A Review of Self-regulated Learning: Six Models and Four Directions for Research

Next, the models will be compared in the following categories. First, the models’ number of cites. Second, all the models are divided into different SRL phases and subprocesses. They are compared here, to extract conclusions. Thirdly, there are three main areas that SRL explores; (meta)cognition, motivation, and emotion; therefore, their positioning in each of the six models is analyzed. And, fourth, the SRL models present significant differences in three major aspects of conceptualization: top–down/bottom–up, automaticity, and context.

Phases and Subprocesses

All of the model authors agree that SRL is cyclical, composed of different phases and subprocesses. However, the models present different phases and subprocesses, and by identifying them we can extract some conclusions. In general terms, Puustinen and Pulkkinen’s (2001) review concluded that the models they analyzed had three identifiable phases: (a) preparatory, which includes task analysis, planning, activation of goals, and setting goals; (b) performance, in which the actual task is done while monitoring and controlling the progress of performance; and (c) appraisal, in which the student reflects, regulates, and adapts for future performances. What is the conceptualization of SRL phases in the two added models? (see Table 2). First, Efklides (2011) does not clearly state an appraisal phase in her model, although she considers that the Person level is influenced after repeated performances of a task. Second, the SSRL model in its version from 2011, although strongly influenced by Winne and Hadwin’s, presents four phases that are similar to Pintrich’s but using different labels. Therefore, the SSRL model classification in the table is the same one that Puustinen and Pulkkinen (2001) proposed for Pintrich’s.

What can be concluded? Even if all of the models considered here except Efklides’, can be conceptualized around those three phases proposed by Puustinen and Pulkkinen, two conceptualizations of the SRL phases can be distinguished. First, some models emphasize a clearer distinction among the phases and the subprocesses that occur in each of them. Zimmerman’s and Pintrich’s models belong to this group, each having very distinct features for each phase. Those in the second group-the Winne and Hadwin, Boekaerts, Efklides, and SSRL (in its forthcoming version) models-transmit more explicitly that SRL is an “open” process, with recursive phases, and not as delimited as in the first group. For example, Winne and Hadwin’s figure does not make a clear distinction between the phases and the processes that belong to each: SRL is presented as a feedback loop that evolves over time. It is only through the text accompanying the figure that Winne and Hadwin (1998) clarified that they were proposing four phases.

One implication from this distinctive difference could be in how to intervene according to the different models. The first group of models might allow for more specific interventions because the measurement of the effects might be more feasible. For example, if a teacher recognizes that one of her students has a motivation problem while performing a task, applying some of the subprocesses presented by Zimmerman at that particular phase (e.g., self-consequences) might have a positive outcome. On the other hand, the second group of models might suggest more holistic interventions, as they perceive the SRL as a more continuous process composed of more inertially related subprocesses. This hypothesis, though, would need to be explored in the future.

(Meta)cognition, Motivation, and Emotion

Next, the three main areas of SRL activity and how each model conceptualizes them will be explored. The interpretation is guided by the models’ figures as they reveal the most important SRL aspects for each author. A classification based on different levels for the three aforementioned areas is proposed (Table 3). It is important to clarify that the levels were conceptualized, not as being close in nature, but rather, as being positions on a continuum.

(Meta)cognition

Three levels are considered with regard to (meta)cognition. The first level includes models with a strong emphasis on (meta)cognition. The first model at this level is Winne and Hadwin’s, in which the predominant processes are metacognitive: “Metacognitive monitoring is the gateway to self-regulating one’s learning” (Winne and Perry, 2000, p. 540). Efklides’ model includes motivational and affective aspects, but the metacognitive ones are defined in more detail at the Task × Person level and are the ones with more substance. Finally, the SSRL model includes in the forthcoming version the COPES architecture from Winne and Hadwin. However, due to the fact that the SSRL 2011 version did not emphasize (meta)cognition, it was decided to locate it after the two more metacognitive models. At the second level are Pintrich’s and Zimmerman’s models. Pintrich (2000) incorporates the “regulation of cognition,” which has a central role along with aspects of metacognitive theory such as FOKs and FOLs. Zimmerman (2000) presents a number of leading cognitive/metacognitive strategies, but they are not emphasized over the motivational ones, as is the case for the models just discussed. At the third level, Boekaerts includes the use of (meta)cognitive strategies in her figures, but does not explicitly refer to specific strategies.

Motivation

A two-level classification is proposed. The Zimmerman, Boekaerts, and Pintrich models are at the first level. Zimmerman’s own definition of SRL explicitly states the importance of goals and presents SRL as a goal-driven activity. In his model, in the forethought phase, self-motivation beliefs are a crucial component; the performance phase was originally described (Zimmerman, 2000) as performance/volitional control, which indicates how important volition is; and at the self-reflection phase, self-reactions affect the motivation to perform the task in the future. According to Boekaerts, the students “interpret” the learning task and context, and then activate two different goal paths. Those pathways are the ones that lead the regulatory actions that the students do (or do not) activate (e.g., Boekaerts and Niemivirta, 2000). In addition, Boekaerts also included motivational beliefs in her models as a key aspect of SRL (see Figure 7). Finally, Pintrich (2000) also included a motivation/affect area in his model that considers aspects similar to those in Zimmerman’s, but Pintrich’s places a greater emphasis on metacognition. It is also important to mention that Pintrich conducted the first research that explored the role of goal orientation in SRL (Pintrich and de Groot, 1990).

The second level includes the SSRL, Efklides, and Winne and Hadwin models. SSRL included motivation in the 2011 version figure and emphasized its role in collaborative learning situations, but without differentiating motivational components in detail. Nevertheless, the authors have conducted a significant amount of research regarding motivation and its regulation at the group level (e.g., Järvelä et al., 2013). Finally, Winne and Hadwin (1998) and Efklides (2011) included motivation in their models, but it is not their main focus of analysis.

Emotion

Three levels are proposed. In the first one (Boekaerts, 1991; Boekaerts and Niemivirta, 2000) emphasizes the influence of emotions in students’ goals and how this activates two possible pathways and different strategies. For Boekaerts, ego protection plays a crucial role in the well-being pathway, and for that reason it is essential for students to have strategies to regulate their emotions, so that they will instead activate the learning pathway. At the second level, Pintrich (2000) and Zimmerman (2000) shared similar interpretations of emotions. They both put the most emphasis on the reactions (i.e., attributions and affective reactions) that occur when students self-evaluate their work during the last SRL phase. In addition, both mentioned strategies to control and monitor emotions during performance: Pintrich discusses “awareness and monitoring” and “selection and adaptation of strategies to manage” (Pintrich, 2000), and Zimmerman stated that imagery and self-consequences can be used by students to self-induce positive emotions (Zimmerman and Moylan, 2009). Nevertheless, in the preparatory phases, neither of them mentions emotions directly. Yet, Zimmerman argues that self-efficacy, which is included in his forethought phase, is a better predictor of performance at that phase than emotions or emotion regulation (Zimmerman, B. J. personal communication with the author, 28/02/2014). The SSRL model includes emotion in its 2011 version figure (Hadwin et al., 2011), but the subprocesses that underlie the regulation of emotion are not specified. Nonetheless, these authors clearly argue that collaborative learning situations present significant emotional challenges, and they have conducted empirical studies exploring this matter (e.g., Järvenoja and Järvelä, 2009; Koivuniemi et al., 2017). Finally, Efklides (2011) and Winne and Hadwin (e.g., Winne, 2011) mention the role of emotions in SRL [e.g., “affect may directly impact metacognitive experiences as in the case of the mood” (Efklides, 2011, p. 19, and she included it in her model at two levels]. However, they do not place a major emphasis on emotion-regulation strategies.

Three Additional Areas for a Comparison

As mentioned earlier, three additional areas in which the models present salient differences were identified.

Top–Down/Bottom–Up (TD/BU)

The first model that included this categorization of self-regulation was Boekaerts and Niemivirta (2000). Top–down is the mastery/growth pathway in which the learning/task goals are more relevant for the student. On the other hand, bottom–up is the well-being pathway in which students activate goals to protect their self-concept (i.e., self-esteem) from being damaged, also known as ego protection. Efklides (2011) also uses this categorization, but with different implications. For her, top–down regulation occurs when goals are set in accordance with the person’s characteristics (e.g., cognitive ability, self-concept, attitudes, emotions, etc.), and self-regulation is guided based on those personal goals. Bottom–up occurs when the regulation is data-driven, i.e., when the specifics of performing the task (e.g., the monitoring of task progress) direct and regulate the student’s actions. In other words, the cognitive processes are the main focus when the student is trying to perform a task.

The other models do not explore this categorization explicitly, although some implicit interpretations can be extracted. This way, there could be a third vision of TD/BU that is based on the interactive nature of Zimmerman’s model and Winne and Hadwin’s. Zimmerman (personal communication to author 27/02/2014) explained:

Historically, top–down theories have been cognitive and have emphasized personal beliefs and mental processes as primary (e.g., Information Processing theories). By contrast, bottom–up theories have been behavioral and have emphasized actions and environments as primary (e.g., Behavior Modification theories). When Bandura (1977) developed social cognitive theory, he concluded that both positions were half correct: both were important. His theory integrates both viewpoints using a triadic depiction. I contend that his formulation is neither top–down [n]or bottom–up but rather interactionist where cognitive processes bi-directionally cause and are caused by behavior and environment. My cyclical model of SRL elaborates these triadic components and describes their interaction in repeated terms of cycles of feedback. Thus, any variable in this model (e.g., a student’s self-efficacy beliefs) is subject to change during the next feedback cycle…. There are countless examples of people without goals who experience success in sport, music, art, or academia and subsequently develop strong goals in the process. Interactionist theories emphasize developing one’s goals as much as following them.

Winne (personal communication to the author 27/02/2014) stated:

I didn’t introduce this terminology because it is limiting. A vital characteristic of SRL is cycles of information flow rather than one-directional flow of information. Some cycles are internal to the person and others cross the boundary between person and environment.

In sum, Zimmerman and Winne do not consider TD/BU to be applicable to their models, as the recursive cycles of feedback during performance generate self-regulation and changes in the specificity of the goals.

As Pintrich’s (2000) model is goal-driven, it could be assumed that it conceptualizes top–down motivation as coming from personal characteristics, as proposed by Efklides (2011). Nevertheless, Pintrich also included goal orientation, which implicates performance and avoidance goals, which has a connection to Boekaerts’ well-being pathway, especially avoidance goals. Therefore, it is difficult to discern with any precision what the interpretation of TD/BU would be for his model. The SSRL model (Hadwin et al., 2011) has not yet clarified this issue, though a stance similar to that of Winne and Hadwin could be presupposed.

Automaticity

In SRL, automaticity usually refers to underlying processes that have become an automatic response pattern (Bargh and Barndollar, 1996; Moors and De Houwer, 2006; Winne, 2011). It is frequently used to refer to (meta)cognitive processes: some authors maintain that, for SRL to occur, some processes must become automatic so that the student can have less cognitive load and can then activate strategies (e.g., Zimmerman and Kitsantas, 2005; Winne, 2011). However, it can also refer to motivational and emotional processes that occur without student’s awareness (e.g., Boekaerts, 2011). Next, some quotations from the models on this topic will be presented to illustrate the different perspectives of automaticity. Winne (2011) stated:

Most cognition is carried out without learners needing either to deliberate about doing it or to control fine-grained details of how it unfolds…Some researchers describe such cognition as “unconscious” but I prefer the label implicit…Because so much of cognitive activity is implicit, learners are infrequently aware of their cognition. There are two qualifications. First, cognition can change from implicit to explicit when errors and obstacles arise. But, second, unless learners trace cognitive products as tangible representations -“notes to self” or underlines that signal discriminations about key ideas, for example-the track [of] cognitive of events across time can be unreliable, a fleeting memory (p. 18).

In addition, Efklides (2011) indicated:

This conception of the SRL functioning at the Task × Person level presupposes a cognitive architecture in which there are conscious analytic processes and explicit knowledge as well as non-conscious automatic processes and implicit knowledge that have a direct effect on behavior (p. 13).

Boekaerts also assumed that automaticity can play a crucial role in the different pathways that students might activate: “Bargh’s (1990) position is that goal activation can be automatic or deliberate and Bargh and Barndollar (1996) demonstrated that some goals may be activated or triggered directly by environmental cues, outside the awareness of the individual” (Boekaerts and Niemivirta, 2000, p. 422). Pintrich (2000) specified: “At some level, this process of activation of prior knowledge can and does happen automatically and without conscious thought” (p. 457). Finally, Zimmerman and Moylan (2009) asserted:

In terms of their impact on forethought, process goals are designed to incorporate strategic planning-combining two key task analysis processes. With studying and/or practice, students will eventually use the strategy automatically. Automization occurs when a strategy can be executed without close metacognitive monitoring. At the point of automization, students can benefit from outcome feedback because it helps them to adapt their performance based on their own personal capabilities, such as when a basketball free throw shooter adjusts their throwing strategy based on the results of their last shot. However, even experts will encounter subsequent difficulties after a strategy becomes automatic, and this will require them to shift their monitoring back from outcomes to processes (p. 307).

Thus, automaticity is an important aspect in the majority of the models. Here, there are three aspects for reflection. First, there are automatic actions that affect SRL; for example, Pintrich (2000) mentioned access to prior knowledge and Boekaerts (2011) discussed goal activation. Second, we can assume that even self-regulation, when it is understood to be the enactment of a number of learning strategies to reach students’ goals, can happen implicitly, as proposed by Winne (2011). This means that students can be so advanced in their use of SRL strategies that they do not need an explicit, conscious, purposive action to act strategically. Nevertheless, this takes practice. Third, some automatic reactions, particularly some emotions, and even some complex emotion-regulation strategies may not be positive for learning (Bargh and Williams, 2007). For example, Boekaerts (2011) mentions that the well-being pathways can be activated even when students are not aware. Therefore, assisting students to become aware of those negative automatic processes could have the potential to enhance self-regulation that is oriented toward learning.

Context

The SSRL model emphasizes not only the role of context, but also the ability of different external sources (group members, teachers, etc.) to promote individual self-regulation by exerting social influence (CrRL) or of groups of students to regulate jointly while they are collaborating (SSRL) (Järvelä and Hadwin, 2013). Zimmerman (2000) did not include context in his Cyclical Phases model, only a minor reference to the specific strategy “environmental structuring.” However, in his Triadic and Multi-level models, the influence of context and vicarious learning is key to the development of self-regulatory skills (Zimmerman, 2013). Boekaerts and Niemivirta (2000) posits that students’ interpretation of the context activates different goal pathways and that previous experiences affect the different roles that students adopt in their classrooms (e.g., joker, geek). For Winne (1996), Pintrich (2000), and Efklides (2011) models, context is: (1) important to adapt to the task demands, and (2) part of the loops of feedback as students receive information from the context and adapt their strategies accordingly. In sum, all of the models include context as a significant variable to SRL. Nevertheless, with the exception of Hadwin, Järvelä, and Miller’s work, not much research has been conducted by the others in exploring how significantly other contexts or the task context affect SRL.