Classroom management functions as a support for instruction and requires the teacher to perform demanding actions (Martin and Sass, 2010). It can be categorized into three facets: monitoring, managing momentum and establishing rules and routines (Gold et al., 2021). Monitoring refers to Kounin’s (1970) dimensions withitness and overlapping. Additionally, it includes strategies aimed at purposefully interrupting classroom disruptions with minimal distraction (Landrum and Kauffman, 2006; Simonsen et al., 2008; van Tartwijk, 2009). Managing momentum relates to the structuring of the lesson. It includes Kounin’s (1970) dimensions smoothness and momentum. Similarly, for a lesson to be successful, it is important to establish a transparent teaching process as well as clarity about tasks, lesson objectives and behavioral expectations – establishing rules and routines (Emmer et al., 1994).

To be able to react appropriately to classroom events relevant to classroom management, teachers must be able to perceive these events in a time-appropriate manner (Seidel and Stürmer, 2014; Gold et al., 2021). Two interconnected sub-processes are key: the noticing of relevant teaching situations and the knowledge-based reasoning of these events in order to be able to react appropriately to the situation (Sherin, 2007; Seidel and Stürmer, 2014). Applying these sub-processes requires declarative, conceptual and case-based knowledge (Berliner, 2001; Stürmer et al., 2013; König et al., 2014). According to the perception-interpretation-decision making model (PID model), professional vision is a situation-specific cognitive skill that mediates between dispositions, professional knowledge, affective-motivational characteristics and classroom performance (Blömeke et al., 2015; Blömeke and Kaiser, 2017). These cognitive processes occur before, during and after teaching in the classroom and thus are an integral part of teachers’ professional actions. The promotion of professional vision and its sub-processes, noticing and knowledge-based reasoning, thus should be one objective of teacher training programs and, recently, the number of studies on this topic has increased (Gold et al., 2021; König et al., 2022; e.g., Weber et al., 2018).

Research findings show that experts – in contrast to novices – can differentiate better between important and unimportant events in the classroom (Star et al., 2011; van den Bogert et al., 2014). Expert knowledge is more elaborate, more interconnected and more accessible than is the knowledge of novices (Borko and Livingston, 1989; Sabers et al., 1991). If disruptive behavior occurs, novices do not try to determine what might be triggering it and do not consider possible solutions to stop it – in contrast to experts. A study on this topic has been performed by Sabers et al. (1991), who evaluated novices’, advanced learners’ and experts’ comments on classroom videos. They observed that, in contrast to experts, novice and advanced learners merely described visual and auditory cues and did not make connections between auditory and visual cues. Those participants made little overall reference to auditory aspects, using them only to describe what they saw. The novices and advanced learners were not able to interpret the spoken language and integrate the meaning of language to the whole observation. In contrast, experts included the teacher’s spoken words and visual observations in their interpretations. To make meaningful interpretations about events in the classroom, experts link the observation of actions to the use of language and incorporate their professional knowledge (Sabers et al., 1991).

Classroom management is considered as a deep structure of teaching quality (Kunter and Voss, 2013). It is, thus, challenging for teachers to perceive events relevant to classroom management since these are on the level of learning processes and cannot be perceived on the surface level on the basis of features such as teaching methods. These challenges are evident in that novices differ from experts in their professional vision of classroom management (Wolff et al., 2015; König and Kramer, 2016; Gold and Holodynski, 2017; Stahnke and Blömeke, 2021). Thus, teacher education should promote the professional vision (of classroom management) of prospective teachers at an early stage through deliberate interventions.

To date, there have been various settings in which student teachers’ professional vision of classroom management has been promoted using classroom videos and classroom transcripts (Weber et al., 2018; Kramer et al., 2020; Gold et al., 2021). Another approach is to compare teaching examples in five steps that can be used specifically to promote the sub-processes of professional vision [describing, interpreting, evaluating, and explaining (Sherin and van Es, 2009; Seidel and Stürmer, 2014)]: description, classification into categories, juxtaposition, summarization, and conclusion (Wedde et al., under review).

In our studies, auditive teaching examples of classroom management were used as contrasting cases for the comparison approach. In our lesson examples, excerpts from classroom scenes were listened to. Each of them contained audio track sequences of a classroom video. The first auditive teaching example represented a less successful, the second auditive teaching example a more successful implementation of classroom management strategies by the teacher. Listening to auditive teaching examples has the advantage of the participants being able to focus on verbal strategies of classroom management. These tasks overcome, for example, the issues expressed in the study by Sabers et al. (1991): novices included the auditive level of videos only descriptively and not analytically in their annotation. Practicing lesson analysis via auditive examples can be an effective way to bring the relevance of language into the focus of learners’ analysis.

Two promising learning formats following the problem-solving prior to instruction approach are invention activities and worked solutions. Comparisons are integrated into these learning formats by using contrasting cases. In the first step of invention activities, learners are given a set of contrasting cases with a prompt to develop a solution to the given problem by comparing the cases. For the worked solution, learners are generally presented with a sample solution for the contrasting cases they need to study as they work through the task (Renkl, 2014). In the subsequent instructional phase, the learners are presented with the canonical solution to the task and instruction is given on the topic to be learned (Schwartz et al., 2011).

In a comparison task, learners need to identify, explain and organize features. A comparison of constructed cases helps highlight the deep features of the concept to be learned (Wedde et al., 2022). Recognizing deep features is a significant part of working through the learning formats. It is also known from research on expertise that experts tend to categorize principles by deep features; novices tend to categorize through surface features (Chi et al., 1981). If the new knowledge of the concept to be learned is stored via deep features, it can be applied better to new situations, and flexible knowledge is developed (Kapur and Bielaczyc, 2012; Loibl et al., 2017). The auditive contrasting cases in this study aim to reduce the complexity of teaching analysis, to make essential strategies of classroom management identifiable through contrasting and to provide negative knowledge about classroom management, i.e., knowledge about non-meaningful pedagogical practices (Oser et al., 2012; Wedde et al., 2022).

Comparing as an integrated task in invention activities can be more effective than simply asking students to find similarities and differences (Chi et al., 2012; Chin et al., 2016). Many studies have shown that invention activities promote the acquisition of conceptual rather than procedural knowledge (Loibl and Rummel, 2014a,b; Loibl et al., 2017; Weaver et al., 2018). Overall, studies contrasting the worked solution with the invention activity showed that the worked solution tended to be superior. The worked solution is considered the preferable format, especially for learners with little expertise and in the context of cognitive load theory, when it comes to task formats with a high level of element interactivity (Sweller et al., 1998). Element interactivity depends not only on the expertise of learners but also on the complexity of the learning material and refers to the elements that learners have to process in working memory to find a solution (Ashman et al., 2020).

In contrast to the worked solution effect, the generation effect, related to invention activity, assumes that learners who have to develop something while working on a task acquire knowledge more sustainably (Bertsch et al., 2007). In their study, Chen et al. (2015) found that the generation effect can occur with low element interactivity and that learners achieved better results when they had to develop something themselves. In contrast, when element interactivity was high, learners achieved better learning outcomes when they worked with a worked solution. This finding indicates that task formats with a low level of element interactivity, which contain prompts to generate something, promote learning.

Studies on the impact of when instruction is given (problem-solving prior to instruction or vice-versa) and on using contrasting cases have shown mixed results. One study (Loibl et al., 2020) could not find any effects: There were no pre- and posttest differences in conceptual understanding between the experimental groups that completed the task either before or after instruction and the task with or without contrasting cases. The authors of that study, thus, highlight that it is unclear which condition is beneficial to learning. They argue that novices, in particular, often did not use the beneficial features of the contrasting cases and may need additional guidance to do so. Learners who work with contrasting cases without additional guidance may tend to discover surface rather than deep features of the concept being learned; they would then perform low in the posttest. However, comparing contrasting cases has been shown to help learners identify more deep features of a concept (Loibl and Rummel, 2014a). Roelle and Berthold (2015) found that the experimental group with additional support for comparing contrasting cases, similar to a worked solution, performed better on the posttest in terms of conceptual knowledge than the other two experimental groups that did not receive additional support for comparing or did not complete a preparatory task. A study involving student teachers on the topic of learning strategies showed that the group working on the worked solution performed better regarding conceptual understanding on the posttest, although in one of these studies, for example, the invention activity resulted in significantly more knowledge gaps after task completion (Glogger-Frey et al., 2015). Those authors concluded that, in domains such as educational studies, working with a worked solution would be particularly suitable for knowledge acquisition and transfer and that long-term knowledge acquisition would have to be checked again in further studies (Glogger-Frey et al., 2015, 2022).

Previous studies have evaluated different aspects of learner solutions from the problem-solving phase regarding invention activities and worked solutions. We summarize the findings of these studies using the term solution quality. These findings lead to the question as to how solution quality is related to learning outcomes. There are different approaches to addressing this question. In one approach, the assumption is that a high solution quality results in advantages for the learning process since the problem-solving phase can support learners in discovering deep features of a concept to be learned (Loibl et al., 2017). This support is beneficial for the learning process because the deeper features are identified during the problem-solving phase, and the learners can focus on the less deep features during the instruction phase (Roll et al., 2011; Loibl and Rummel, 2014a). Alternatively, other researchers assume that failure during the problem-solving phase is beneficial for the learning process. This approach is called productive failure (Kapur and Bielaczyc, 2012). Important in productive failure is that learners become aware of their knowledge gaps and that, during instruction, they receive an explanation as to why a solution does not work (Sinha et al., 2021).

Previous studies also show a mixed picture of how solution quality relates to learning outcomes. In their study with psychology students, Wiedmann et al. (2012) found that high solution quality in the solution attempts correlated positively with procedural and conceptual knowledge in a follow-up quiz. In their experiment with student teachers on learning strategies, Glogger-Frey et al. (2015) only assessed the correctness of the solution attempts of the experimental group invention activity. The results also showed a positive correlation between the solution quality and posttest scores. In another study with student teachers, Glogger-Frey et al. (2022) found no correlations between the number of solution attempts or the number of optimal or suboptimal solution attempts for the experimental group invention activity. A further study also evaluated solutions only for the experimental condition invention activity. The study found that the quality of the solutions correlated positively with the results in the posttest (Roelle and Berthold, 2015). In their study on the problem-solving prior to instruction approach, Loibl and Rummel (2014a) found that, although the solutions of the group working with contrasting cases were of higher quality, the quality was not related to conceptual knowledge in the posttest. A recent study conducted by Sinha et al. (2021) with students in a course on data science yielded results in line with the productive failure approach: Learners who worked with a failure-driven task (learners were prompted toward an incorrect solution) during the problem-solving phase achieved higher conceptual understanding in the posttest than learners who worked with a success-driven task (learners who were prompted toward a correct solution). Hence, solving a task less successfully may well lead to better learning outcomes than solving a task more successfully.

The aforementioned studies, which are representative of research on the relationship between solution quality and learning outcome, show a mixed picture (Sinha et al., 2021). The findings suggest either a positive relationship between solution quality and learning outcomes (Wiedmann et al., 2012; Glogger-Frey et al., 2015; Roelle and Berthold, 2015), a low solution quality and better learning outcomes (Sinha et al., 2021), or no correlations between solution quality and learning outcomes (Loibl and Rummel, 2014a). Moreover, it remains to be clarified to what extent these results can be replicated for educational research. The studies mentioned above were mostly conducted in the scientific and mathematical fields. Similarly, limited studies have shown that the worked solution is the more effective task format for educational studies (Glogger-Frey et al., 2022). The results of our studies up to now indicate that, at the process level, the worked solution task format is more appropriate (Wedde et al., 2022, under review).

Study 1 examined the extent to which the learning format can initiate professional vision of classroom management in the short term by comparing the two experimental conditions and investigating the extent to which there is a positive or negative correlation between solution quality and learning outcomes. Teacher education studies on invention activities and worked solution with a high element interactivity found that the worked solution condition is more suitable (Glogger-Frey et al., 2015, 2022). Other study results also proved that, for learning material with a high element interactivity, the worked solution might be the preferable format (Chen et al., 2015). For RQ1, we formulated the following hypothesis.

Overall, the problem-solving phase is assumed to be related to the posttest. Therefore, we included the variables of analytical and content-related solution quality in this study (see Section 5.2.4 for a description of these variables; Wedde et al., 2022, under review). However, it remains to be clarified whether these correlations between the variables of analytical and content-related solution quality are positive or negative. Negative correlations would speak for a productive failure effect (Kapur and Bielaczyc, 2012). Therefore, we also formulated non-directional hypotheses to help answer RQ2.

We were able to demonstrate in Study 1 that both learning formats promoted professional vision of classroom management. However, Study 1 did not show which of the two experimental groups was more likely to initiate long-term skill acquisition. Hence, we conducted Study 2 with a new sample. To gain detailed insights into these task formats, we included a follow-up survey to examine the extent to which sustained learning can be fostered by the two task formats. According to the generation effect, it can be assumed that working on an invention activity is more likely to promote sustainable learning than studying a worked solution (Bertsch et al., 2007). However, we assume a high element interactivity for our task material, which is why it is questionable whether the generation effect would occur (Chen et al., 2015). Given the inconsistent evidence from studies on this issue, we also formulated non-directional hypotheses.

Table 7 shows the descriptive statistics of the two experimental groups and the entire sample for pretest, posttest and follow-up. Both experimental groups improved at the posttest compared with the pretest, but the two experimental groups did not differ significantly in the posttest, F(1,51) = 0.06 p = 0.82, η² = 0.00. There was a highly significant effect of pretest, F(1,51) = 31.74, p < 0.001, η² = 0.38. Therefore, hypothesis H1 is rejected. In the follow-up, the experimental group IA performed unchanged from the posttest, whereas the WS group’s test performance declined and was even below the pretest value. An ANCOVA showed a highly significant effect of experimental group, F(1,51) = 11.61, p = 0.001, η² = 0.19, and a highly significant effect of pretest, F(1,51) = 28.87, p < 0.001, η² = 0.36. Thus, hypothesis H2 can be confirmed.

The objective of these studies was to determine which of two task formats, IA and WS, was preferable for promoting professional vision amongst first-year students. The two formats were compared using contrasting teaching examples. To this end, we determined that both experimental groups improved similarly from pre- to posttest and that there were no significant differences between the two experimental groups at posttest in either Study 1 or Study 2 (RQ1). Initially, this result is, indeed, positive, as both experimental groups demonstrated higher professional vision of classroom management. However, neither of the posttest in our studies showed an advantage for one of the two task formats. This finding contrasts to those of other studies that found WS to be the superior task format, especially for tasks with a high element interactivity, which we assumed for the learning material used in this study (Chen et al., 2015; Glogger-Frey et al., 2015; Roelle and Berthold, 2015; Glogger-Frey et al., 2022).

To investigate how sustainable the skill acquisition was, we conducted a follow-up in Study 2. Here, the results of the experimental group WS did not differ significantly from the results of the experimental group IA at posttest. However, both groups differed highly significant at follow-up; the experimental group IA demonstrated consistent results from posttest to follow-up and the results of the experimental group WS decreased from posttest to follow-up (RQ3). This finding would indicate a generation effect (Bertsch et al., 2007). As a result, for the long-term promotion of professional vision of classroom management, IA is the preferable format. It can be assumed that, since the group members developed the categories themselves, deeper cognitive processes were triggered, which resulted in their longer-term ability to perceive classroom management relevant events professionally. The results of this study suggest that when element interactivity is high, more sustained learning occurs with an invention activity.

A previous study found that students in an IA experimental group perceive more knowledge gaps while working on the task than those in a WS experimental group (Glogger-Frey et al., 2015). We could not replicate this result with Study 1, as there were no differences between the two experimental groups. Similarly, awareness of knowledge gaps was not related to the posttest.

The correlations between the variables of the analytical and content-related solution quality of the task showed a mixed picture for the two experimental groups in Study 1 (RQ2). The analytical solution quality of the experimental group WS was positively significantly related to the posttest. This finding may suggest that working through the steps of an effective comparison could initiate professional vision (Wedde et al., under review). Yet, the variables of the content-related solution quality were not significantly related to the posttest of the experimental group WS. Thus, no significant correlations between the number of surface features and the posttest could be demonstrated for the experimental group WS or for the experimental group IA (H3a). Furthermore, no significant correlation existed between analytical solution quality and posttest for experimental group IA (H3d). Nevertheless, two (almost) negatively significant correlations existed between the number of mentioned categories (H3c) as well as the number of deep features (H3b) and the posttest for experimental group IA. Thus, the fewer categories or the fewer deep features used for comparison, the better students performed on the posttest. This finding would suggest a productive failure effect for experimental group IA (Kapur and Bielaczyc, 2012).

With this result, it would have been expected that the experimental group IA would perceive correspondingly more knowledge gaps after task completion, which would have been a possible explanation for the results of this study. The assumption would be that students in experimental group IA are more likely to fail to develop an optimal solution to the task. This failure would lead them to perceive correspondingly more knowledge gaps when working on the task. During instruction, they would then be able to fill previously perceived knowledge gaps, which could lead to more sustained learning as well as networked and organized knowledge. For the experimental group WS, this explanation would mean that, although the students achieved a higher solution quality during the problem-solving phase (Wedde et al., 2022, under review), they perceive fewer knowledge gaps after working on the task and are thus less able to fill these knowledge gaps during instruction, which is why long-term learning is less likely to occur.

The low level of students’ awareness of knowledge gaps in Study 1 cannot contribute to the explanation of these results. This failure may also be due to collection through the standardized items used. In the future, less abstract items or interviews could be used to collect data on the awareness of knowledge gaps following the task completion. According to the statistical power analysis, the estimated sample size for a medium effect size had to be larger than the actual sample in Study 2, limiting the significance of Study 2’s findings. Sample dropout is another limitation of the two studies. In Study 1, the group excluded by the video filter demonstrated higher professional vision of classroom management in the pretest. In Study 2, the group excluded by the video filter demonstrated a lower professional vision in the posttest than the included group. This difference raises the question of how the results of the two studies would have been if these sub-samples had compliant video times and thus did not need to be excluded. Furthermore, no control group was used in this study, so it cannot be ruled out that the results are a test–retest effect. Thus, no results are available on how a control group that had not worked on a comparative task would have developed in their professional vision of classroom management. Additionally, the sample was drawn from only one institution, which also limits the study’s generalizability.

Overall, our study was able to show that there is still unclarity in educational science as to which of the two task formats, IA and WS, is preferable. However, this preference also depends significantly on the learning object. When it comes to developing conceptual knowledge on a topic, WS may be superior to IA (Glogger-Frey et al., 2015, 2022). However, if a situation-specific cognitive skill based on declarative and conceptual knowledge is to be fostered beyond conceptual knowledge, IA may well be the preferable learning format, as our studies have shown. To address in detail the question of the appropriate task format, further studies should examine the extent to which comparing contrasting teaching examples specifically promotes conceptual knowledge of classroom management. Likewise, constructs of affect motivation, which are also the basis of professional vision, could also be included in further analyses (Blömeke and Kaiser, 2017). Additionally, the results showed that comparing contrasting auditive teaching examples positively influences student teachers’ professional vision of classroom management. Future research could also compare auditive contrasting cases with video-based cases to determine which cases are more likely to initiate student teachers’ professional vision at the beginning of their studies with low cognitive load. However, further studies should examine the extent to which the productive failure effect can be replicated. Performing such studies could address the question of how the effect occurred in the context of our study, although previous studies in the field of educational science found contrary results (Glogger-Frey et al., 2015, 2022). In addition, the effect found in Study 2 would have to be verified in a follow-up with a larger sample and with students from different institutions.

In Study 2, the IA experimental group had better results in the follow-up test. The reasons for this difference between the two groups are currently speculative. A further study should examine the differences in more detail. Such a study could, in particular, investigate whether deeper cognitive processes had occurred in the IA group, a suggestion put forward but not substantiated in our previous studies (Wedde et al., 2022, under review). For this purpose, a further study would have to be conducted to determine the reasons for the better performance of the IA experimental group. Nevertheless, we were able to demonstrate that auditive teaching examples are appropriate for promoting professional vision of classroom management among first-year students. These examples reduced the complexity of the analysis during the acquisition phase. Whether this reduction in complexity also succeeds in bringing the significance of language into focus when novices analyze teaching situations must be examined in further studies. To this end, it would be necessary to specifically examine in the posttest and follow-up to what extent the study participants refer to language elements in their professional vision and use these to interpret the visuals (Sabers et al., 1991).