related three types of challenges – motivational, external, and social – to transitions across three topics of SSRL – cognitive, motivational environment and time management. Transitions across each pair of challenge-SSRL topic were mined for patterns within groups. Discourse in groups that performed well on the assigned essays was compared to discourse in groups that produced relatively weaker essays. As might be expected, groups in which SSRL focused on cognitive and motivational challenges performed better. While social challenges were not ignored in these groups, neither were social challenges keys to predicting achievement. Lower performing groups focused more on managing external challenges. The researchers described these groups as engaging in more ‘‘superficial’’ SSRL. A key factor distinguishing higher-performing groups from lower-performing ones was flow when managing challenges. Higher performing groups relatively constantly dedicated attention to managing challenge throughout their meetings. Lower performing groups showed a more punctuated pattern.
3
Process mining is a powerful new methodology. I predict it will significantly advance research in CSCL and, in particular, research on regulation of activities. Notably, process mining can augment research about any kind of regularized transitions that can be identified in regulated learning: topic flow, communication patterns, and tool use in CSCL environments. A particularly interesting potential for process mining methodology will be forming real-time feedback to groups that can mirror its processes. Lee, O’Donnell, and Rogat (2015) probed how exchanges among group members about planning, monitoring and evaluation shaped the trajectory of a group’s approaches to analyzing scenarios and connecting theory with applications in a course on educational psychology. Their data were chat logs generated as students physically distant from one another synchronously using an online tool. A feature of this study, somewhat like Malmberg et al.’s, was that students were prompted at the outset of a session to consider (a) three planning processes: planning, goal setting and role assignment; (b) two monitoring processes: content monitoring and task monitoring; and (c) two topics for evaluation: task and content. Lee et al. coded conversational turns for social regulation which they characterized as ‘‘attempts to regulate the group’s conceptual understanding of content and to regulate the task’’ (pp. XX). They discriminated other regulation by one group member from SSRL where two or more group members equally contributed information that was intended to orient group processes. Data from two selected groups were intensely analyzed. These two groups were matched on various demographic characteristics as well as the amount of overall participation by each group member. One group was selected because it displayed SSRL. In the second group, one member’s contributions in the latter two of three chats led to the researchers characterizing this group as other regulated. Using social network analysis, centrality indexes were calculated for each member of each group. These statistics supported the a priori classifications of one group as engaging SSRL and the second group as being other directed. As might be expected, content monitoring was prevalent in the SSRL group despite what the researchers’ viewed as a demanding discourse environment created by the synchronous chat environment. A counter-intuitive finding was that other regulation was not inherently counterproductive. In the other-directed group that Lee et al. examined, a skilled other regulator guided the group to engage in high quality content monitoring. Regardless of the form of regulation, planning played an important role in moving the group toward its goals. Like Lajoie and colleagues’ project, participants in Lee et al.’s study were not in face-to-face contact. Citing Hurme, Palonen, and Järvelä (2006) and Volet, Vauras, and Salonen (2009), Lee et al. considered this was influential because ‘‘group members have limited access to the social and affect reactions from their peers when interacting via a technological interface’’ (p. XX). Textual signals may be misinterpreted in ways that lead participants to misinterpretations of partners’ meaning which, in turn, may shape SSRL and even CoRL.
3. Expanding frontiers in conceptualizations, methodologies and practice The field is progressively sharpening a shared conceptualization of forms of regulation that adapt an individual’s and a group’s work toward goals. While there is not precision of the sort chemists have when they stipulate conditions of ‘‘standard temperature and pressure’’ for an experiment, the work published here illuminates not only that metacognition is a key ingredient in regulated learning but also that attention to standards used in metacognitive monitoring are key (see Winne, 2011).
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
4
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
Trace data (see Winne, 2010) play a critical role in revealing to researchers the standards group members use in metacognitive monitoring. The empirical studies here analyze both the semantics within discourse as well as patterns across discourse events. This affords probing for events representing metacognitive monitoring and tracking shifts in metacognitive control, such as between facilitative and directive regulation, over kinds of information contributed and social factors that give texture to those events. In several ways, the theoretical analyses and operational definitions represented in this special issue emphasize that social information – information about relations among individuals – should not be neglected. Fusing elements of the cognitive domain with the social domain is not entirely novel – it is prominent in much research in social psychology – but the ‘‘shape’’ of socio-cognitive amalgamations is fresh as regards regulating activity an individual undertakes as well as regulating ‘‘by and for’’ the collective. Advantages of technologies, such as providing group members with an online form that shapes a group’s consideration about standards for metacognitively monitoring group work, are nicely illustrated. As previously noted, technologies also afford gathering ‘‘all the data’’ under many circumstances, e.g., what was said via text entry in an asynchronous chat. Those same technologies also systematically force excluding some data, e.g., expressive cues that would be available in a face-to-face situation. Prudence will be needed in designing future research with respect to choosing media for information exchange. There is no golden rule that defines the ‘‘right’’ data at the optimal grain size for examining hypotheses or for mining for information in CSCL. Relatively novel analytic tools are displayed in this set of studies. Using social network methods to verify what appears ‘‘obvious’’ about structural properties of information exchange will be useful in a wider range of studies investigating CoRL and SSRL. Also, sophisticated methods for examining structures in data, such as process mining and pattern mining, offer great promise for investigating adaptations in information exchanged as well as forms of exchange that manifest metacognitively-driven regulation of learning. The real advantage of technological tools may be that they exponentially reduce need to ‘‘hand code’’ discourse. If classification of discourse by algorithm can be realized, the practical need to limit research to a few groups working over short time frames may be overcome. Larger and more diverse samples may open windows on a wider landscape of CoRL and SSRL within CSCL. As I noted in reviewing several of the studies, an exciting horizon for research in CSCL is operationalizing interventions in the form of pattern-recognizing software agents that intervene to steer CoRL and support its transformation into SSRL. Real-time analysis of discourse (and, perhaps other channels for data such as galvanic skin response, facial recognition of emotions and eye track data;
see Azevedo 2015) may be available in a not too distant future. Upon arrival and once the challenges of integrating such complex data are met, these new technologies will afford a significant upgrade in information for characterizing CoRL and SSRL. There are yet further questions needing address that lie beyond the scope of this collection. In concluding remarks about messages published in several chapters in their edited book on peer learning, editors O’Donnell and King (1999, p. 316) observed ‘‘the importance of not mistaking immediate change in performance as evidence of conceptual change’’ (p. 316). This caution rings true for reports in this collection. Discourse is a surface marker or trace of underlying and ultimately unobservable cognitive events. Clarifying links between surface messages and underlying constructs remains a key goal in studies of group interaction. References Azevedo, R. (2015). Defining and measuring engagement and learning in science: Conceptual, theoretical, methodological and analytic issues. Educational Psychologist, 50(1), 84–94. Hurme, T.-R., Palonen, T., & Järvelä, S. (2006). Metacognition in joint discussions: An analysis of the patterns of interaction and the metacognitive content of the networked discussions in mathematics. Metacognition and Learning, 1(2), 181–200. Lajoie, S. P., Lee, L., Poitras, E., Bassiri, M., Kazemitabar, M., Cruz-Panesso, I., et al. (2015). The role of regulation in medical student learning in small groups: Regulating oneself and others’ learning and emotions. Computers in Human Behavior. http://dx.doi.org/10.1016/j.chb.2014.11.073. XX, pp–pp. Lee, A., O’Donnell, A. M., & Rogat, T. K. (2015). Exploration of the cognitive regulatory sub-processes employed by groups characterized by socially shared and other-regulation in a CSCL context. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2014.11.072. Malmberg, J., Järvelä, S., Järvenoja, H., & Panadero, E. (2015). Promoting socially shared regulation of learning in CSCL: Progress of socially shared regulation among high- and low-performing groups. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2015.03.082. Miller, M. F.W., & Hadwin, A. F. (2015). Scripting and awareness tools for regulating collaborative learning: Changing the landscape of support in CSCL. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2015.01.050. O’Donnell, A. M., & King, A. (1999). Concluding remarks. In A. M. O’Donnell & A. King (Eds.), Cognitive perspectives on peer learning (pp. 313–317). Mahwah, NJ: Lawrence Erlbaum. Rogat, T. K., & Adams-Wiggins, K. R. (2015). Interrelation between regulatory and socioemotional processes within collaborative groups characterized by facilitative and directive other-regulation. Computers in Human Behavior, XX pp–pp. http://dx.doi.org/10.1016/j.chb.2015.01.026. Volet, S., Vauras, M., & Salonen, P. (2009). Self- and social regulation in learning contexts: An integrative perspective. Educational Psychologist, 44(4), 215–226. Winne, P. H. (2010). Improving measurements of self-regulated learning. Educational Psychologist, 45, 267–276. Winne, P. H. (2011). A cognitive and metacognitive analysis of self-regulated learning. In B. J. Zimmerman & D. H. Schunk (Eds.), Handbook of self-regulation of learning and performance (pp. 15–32). New York: Routledge. Winne, P. H., Zhou, M., & Egan, R. (2011). Designing assessments of self-regulated learning. In G. Schraw & D. H. Robinson (Eds.), Assessment of higher-order thinking skills (pp. 89–118). Charlotte, NC: Information Age Publishing.
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
Contents lists available at ScienceDirect
Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh
What is the state of the art in self-, co- and socially shared regulation in CSCL? q Philip H. Winne ⇑ Simon Fraser University, Canada
a r t i c l e
i n f o
Article history: Available online xxxx
a b s t r a c t Articles in this special issue on regulation of learning in computer-supported collaborative learning apply tools across the spectrum of qualitative and quantitative methods to investigate self-, co- and socially shared regulation of learning. As well, a careful consideration of each of these constructs is provided. I briefly review these contributions to identify unique and forward-looking approaches to research in this vibrant area of research. A particular opportunity is recommended for future research regarding the use of process mining, sequence mining, social network analysis and an as-yet to be invented amalgam of these methods in constructing intelligent software agents that could guide participants in CSCL to assemble an optimum mix of self-, co- and socially shared regulation of learning. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction This special issue includes five diverse contributions, each addressing from a different perspective the topic of how learners regulate learning in a group. The collection illustrates a variety of methods in contexts where software supports and constrains what information learners share and how they share it. Amidst an extensive background of research on the more general topic of computer supported collaborative learning (CSCL), these five shine light on matters that I recommend be researched more thoroughly. To set a stage for those recommendations about researching regulated learning and CSCL, I first encapsulate each contribution and select some features they exhibit.
2. Comments on reports in this special issue Miller and Hadwin (2015) tackle with helpful clarity the central matter of defining forms of regulated activity. They begin with an expansive assertion: Activity in groups is regulated when there is ‘‘. . . intentional, goal directed metacognitive activity in which learners and groups take strategic control of their actions (behavior), thinking (cognitive), and beliefs (motivation, and emotions) q Author note: This work was supported by grants to Philip H. Winne from the Canada Research Chairs Program and the Social Sciences and Humanities Research Council of Canada SRG 410-2011-0727. ⇑ Address: Faculty of Education, Simon Fraser University, Burnaby, British Columbia V3H 4R2, Canada. Tel.: +1 778 782 4858. E-mail address: [email protected]
in the context of dynamic social interactions’’ (p. XX). I foreground two features of their claim. First, consider Miller and Hadwin’s definition of socially shared regulated learning (SSRL) as occurring when a ‘‘group collectively regulates their thinking, behaviour, motivation, emotions in the joint task’’ (p. XX, Table 1). For Miller and Hadwin, a task involves a goal, a plan for achieving that goal, skills for working cooperatively and collaboratively, and standards for metacognitively monitoring any or all of these. This begs a question: How does a group transition from a collection of individuals to acting as a collective? Consider three possibilities. The first possibility is that, before the group is formed, each member of that future group has nearly identical knowledge, motivational stances and emotional connections to a task. In other words, each of the group’s members has previously achieved mastery of content, and there are clear and widely shared sociocultural norms and values. In authentic instructional settings and research contexts, the former is improbable. The latter is often assumed but rarely corroborated by data about that particular group and seems improbable when groups are purposively formed to introduce diversity of members’ views about the task. A second path for a group to reach the status of a collective is when one or fewer than all members of a group exercise co-regulation of learning (CoRL). Miller and Hadwin describe CoRL as when ‘‘Individual(s) temporarily guide, prompt, nudge and support each other‘s self-regulation of thinking, behaviour, and beliefs in the joint task’’ (p. XX, Table 1). In this case, at least some members of a group migrate from a prior state of not belonging to the collective to an initial and perhaps steady state of SSRL. The key parameter that may shape shifts across self-regulated
http://dx.doi.org/10.1016/j.chb.2015.05.007 0747-5632/Ó 2015 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
2
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
learning (SRL), CoRL and SSRL is new information that is introduced into the group context by members or via resources that members consult. An third and important alternative to this second path to forging a collective has become possible as a result of recent research on quasi-intelligent software agents, often realized as an avatar. In this case, new information is introduced to the group not because any member of the group is intentional, strategic, goal-directed or metacognitively attentive. Rather, one or more members of the group recognizes the tactical or strategic value of information introduced by a software agent. What is noteworthy about this case is it affords rigorous experimental control when investigating standards group members use in monitoring information introduced into group work. When new information shifts members’ work from individual to shared, this is CoRL triggered by the agent. When the group transforms from cooperative forms of work, where ‘‘partners split the work, solve sub-tasks individually and then assemble the partial results into the final output’’ (Dillenbourg, 1999, p. 8), to collaboration, where ‘‘partners do the work ‘together’’’ (Dillenbourg , 1999, p. 8), SSRL is in play. I note another central feature of Miller and Hadwin’s definition of SSRL. It is that metacognition is integral in all three forms of regulated learning. The scope of this claim needs mapping. Metacognition in regulated learning (see Winne, 2011) arises within an individual. It could be a ‘‘collective act’’ under two strict conditions. One is that group members hold identical standards for metacognitive monitoring. The other is overlap of group members’ options for exercising metacognitive control; i.e., not every member must have exactly the same set of skills from which to choose in applying metacognitive control to the task or to managing the group, but some members must have some of the same skills. As I previously noted, the likelihood that group members share standards and skills is generally low. To the extent this is true, and given the fact that instances of CoRL and SSRL arise, it suggests researchers should use a lens of metacognition to illuminate the evolution of CoRL and SSRL over time as a group works. Elsewhere, colleagues and I described challenges in measuring features of regulated learning (e.g., Winne, 2010; Winne, Zhou, & Egan, 2011). Software systems that provide a medium for collaborators’ work can record comprehensive trace data needed to address some of these challenges. Mapping how metacognition is manifested by members of groups and how groups and their members oscillate across SRL, CoRL and SSRL should be a prime focus in future research. The sorts of systems and tools Miller and Hadwin describe will play key roles in these endeavors, as will approaches applied in the study by Lajoie, Lee, Poitras et al.’s study (2015). Lajoie and her colleagues (2015) examined the sensitive and challenging task of transmitting bad news of kinds that physicians sometimes must convey to patients. They explored how online collaboration with veteran physicians and medical students unfolded as the latter sought to learn this delicate craft using a synchronous conferencing system. Chat windows were the medium for exchanging comments between students and the facilitator using a problem-based learning protocol when discussing about video episodes showing how a physician communicated bad news. Transcripts of turns (successive contributions to the chat) formed the corpus analyzed after turns were coded to reflect metacognitive activities, co-regulation and socio-emotional interactions. The researchers sought to identify patterns that ‘‘represented sensible sequences of metacognitive activities’’ as a way to open a window onto ‘‘metacognitive strategies that contribute to learning’’ (p. XX). Lajoie and her colleagues’ view of co-regulation requires ‘‘purposeful mediation of planning, monitoring, evaluating or changing specific beliefs and strategies for motivation, cognition or behavior’’ through verbally contributions to the task at hand (p. XX).
They note that co-regulatory activities can be productive in two ways: moving the group toward shared goals, or turning the group’s focus away from unproductive work and toward a more gainful approach to reaching objectives. Over two separate sessions, discourse turns were coded as metacognitive if they represented moves for orienting, planning, executing, monitoring, evaluating and elaborating aspects of communicating bad news. Co-regulatory contributions were coded as activating or confirming when they facilitated collaboration, or as slowing or changing if turns inhibited co-regulation. Socio-emotional interactions were coded in several sub-categories nesting under the general division of positive vs. negative socio-emotional interactions. Lajoie’s team applied state-of-the-art data mining algorithms to coded events in the corpus to identify patterns across reciprocating turns that representing co-regulation as ‘‘a complex construct that consists of both cognitive and metacognitive activities . . . in which multiple group members contribute to the task at hand’’ (p. XX). In a first stage, they mined codes for sequences in a way that spanned levels of granularity. This afforded stronger representations of context for transitions across three pairings: co-regulatory to metacognitive events, co-regulatory to socio-emotional events, and metacognitive to socio-emotional events. In stage 2 of the analysis, a heat map was constructed to visualize these relationships and aid interpretation. Within the corpus, discourse changed across sessions and showed patterns within sessions. Metacognition increased over time. The first half of sessions evidenced more orientation and planning while the second half showed more fluid shifts with a greater emphasis on monitoring, evaluating and elaborating. Co-regulation and socio-emotional events were similar in quantity but changed pattern across sessions. Cohesion increased and correlated with fewer change-related co-regulatory moves. Cultivating social presence elevated community. Heat maps of patterns showed that activating a new but related topic or completely changing topics led to co-regulatory events. In these instances, ‘‘group members ease[d] cognitive demands by sharing metacognitive demands’’ (p. XX). Lajoie and colleagues’ analyses demonstrate how advanced quantitative methods can be used to compare discourse to statistically expected likelihoods of occurrence; and how co-regulation manifests as a contextualized social exchange involving a mix of cognitively-, metacognitively- and socio-emotionally-referenced information. These sophisticated interpretations were made possible by the advanced quantitative methods and modern information visualization this team used. In real life, as well as in research settings, not all group work is productive. Why? Rogat and Adams-Wiggins (2015) approached this question by probing fine-grained features of information that group members introduce into the group’s shared environment. They distinguished two forms of CoRL. Each draws differently on a view that group members have limited resources for attending to information. Directive CoRL is characterized when standards for metacognitive monitoring lead to superficial monitoring of whether products meet goals. In directive CoRL, a leader strives to sustain elevated social status and diminish others’ contributions to the group’s task. Social exchange in directive CoRL disrespects some group members. The consequence is an erosion of group cohesion. In contrast, when CoRL is facilitative, the leader emphasizes standards for metacognitive monitoring that focus on understanding. In the social plane, the leader promotes an inclusive atmosphere for information exchange. This promotes group cohesion and elevates group productivity. To investigate these relationships, Rogat and Adams-Wiggins analyzed discourse in two small groups of grade 7 students who worked on inquiry-based science tasks. The reseachers purposively
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
selected groups to contrast facilitative versus directive CoRL. Expected correlations between the forms of CoRL and attention to information were observed. Rogat and Adams-Wiggins’ study demonstrates the value of having ‘‘all the data’’ when investigating in fine-grained detail how a group’s discourse evolves over time. Their work is helpful in pointing out need to investigate not only ‘‘what works’’ but ‘‘what interferes with work.’’ Attending to the latter can set a stage for interventions that can help groups avoid downward spirals that undermine productivity and reduce group cohesion. In this regard, future research might focus on developing algorithms to detect features of discourse that mark counterproductive directive CoRL in text exchanges or, when software is more advanced, on-the-fly transcriptions of participants’ speech. More general algorithms driving software agents that can intervene to steer groups toward facilitative CoRL and model productive SSRL are natural extensions. Rogat and Adams-Wiggins findings, though clearly limited by context and group size, plant a seed for these next stages of research on CSCL. Malmberg, Järvelä, Järvenoja, and Panadero (2015) studied how 30 groups of students in a teacher education program navigated challenges in group work across six weekly meetings as they developed a group essay on assigned topics. These researchers focused on events that represented SSRL. A premise underlying their study was that learners must be vigilant in monitoring how information is introduced and how it relates to task, to self and to social qualities of work. Interestingly, they also speculated that focusing cognitive resources on anticipating potential challenges in the group may impede progress. Expressed in theoretical terms of my own, members must be clear about and share standards for metacognitively monitoring information related to these three spheres of work, and they must avoid cognitive overload that interferes with tracking and exercising productive SSRL. Malmberg et al. examined structure in discourse using a sophisticated quantitative method called process mining. They sought to identify patterns in temporally evolving data and markers of a pattern’s onset and offset. Process mining also supports comparing patterns and markers to a standard. This opens a window onto statistical tests of how well group’s process corresponds to a theoretical model. To avoid an excess of variance in their data, as well as to scaffold collaboration, Malmberg et al. invited each group to collectively complete two standard online forms, ‘‘OurPlanner’’ and ‘‘OurEvaluator.’’ The former prompted group members to plan how they would collaborate and how they would avoid challenges. The latter shaped a group’s reflections on challenges encountered and how collaboration played out in those situations. These interventions shaped data a priori. Findings emerging from the discourse data were shaped a posteriori to fit a grid that related three types of challenges – motivational, external, and social – to transitions across three topics of SSRL – cognitive, motivational environment and time management. Transitions across each pair of challenge-SSRL topic were mined for patterns within groups. Discourse in groups that performed well on the assigned essays was compared to discourse in groups that produced relatively weaker essays. As might be expected, groups in which SSRL focused on cognitive and motivational challenges performed better. While social challenges were not ignored in these groups, neither were social challenges keys to predicting achievement. Lower performing groups focused more on managing external challenges. The researchers described these groups as engaging in more ‘‘superficial’’ SSRL. A key factor distinguishing higher-performing groups from lower-performing ones was flow when managing challenges. Higher performing groups relatively constantly dedicated attention to managing challenge throughout their meetings. Lower performing groups showed a more punctuated pattern.
3
Process mining is a powerful new methodology. I predict it will significantly advance research in CSCL and, in particular, research on regulation of activities. Notably, process mining can augment research about any kind of regularized transitions that can be identified in regulated learning: topic flow, communication patterns, and tool use in CSCL environments. A particularly interesting potential for process mining methodology will be forming real-time feedback to groups that can mirror its processes. Lee, O’Donnell, and Rogat (2015) probed how exchanges among group members about planning, monitoring and evaluation shaped the trajectory of a group’s approaches to analyzing scenarios and connecting theory with applications in a course on educational psychology. Their data were chat logs generated as students physically distant from one another synchronously using an online tool. A feature of this study, somewhat like Malmberg et al.’s, was that students were prompted at the outset of a session to consider (a) three planning processes: planning, goal setting and role assignment; (b) two monitoring processes: content monitoring and task monitoring; and (c) two topics for evaluation: task and content. Lee et al. coded conversational turns for social regulation which they characterized as ‘‘attempts to regulate the group’s conceptual understanding of content and to regulate the task’’ (pp. XX). They discriminated other regulation by one group member from SSRL where two or more group members equally contributed information that was intended to orient group processes. Data from two selected groups were intensely analyzed. These two groups were matched on various demographic characteristics as well as the amount of overall participation by each group member. One group was selected because it displayed SSRL. In the second group, one member’s contributions in the latter two of three chats led to the researchers characterizing this group as other regulated. Using social network analysis, centrality indexes were calculated for each member of each group. These statistics supported the a priori classifications of one group as engaging SSRL and the second group as being other directed. As might be expected, content monitoring was prevalent in the SSRL group despite what the researchers’ viewed as a demanding discourse environment created by the synchronous chat environment. A counter-intuitive finding was that other regulation was not inherently counterproductive. In the other-directed group that Lee et al. examined, a skilled other regulator guided the group to engage in high quality content monitoring. Regardless of the form of regulation, planning played an important role in moving the group toward its goals. Like Lajoie and colleagues’ project, participants in Lee et al.’s study were not in face-to-face contact. Citing Hurme, Palonen, and Järvelä (2006) and Volet, Vauras, and Salonen (2009), Lee et al. considered this was influential because ‘‘group members have limited access to the social and affect reactions from their peers when interacting via a technological interface’’ (p. XX). Textual signals may be misinterpreted in ways that lead participants to misinterpretations of partners’ meaning which, in turn, may shape SSRL and even CoRL.
3. Expanding frontiers in conceptualizations, methodologies and practice The field is progressively sharpening a shared conceptualization of forms of regulation that adapt an individual’s and a group’s work toward goals. While there is not precision of the sort chemists have when they stipulate conditions of ‘‘standard temperature and pressure’’ for an experiment, the work published here illuminates not only that metacognition is a key ingredient in regulated learning but also that attention to standards used in metacognitive monitoring are key (see Winne, 2011).
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007
4
P.H. Winne / Computers in Human Behavior xxx (2015) xxx–xxx
Trace data (see Winne, 2010) play a critical role in revealing to researchers the standards group members use in metacognitive monitoring. The empirical studies here analyze both the semantics within discourse as well as patterns across discourse events. This affords probing for events representing metacognitive monitoring and tracking shifts in metacognitive control, such as between facilitative and directive regulation, over kinds of information contributed and social factors that give texture to those events. In several ways, the theoretical analyses and operational definitions represented in this special issue emphasize that social information – information about relations among individuals – should not be neglected. Fusing elements of the cognitive domain with the social domain is not entirely novel – it is prominent in much research in social psychology – but the ‘‘shape’’ of socio-cognitive amalgamations is fresh as regards regulating activity an individual undertakes as well as regulating ‘‘by and for’’ the collective. Advantages of technologies, such as providing group members with an online form that shapes a group’s consideration about standards for metacognitively monitoring group work, are nicely illustrated. As previously noted, technologies also afford gathering ‘‘all the data’’ under many circumstances, e.g., what was said via text entry in an asynchronous chat. Those same technologies also systematically force excluding some data, e.g., expressive cues that would be available in a face-to-face situation. Prudence will be needed in designing future research with respect to choosing media for information exchange. There is no golden rule that defines the ‘‘right’’ data at the optimal grain size for examining hypotheses or for mining for information in CSCL. Relatively novel analytic tools are displayed in this set of studies. Using social network methods to verify what appears ‘‘obvious’’ about structural properties of information exchange will be useful in a wider range of studies investigating CoRL and SSRL. Also, sophisticated methods for examining structures in data, such as process mining and pattern mining, offer great promise for investigating adaptations in information exchanged as well as forms of exchange that manifest metacognitively-driven regulation of learning. The real advantage of technological tools may be that they exponentially reduce need to ‘‘hand code’’ discourse. If classification of discourse by algorithm can be realized, the practical need to limit research to a few groups working over short time frames may be overcome. Larger and more diverse samples may open windows on a wider landscape of CoRL and SSRL within CSCL. As I noted in reviewing several of the studies, an exciting horizon for research in CSCL is operationalizing interventions in the form of pattern-recognizing software agents that intervene to steer CoRL and support its transformation into SSRL. Real-time analysis of discourse (and, perhaps other channels for data such as galvanic skin response, facial recognition of emotions and eye track data;
see Azevedo 2015) may be available in a not too distant future. Upon arrival and once the challenges of integrating such complex data are met, these new technologies will afford a significant upgrade in information for characterizing CoRL and SSRL. There are yet further questions needing address that lie beyond the scope of this collection. In concluding remarks about messages published in several chapters in their edited book on peer learning, editors O’Donnell and King (1999, p. 316) observed ‘‘the importance of not mistaking immediate change in performance as evidence of conceptual change’’ (p. 316). This caution rings true for reports in this collection. Discourse is a surface marker or trace of underlying and ultimately unobservable cognitive events. Clarifying links between surface messages and underlying constructs remains a key goal in studies of group interaction. References Azevedo, R. (2015). Defining and measuring engagement and learning in science: Conceptual, theoretical, methodological and analytic issues. Educational Psychologist, 50(1), 84–94. Hurme, T.-R., Palonen, T., & Järvelä, S. (2006). Metacognition in joint discussions: An analysis of the patterns of interaction and the metacognitive content of the networked discussions in mathematics. Metacognition and Learning, 1(2), 181–200. Lajoie, S. P., Lee, L., Poitras, E., Bassiri, M., Kazemitabar, M., Cruz-Panesso, I., et al. (2015). The role of regulation in medical student learning in small groups: Regulating oneself and others’ learning and emotions. Computers in Human Behavior. http://dx.doi.org/10.1016/j.chb.2014.11.073. XX, pp–pp. Lee, A., O’Donnell, A. M., & Rogat, T. K. (2015). Exploration of the cognitive regulatory sub-processes employed by groups characterized by socially shared and other-regulation in a CSCL context. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2014.11.072. Malmberg, J., Järvelä, S., Järvenoja, H., & Panadero, E. (2015). Promoting socially shared regulation of learning in CSCL: Progress of socially shared regulation among high- and low-performing groups. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2015.03.082. Miller, M. F.W., & Hadwin, A. F. (2015). Scripting and awareness tools for regulating collaborative learning: Changing the landscape of support in CSCL. Computers in Human Behavior, XX, pp–pp. http://dx.doi.org/10.1016/j.chb.2015.01.050. O’Donnell, A. M., & King, A. (1999). Concluding remarks. In A. M. O’Donnell & A. King (Eds.), Cognitive perspectives on peer learning (pp. 313–317). Mahwah, NJ: Lawrence Erlbaum. Rogat, T. K., & Adams-Wiggins, K. R. (2015). Interrelation between regulatory and socioemotional processes within collaborative groups characterized by facilitative and directive other-regulation. Computers in Human Behavior, XX pp–pp. http://dx.doi.org/10.1016/j.chb.2015.01.026. Volet, S., Vauras, M., & Salonen, P. (2009). Self- and social regulation in learning contexts: An integrative perspective. Educational Psychologist, 44(4), 215–226. Winne, P. H. (2010). Improving measurements of self-regulated learning. Educational Psychologist, 45, 267–276. Winne, P. H. (2011). A cognitive and metacognitive analysis of self-regulated learning. In B. J. Zimmerman & D. H. Schunk (Eds.), Handbook of self-regulation of learning and performance (pp. 15–32). New York: Routledge. Winne, P. H., Zhou, M., & Egan, R. (2011). Designing assessments of self-regulated learning. In G. Schraw & D. H. Robinson (Eds.), Assessment of higher-order thinking skills (pp. 89–118). Charlotte, NC: Information Age Publishing.
Please cite this article in press as: Winne, P. H. What is the state of the art in self-, co- and socially shared regulation in CSCL?. Computers in Human Behavior (2015), http://dx.doi.org/10.1016/j.chb.2015.05.007