Key research papers on self-regulated learning

How the literature is organised

Self-regulated learning research took shape around 1990. Zimmerman (1990) gave an overview of self-regulated learning and academic achievement. Pintrich and De Groot (1990), working with 173 seventh graders across fifteen science and English classes, found self-efficacy and intrinsic value positively related to cognitive engagement and performance, and self-regulation, self-efficacy and test anxiety the best predictors of performance in regression — with the useful negative result that intrinsic value had no direct influence on performance, though it was strongly related to self-regulation and strategy use regardless of prior achievement.

During the 1990s and 2000s several models were developed in parallel. Butler and Winne (1995) synthesised a model in which feedback is inherent in self-regulation and monitoring is the hub of self-regulated cognitive engagement, and argued that feedback research and SRL research should be coupled. Pintrich (2000) covers the role of goal orientation, Pintrich (1999) the role of motivation, and Pintrich (2004) offers a conceptual framework for assessing motivation and SRL in college students. Pekrun et al. (2002) added academic emotions, building the Academic Emotions Questionnaire around enjoyment, hope, pride, relief, anger, anxiety, shame, hopelessness and boredom, and showing them related to motivation, learning strategies, self-regulation and achievement.

Because there are many models with different vocabularies, newcomers find the theory confusing. Panadero (2017) addresses that directly by comparing six models on history, phases, how each treats cognition, motivation and emotion, and the instruments built from each, then drawing out four directions for research. In higher education practice, the most-cited work after Pintrich and De Groot is Nicol and Macfarlane-Dick (2006), which reinterprets formative assessment through self-regulation to derive seven principles of good feedback, on the argument that students are already generating their own feedback and that teaching should build on it.

Main debates

Measurement is the persistent problem. The models describe a process that unfolds across phases, while most studies measure it once with a self-report questionnaire. Learning analytics is the obvious alternative, drawing on what students actually do on a platform, but the evidence for it is not yet strong: Heikkinen et al. (2023) reviewed 56 studies applying learning analytics interventions to support SRL and found that only 46% showed a positive impact on learning, and only four covered all three phases of the SRL cycle. Their recommendation — design interventions for all phases, not just one — is a fair way to frame a study.

A second question is where SRL support belongs. Xu et al. (2023), scoping 163 studies of SRL in online and blended environments, report that most did not address the preparatory and planning phases at all, and that research on children's and adolescents' strategies is the urgent gap. Their meta-analysis (Xu et al., 2023, in Behaviour and Information Technology) compares effects across educational levels, STEM and non-STEM subjects, and learning contexts, which is the comparison to make rather than assuming SRL transfers.

Where recent work is heading

Two strands stand out. One is support through technology. Chang et al. (2023) propose design principles for AI chatbots built on Zimmerman's framework — teaching prompting, reverse prompting, and analytics that let learners reflect. Ng, Tan and Leung (2024) compared a generative AI chatbot with a rule-based one across three weeks with 74 secondary students and found the generative one improved science knowledge, behavioural engagement and motivation. Lee et al. (2024) went further and built the guidance in: their ChatGPT aid gives hints rather than answers, and in a randomised trial with 61 undergraduates it produced better SRL, higher-order thinking and knowledge construction than ordinary ChatGPT use. Jin et al. (2023) asked 16 students what they wanted from such tools and found they saw AI as useful for metacognitive, cognitive and behavioural regulation — but not for regulating motivation. Lan and Zhou (2025) synthesise 14 studies and warn that whether the agency stays human-centred or becomes AI-centred changes the SRL model itself.

The other strand came from the pandemic. Pelikan et al. (2021), surveying 2,652 Austrian secondary students during school closures, found those who felt competent used goal setting, time management and metacognitive strategies more, procrastinated less and needed less support, although all students faced the same challenges. Hong, Lee and Ye (2021) report the complementary finding that procrastination predicted online self-regulated learning and learning ineffectiveness during lockdown.