Frontline Learning Research

Mechanistic explanations in the learning sciences as common ground for a more cumulative knowledge production

Sten Ludvigsen — 2026-09-14

Human learning is studied at multiple levels and timescales, and different research communities within the interdisciplinary field of learning research have accumulated specialized knowledge about various facets of the learning phenomenon. However, productive dialogue between such communities is also important for a continuous maturation of the field, yet this dialogue is difficult to achieve because it is dependent upon sufficient common ground in terms of how the issue of explanation is approached epistemologically and methodologically. The aim of this article is to discuss how mechanisms and mechanistic explanations, rooted in new mechanistic philosophy, can constitute a sufficient common ground for a more cumulative knowledge production in the learning sciences and learning research. The argument is that a mechanistic stance, as a meta perspective, can improve the quality of the learning research because it allows for the development of novel explanatory models that can connect multiple levels and timescales in the study of learning, and thus strengthen connections between specialized research communities and their accumulated knowledge.

Decomposing the dynamics in time series data with spectral analysis

Helena J. M. Pennings — 2026-09-14

Many educational and social phenomena are dynamic and change over time. To study such phenomena, intensive longitudinal data and time-series analyses are essential. Yet such methods remain largely underused in educational sciences, due to their perceived complexity and the dominance of group-level prediction. Complex Dynamic Systems (CDS) perspectives offer promising tools for studying educational phenomena that unfold over time, addressing limitations of traditional nomothetic approaches. CDS emphasises idiographic methods that capture individual, context-dependent processes and within-person change. This paper introduces an accessible approach to CDS research by explaining and illustrating how time-series decomposition with spectral analysis can reveal trends, cycles, and level of synchronisation. By breaking down time-series into interpretable components, researchers can better understand dynamic educational processes and avoid misrepresenting complex phenomena. The current paper illustrates the application of time-series decomposition and spectral analysis with time series data of teacher behaviour, student behaviour, and teacher physiology in four classrooms. The application of time-series analysis is discussed considering the differences in the teachers’ dynamic profiles as well as the potential to study a large variety of other educational topics.

The science-practice gap in education: knowledge conceptualization, knowledge production, and knowledge transfer problems

Hanke Korpershoek — 2026-09-14

There is a disconnection between the knowledge that researchers produce and the knowledge that practitioners use in their daily routines. This so-called science-practice gap is one of the main scientific challenges in this era in applied fields such as educational sciences, because it implies that practitioners fail to adopt evidence-informed practices. This conceptual paper presents a theory synthesis with the aim to unravel the complex challenge of reducing the science-practice gap in education. Based on an integrative conceptual literature review, I argue that new leads to reduce the science-practice gap in education lie in a fundamental understanding of (a) the knowledge conceptualization problem, (b) the knowledge production problem, and (c) the knowledge transfer problem. We face complex societal challenges that can only be solved by collective engagement and collaboration between researchers and practitioners to address these three fundamental problems. In this paper, I propose that the interchange between science and practice should start with the question “What knowledge do we need to improve educational practice?”.

Algorithmic Justice in Education Through De-Biasing: Towards Politically Actionable Evidence That is Rooted in Identity Theory and De-Colonial Thought

Adrian Grimm — 2026-09-14

As Artificial Intelligence (AI) algorithms are increasingly used in education, research shows that the use of these algorithms is not without cost. Instead, AI algorithms are prone to biases which are discussed a lot in various domains. The core strength of this contribution is to anchor the discussion of biases in the specific domain of physics education and to discuss the biases in front of a description of the domain-specific inequalities along physics identity development of students. The database consists of the written answers of 527 students to around 30 items from a five-week-period of physics classes in a digital learning environment. Two concrete biases of AI algorithms in physics education and possible approaches to identify and reduce these biases are investigated quantitatively. In a critical discussion from a feminist and de-colonial perspective, it is highlighted that the chosen approaches seem to have promising potentials to mitigate negative effects on under-served students´ physics identity development. Besides, relevant limitations lead to conclusions that additional counter-measures are needed in order to break out of the vicious cycle of reproduction of historically grown inequalities in physics education. The domain-specific analysis can serve as orientation for other domains as well in order to tackle the challenges of AI algorithmic bias effectively and efficiently.