Effects of self-assessment on self-regulated learning and self-efficacy: Four meta-analyses

Highlights

• Self-assessment effects on self-regulated learning (SRL) and self-efficacy were explored.

• 19 studies (2305 students) were included in four different meta-analyses.

• Effects sizes from the three meta-analyses on SRL were 0.23, 0.65 and 0.43.

• The effect size from the meta-analysis on self-efficacy was 0.73.

Abstract

This meta-analytic review explores the effects of self-assessment on students' self-regulated learning (SRL) and self-efficacy. A total of 19 studies were included in the four different meta-analyses conducted with a total sample of 2305 students. The effects sizes from the three meta-analyses addressing effects on different measures of SRL were 0.23, 0.65, and 0.43. The effect size from the meta-analysis on self-efficacy was 0.73. In addition, it was found that gender (with girls benefiting more) and certain self-assessment components (such as self-monitoring) were significant moderators of the effects on self-efficacy. These results point to the importance of self-assessment interventions to promote students’ use of learning strategies and its effects on motivational variables such as self-efficacy.

Introduction

Student self-assessment has been one of the main areas of research in contemporary education and educational psychology research since the seminal reviewing work by Nancy Falchikov and David Boud in the late 80's (Boud & Falchikov, 1989; Falchikov & Boud, 1989). Research on self-assessment has also transcended these fields of research, extending into fields such as social psychology (e.g. Dunning et al., 2004, Kruger and Dunning, 1999) and medical education (Eva & Regehr, 2005; 2008). The educational research on self-assessment is currently going through an important phase, as shown by recent publications reviewing the accumulated empirical evidence and proposing a new agenda for self-assessment (e.g. Brown and Harris, 2013, Panadero et al., 2016a).

As will be shown in this meta-analysis, the relationship between the constructs self-assessment, self-regulated learning (SRL), and self-efficacy has been the object of empirical research for at least twenty years. This relationship is both reciprocal and intricate: self-assessment is conceptualized as a learning regulatory strategy (Nicol and McFarlane-Dick, 2006, Panadero and Alonso-Tapia, 2013, Paris and Paris, 2001); SRL is dependent on self-assessment – via self-monitoring and self-evaluation – to support student learning (Butler and Winne, 1995, Zimmerman and Moylan, 2009); and self-efficacy is thought to enhance students' activation and use of regulatory strategies, such as monitoring and evaluation (Schunk and Ertmer, 1999, Pajares, 1996, Pajares, 2008). Furthermore, self-assessment might increase the perceived capability among students, which could affect students’ self-efficacy (Andrade, Wang, Du, & Akawi, 2009). In the coming sections of the theoretical framework we will present self-assessment, SRL, and self-efficacy along with the moderating variables included in this meta-analytic review.

In a recent state-of-the-art review, self-assessment was defined as a “… wide variety of mechanisms and techniques through which students describe (i.e., assess) and possibly assign merit or worth to (i.e., evaluate) the qualities of their own learning processes and products” (Panadero et al., 2016a, Panadero et al., 2016b, p. 2). According to this definition, self-assessment is about students assessing their own work; not about signaling their perceived understanding to the teacher through “traffic lights” or evaluating their satisfaction with the instruction. Also important to notice is the emphasis on a “wide variety of mechanisms”, which acknowledges that there are different ways to implement self-assessment in the classroom. As a matter of fact, Panadero et al., 2016a, Panadero et al., 2016b found 20 different categories of self-assessment implementations in their review of different self-assessment typologies. For example, a very simple form of self-assessment is to award a grade/mark to own work (sometimes called “self-evaluation¹” or “self-grading”). A more complex form of self-assessment may involve a rigorous analysis of strengths and weaknesses, as well as the formulation of formative feedback, in relation to explicit criteria (Andrade, 2010).

Actually, a large number of calls have been issued recently for moving away from the simpler forms of self-assessment, where students are merely asked to score or grade themselves, as opposed to making qualitative judgments about their own performance (Andrade, 2010, Boud and Falchikov, 1989; Eva & Regehr, 2005). In particular, the formative assessment agenda has contributed to changing the focus of self-assessment research. Self-assessment is a fundamental component of formative assessment since, as stated by Sadler (1989), it is ultimately the student herself that has to “close that gap” between a current performance (as revealed by assessment) and the desired standard. A student who only follows the teachers’ prescription without understanding its purpose will not learn to monitor and self-adjust her work. As emphasized by Black and Wiliam (1998), self-assessment is therefore not “an interesting option or luxury” (p. 54–55), it is essential to productive learning, and empirical research supports this idea.

Whether self-assessment has an impact on student learning has been explored in a number of studies. For instance, in a narrative review of research about self-assessment, Topping (2003) concluded that there is evidence that self-assessment can result in improvements in the effectiveness and quality of learning. In a more recent publication, Brown and Harris (2013) come to a similar conclusion by reviewing 23 studies, including a wide variety of operationalizations of self-assessment. The effects range from −0.04 to 1.62 (Cohen's d) with a median effect between 0.40 and 0.45. The authors also note that self-assessment seems to improve student performance across a range of grade levels and subject areas, but that it seems to be the implementation and the complexity of the self-assessment intervention, rather than the type, which generates the positive effects. As an example, rubric guided judgement as a type of self-assessment, has been shown to result in both very high effect sizes, as well as very low (and even negative) effect sizes. While students in the latter case (i.e. low effect sizes) participated in two self-assessment lessons, during which they used a rubric for essay writing to assess the quality of their drafts (Goodrich Andrade & Boulay, 2003), the students in the former (i.e. high effect size) also received rubrics articulating assessment criteria for essay writing. However, these students also participated in generating a list of criteria from a model paper (Andrade, Du, & Wang, 2008).

Besides rubrics, which have been suggested to support student learning if combined with self-assessment or other meta-cognitive activities (Panadero & Jonsson, 2013), feedback has also been shown to influence the relationship between self-assessment and learning. In a meta-analysis by Sitzmann, Ely, Brown, and Bauer (2010), the correlation between self-assessment and learning was stronger for courses that included feedback (r = 0.28) than for courses that did not include feedback (r = 0.14). Furthermore, when students self-assessed once, or on multiple occasions without receiving feedback, the relationship with learning was weaker (0.29 and 0.30 respectively) as compared to situations where students received external feedback on their accuracy (r = 0.51).

The abovementioned educational gains from self-assessment are suggested to be related to the enhancement of ownership of learning and the use of SRL strategies. This means that self-assessment is thought to contribute to student learning by, for instance, enhancing the clarity of the learning goals, involving students in monitoring the learning process, and facilitating reflection about the final product or learning outcome (Brown and Harris, 2013, Nicol and McFarlane-Dick, 2006, Panadero and Alonso-Tapia, 2013).

Although the relationship between self-assessment and SRL strategies has been claimed theoretically since the beginning of the formative assessment agenda (e.g. Black & Wiliam, 1998), there is a need to review what is known from empirical research. For example, Topping (2003) concluded that the evidence for self-assessment affecting students’ SRL was “encouraging” and more research was needed. In a similar vein, Brown and Harris (2013) conclude that the research evidence for the connection between self-assessment and SRL is “not robust” and that it is still unclear which students benefit from training in self-assessment.

Regarding self-efficacy, according to social cognitive theory (Bandura, 1986), there are four factors increasing an individual's self-efficacy, which are own experiences of successful performance, watching others succeed, encouragement, and physiological factors. The basic presumption as to why self-assessment has an effect on students' self-efficacy is that by gaining a deeper understanding of the requirements of the task at hand, students are likely to perform better and therefore to experience successful performance. This, in turn, is thought to trigger feelings of worth and a perception of improved capability, which will finally impact on the level of self-efficacy (e.g. Paris & Paris, 2001).

In the self-assessment literature, however, the findings regarding self-efficacy are mixed. As an example, in a study by Andrade et al. (2009) generating a list of criteria from a model essay and using a rubric to self-assess drafts, was shown to increase students' self-efficacy. However, on average, all students' self-efficacy increased, including students in the control group. The increase was larger in the treatment group, but not significantly. There was also a difference between genders, where girls' average self-efficacy for writing tended to be higher as compared to the boys, especially in the beginning. Since the boys' self-efficacy increased more, as compared to the girls, there was not a statistically significant difference at the end of the intervention. Similarly, in a series of studies on the impact of rubric supported self-assessment on students’ self-efficacy, Panadero (2011) found that self-efficacy was impacted by the use of rubrics, but only in one of the three studies. The mixed findings makes it interesting to explore this relationship further through meta-analytic methodology. In the coming sections, the two dependent variables are presented in more detail.

According to one of the most widely used definitions, SRL is: “self-generated thoughts, feelings, and actions that are planned and cyclically adapted to the attainment of personal goals” (Zimmerman, 2000, p. 14). Self-regulated learning has become one of the most prevalent educational theories to explain students’ achievement as it includes a large number of variables related to learning, such as goal orientation, task specific strategies, metacognitive strategies, attribution theory, etc. (e.g. Panadero, 2017). As an example, Richardson, Abraham, and Bond (2012) performed a meta-analysis based on 11 different SRL components (such as rehearsal, effort regulation, and help seeking), showing that the use of SRL strategies was a significant predictor of academic performance. Similar results have also been reported elsewhere (e.g. Broadbent & Poon, 2015).

There are several models of SRL, but most of them include a preparatory phase, a performance phase, and an appraisal phase, each consisting of different sub-processes (Panadero, 2017). In Zimmerman’s (2000) cyclical model, which is one of the most cited in the SRL literature, these phases are called forethought, performance, and self-reflection.

A number of previous meta-analyses have investigated the influence of SRL on student learning and performance (for a meta-review, see Panadero, 2017). For instance, Hattie, Biggs, and Purdie (1996) examined 51 intervention studies aiming to enhance student learning through SRL. In their meta-analysis, the overall effect size for student performance was 0.57. Dignath, Büttner, and Langfeldt (2008) analyzed the effects of SRL on primary school students’ academic achievement, cognitive and metacognitive strategy application, and motivation. Results from 48 comparisons show that SRL interventions were effective, even at the primary school level (the overall mean effect size for academic performance was d = 0.62). For motivational aspects (including self-efficacy), the mean effect size was even higher (d = 0.76). In a later study (Dignath & Büttner, 2008), encompassing both primary- and secondary school (357 effect sizes in total), the mean effect size for academic performance was 0.61 for primary school and 0.54 for secondary school, suggesting that SRL can be fostered effectively at both primary and secondary school level. However, the effect for motivational aspects was considerably lower for secondary students.

In the abovementioned meta-analyses, SRL strategies are clustered into larger categories, making it impossible to disentangle the effects from individual SRL components. In a study by Sitzmann et al. (2010), however, 16 different SRL components were investigated (k = 430), reporting that goal level, persistence, effort, and self-efficacy were the constructs with the strongest effects on learning for adults.

As mentioned above, a strong theoretical connection has been suggested between SRL and self-assessment (e.g. Andrade, 2010, Panadero and Alonso-Tapia, 2013). For example, two of the SRL subprocesses in Zimmerman's (2000) model (self-monitoring and self-evaluation) have a clear similarity to self-assessment, since they are oriented towards assessing own performance. Furthermore, self-assessment has been proposed to be key for the internalization of standards, so that students can regulate their own learning more effectively (Paris & Paris, 2001, p. 95), which involves the first phase of SRL (forethought).

Consequently, self-assessment does not only affect the self-reflection phase, but also the forethought phase (for instance when providing the students with assessment criteria, so that they are able to set realistic goals for the task) and the performance phase (since monitoring can be done with more accuracy, as there is a clearer understanding of the final product/learning outcome) (Andrade, 2010, Panadero and Alonso-Tapia, 2013). The importance of integrating planning with self-assessment related processes has been shown by Dignath et al. (2008), who used meta-analytic methodology to explore SRL interventions in primary school settings. Findings show that metacognitive interventions aiming at a combination of planning and monitoring (P&M), or planning and evaluating, were the most successful to enhance students’ strategy use (d = 1.50 and 1.46 respectively) and had significant effects on motivational outcomes (d = 0.58 and 1.59), as well as academic performance (P&M d = 0.78) (Dignath et al., 2008 p. 115, Table 9). It has therefore been argued that interventions to promote self-assessment should be initiated before students start performing the task, for instance by providing the students with assessment criteria, so that the students can plan, monitor, and evaluate with the help of these criteria (e.g. Andrade and Valtcheva, 2009, Jonsson, 2014). One of the most direct pieces of empirical evidence for self-assessment having an effect on all phases of the SRL cycle comes from a study by Panadero and Romero (2014). In this study, the use of explicit assessment criteria was shown to have a significant impact on the forethought phase (η² = 0.257), the performance phase (η² = 0.084), and the self-reflection phase (η² = 0.217). As can be seen, the preparatory/forethought phase activation of learning strategies was affected the most.

Self-efficacy is the belief about the personal capabilities to perform a task and reach the established goals (Bandura, 1997). The concept was introduced by Bandura (1977), who also made developments within the social cognitive theory (1986), making a major impact in education and educational psychology (van Dinther, Dochy, & Segers, 2011). Self-efficacy has been found to be the strongest predictor of academic performance in tertiary education in two meta-analyses (Richardson et al., 2012, Robbins et al., 2004), and a more recent study also report effect sizes of similar magnitude (Honicke & Broadbent, 2016).

The influence of self-efficacy on the conceptualization and development of SRL has been crucial (Zimmerman, 2000). Over the years, self-efficacy has become one of the most important variables not only in research on motivation, but also in research on SRL (e.g. Schunk & Usher, 2011), and self-efficacy has therefore been incorporated into SRL models (Panadero, 2017). For example, self-efficacy is an essential sub-process in the models by both Zimmerman (2000) and Pintrich (2000). As will be evident from this meta-analysis, self-assessment literature includes a significant number of studies analyzing self-efficacy in isolation from SRL (e.g. Sitzmann et al., 2010, reporting a moderate correlation between self-assessment and self-efficacy based on 32 effect sizes), and to a lesser extent, in combination.

There are a number of variables that may influence the effects of self-assessment on SRL and self-efficacy explored in the literature. The four moderating variables included in this meta-analysis are presented below.

First, gender is a factor that is likely to influence the effects of self-assessment on SRL and self-efficacy, since gender differences have been reported for all three variables. However, the findings are not conclusive for all variables and the amount of research evidence also differs greatly. Starting with the effects of gender on self-assessment, this has not been extensively studied (e.g. Brown and Harris, 2013, Wright and Houck, 1995 and what is known is mostly in relation to the accuracy of self-scoring. Gender effects on self-assessment is therefore an under-researched area and several reviews have recommended future research to explore this effect (e.g. Boud and Falchikov, 1989, Brown and Harris, 2013, Panadero and Jonsson, 2013). There is also evidence supporting the existence of gender differences in SRL, which, according to Bussey (2011), do not relate to differences in SRL capabilities, but to self-efficacy and expectations (i.e. two SRL sub-processes). As opposed to self-assessment, gender differences in relation to self-efficacy have been explored in a number of reviews and meta-analyses. In a meta-analysis by Whitley (1997), it was reported that males had a self-efficacy for computers, which was 0.41 standard deviations above the average for females. Pajares conducted two narrative reviews (2003, 2005) on self-efficacy for writing and mathematics, finding that males tended to have higher self-efficacy for mathematics, while females tended to have higher self-efficacy for writing during middle school, but that this difference tend to decrease at older ages. However, these results need to be re-interpreted in the light of the meta-analysis by Huang (2013), which subsumes the work by Whitley (1997) and also uses a more rigorous meta-analytical methodology as compared to Pajares (2003, 2005). The findings from Huang (2013) were a general effect of 0.08 favoring males, therefore a “small difference” as interpreted by the author, and differences across subjects. Gender is therefore a crucial variable to be considered in the current meta-analytic review.

Second, another factor potentially influencing the effects of self-assessment on SRL and self-efficacy is age (or educational level) of the students. In relation to self-assessment, such research is more or less non-existing and no previous studies comparing self-assessment skills across different ages have been found, although there have been calls to approach self-assessment training as a skill that need practice to develop (Panadero et al., 2016a, Panadero et al., 2016b). In relation to the development of SRL strategies across age and educational level this is a growing area of research. For instance, studies show that already young students can be taught and develop SRL strategies (Perry and Rahim, 2011, Whitebread et al., 2007). Furthermore, meta-analyses by Dignath and Büttner, 2008, Dignath et al., 2008, and Hattie, Biggs, and Purdie (1996) found differential effects of SRL interventions aiming for primary and secondary school students. First, effect sizes for academic achievement from interventions aiming to enhance SRL were larger for younger students (i.e. primary- and lower secondary school) than for older students (i.e. secondary- and higher education students) (Dignath and Büttner, 2008, Hattie et al., 1996).² Second, Dignath and Büttner (2008) found a larger effect on motivational outcomes for primary students, as compared to secondary students, from SRL interventions. And finally, when exploring strategy use, the effects were reversed (i.e. secondary education students benefited more, as compared to primary students). Age was also included as a variable in the meta-analysis by Huang (2013), who reported that gender differences in self-efficacy tend to increase as age increases.

Third, as reported by Panadero et al., 2016a, Panadero et al., 2016b, there is a large number of different self-assessment practices, which might have differential effects. As an example, in the list of effect sizes for the relationship between self-assessment and learning, as presented by Brown and Harris (2013, p. 382), studies with a similar design can be found both at the top and the bottom of the list (e.g. self-assessing writing with a rubric), but which differs in the comprehensiveness of the self-assessment intervention. As suggested by for instance Panadero and Jonsson (2013), as well as the work by Sitzmann et al. (2010), different self-assessment components may therefore have differential effects. Furthermore, from previous research, rubrics could be expected to have a larger impact on performance/avoidance SRL as compared to learning SRL. It is therefore of great need to compare the effects of self-assessment interventions with different components and intensity in relation to effects on SRL and self-efficacy.

A fourth potential moderating variable is the agent who implements the intervention (i.e. the teacher or the researcher). This moderator comes from SRL research, where it has been found that when the SRL interventions were conducted by the researcher, the effect size was higher as compared to when teachers were in charge of the implementation (Dignath et al., 2008). Here it will be explored for both SRL and self-efficacy.

The choice of measurement instruments for SRL and self-efficacy is also a potential moderating variable, since different types of instruments have been shown to provide different results. For instance, in the study by Dignath et al. (2008), intervention studies using questionnaires reported a higher impact on SRL, as compared to other types of measurements (e.g. multiple choice test). However, which instruments to use is a controversial issue in the field. There is a critique of self-reported data (Boekaerts and Corno, 2005, Veenman, 2011), as well as a defense for other types of self-reported data (Samuelstuen & Bråten, 2007). There is also a tension between off-line and online measurements (Winne & Perry, 2000), as well as suggestions for new ways of measuring SRL (Panadero, Klug, & Järvelä, 2016). In this meta-analysis, SRL measurements have not been used as a moderating variable. Instead, due to its importance, this variable has been decompounded into three different (dependent) SRL variables.

This tension between different measurement instruments is not visible in the self-efficacy literature. As shown in the meta-analysis by Honicke and Broadbent (2016), self-report has been the primary method to measure self-efficacy. Specifically, the Motivated Strategies for Learning Questionnaire (MSLQ) (Pintrich, Smith, Garcia, & McKeachie, 1991) has been the most frequently used instrument. An interesting observation is therefore that even though the MSLQ was created as an instrument to measure SRL, it is to a large extent used to measure self-efficacy. In sum, the implication for this meta-analytic review is that there was no need to make a distinction between different measurement instruments for self-efficacy, since all studies in the sample used the same measurement method (i.e. questionnaires).

According to the research reviewed above, the empirical support for the effects of self-assessment on students’ SRL strategies and self-efficacy is seen as promising, but not conclusive. There are, however, some concerns making it difficult to judge the validity of this conclusion. First, there is a number of studies about the effects of self-assessment on SRL that are not included in the abovementioned narrative reviews by Topping (2003) and Brown and Harris (2013). Second, those reviews do not make any distinction between: (a) different research designs (such as qualitative research, quasi-experimental, and experimental design); (b) different measures of SRL; or (c) SRL and self-efficacy. Thirdly, the review by Brown and Harris (2013) only covered K-12, and provided meta-analysis only for the relationship of self-assessment to academic achievement.

Given the inconclusive findings from the primary research, and the limitations from recent reviews, this study aims to use meta-analytic methodology to explore the evidence accumulated on the effects of self-assessment on SRL and self-efficacy. Specifically, this study will explore the following research questions:

• RQ1: Do self-assessment interventions have an effect on students' SRL?

  • RQ1a. Is there a differential effect based on different SRL instruments and constructs?

• RQ2: Do self-assessment interventions have an effect on students' self-efficacy?

• RQ3: Do the moderating variables gender, age/educational level, self-assessment intervention, and implementation agent influence the effects of self-assessment on students' SRL and/or self-efficacy?