Research suggests that varied instructional approaches contribute to better learning outcomes: Some students thrive on visual aids, others on lectures—two examples among countless methods educators can use in lessons.
Still another matter is the learning processes that contribute to learning outcomes, an equally nuanced and individualized topic.
Can researchers quantify learning processes through data?
That was one area of focus in an academic paper led by Assistant Professor Elizabeth Cloude, published by the British Journal of Educational Technology.
Cloude and colleagues examined how cognitive, affective, metacognitive and motivational processes (or how learners thought, felt, reflected and stayed engaged throughout their learning) factored into learning. The scholars used a mobile game, “Antidote COVID-19,” and measured learning processes in the moment during gameplay via physiological sensing markers and concurrent verbalizations.
“This study underscored that learning is non-linear. It’s complex and contextually based,” said Cloude, the paper’s lead author.
Study outcomes
The scholars found that the structure of how individuals transition between cognitive, affective and metacognitive (CAMeta) states matters for learning outcomes. The scholars found that what CAMeta states participants exhibited didn’t matter so much as the pattern with which they regulated those states.
“Learning depends on a balance of stability and flexibility,” Cloude explained.
When participants were given several challenging tasks in the game environment, it was common for them to express confusion or frustration about how to proceed. Participants thrived when they struck a balance between stability and flexibility in their transitions (what researchers called “optimal adaptivity”). They avoided becoming trapped in rigid or repetitive states, and they also did not have chaotic or unpredictable shifts in regulation. Instead, they were more stable: keeping a level trajectory to their goals and overall strategy for game progress, while still adjusting their CAMeta states when faced with feedback, setbacks or changing task demands. This optimal adaptivity helped them sustain continual progress and, ultimately, successful gameplay.
But it wasn't just about getting through the challenge; it was about the road taken.
When participants were more chaotic in their approaches, repeatedly trying different tactics and ways of reasoning, their patterns of thinking and responding became increasingly scattered. Therefore, their learning trajectory was not as strong. The participants were flexible but disorganized, leading to poorer outcomes. Other participants never changed their strategy at all, showing stability but also rigidity, and also had poor learning outcomes.
Scholars also collected participants’ physiological data, including analyzing facial expressions of excitement (a proxy of motivational effort). They found that these facial changes as a result of stimulation were not strong predictors of learning outcomes. Instead, the strongest signal came from patterns in how learners thought and felt, and how they adjusted their strategies over time.
Self-regulation and learning
“With learning, we can think in linear terms—the more you do something, or the more strategies you have, the better you’ll become at doing it and the more learning you’ll have. But that’s not always the case,” Cloude said.
In fact, their data suggested that persistence—keeping at the task—only explained 9% of the overall change in learning gain. “This idea that more learning behaviors or strategies alone equates to greater learning gain is too simple,” Cloude added. “Our findings suggest that learning is not about increasing the frequency of behaviors or strategies, but about using them adaptively. The relationship appears curvilinear, where both too little or too much of a behavior (or state) may be suboptimal. What matters is whether learners can flexibly deploy strategies in response to the demands of the task, the context, and their own individual needs.”
This paper fills some of the gap, suggesting that self-regulation (how we manage ourselves and our reactions while we’re learning) is significant to learning outcomes. The authors found that the way learners regulated different thoughts, emotions and learning strategies over time explained 31% of the differences in learning gains.
Cloude says the data show that there are more insights that scholars need to uncover. Though the data found a part of what makes learning work, it was only a piece of the puzzle. More data from the same research could help. It is the first of several papers from Cloude’s work as a recipient of the Marie Skłodowska-Curie Postdoctoral Fellowship (supported by the European Union). She partnered with Stefan E. Huber (University of Graz, Austria); Jingwei Wei, Bianca Esmanhoto, Muhterem Dindar and Kristian Kiili (Tampere University, Finland); as well as Manuel Ninaus (University of Graz, Austria and University of Tübungen, Germany) on this study, through connections she made during the fellowship. Other papers are in progress, either under review or not yet submitted for publication.
What's next
Cloude, who joined MSU in 2025, has established the CLOUDS (Computational Learning Organization Using Digital Software) Laboratory and is back in the gaming environment to further explore what makes learners and learning thrive in STEM classrooms. Her next project focuses on Minecraft education and artificial intelligence. She aims to build a collaborative, community-based research program to understand how students learn while also co-designing and co-developing curricular changes along with study participants.
It’s all to help define and curate an optimal learning environment, one that includes stability, flexibility, curiosity and, coming soon, Minecraft.