Reflective practice is a longstanding ideal for good teaching. It involves a critical evaluation of one’s perceptions, continuous learning, and the application of insights from reflection (Schön, 1983). However, ‘reflection’ has long lacked a universal definition (Collin et al., 2013). This ambiguity allows ‘reflection’ to encompass discovering ideas, solving problems, evaluating situations, and critiquing theories (Beauchamp, 2006). Recent literature has sought to sharpen the concept of reflection in teacher education. For example, Arendt et al. (2025) propose an integrated model that treats reflection as an input–process–output cycle: teachers start from pedagogical experiences (input), engage in processes such as describing, evaluating, and generating alternatives (process), and aim at professional development (output). In general, models of reflection emphasize that reflection differs from mere analysis through its reliance on self-reference and justification with professional knowledge.

In this article, we conceptualize one specific aspect of reflection in detail: the active construction of knowledge on teaching— encompassing both research-based knowledge acquired through formal teacher education and practical knowledge developed through personal experience in educational settings. We focus on university-based teacher education and regard reflective practice as a means to support PSTs’ ability to apply their academically acquired knowledge base to the analysis and interpretation of classroom events (Fives et al., 2017). For example, PSTs can learn to describe and interpret pedagogical situations using conceptual frameworks, empirical findings, and disciplinary norms drawn from educational science coursework (van Es and Sherin, 2002). Approaches to foster PSTs’ reflective practice often involve inquiry processes (e.g., Cochran-Smith and Lytle, 1999) or dilemma-like pedagogical situations (e.g., Caspari-Gnann and Sevian, 2022; Kavanagh et al., 2020) to equip future teachers with adaptable and informed decision-making skills.

Models of epistemic cognition (Chinn et al., 2011, 2014) have increasingly been applied to the complexities of teaching situations. For example, Lunn Brownlee et al. (2017) introduced the 3R-EC framework (Reflection, Reflexivity, Resolved Action for Epistemic Cognition), using epistemic cognition as the starting point for supporting a variant of teachers’ reflective practice. The framework emphasizes the role of reflective thinking in fostering deeper understanding and transformation in teaching practices, thereby enhancing the overall educational experience for students. Epistemic cognition refers to “all kinds of explicit or tacit cognitions related to epistemic or epistemological matters” (Chinn et al., 2011, p. 141). In teaching, it pertains to the processes and mediational mechanisms employed when teachers engage in knowledge construction (e.g., Buehl and Fives, 2016). PSTs’ epistemic cognition may influence their understanding of coursework and the resulting knowledge of teaching and learning (Lunn Brownlee et al., 2017). One notable framework is the AIR model (Aims, Ideals, Reliable Processes) by Chinn et al. (2014) and Chinn and Rinehart (2016), has been adapted specifically to teaching (Buehl and Fives, 2016; Lunn Brownlee et al., 2017). The AIR model suggests that in processing information, people set individual aims for the epistemic (i.e., knowledge) products they hope to generate. The framework then differentiates the values people place on these aims, their ideals for evaluating their success at arriving at their aims, and processes that can be considered reliable for achieving their aims (reliable epistemic processes). In contrast to the other dimensions of the AIR model, these processes are more directly observable in individuals’ pursuit of knowledge aims. In studies that build on the AIR model, researchers identified reliable epistemic processes employed by teachers in classroom assessment (Barnes et al., 2020), in providing dialogic feedback (Rinehart et al., 2022), and in learning through professional development (Fowler et al., 2022).

However, theoretical literature on epistemic cognition assumes that effective performance requires not so much the use of one single epistemic process, but the use of an array of different processes depending on the questions at hand (Barzilai and Chinn, 2018; Chinn et al., 2014). Understanding which epistemic processes PSTs use—and when and how they use them—could offer insight into when and how educational sciences knowledge is perceived as useful or usable. For instance, selecting relevant theories, critically evaluating research, or integrating course content with teaching experiences are not just cognitive achievements but indicators of educational knowledge being valued and applied in meaningful ways.

Previous research also shows that processes of knowledge construction—and the ways in which PSTs activate, interpret, and apply educational science knowledge—may vary between individual and group formats (Csanadi et al., 2021; Kuhn, 2015). In a study by Kuhn and Moore (2015), students engaged in a dialogic task used more of the available knowledge to counter positions opposing their own and also counted more on personal experience than students working on individual tasks. In the individual task, the students used more of the available information sources overall. This finding is corroborated by Csanadi et al. (2021), who found PSTs working collaboratively to engage more in the generation of hypotheses and in the evaluation of claims than those working individually. However, these findings are primarily derived from experimental studies and have not yet been supported by evidence from authentic, contextualized settings in teacher education.

Knowledge construction can encompass a range of activities such as identifying problems, generating hypotheses, or utilizing different sources of evidence and methods of analysis. Each of these processes play a distinct role in shaping PSTs’ approaches to reflective practice. However, besides identifying different types of processes, linking the AIR model to teaching knowledge construction requires considering the processes’ quality, that is, an estimation of how well generally reliable processes are implemented in a specific task (Rinehart et al., 2022). Procedures that are deemed reliable as a general principle, like using scientific educational literature to gain insight into classroom situations, can vary in proficiency and lead to different outcomes.

Teacher educators need to be able to assess how PSTs engage in processes of knowledge construction during common reflection exercises. Evaluating their understanding of complex classroom situations requires clear and reliable indicators to distinguish and evaluate the processes they implement. Such indicators also respond to calls for more reflective practice in teacher preparation (e.g., Alexander, 2017; Hartmann et al., 2021). As in many complex domains, reflective practice in teacher education involves connecting various sources of information to the specific context of the situation, using these sources to understand pedagogical practice in its complexity.

In addition to distinguishing different quality dimensions for individual processes, it is important to consider their interconnections. This aligns with research findings in teacher education that emphasize the role of integrating different sources of knowledge—e.g., past experiences, theoretical and research literature, testimony, observation, personal reflection (Harr et al., 2015; Renkl, 2022; Shulman, 1987). Effective integration connects different types of processes, creating a richer understanding. In contrast, processing sources merely sequentially without integrating them may be less effective (Hartmann et al., 2021). Knowledge integration might also comprise reflectively linking theoretical knowledge with personal teaching experiences (Kim and Klassen, 2018). For these reasons, combining knowledge construction processes could serve as another indicator of better process implementation (Chinn et al., 2014).

This study was situated in a two-year Master of Education program for secondary school teachers in a mid-size university in Germany. The program includes a mandatory four-day course to prepare PSTs for their practical term, followed by a 5-month practical placement in secondary schools. The course aimed to familiarize participants with models of reflective practice (e.g., van Es and Sherin, 2002), helping them to question and analyze authentic pedagogical situations by connecting them to theory and research.

Case vignettes of classroom situations were presented in both video and written formats. The participants received guidance on using scientific databases. PSTs were tasked to: (a) identify essential characteristics of the situation, (b) formulate plausible explanations for what happened during the situation, (c) use professional literature to support their explanations, and (d) use their explanations to arrive at practical conclusions and alternatives for action.

PSTs concluded the course by choosing between writing an essay (EW condition) or working collaboratively in small group discussions (GD condition). All participants consented to data analysis and completed a short background questionnaire.

Twenty-nine German PSTs (M_Age = 27.58, SD_Age = 5.65, 69% female, 3.4% unreported gender) were recruited from two parallel courses. Participants in the two conditions differed marginally with respect to group size (15 in the EW condition, 14 in GD condition), and gender (eight female in the EW condition, 12 female in the GD condition, one participant with unreported gender).

During the practical phase, PSTs observed experienced teachers and taught their first lessons. In the fourth month, all PSTs analyzed the same complex classroom vignette (Supplementary material), which was open to multiple interpretations. The PSTs analyses served as the primary data source for our study. Each participant began with an individual literature search to find educational sources relevant to the pedagogical situation.

In the EW condition, participants wrote essays summarizing their reflections and submitted their reference lists, explanations, and conclusions to instructors. In the GD condition, after the literature search, participants brought their reference lists to small group meetings of 4–5 members at the university. Guided by the same instructions as the EW participants, plus additional ones (Appendices B, C), they collectively discussed possible explanations. These 30- to 40-min discussions were recorded and transcribed using a clean verbatim transcription protocol. Afterwards, GD participants individually wrote a short conclusion essay with prompts identical to those in the EW condition.

Participants chose a reflection format (EW or GD) based on preference and motivation. To mitigate potential self-selection effects, we used background measures to check for differences between the conditions in areas we considered most likely to influence our results. Internal consistency, means and standard deviations for all measures described below are shown in Table 1.

Additionally, we compared the reference lists that the participants prepared for the task. Each student provided a list with three to 18 references, totaling 29 lists. These references consisted of common university textbooks, scientific articles, and practical teaching guides. We reviewed these lists for the number and type of sources selected to explain the pedagogical situation.

Analyses revealed no differences in responses to any of the standardized measures. Regarding the reference lists, we found that on average, participants in both conditions used a similar number of external literature sources (M_GD = 7.42; SD_GD = 2.68; M_EW = 6.87; SD_EW = 3.8), and similar types of literature, t-tests revealed no significant differences.

For our analysis, we used the explanatory sections of the essays and the GD transcripts, excluding participants’ conclusions. To analyze transcripts, we isolated each participant’s contributions by removing others’ comments. This eliminated conversational context but provided insight into individual processes. Essays lengths ranged from 4,625 letters to 15,788 letters. Participants’ individual contributions in the GD condition ranged from 2,799 to 10,159 letters.

Each participant was assigned a code linking their utterances in the transcripts to their reference lists and background information. In the EW condition, participants provided personal codes to match essays with other data sources.

On average, we found 29 segments indicating knowledge construction processes in the essays and 32 segments in individual contributions to the transcripts. Over 50% of the total number of letters in essays were assigned to the category referring to literature (Figure 1). In transcripts, we coded almost 40% of the total number of letters as generating explanatory hypotheses. The two inductively added categories referring to details of the situation description and referring to current emotional experiences had considerably shorter segment lengths, only making up a small proportion (less than 6%) of the total number of letters used in both conditions.

Differences regarding process implementation quality are displayed in Figure 2. More than 70% of the letters used in the essays were coded as indicative of evolving analytic implementation. In contrast, the oral statements of the GD participants used more letters coded as reflective. Both conditions were similar in their use of superficial process implementations.

Two examples from EW (“Lisa”) and GD (“Tom”) conditions are displayed in Figures 3, 4. Selected for their sequential patterns with little deviation from their condition’s norm, both exhibit characteristic behaviors for their respective conditions. Despite his slightly atypical lower number of reflective processes, Tom represents the GD cases especially well through his use of medium quality and varied processes, thus highlighting this aspect of our data on the GD condition.

EW participants often used the same kinds of processes throughout large text portions. When changing between process types, they would usually remain within the new process for several sentences. Lisa began her essay by establishing a connection between her focus on feedback in the explanation of the situation vignette and her own past teaching experiences:

[…] Lisa went on to describe her emotional experience, in her case a feeling of interest, in this aspect of the situation:

She then explicated the focus on feedback by defining the term and delving into the literature on various aspects of feedback and motivation. She did not use these sources to contrast any conflicting ideas on the issue, but correctly and transparently presented several theoretical distinctions and empirical findings:

She concluded the explanatory part of her essay by returning to past experiences she connected to the situation described in the vignette:

Regarding process implementation quality, all of these examples were coded as evolving analytic, as they were clearly intended to construct an understanding of the situation and connected different sources but did not involve any form of questioning either the literature, the situation, or her own interpretations of the situation.

By contrast, participants in the GD condition not only used a wider variety of different processes, but also tended to switch more between different processes to argue for a specific understanding. Tom (Figure 4), mixed references to past experiences and explanations of the situation in his first recorded statement in the group discussion:

As these statements rested mostly on preconceived judgments, we coded them as superficial. Later, Tom would interweave ideas for alternative courses of action, references to the situation description, and explanations of the situation:

In the course of the discussion, Tom connected aspects of the situation to literature and to experiences from his own forays into teaching. In some cases, he questioned his own or his fellow discussants’ first interpretations of the situation and thus exhibited reflective processes. For example, when discussing in how the teacher in the pedagogical situation provides feedback to her students, he showed awareness of different possible interpretations of a detail of the situation:

His explanation of this difference was cut short by another student. Overall, Tom remained mostly on the levels of superficial and evolving analytic process implementations: He used the information he had gathered from scientific literature for a straightforward explanation of various aspects of the situation but did not show any signs of weighing or questioning these explanations.

Based on our qualitative results, we saw descriptively that participants in the GD condition use a wider variety of processes, both with regard to their total number as well as the number of times they switched between different kinds of processes. We explored these descriptive findings using t-tests. Regarding the number of different processes used in the two reflection formats, a two-tailed t-test showed a significant difference between the conditions (Table 3). On average, members of the GD condition used a larger number of processes. We found similar results when we looked at the number of switches between different process types.

While these results might point to a potential for a more comprehensive process enactment in the GD condition, we wanted to further explore differences between the two conditions. We therefore investigated whether the conditions differed regarding our ratings of process implementation quality. We tended to observe more processes of the highest implementation quality, called reflective in the GD condition. Using a one-factor MANOVA on the standardized number of characters in each process quality category, we found a statistically significant difference between the reflection formats for three degrees of process implementation [F(3, 25) = 10.765, p < 0.001, partial η² = 0.564, Pillai’s V = 0.564]. Between-subjects effects show significant differences for evolving analytic processes and reflective implementation. This was confirmed by two-sided t-tests (Table 4).² No significant differences were found for superficially implemented processes. Effect sizes for evolving analytic and reflective codes were large (d = −1.59 and d = 1.15).

A one-factor MANOVA with the condition as the independent variable showed a significant difference in the standardized numbers of letters in the three dominant knowledge construction processes: referring to literature, referring to past experience, and generating explanatory hypotheses [F(3, 25) = 11.174, p < 0.001, partial η² = 0.573, Pillai’s V = 0.573]. The between-subjects effects show significant (p < 0.05) influences of the referring to literature and generating explanatory hypotheses variables (Table 5). This was confirmed by subsequent Bonferroni-corrected one-tailed t-tests. Standardized values for referring to literature, and generating explanatory hypotheses were both significant. In contrast, differences in referring to past experience were found to be non-significant. Since the normality assumption was violated in the variables referring to literature and referring to past experience, results were additionally confirmed by the removal of outliers and by Mann–Whitney U tests: Mdn_GD = 7.71, Mdn_EW = 21.8, U(14, 15) = 207, z = 4.45, p < 0.001 and Mdn_GD = 16.93, Mdn_EW = 13.2, U(14, 15) = 78, z = −1.18, p = 0.252. Both referring to literature (Cohen’s d = −1.77) and generating explanatory hypotheses (d = 0.86) showed large effect sizes, with participants in the GD condition demonstrating more frequent uses of these processes in comparison to the EW group.

Conceptually, teachers’ knowledge construction involves processing both their own teaching decisions as well as their students’ learning (Buehl and Fives, 2016). Our study focused on PSTs as learners about different aspects of teaching. We chose this limited focus because we assume that teachers’ knowledge construction regarding their teaching role should better be studied in the environment of practicing, in-service teachers rather than on the basis of the mostly hypothetical reflections of PSTs still enrolled in university programs. To address this issue, different research designs would be necessary.

Our small sample size limits generalizations to a broader PST population and precluded an examination of possible biases from the slight gender imbalance. Moreover, self-selection bias may exist, as PSTs chose their preferred reflection condition. Although we compared the two groups on several measures, we cannot rule out that unobserved differences (e.g., talkativeness or discussion affinity, writing self-efficacy, comfort with public reasoning) may have influenced both format choice and process use. Our design privileges ecological validity over causal identification; results should be interpreted as descriptive of authentic practice under choice, not as causal effects of format. Interpreting results is further complicated by possible methodological biases associated with the two reflection modes (EW and GD). While GD transcripts may be regarded as proximal indicators of ongoing processes, individual reflective essays might represent final products rather than the processes of reflective engagement. Before writing an essay, PSTs might have engaged in several steps of knowledge construction that may be invisible in their written essays. Thus, the GD transcripts may more directly reveal underlying processes compared to essays.

Additionally, essay-writing evokes a more academic mindset that peer discussions (Kuhn, 2009). PSTs might feel obligated to use more scientific sources and express arguments in a more sophisticated way. However, this might not necessarily reflect its perceived practical utility. As previous research has shown, PSTs often struggle to connect abstract concepts from educational sciences to concrete teaching actions unless these connections are explicitly scaffolded (Zaragoza et al., 2021). In our study, we were unable to collect think-aloud protocols for the participants in the EW condition that might have revealed earlier phases of reflective thinking. The GD task is also intrinsically multi-layered: it integrates written artifacts (prepared notes/bibliographies) with oral deliberation, and participants may alternate between these modes during the session. These modality and task-structure differences likely contributed to the observed process profiles and should be considered when interpreting between-format contrasts. To replicate and extend our findings, we suggest incorporating different kinds of individual process data such as annotated outlines, planning notes, or concept maps.

GD transcripts also posed challenges in identifying individual participants’ knowledge construction processes. We adopted an individual-contribution lens to align modalities (one essay ≈ one individual’s contribution set), but we acknowledge that this choice ignores within-group clustering. Even though we segmented discussions into individual utterances, participants’ arguments build on others’, making them not fully independent. This might be problematic when several participants engaged in similar knowledge construction processes but only one of them demonstrates these explicitly. In this case, we would only count this process as part of the engagement of one individual PST. Our data suggests that this is not a general pattern, as there were proportionally more codes in GD transcripts than in the essays. Still, p-values in GD/EW contrasts may lead to over-interpretation. We emphasize converging patterns and effect sizes, and were cautious when drawing inferences from these results. To address selection and dependence concerns while preserving ecological validity, follow-up studies should prioritize: (a) random assignment to EW/GD or within-subjects cross-over designs; (b) designs that hold preparatory work constant across formats; and (c) analyses that model clustering when using group-embedded data.

As noted in the Methods section, character counts were used to approximate elaboration across German written and spoken language; this choice has limitations (e.g., sensitivity to transcript conventions). While proportional reporting reduces scale dependence, different metrics (words, speaking time) may capture distinct aspects of production. Our core claims center on relative profiles (which processes and qualities are foregrounded) rather than absolute quantities. Another limitation of our analysis approach is that the use of one-sided tests was data-driven rather than theory-driven. Consequently, the findings should be interpreted with caution and considered preliminary.

Despite these limitations, our study’s results remain significant due to the authenticity of our comparative approach within real-world teacher education contexts. Alternative methodologies might offer more direct comparability but would sacrifice the naturalistic setting crucial for understanding reflective processes in actual practice. Our findings highlight trade-offs in research design and underscore the value of understanding reflective practices as they naturally occur.

Achieving continuous reflective practice in teaching requires a deeper understanding of PSTs’ knowledge construction processes. We consider our study and coding scheme a starting point to assess types and implementation qualities of knowledge construction processes directly linked to the explanation of complex pedagogical situations likely to occur in school practice. In doing this, we offer insight into how to identify and evaluate PSTs’ process enactment on ecologically valid tasks and into how different reflective tasks might support or constrain PSTs’ tendency to integrate different forms of knowledge into classroom reasoning. This may help educational researchers and teacher educators, and, in the long run, teachers’ classroom practice. Educational researchers might adapt the coding scheme for various settings and reflection formats.

Teacher educators can use our findings to select reflection formats depending on their courses’ specific aims. If the aim focuses on applying a theoretical approach or an empirical finding to a pedagogical situation, an individual task like essay-writing might be preferable. However, if teacher educators aim to elicit a combination of various processes, linking educational science knowledge with experiential knowledge, a peer discussion might be more appropriate. Our results suggest that well-structured peer discussions where students are required to come with a prepared reference list support more elaborate process implementations and unfold a larger array of processes without distracting from epistemic focus. This underscores the need for teacher educators to explicitly teach PSTs how they are supposed to implement aspects of professional reflection in elaborate ways. Such specificity is crucial for understanding complex classroom situations and influencing professional judgment and decision-making in meaningful ways to become reflective practitioners (Schön, 1983)—that is, professionals who can interpret complex teaching situations through a thoughtful integration of theory and evidence from educational science with their context-specific judgment.