Over the past 25 years, the question of how learners regulate their own motivation for studying has attracted increasing attention in educational psychology. This research draws on diverse definitions of motivation and its control, operationalizations, study designs, samples, and correlates. Simultaneously, there is a lack of comprehensive syntheses of this research field to date, making it difficult to grasp what we already know about motivational regulation, from which sources this knowledge stems, and what we do not yet know. As a scoping review and evidence and gap map of studies on motivational self-regulation of learners to date, this study summarizes findings on motivational self-regulation and identifies understudied gaps in the field, thus making existing research accessible and informing future research.

Theory and Research Question

Motivational self-regulation as learners’ active behaviours and thoughts to enhance and maintain their motivation for studying (Wolters, 2003) is an important aspect of self-regulated and life-long learning (e.g. Pintrich, 2004). However, despite growing publication numbers, we are lacking comprehensive overviews over the current state of evidence on motivational regulation summarizing what we know about motivational regulation, whether these findings may or may not generalize across samples, contexts, or methods of assessment, and which aspects remain understudied. The aims of this study are therefore to identify 1. how many studies reported associations or effects between motivational regulation and other variables in its nomological network, as well as the direction of these relations (positive, negative, non-significant), 2. which samples these studies examined (e.g. educational settings, gender distributions, and sample ethnicity), 3. which contexts have been considered (e.g. analogous or digital learning settings, motivational regulation in general/ for specific subjects), 4. which methodological approaches were used (operationalizations of motivational regulation, cross-sectional, longitudinal, or experimental study designs; inter- or intraindividual analytic approaches, and publication status). Such an overview of what we already know about motivational regulation, the sources of this knowledge, and systematically understudied aspects is important to facilitate access to this knowledge to interested researchers and practitioners, to neither under- nor overinterpret existing evidence, and to rapidly and strategically develop future research programs.

Method

Following existing recommendations, we conducted a scoping review (Peters et al., 2020; Polanin et al., 2019; White et al., 2020). Several databases, conference programmes, journals, and overview texts relevant to the field were searched using predetermined key terms and, where possible, search algorithms. Several coders received training for screening and a subset of studies was screened by all coders to ensure and report interrater reliabilities. For more details, the study’s preregistration can be obtained from https://doi.org/10.17605/OSF.IO/QRMYC.

Preliminary Results and Discussion

A total of K = 17 867 studies were identified out of which 6032 were duplicates. An additional 10 078 studies were excluded after screening titles and abstracts based on predetermined inclusion and exclusion criteria. Currently, the remaining 1757 studies are screened for in- and exclusion based on full texts and will afterwards be coded based on a preregistered coding handbook using EPPI-Reviewer-Web (Thomas et al., 2022). These codes will be displayed in an interactive evidence and gap map (Miake-Lye et al., 2016; Saran & White, 2018; Snilstveit et al., 2016), available by August 2023.

Retesting can foster students' learning in digital learning systems (DLS). However, the repeated presentation of the same questions in practice testing induces students to apply superficial recognition strategies and to avoid elaborated learning in DLS. This could potentially harm the learning effect of practice testing in the DLS. The goal of the present study was to prevent shallow processing of questions by presenting variations of questions. These variations differ in details (e.g., negations) and the corresponding correct answer choices. By comparing questions with vs. without such variations, we observed strong effects on the number of correctly answered questions and mean processing time. The results show that superficial processing of repeatedly presented questions exist. Variations of questions can increase desirable difficulties.

The study analyses data from self-regulated real-world learning of students with meaningful materials. The data are obtained from a DLS provided in an introductory first-semester lecture. A DLS can support learning (Kornell, 2009; Endres et al., 2017). Comprehension-enhancing learning questions with corrective feedback promote elaboration. Repeated retrieval strengthens memory. Both features are thought to be effective (Naujoks et al., 2022; Endres et al., 2017). However, the effectiveness depends on how the DLS is used. Observations with previous cohorts showed that a considerable number of students achieved a high DLS learning success index, but a poor exam score. This suggests that these students learned mainly through repetition and recognition and avoided elaborate learning.

For the 2022 cohort, answer options were copied and small variations (e.g. negations) were applied for about half of the questions in the DLS. Thus, an answer option that was correct the first time might be wrong the second time if the varied option was selected. We expected the tendency of students to read superficially in repeated testing to be reflected in more wrong answers to the varied questions. The variation was also intended to increase desirable difficulties (Clark & Bjork, 2014; de Bruin et al., 2023). We expected desirable difficulties to be reflected in longer reading times on all questions because students would not know which questions were varied. In 2022, N = 114 students used the DLS and answered M = 1349.80 (SD = 1190.18) exercises on average.

The mean accuracy for questions with variability was lower than the accuracy for questions without variability, t(113) = -11.35, p < .01, d = -1.06. This result confirms that these variations induce difficulty, suggesting that students overlooked small variations in repeated testing. However, students of the 2022 cohort had longer reading times on questions in general than students of the 2021 cohort consisting of N = 155 students with M = 1401.97 (SD = 1511.58) exercises on average, t(234.18) = 4.95, p < .01, d = .62. This suggests that students realized through the provided feedback that variations in answer options existed, and read slower.

Whereas the implementation of variability to the answer options induced difficulty, the difficulty can have the desirable effect of provoking longer processing times. However, more research is needed to reveal whether this longer processing is due deeper semantic processing of answer options.

Much attention has been given to fostering self-regulated learning (SRL) in computer-based learning environments (CBLE). One approach is by means of metacognitive prompts. For the present study, such prompts have been developed for students in lower secondary school. The aim was to examine which activities of SRL as well as in which temporal patterns these activities are shown and whether the prompts foster the students’ SRL and academic achievement. Using a pre-post experimental design, students (N = 34) received prompts or no prompts. Using the think-aloud method, findings indicate that planning activities are most frequent and a pattern of SRL activities was detected. No significant differences between the groups were identified. Conclusions concerning the design of metacognitive prompts are drawn.

1. Self-regulated learning in computer-based learning environments

Findings show that skills to orient, plan and reflect develop early at a basic level, but might not be carried out spontaneously (Veenman et al., 2006; Markman, 1979). If students are unable to apply strategies of SRL, learning with CBLEs can be detrimental (Winters, Greene & Costich, 2008). An effective approach to foster students’ SRL in CBLEs is seen to be by means of metacognitive prompts (Guo, 2022). A prevalent issue is, however, that students do not react to the prompts as intended (Furberg, 2009). Well-crafted support, specialized on the target group, is therefore needed. For the present study, prompts to foster planning and evaluation activities following Winne and Hadwin’s SRL model (1998) have been designed for an authentic CBLE for lower secondary school students. The aim of the study was to examine which activities of SRL (RQ1) as well as in which temporal patterns these activities (RQ2) are shown and whether the prompts foster the students’ SRL and academic achievement (RQ3).

2. Method

German lower secondary school students were asked to think aloud in a 30 minutes learning session in the CBLE. Stratified randomization was used to allocate them into groups with prompts (n = 18) or without prompts (n = 16). The utterances were subsequently transcribed and coded in activities of SRL. By means of process mining, patterns of SRL activities were analyzed. After a learning session, a recall test was completed.

3. Results

Regarding RQ1, a total of 8578 activities was coded from the sample. Planning activities occurred most frequently (M = 108,5; SD = 38,38). The substrategy searching and collecting information was uttered most often (M = 75,18; SD = 28,9). Regarding RQ2, process models demonstrate patterns of SRL activities, most students starting with orientation or planning and ending with evaluation and reflection. Regarding RQ3, no significant differences between the groups in terms of SRL activities or academic achievement could be found.

4. Discussion and Implications

Firstly, the prompts’ design was based on the assumption that students already possess some basic knowledge on SRL. The findings suggest that more information on SRL needs to be presented. Secondly, it seems that prompts should be designed in in a way that students are forced to interact with the prompts. Thirdly, the findings suggest that students need more routine to get accustomed to the prompts. One learning session might not be enough to yield significant changes in the students’ SRL or academic achievement.

Self-regulated learning (SRL) strategies are a key predictor of academic performance. In the present work, we compare the predictive power of self-reported SRL strategies on learning behavior in a digital learning environment and subsequent exam performance. With our research program involving four cohorts of N = 452 teacher students at a German university using a digital learning system (DLS) for practice testing, we show that SRL strategies predict differences in learning behavior in the DLS, namely the total amount of learning activities, but not in exam performance. On the other hand, indicators of interindividual differences in learning behavior in the DLS, namely the distribution of learning activities and also the number of learning activities, are predictors of exam performance.

SRL is a key predictor of academic performance (Schunk & Zimmerman, 2023). However, students often struggle to initiate and maintain SRL (Klingsieck, 2013; Steel, 2007), also while using DLS (Azevedo et al., 2011; Winters et al., 2008). Thus, we proposed that higher self-reported SRL strategies should predict higher overall learning activities (H1a), more distributed learning activities (H1b), and less procrastination (H1c) in a DLS. Higher SRL strategies should also lead to higher exam performance (H2). Furthermore and based on respective literature (e.g., Goda et al., 2015; Kornell, 2009), higher overall learning activities (H3a), more distributed learning activities (H3b), and less procrastination (H3c) in the DLS should also predict exam success. We also proposed that the effects of SRL strategies on exam performance should be mediated by the respective learning patterns (H4a-H4c). All hypotheses were preregistered (https://doi.org/10.17605/OSF.IO/Q8H2V).

We assessed learning data and exam performance from four cohorts of first-semester teacher students (N₂₀₁₉ = 150, N₂₀₂₀ = 141, N₂₀₂₁ = 76, N₂₀₂₂ = 85) using a DLS providing practice exercises for practice testing. We assessed their SRL strategies using the LIST-K questionnaire (Klingsieck, 2018) at the beginning of the semester. Using the log data from the DLS, we assessed the number of processed exercises (overall learning activities; M = 1221.65, SD = 1326.88), their distribution over the semester, and information when participants reach 50% of their cumulated individual learning activities (procrastination).

Using multilevel regression analyses, we revealed that higher SRL strategies lead to higher overall learning activities, ß = .15, p < .01, supporting H1a. For H1b we found no association, ß = .05, p = .23, and we found a contradictory association of SRL strategies and procrastination, ß = .09, p = .05. Strategies were not predictive for exam success, ß = -.01, p = .85, resulting in no support for H2. However, we found that more distributed learning, ß = .267, p < .01, and increased learning intensity, ß = .10, p = .05 was associated with exam performance, but not procrastination, ß = -.044, p = .52. Hence, H3a and H3b were supported. Without a main effect of SRL strategies on exam success, we were not able to support H4a-H4c.

We discuss the implications for theory and practice with special regard to the finding that SRL self-reports did predict learning patterns but not exam performance, while the DLS explained additional variance in exam performance.