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

The aim of this review was to compare educational psychology models of SRL. To achieve this goal, the included models have been presented and compared. Next, some final conclusions will be presented.

Meta-Analytical Empirical Evidence for the Models

All of the models have empirical evidence that supports the validity to some of their main aspects. However, because the SRL models share a high number of processes, there is a significant overlap in the empirical evidence. For example, self-efficacy is a crucial variable for some SRL models (e.g., Pintrich, 2000; Zimmerman, 2000). Thus, if we consider van Dinther et al. (2010) review on factors affecting self-efficacy in higher education students, it has implications for all those models that emphasize self-efficacy as a crucial SRL process. For this reason, trying to disentangle each individual empirical contribution tailored to a specific SRL model and applying it to the other five models would be very complex. As a consequence, the analysis will focus on more transversal findings, which stem from the meta-analyses conducted in the SRL field.

Three meta-analyses have been conducted with the main aim to study SRL effects (Dignath and Büttner, 2008; Dignath et al., 2008; Sitzmann and Ely, 2011) and a fourth meta-analysis have explored learning skills interventions with direct implications for SRL research (Hattie et al., 1996). First, regarding Hattie et al. (1996), they did not explore differential effects of SRL theories or models. Despite this limitation, one interesting conclusion was that: “it is recommended that training (a) be in context, (b) use tasks within the same domain as the target content, (c) and promote a high degree of learner activity and metacognitive awareness” (Hattie et al., 1996, p. 131). Secondly, Dignath’s work explored the effects of SRL interventions in primary-school students (Dignath et al., 2008), while Dignath and Büttner (2008) added secondary-education students. As extracted from the latter, regarding primary school results, effects size on academic performance were higher if “the intervention was based on social-cognitive theories (B = 0.33) rather than on metacognitive theories (reference category)” plus “if interventions also included the instruction of metacognitive (B = 0.39) and motivational strategies (B = 0.36)” (p. 246). In the same study, effects sizes for secondary education results about the same variables were found to be higher:

“if the intervention was based on metacognitive theoretical background (reference category) rather than on social-cognitive (B = -1.41) or motivational theories (B = -0.97)” plus “if the intervention focused on metacognitive reflection (B = 0.82) or motivation strategies (B = 0.56) rather than on cognitive strategies (reference category), but higher for interventions promoting cognitive rather than metacognitive strategies (B = -0.64)” (p. 246).

The last meta-analysis, by Sitzmann and Ely (2011), focused on how adults regulate their learning in two settings: higher education and the workplace. These authors did not include the SRL theoretical background as a moderator, as Dignath and colleagues did. However, they extracted three key results that apply to the aim of this review. First, the authors found that the constructs that there were included in more SRL theories were the ones that had stronger effects on learning (p. 433). Second, the overlap of the empirical results indicates that significant relationships exists between the different models. Third, “most of the self-regulatory processes exhibited positive relationships with learning, goal level, persistence, effort, and self-efficacy having the strongest effects. Together these four constructs accounted for 17% of the variance in learning after controlling for cognitive ability and pre-training knowledge” (p. 438).

Three main conclusions can be extracted from these four meta-analyses. First, SRL is a powerful umbrella to anchor crucial variables that affect learning, offering, at the same time, a comprehensive framework that explains their interactions. Second, SRL interventions are successful ways to improve students’ learning, if properly designed. Third, SRL interventions have differential effects based on the students’ educational level.

Self-regulated learning interventions that are grounded in socio-cognitive theory (Bandura, 1986) have a higher impact when used at earlier educational stages (e.g., primary), and when a framework is provided for students and teachers (Dignath et al., 2008). It has been hypothesized that this probably happens because socio-cognitive models (e.g., Zimmerman, 2000) are more comprehensive and easier to understand (Dignath et al., 2008). In addition, these models contain motivational and emotional aspects, which are more salient for academic performance during primary education (Dignath et al., 2008). When it comes to more mature students (i.e., those in secondary-education), they benefit more from interventions including more metacognitive aspects (Dignath and Büttner, 2008). This is probably due to the increased performance of cognitively demanding tasks in which it is necessary to use more specific strategies (Dignath and Büttner, 2008). Therefore, it could be hypothesized that metacognitive models (e.g., Efklides, Winne, and Hadwin) would have a higher impact at this educational level. Finally, the results from higher education and workplace trainees (Sitzmann and Ely, 2011) show that the four biggest predictors that were found—goal level, persistence, effort, and self-efficacy—have a significant motivational value and are all comprehended in the socio-cognitive theory. These results align with those of Richardson et al. (2012), who found that (a) self-efficacy was the highest predictor, (b) goal-setting strategy boosts effort regulation, and (c) multifaceted interventions may be more effective (pp. 375–376); and with the results of Robbins et al. (2004), where the best predictors for GPA were academic self-efficacy and achievement motivation. Therefore, there seems to be a tendency for higher education students to have better results if the interventions are aiming at motivational and emotional aspects—i.e., self-efficacy and goal setting. It could, thus, be hypothesized that models with an emphasis on motivation and emotion (e.g., Boekaerts, Pintrich, and Zimmerman) might have a higher impact. Nevertheless, it is important to emphasize that the conclusions regarding higher education students are not build on meta-analyses that explicitly compared different SRL models (i.e., theoretical backgroud) interventions as are the ones for primary and secondary education.

Finally, there was another interesting finding from Dignath and colleagues meta-analyses. They found that SRL interventions, including group work, were detrimental for primary students but that they had positive impact in secondary education students (Dignath and Büttner, 2008). This finding implies that the implementation of SSRL model interventions should be carefully conducted in primary education to try to maximize positive effects. Additionally, SSRL interventions might be needed in secondary and higher education, where the amount of group work increases.

Educational Implications

Four educational implications will be discussed. First, if we examined the psychological correlates (e.g., self-efficacy, effort regulation, procrastination) that influence academic performance (Richardson et al., 2012), the conclusion is that the vast majority of these correlates are included in the SRL models. Additionally, SRL interventions promote students’ learning (Dignath et al., 2008; Rosário et al., 2012). Therefore, a first implication is that teachers need to receive training on SRL theory and models to understand how they can maximize their students’ learning (Paris and Winograd, 1999; Moos and Ringdal, 2012; Dignath-van Ewijk et al., 2013). There are three ways of intervening. First, pre-service teachers should receive pedagogical training for their future adaptation to the workplace. There is a significant number of studies conducted with pre-service teachers (e.g., Kramarski and Michalsky, 2009; Michalsky and Schechter, 2013), however, there is need for longitudinal studies exploring the final outcomes of such training when they start working. Second, in-service teachers also need to receive training on SRL, as they most probably did not receive any during their pre-service preparation (Moos and Ringdal, 2012). Third, teachers should gain SRL expertize themselves as learners, as this will impact their knowledge and pedagogic skills (Moos and Ringdal, 2012).

A second implication relates to how to teach SRL at different educational levels. Different models work better at different educational levels (Dignath and Büttner, 2008). Furthermore, another review shows that teachers at different educational levels used different approaches to SRL (Moos and Ringdal, 2012), but not in the expected direction. These authors found that: (a) higher education teachers tend to focus on the course content, providing limited opportunities for scaffolding SRL; (b) secondary teachers offer more of those opportunities but do not formulate explicit instructions in terms of SRL; and (c) primary teachers implement more SRL practices. There is, therefore, a misalignment between what SRL research says about its implementation at different educational levels (Dignath and Büttner, 2008), and what teachers actually do in their classroom (Moos and Ringdal, 2012). This brings us back to the previous implication: more teacher training is then needed, however, it needs to be tailored so that the interventions take the differential effects of SRL models into account.

A third implication is related to creating environments that leads students’ actions toward learning. All of the models consider SRL as goal-driven, so students’ goals direct their final self-regulatory actions. However, as Boekaerts (2011) argues, students also activate goals not oriented to learning (well-being pathway) and, as a consequence, students might self-regulate toward avoidance goals (e.g., pretending they are sick to miss an exam) (Alonso-Tapia et al., 2014). There is a line of research that explores how teachers can create a classroom environment that is conducive toward learning goals (Meece et al., 2006; Alonso-Tapia and Fernandez, 2008). Educators need to maximize the learning classroom climate for SRL to promote learning.

Fourth, a SRL skill developmental approach is more beneficial for learning. We already know that SRL skills develop over time with practice, feedback, and observation (Zimmerman and Kitsantas, 2005). We also know that students experience a high cognitive load when performing novel tasks, as claimed by cognitive load theory (Sweller, 1994). If we consider what we know on how to design instructional environments to minimize the impact of cognitive load (Kirschner, 2002), then a SRL skill developmental approach should be chosen. Such an approach would consider the four stages for acquisition of SRL, formulated in Zimmerman’s Multi-Level model (Zimmerman and Kitsantas, 2005): observation, emulation, self-control (including automaticity), and self-regulation. This approach will maximize SRL skill development and has been proposed for self-assessment, which is a crucial process for SRL (Panadero et al., 2016).

Future Research Lines

Four future lines will be discussed. First, a call for a connection between the empirical evidence on SRL and meta-analytic evidence of correlates of learning and academic performance should be issued (e.g., Hattie et al., 1996; Richardson et al., 2012). As already argued, SRL models are comprehensive models. Therefore, the validation of the models becomes complex, as it requires either (a) conducting one study with a very large number of variables, or (b) conducting a number of studies with a narrower approach. However, if future research combines conclusions from previous meta-analyses (e.g., Hattie, 2009) and SRL models validational studies, we could advance our understanding of SRL and test even more specific SRL models’ differential effects. This attempt to combine meta-analytic and primary research studies should lead to the construction of a meta-model of SRL, including all SRL areas and interconnecting the existing models. A preliminary attempt can be found in Sitzmann and Ely (2011), however, it needs to be developed further.

Second, more fine-grained studies should also be conducted to understand how the specifics of SRL work. Although SRL models provide a quite specific picture of their processes, there is still much needed to understand SRL mechanisms more precisely (e.g., how self-reflection works, interactions that leads to attributions). This could be achieved through solid experimental designs controlling for strange variables.

Third, longitudinal research on the development of SRL skills throughout the life span is needed, especially regarding how SRL applies to adults in their workplace (Sitzmann and Ely, 2011). There is a compelling amount of research on SRL development during formal education years (Paris and Newman, 1990; Ley and Young, 1998; Núñez et al., 2013). However, we need to further implement longitudinal studies that cover a significant amount of years, and emphasize the role of SRL in adult life. In terms of the latter, our call would be to first test if the available models are valid, rather than developing a new SRL model (Sitzmann and Ely, 2011). Additionally, more longitudinal research on SRL, which focuses on its development during more specific and shorter periods of time, is needed. For example, studies that focus on one specific crucial academic year (e.g., first year of university). The research on learning diaries (Schmitz et al., 2011) is a very promising stream of research that allows for extraction of information in a longitudinal manner.

Fourth, new insights into SRL processes will come from recent developments in SRL measurement (Panadero et al., 2015b). The introduction of computers in SRL research, not only to measure but also to scaffold SRL, is showing promising results. This will provide more tailored interventions and learning environments over the coming years, which should be integrated into the existent body of knowledge.