Introduction

In the dynamic realm of modern education and workforce readiness, supporting the development of learners and competencies takes centre stage. Teamwork capabilities are crucial to equip learners and the workforce with the capacity to skilfully tackle intricate challenges that define our ever-changing world (Lamb et al., 2017; Universities Australia, 2019). One of the unique characteristics of teams is their ability to achieve and maintain high levels of team cohesion, an essential component of these capabilities (Beal et al., 2003; Grossman et al., 2021; Kozlowski et al., 1999). Cohesion, as Festinger (1950) defines it, is “the resultant of all forces acting on all the members to remain in the group” (p. 274). Contemporary research acknowledges cohesion as a multidimensional construct encompassing task cohesion, reflecting group members’ cooperation towards shared goals, and social cohesion, representing the interpersonal bonds within the group (Mullen & Copper, 1994).

In collaborative learning, cohesive group members are more likely to engage and support each other, leading to improved learning outcomes (Kreijns et al., 2003). Cohesion fosters leadership, satisfaction, communication and engagement, promoting learning, and naturally emerges when groups share a common goal (Garcia & Privado, 2020; Miyake & Kirschner, 2014; Xie et al., 2019). Task cohesion, which is crucial for team performance outcomes (Grossman et al., 2021), positively influences learning and behaviours in collaborative settings (Miyake & Kirschner, 2014; Zamecnik et al., 2022). When teams are committed to a task, members actively contribute to its completion (Miyake & Kirschner, 2014), achieved through the collaborative efforts of individuals working towards a shared objective within a specified timeframe (Zaccaro et al., 1995). Therefore, tracking these efforts can assess the effectiveness of collaborative learning, identify group dynamic issues, and provide support to improve and maintain cohesion (Zamecnik et al., 2023).

Nevertheless, there remains a significant gap in research investigating the changes in cohesion over time. There is a potential to identify interventions that enhance cohesion, ultimately leading to more successful and cohesive collaborative groups (Zamecnik et al., 2023). With a comprehensive understanding of how cohesion changes, the range of available approaches for improving cohesion in small groups over time and thus maximising their learning outcomes is expanded (Zamecnik et al., 2022). Current research in collaborative learning often positions task or social cohesion as a secondary consideration, with greater emphasis usually placed on other dimensions, such as learning processes (Zamecnik et al., 2022). This focus on learning processes shapes the trajectory of research and practice in collaborative learning, including investigations into technologies and methodologies (Cress et al., 2021; Lämsä et al., 2021). However, it is essential to recognise the importance of emergent group behaviours, such as their cohesion, to shed light on team and learning outcomes in collaborative learning.

Temporal network motifs, in contrast, provide a complementary perspective that enables insights into how trace data can be used to understand the level of task cohesion in team learning environments (Zamecnik et al., 2022). The capacity to approximate task cohesion in teams over time can be achieved by analysing the synchronisation of team members’ interactions with learning resources through analysis of temporal network motifs (Zamecnik et al., 2023). These networking approaches offer a valuable means to capture the intricate dynamics of cohesion within team settings, shedding light on the interplay between external influences and internal cohesion mechanisms (Santoro et al., 2015). By harnessing temporal motif analysis, researchers can gain a deeper understanding of how cohesion evolves in response to various factors, thus contributing to the refinement of strategies for fostering effective teamwork and enhancing learning outcomes.

While temporal motif analysis has provided valuable insights and contributed to shaping the research agenda by using unobtrusive measures to approximate task cohesion within teams in learning environments, a significant challenge arises concerning the construct validity that underpins this work (Zamecnik et al., 2023). Utilising established research tools such as questionnaires to assess team task cohesion and data-driven approaches capturing team behaviour indicative of task cohesion offers a robust pathway to establish construct validity (Salas et al., 2015; Santoro et al., 2015). This process builds towards the accurate representation of the intended theoretical construct and elevates the credibility of research findings. Consequently, this approach empowers instructors to prompt behavioural transitions conducive to enhancing the task cohesion of teams within collaborative learning settings (Zamecnik et al., 2022).

Built upon a validated framework for assessing task cohesion in learning environments (Zamecnik et al., 2023), this study aims to investigate the dynamics of team cohesion in educational settings, specifically focusing on how team members’ perceptions of task cohesion evolve over time. Using two assessment points from the participants’ work-integrated learning (WIL) studies, the study examined the perceived changes in team task cohesion during collaborative learning activities. The research identifies key groups (clusters) of teams and their starting task cohesion perceptions, characterising them based on their involvement in learning activities, performance outcomes and group synchronisation. By leveraging a validated framework for assessing task cohesion in learning environments (Zamecnik et al., 2023), the research explores shared task cohesion perceptions among team members. Through this analysis, the study provides valuable insights into the dynamics of team cohesion in educational contexts and offers practical strategies for instructors to foster team cohesion through targeted interventions. We focus on task cohesion due to the limited research in this area. Despite its potential significance in collaborative learning environments, task cohesion is often overlooked. We see an opportunity to examine task cohesion more closely, particularly regarding how group members engage in collaborative learning tasks. By exploring task cohesion in-depth, we aim to contribute to understanding its role in collaborative learning outcomes.

Research aim and questions

We apply a mixed-methods approach using temporal network motifs and surveys to examine the development of task cohesion in team learning contexts. To our knowledge, no studies have investigated this synergy for task cohesion to date. Temporal network motifs could benefit more from surveys when connecting trace data to build construct validity. Therefore, by methodologically collecting task cohesion scores of teams and combining them with the survey data, we extended the existing methodology by Zamecnik and colleagues (2023) to provide a more valid evaluation of team cohesion. We focus on examining the alignment of teams with different task cohesion perceptions characterised by task cohesion scores, engagement, performance and group synchrony using temporal network motifs. By examining the differences between task cohesion scores and teams’ perceptions, we gain insight into these transitions in a WIL environment and how they are characterised by learning engagement, performance and group synchrony. To achieve this goal, we address the following research questions:

  • RQ1: During the collaboration in WIL, what transitions occur in both task cohesion and the perception of cohesion within teams?

For RQ1, we seek to uncover how both task cohesion and its perception change over collaborative learning activities. We expect that teams who experienced higher levels of cohesion at the start of their collaborative experience will likely maintain it, promote and share similar cohesive perceptions over time. Kozlowski and Chao (2012) indicate that teams who can establish common ground quickly maintain cohesion and perform well, while teams that respond with lower cohesion at the start of their collaborative experience will likely have misaligned perceptions over time. This research question aims to examine how these dynamics unfold in a collaborative learning context, group teams into transition groups and examine those relationships with the goals of the second research question.

  • RQ2: During the collaboration in WIL, how do variations manifest in overall learning engagement, performance and group synchrony?

For RQ2, we connect the transition groups that are distinguished by their task cohesion and perceived cohesion to uncover the extent of their signatures with learning engagement, performance and group synchrony over the course of collaborative learning activities. We expect that teams with high levels of cohesion and similarly aligned cohesive perceptions will have higher learning engagement, performance and group synchrony compared with those with lower task cohesion. Zamecnik and colleagues (2023) suggest that task cohesion will manifest as a result of learning engagement, performance and group synchrony. This research question aims to examine to what extent these manifestations align with the self-reports of task cohesion over the course of collaborative learning activities.

Theoretical background

Cohesion and collaborative learning

Team cohesion has been extensively researched in the contemporary literature, examining the relationship between team effectiveness and performance (Beal et al., 2003; Kozlowski & Chao, 2012; Salas et al., 2015). This study emphasises task cohesion, highlighting the importance of collaborative goal completion, involving the combined efforts of multiple individuals working towards a common objective within a defined timeframe (Zaccaro et al., 1995). One of the main dimensions of team cohesion that has been shown to consistently and positively predict team performance across various domains (such as military, health, education, sports and extreme environments) is task cohesion (Dimas et al., 2021; Grossman et al., 2021; Salas et al., 2015; Zamecnik et al., 2022). A highly task cohesive team will persist in the face of stresses and obstacles (Zaccaro et al., 1995). Elevated task cohesion among members often stems from shared objectives and a readiness to collaboratively work as a cohesive unit, characterised by coordinated group efforts rather than merely relying on personal friendships (MacCoun et al., 2006). An indicator of strong team cohesion is their high levels of engagement in undertaking a series of activities necessary to complete a task (Forsyth, 2021; Marks et al., 2001; Tesluk et al., 1997; Van Swol & Kane, 2019). Constructive exchanges and task independence can even help teams overcome negative relations and mitigate their impact on team performance (de Jong et al., 2014).

The cohesion in teams is dynamic, and it can fluctuate based on external forces or events that may disturb the common ground of the team (S. W. J. Kozlowski & Chao, 2012; Salas et al., 2015). For example, performance results viewed by the team can increase or decrease motivation, impacting cohesion and overall team performance (Karau & Williams, 1997). Concerning a team’s behavioural interaction intensity, processes that exhibit positive reciprocal relationships or rhythms also impact cohesion. According to Kozlowski and Klein (2000), teams develop common ground and attitudes about team cohesion through a process involving challenged perceptions and the convergence of perspectives. Within collaborative learning, teams often exhibit shared patterns of behaviour in how they interact, communicate and work together to achieve goals. Behavioural interactions exemplify task cohesion, signifying commitment, unity, conflict resolution and a collective drive to achieve a common goal (Carron et al., 1985). For our study, we specifically focus on understanding and analysing the task cohesion dimension of teams as it pertains to collaborative learning environments.

Measures and analysis of task cohesion in collaborative learning

The analysis and assessment of task cohesion in collaborative learning contexts is mainly done by cross-sectional surveys that stem from Carron and colleagues’ (1985) early group environment questionnaire (Zamecnik et al., 2022). Typical questions used to measure task cohesion from respondents include statements such as “Our team is united in pursuing its performance goals” or “This team does not provide sufficient opportunities for me to enhance my personal performance.” These statements are rated on a Likert scale with five or seven subscales for each question. While surveys offer flexibility, cost-effectiveness, and generalisability, enabling researchers to grasp the broad characteristics of a given population and ensure construct validity (Shaughnessy et al., 2000), they also present inherent weaknesses that require careful considerations. The limitation of surveys lies in their static nature and limited fidelity in capturing the dynamic development of cohesion over time (Kozlowski & Chao, 2012; Visser et al., 2000). The multifaceted nature of cohesion underscores the need for its comprehensive and accurate measure (Carron et al., 1985; Salas et al., 2015). There have been calls to make the analysis and measurement of cohesion more objective (Kozlowski & Chao, 2018; Salas et al., 2015; Santoro et al., 2015; Zamecnik et al., 2022).

Recent endeavours have sought to broaden the horizon of harnessing learning analytics and psychometric approaches to develop methods that enable the assessment of cohesion using learner trace data (Zamecnik et al., 2023). Analysing fine-grained data in learning analytics (LA) research has continuously supported making inferences about academic performance and learner autonomy (Winne, 2020). For example, Zamecnik and colleagues (2023) used temporal network motifs, which are recurring patterns that occur in time-varying networks (Paranjape et al., 2017), to understand how cohesion unfolds by analysing the group synchrony of teams using learner trace data. Synchrony, as observed in groups displaying such patterns, plays a crucial role in fostering group cohesion (Kerr & Bruun, 1983) and enabling them to effectively carry out their tasks (Wiltermuth & Heath, 2009). Similar sets of conceptualisations and methods have been proposed to study cohesion with networked and data science approaches (Kozlowski & Chao, 2012; Mohammed et al., 2021; Santoro et al., 2015). They can help depict the representation of group members making an effort and commitment to the task in collaborative learning using learner trace data (Miyake & Kirschner, 2014). By offering objective and tangible measures that mitigate survey bias and enable a more accurate assessment of students’ teamwork (Winne, 2020).

Despite their limitations, surveys remain a powerful tool for providing context and depth to quantitative findings. When combined with data-driven methods, such as in the case of mixed-methods approaches, surveys play a crucial role in examining how teams collaborate in educational settings (Zamecnik et al., 2022). By integrating surveys with data-driven techniques, researchers can gain a deeper and more comprehensive understanding of teamwork, matching the diverse aspects addressed in their studies. This combined approach allows researchers to leverage the strengths of both surveys and behaviour-based assessments, leading to a more nuanced understanding of team collaboration in educational contexts. These efforts offer valuable alternatives to traditional methods of assessing cohesion, providing instructors with actionable insights to enhance team cohesion across different contexts and over time.

Method

Study design

The study centred on participants engaged in a prominent WIL environment offered by a leading educational technology provider. This online platform is designed to deliver experiential learning programs to enhance student employability and foster industry engagement. Within this platform, teams collaborate through various prefabricated materials concerning activities with courses, forums, feedback and grading tools, mirroring functionalities commonly found in traditional learning management systems (LMS). The program is a full-time, standalone, 3-week initiative focusing on improving learners’ job-related skills through team projects, coaching, and professional development activities (Cooper et al., 2010). In this WIL program, teams learn how to collaborate and work effectively to support the project’s completion. Teams worked towards a project set out by the client while using learning resources accessed from the WIL platform and mentors to support them in that learning journey. Learners who participated in the program are from a mix of higher education and industry. The content for weeks 1, 2 and 3 focused on numerous activities (Table 1) carried out by teams, which facilitated the completion of project planning, drafting, and report development, respectively. The learners were all adults from diverse age groups, genders and demographics. Due to privacy and ethics, as part of our initial data collection agreement, the specific details of age, gender, educational background and demographics were anonymised. The prohibition of using personally identifiable information prevented the distribution analysis of the data mentioned above. Despite the lack of information concerning the profile of the learners, these details do not hinder the analysis results.

Table 1 Overview of the engagement activity types and descriptions
Full size table

During the program, students applied their classroom knowledge by collaborating closely with industry mentors and clients, gaining practical skills (Patrick et al., 2008). Students were randomly grouped into teams, with an average group composition of five members per team, to strategise and complete projects using learning materials. The research covered cohorts from late 2021 and early 2022, totalling 944 learners in 194 teams across 15 program sessions. Instructors, comprising industry mentors and clients, evaluated deliverables using specific criteria on a scale of 0.00–1.00 and offered detailed feedback. Teams interacted with different activities (Table 1) on the platform, with each team member’s interactions recorded as log data.

The collaborative learning approach has proven benefits in enhancing critical thinking, problem-solving and teamwork skills, which are essential for success in collaborative work environments (Johnson & Johnson, 1991). Collaborative learning fosters shared responsibility and accountability, leading to more engaged and motivated learners (Slavin et al., 2003). The study design in the WIL platform incorporated various collaborative learning activities, including group discussions, peer feedback sessions and collaborative projects. These activities served as catalysts for active engagement, fostering a culture of knowledge-sharing and teamwork among students. Notably, mentors and clients evaluated teams collectively for each of the three deliverables in this project, their assessments based on the collaborative activities detailed in Table 1 and the quality of their weekly project deliverables. The performance metrics are derived from evaluations by both mentors and clients, who assess teams based on their cohesion, collaboration and the excellence demonstrated in each deliverable.

To support collaboration, clear guidelines and expectations were provided for teamwork, regular feedback from instructors and peers and opportunities for reflection on the collaborative process. These scaffolds were designed to enhance student’s ability to work effectively together and improve learning outcomes.

RQ1 – perceptions of task cohesion and transitions

Data collection

The study utilised survey data administered by the technology provider at the end of weeks 1 and 3. The questions on team cohesion were based on four survey items from the Carless and De Paola (2000) study, altered to be relevant within the WIL context. Teams rated responses on a 5-point Likert scale (1: strongly disagree, 5: strongly agree) for the following four questions:

  • Commitment: “I am content with my team’s commitment and contribution level.”

  • Conflict resolution: “Our team effectively resolves conflicts as they arise.”

  • Receptive: “Our team welcomes feedback and takes appropriate action.”

  • United: “Our team is united in pursuing program goals and outcomes.”

The self-reports were perceived as accurate, and the instructors found no major concern when reading the responses provided by the team members.

Measuring and evaluating perceptions of task cohesion and their transitions

To answer our first research question, the first step of the analysis examined the internal consistency of the surveys using Cronbach’s alpha reliability coefficient (Cronbach, 1951). We then used a weighted Kappa conceptually similar to Fleiss’s (1971) for our calculation. It measures the degree of agreement or disagreement between raters’ categorical responses with our survey items. The weighted Kappa values provide insights into the level of consensus or disagreement among the raters, incorporating weights to account for varying levels of disagreement. For example, teams with agreement below a weighted Kappa of 0.33 were labelled with different (diverging) cohesion perceptions. In contrast, teams with agreement above a weighted Kappa of 0.33 were labelled with similar (converging) task cohesion perceptions. Given that a team’s perception is not a constant throughout the 3-week program, we also examined the changes in team cohesion perceptions between Week 1 and Week 3.

RQ2 – manifestations of engagement, performance and group synchrony

Data collection

The data supporting our second research question comprised log records documenting student interactions with pre-designed course materials, categorised into 10 distinct activity types, detailed in Table 1. These learner trace data were obtained from our industry partner’s online WIL platform, utilising their AWS cloud infrastructure.

For instance, the skills activity type encompassed descriptions such as “Team professional development”, “Conflict resolution”, and “Networking”. The project and assessment activity types involved specific learning-supportive and project-submission activities. These logs were collected and pre-processed to the extent that they are useful for analysis.

Measuring and evaluating engagement, performance and group synchrony

To answer the second research question, we examined the differences in the performance of teams with different patterns of task cohesion perception change. We used assessment score outcomes provided by the client and mentor for each weekly deliverable that the team completed as measures of the team’s performance. These scores are determined collectively by the mentor and client on the basis of how well the team achieved those outcomes (see Section "Performance" for an explanation). The engagement of teams was based on the teams’ frequency with the activity types presented in Table 1. Team engagement was measured collectively by counting the number of times that each team member interacted with the activity types using statistical approaches. Engagement refers to how team members interacted with the online learning content within the WIL platform by accessing content relative to achieving the weekly deliverables (see Section "Engagement" for an explanation). Lastly, group synchrony was estimated on the basis of the temporal sequences and patterns of team members’ interactions with learning activities over time (see Section "Group synchrony" for an explanation). These interactions help us identify recurring patterns within teams, allowing us to infer their level of task cohesion. A non-parametric significance test was conducted to assess the significant differences among the four transition groups in terms of performance, engagement and group synchrony manifestation, followed by post hoc analysis using the Dunn test with Bonferroni correction (see Section "Group synchrony" for an explanation).

Network of team interactions and motif extractions

After outlining the key components of our analytical approach, we now delve into a detailed examination of group cohesion. This aspect of our analysis was particularly significant for our second research question, as it shed light on how teams evolve in their collaborative learning journey. We adopted an analytical approach by Zamecnik and colleagues (2023) to measure group synchrony using temporal network motifs to explore its alignment with the four transitions. The analysis and data visualisation were performed in the R programming language. Similarly, we analysed team interactions in a temporal motif network using the Stanford Network Analysis Platform (SNAP) library (Leskovec & Sosič, 2016). Nodes represented team members, and edges indicated subsequent interactions with the engagement activities outlined in Table 1. For example, an edge between student A and B, would indicate that there was an interaction of student A with a particular activity that was followed by an interaction of student B with the same activity. We focused on a group of six specific activity types, excluding those related to the final assessment submission, and represented interactions between team members and this group of activities using edges in the graph. Activity types were grouped to understand how these sets of activities contribute to task cohesion, allowing deeper examination into overall patterns and dynamics of interaction within the group. Doing so allows us to study how the group collaborates and coordinates its efforts to achieve common goals. Each edge also included a duration attribute, capturing the time before subsequent interactions occur. Temporal motifs offer a means to characterise teams on the basis of the duration and frequency of interactions with specific activity types, forming the motifs. By analysing these motifs, we can infer each team member’s level of engagement and commitment to the team’s shared cognition. This approach provides insights into how teams collaborate and coordinate their efforts, shedding light on team interaction and cohesion dynamics. We considered interactions within 1 day to calculate motifs due to the program’s daily engagement expectations. This analysis provided insights into how each motif characterised team clusters and enhanced our understanding of team synchrony and cohesion.

The methodological choice of using temporal network motifs, as introduced by Zamecnik et al. (2023), offers a promising approach to uncovering the dynamic complexities of task cohesion. Their results, as noted, are encouraging, which motivated us to apply this technique in our study. By doing so, we aim to validate its effectiveness further, as it has the potential to infer task cohesion as it naturally occurs in collaborative learning settings. This choice allows us to delve deeper into the nuanced interactions among team members, providing valuable insights into the evolving nature of teamwork and collaboration.

Figure 1 illustrates different motifs with specific colours. Blue represents all four Dyad motifs, capturing pairs of students (orange and green) interacting with the same learning resource. Light blue represents three possible Switch Triad motifs, where two members (orange and purple) have engaged with a learning resource previously engaged by a third team member (green) but not together. Red represents six possible Chaotic Triad motifs, where three team members (green, orange and purple) engage on the same resources in a non-sequential order. Orange represents two possible Ordered Triad motifs, where three members engage on the same resources sequentially. Non-highlighted represents 21 possible star triad motifs with a similar description to switch triad, but only involving two members (green, orange, or others) interacting back and forth with the same activity type.

Fig. 1
figure 1

Grouping of temporal motif codes into five motif groups, where four are highlighted and one is transparent (Paranjape et al., 2017); Blue: Dyads motifs, Aqua: Triad Switch motifs, Red: Chaotic Triad motifs, Orange: Ordered Triad motifs, and Transparent: Triad Star motifs (Zamecnik et al., 2023)

Full size image

In Fig. 1, which depicts the temporal network motifs by Paranjape et al. (2017), we primarily observe a directed graph. However, there are instances, such as motifs M5,3 and M5,1, where partially undirected and fully undirected edges are present, respectively. The edge constraints between vertices are crucial for capturing the various motifs and understanding the underlying structures that form certain patterns. These patterns are important for helping us understand how team members interact sequentially with the activity types.

Results

RQ1: task cohesion, cohesive perceptions and their transitions

The first part of the analysis involved the measure of the internal consistency and reliability of the four Week 1 and Week 3 survey items using Cronbach’s alpha (Cronbach, 1951). These results yielded alpha estimates that exceeded reliability alpha = 0.70 (Nunnally, 1978). The reported results of alpha 0.93 (Week 1) and 0.94 (Week 3) are expected, as the items in the survey are of similar context but not redundant.

Figure 2 shows differences in scores for four task cohesion questions between teams with similar and different cohesive perceptions. Not surprisingly, teams with similar cohesive perceptions reported more consistent task cohesion scores with very few outliers. In contrast, the teams with different cohesive perceptions did not rate task cohesion scores in a similar manner, having far wider disparity in scores and more outliers. Looking at the average cohesion scores of the two groups, we see that teams with similar perceptions reported, on average, higher task cohesion scores than teams with different team cohesion perceptions. A non-parametric test using Kruskal–Wallis confirmed significant differences in terms of cohesive perceptions between two groups of teams for each of the four items (Kruskal & Wallis, 1952) p < 0.0001. These results indicate that task cohesion scores are higher for teams with similar cohesive perceptions and lower for teams with different cohesion perceptions.

Fig. 2
figure 2

Box plot of task cohesion scores along the y-axis, the four question items along the x-axis, and the comparison between Week 1 and Week 3 of cohesive perceptions that are similar and different

Full size image

Table 2 summarises transition groups on the basis of their characteristics in Week 1 and Week 3, categorising them as Aligned, Multifaceted, Regressive, or Progressive, and includes the number of teams, average team size and their standard deviations (σ) for each group. On the basis of the results, the perceptions of team cohesion in Weeks 1 and 3, teams were categorised into one of the four groups:

  1. 1.

    Aligned (SS) – Week 1: Similar; Week 3: Similar

  2. 2.

    Multifaceted (DD) – Week 1: Different; Week 3: Different

  3. 3.

    Regressive (SD) – Week 1: Similar; Week 3: Different

  4. 4.

    Progressive (DS) –Week 1: Different; Week 3: Similar

Table 2 The number of teams for each of the four possible transition groups between Week 1 and Week 3
Full size table

The Aligned (SS) group consists of teams with similar perceptions that remain unchanged from Week 1 to Week 3. They are called ‘Aligned’ because their cohesion perceptions align consistently over time. The Multifaceted (DD) group consists of teams with different perceptions that remain consistent from Week 1 to Week 3. They are named ‘Multifaceted’ because their cohesion perceptions are diverse but remain consistent throughout the study period. The Regressive group (SD) consists of teams with initially similar perceptions that transition to different perceptions from Week 1 to Week 3. They are termed ‘Regressive’ because their initial cohesion perceptions regress from each other over time. Finally, the Progressive (DS) group are teams with initially differing perceptions that transition to similar perceptions from Week 1 to Week 3. They are named ‘Progressive’ because their cohesion perceptions progress towards alignment over time.

These four distinct transition groups are delineated by their consistent perceptions. The number of teams (N) for the Aligned, Multifaceted, Regressive, and Progressive groups were 43, 72, 25, and 54, respectively. Notably, all transition groups had an average team size of five members, indicating that team compositions were comparable. The Aligned (SS) group has the highest average team size, 5.39, with a standard deviation (σ) of 0.98, suggesting relatively low variability in team sizes within this group. In contrast, Regressive (SD) has the lowest average team size, 4.74, with a higher standard deviation (σ) of 1.17, indicating greater variability in team sizes within this group.

RQ2: cohesive perceptions characteristics with performance, engagement and group synchrony

Performance

Table 3 presents performance scores for three main assignment projects (i.e. plan, draft and report) for each transition group.

Table 3 The assessments and associated performance scores as means and standard deviations (i.e. sigma) for each perception group between Week 1 and Week 3
Full size table

The performance scores for all four perception groups on Week 1 were approximately the same, indicating no differences between teams with different perceptions of team cohesion. However, the performance score results for the perception groups on Week 3 were different, indicating a notable change in their performance. The Aligned (SS) group had stable cohesion perceptions over time, increasing scores from 0.84 to 0.89. The Multifaceted (DD) group maintained different baseline perceptions over time, decreasing scores from 0.84 to 0.77. The results indicated that teams who maintained similar cohesive perceptions increased their performance, and teams who maintained different cohesive perceptions decreased their performance. For the Regressive (SD) and Progressive (DS) groups that transition (i.e. where cohesive perceptions transition from similar to different and vice-versa), their performance scores remain close, ranging from 0.84 to 0.86 for Week 1 and Week 3. The Kruskal–Wallis test was performed as a follow-up analysis to examine the differences between the four perception groups concerning their Week 1 and Week 3 scores. The results for Week 3 scores show significant differences between the perception groups (χ2 (3, N = 194) = 26.472, p < 0.000), with a moderate magnitude (0.124). As a result, a post hoc analysis for Week 3 score was performed using the Dunn test with Bonferroni correction for the adjusted p value in Table 4.

Table 4 Post hoc analysis of the differences in assessment performance between the four groups of teams based on their transition of team cohesiveness
Full size table

Engagement

For both Week 1 and Week 3, all four groups behave similarly. However, between Week 1 and Week 3, there were changes for all of them, as shown in Table 5. The non-parametric Kruskal–Wallis test was performed to check for significant differences between four transition groups at Week 1 and Week 3 among the types of activities. All engagement activity types except the feedback (χ2 (3, N = 194) = 14.737, p < 0.002) for Week 1 had a large magnitude (0.06), and feedback (χ2 (3, N = 194) = 13.671, p < 0.003) for Week 3 had a small magnitude (0.056).

Table 5 The mean summary of team activity for each of the transition groups during the 3-week program
Full size table

The only significant differences between the four transition groups were observed for the feedback activity. The engagement with most other learning activities was not different between the transition groups. As a result, the post hoc analysis test was conducted with pairwise comparisons using Dunn test with Bonferroni correction to check between the transition groups to determine how the feedback activity differs between each group.

Table 6 shows the significant differences between transition groups Aligned - Multifaceted, Multifaceted – Regressive and Multifaceted – Progressive for Week 1. For Week 3, there are significant differences between transition groups Aligned - Multifaceted and Multifaceted - Progressive. Teams with different cohesive perceptions have comparatively lower feedback interactions than teams with similar cohesive perceptions. Teams who are not cohesive throughout the 3-week program have significantly lower interactions with feedback content than other transition groups.

Table 6 Post hoc analysis between transition groups and the mean summary of the mid-test and post-test regarding the feedback activity feature with pairwise comparisons using the Dunn test with Bonferroni correction
Full size table

Group synchrony

The study examined the differences between the transition groups regarding the development of temporal network motifs observed during their interactions in the learning platform. Figure 3 presents the total number of motifs for each of the five groups of temporal motifs. The most frequent motifs were star motifs, followed by the Switch, Ordered, Chaotic and Dyad motifs. The results reveal the presence of Star triad interactions, where interactions with a learning resource of two team members follow interactions with the same resource by the third team member. Similarly, there are triadic Switch interactions, where interactions of a single team member with learning resources are followed by interactions of two other team members (whose interactions do not follow each other’s interactions). In contrast, triadic interactions where all three members have interacted with learning resources together in a sequential (Ordered triad) and non-sequential (chaotic triad) order occur to a lesser extent than in the other motif groups. The Dyads are the lowest occurring motif, where teams mostly form dyadic interactions in pairs, where two out of three team members interact with learning resources.

Fig. 3
figure 3

Total summary of the grouped temporal motif counts

Full size image

After examining the overall distribution of temporal network motifs, the distribution for each perception group individually was investigated. Figure 4 presents the normalised distribution of the five temporal motif groups for each perception group, with some interesting differences between them. Looking at the star triad motifs, each group had similar average normalised values. Only the Aligned group has the highest normalised values for Switch Triad motifs, whereas the other groups are around the same. A different pattern for the Ordered Triad motifs is observed, with the Regressive and Multifaceted groups having the highest normalised values and the Regressive and Progressive groups having the lowest. In contrast, for the Chaotic Triad motifs, the Aligned and Multifaceted groups are similar and below the Regressive groups and above the Progressive groups. Finally, clear differences for the Dyadic motifs were observed with the Multifaceted and Progressive groups, with the highest normalised motifs and the lowest for the Aligned and Multifaceted groups. To further investigate the differences in temporal network motifs between the four perception groups, five Kruskal–Wallis tests (Kruskal & Wallis, 1952) were used, one for each network motif group. The group assignment is the single independent variable, and the frequencies of five temporal network motif groups are the dependent variables, as seen in Table 7.

Fig. 4
figure 4

Normalised distribution of temporal network motif groups for each perception group

Full size image
Table 7 Kruskal–Wallis test between the teams and each motif group with their size effects
Full size table

The results in Table 7 show significant differences between perception groups for each motif group. The reported size effects are shown where Dyads and Star Triads have moderate magnitudes, while Chaotic, Switch and Ordered Triads have small magnitudes. To further investigate the groupwise difference in the temporal motif instances, the Dunn test as a post hoc analysis was used after the Kruskal–Wallis test (Dinno, 2015). Table 8 below presents the results of the post hoc analysis.

Table 8 Post hoc analysis between clusters and the normalised number of the five grouped temporal motif instances using the Dunn test with Bonferroni correction showing Bonferroni adjusted p-values
Full size table

Table 8 compares four perception groups of teams to examine the differences concerning five temporal network motifs based on the results in Fig. 4. The results have shown that several perception groups differ significantly regarding Dyads, Chaotic and Star Triad motifs.

The results depict temporal network motif frequency transitions across perception groups over time, shedding light on team cohesion and interaction patterns concerning learning resources. Figure 5 illustrates the portion of instances for each perception group’s motif categories over 3 weeks. The x-axis represents program days, and the y-axis shows the total instances for categorised motifs, labelled by shape and colour in the legend.

Fig. 5
figure 5

Overview of the portion of instances (normalised) over 3 weeks (21 days) for each motif group per perception group

Full size image

Aligned teams demonstrate a gradual increase in temporal motifs, especially ligned with Switch and Ordered Triads, reflecting a commitment to learning goals. Dyadic motifs are lower, indicating a preference for triadic interactions and shared commitment. In the Multifaceted group, there was an early rise in motifs, particularly Ordered and Star Triads, transitions to Dyadic motifs in Week 2, and shifting back to triadic in Week 3, though shared commitment could be limited. Regressive teams mirror the Aligned group’s plateaued motif distribution in Week 1, with substantial triadic spikes dominated by various triads. The Progressive group initially lacks motif development, gradually aligning with the Multifaceted pattern by Week 2, leading to a dominance of Star and Ordered Triads by Week 3, suggesting the developing shared commitment.

Discussion

RQ1: transitions in task cohesion and perceptions over time

In this work, we build on the exploratory research by Zamecnik and colleagues (2023), using a mixed-methods approach to examine the alignment between learner trace data and self-reported survey responses. In this study, we introduced a questionnaire to gauge the task cohesion scores of teams. Through the use of statistical techniques, we identified teams with similar or contrasting perceptions regarding their task cohesion. To understand how perceptions transitioned over time, we grouped them into four perception groups: Aligned, Multifaceted, Regressive and Progressive.

For the first research question, the findings depict that teams with similar perceptions also rate their team with high task cohesion. Teams with different perceptions rate their task cohesion mixed and generally lower in comparison. These findings support the theoretical view that shared perceptions or group agreement will likely have higher task cohesion (Klein et al., 2001; Kozlowski & Klein, 2000). However, it does not necessarily mean that teams with high task cohesion have identical perceptions, but a consensus is that it promotes perception. Another important finding is that half of all teams reported divergent perceptions in week one. However, in week one, teams tended to have similar perceptions of task cohesion. This phenomenon explains that new teams are unlikely to have shared perceptions of task cohesion early in their project (Kozlowski & Klein, 2000). As they build upon their interpersonal cohesion, we observe an increase in shared perceptions of task cohesion (Picazo et al., 2015). The findings also indicate that teams who maintain an aligned perception at the beginning will likely stay with that perception with high task cohesion scores due to established grounds that only magnify cohesion over time (Kozlowski & Chao, 2012). Smaller cases of teams are regressive, with conflicting perceptions over time that may be due to stresses within the teams, fluctuating behaviours or external factors impacting team members (Kozlowski & Chao, 2018).

Teams with low task cohesion, indicated by multifaceted perceptions, tend to exhibit higher deviations in team sizes compared with teams with aligned perceptions. This higher deviation may suggest that team members have dropped out as a result of potential conflicts, which is a common response when teams are not in agreement (Karau & Williams, 1997; Kozlowski & Chao, 2018; Latané et al., 1979). Such a lack of shared perceptions or low task cohesion can compromise team performance, undermining the benefits of collaborative learning. As a result, these teams are disadvantaged compared with more cohesive teams that exhibit higher cohesion levels. Consequently, further investigation is necessary to pinpoint heterogeneous teams and forecast the signs that may lead to disagreements. By identifying these underlying causes, educators and instructors can offer targeted support to the most at-risk teams, helping to alleviate the negative impacts on their collaborative learning experience.

RQ2: group perception characterisations with performance, engagement and group synchrony

Teams that exhibit higher task cohesion scores are recognised for their correlation with positive performance outcomes (Grossman et al., 2021). Such cohesion often translates into more effective collaboration, leading to enhanced results and a more productive learning environment. Despite our findings concerning the four group perceptions described in the results, the team participants arguably performed very well based on their scores. Upon closer examination, there are indications that some perception groups perform better than others over time. Our findings reveal a clear contrast in performance outcomes between different types of teams. Specifically, teams that maintain aligned perceptions tend to achieve higher performance outcomes. In contrast, teams characterised by multifaceted perceptions often experience a decrease in performance outcomes, even if they initially achieve a satisfactory result. This highlights the importance of cohesion and shared understanding within a team for sustained success. Given that the scores of task cohesion were generally higher, including performance outcomes, it was no surprise that engagement might be relatively similar. Our results affirm this, albeit significant differences in engaging with feedback for teams with multifaceted perceptions compared with other perception groups. Given the established connection between feedback and team learning and cohesion (Gabelica et al., 2012), it stands to reason that teams actively seeking feedback for improvement would see a noticeable impact on their performance and cohesion. This underscores the essential role of continuous feedback in fostering growth and unity within a team. Furthermore, follow-up studies are needed to explore why such groups are not pursuing feedback as much as the other groups to improve their learning.

Given the nature of teamwork, the calculation of team averages from learner trace data may not be enough to characterise cohesion even in a longitudinal sense (von Treuer et al., 2018), which might explain why engagement with learning activities does not depict such a scenario based on the four perception groups in this study. However, exploring team cohesion based on their temporal motif characterisations (i.e. the five motif groups) shows intriguing insights and offers an alternative approach to examine task cohesion beyond the summarisation of engagement counts.

For the second research question, the findings of our study are similar and expand the previous works of Zamecnik and colleagues (2023), who indicated that teams displaying higher patterns of triadic motifs over time are likely indicators of task cohesion. We also found that these teams with higher triadic interactions display higher performance scores, which corroborates the work by Zamecnik et al. (2023). Similarly, the dyadic motifs are a characterisation for teams that maintain differences in perceptions (i.e. Multifaceted) and have lower performance and task cohesion scores. In contrast, teams with similar cohesive perceptions (i.e. Aligned), characterised by triadic motifs, have higher performance and task cohesion scores. Therefore, there is merit in suggesting that, in this study, the use of temporal network motifs seems generalisable and shows promise as an indicator or proxy for task cohesion in these WIL environments. The findings of this study help support and validate the theory put forth by Zamecnik and colleagues (2023) that integrating temporal network motifs with survey data (i.e. mixed methods) is a robust approach for studying task cohesion using learner trace data. The self-reported task cohesion scores and the statistical techniques employed to analyse the teams’ perceptions align with the characterisations of temporal network motifs, as Zamecnik et al., (2023) claimed.

An insight also emerged regarding the role of perception alignment and motif compositions in shaping the dynamics of collaborative learning teams. The findings revealed that teams characterised by aligned perceptions and triadic motif compositions tend to exhibit behaviours indicative of positive interdependence. In such teams, members rely on each other to achieve their learning goals, fostering a sense of shared responsibility and mutual support (Laal, 2013). These findings align closely with the seminal work of Wang and Hwang (2012), underscoring the importance of positive interdependence in promoting task cohesion within collaborative learning environments. Conversely, teams displaying multifaceted perceptions and dyadic motif compositions often demonstrate independence-related behaviours. In these teams, members may prefer to pursue their learning objectives individually, potentially undermining the cohesive teamwork necessary for optimal collaborative learning outcomes. However, challenges occur when characterising the motif category (i.e. ordered triad, chaotic triad, star triad, switch triad and dyad) of each perception group. For certain teams, the motif compositions over time may be indicative of interdependence (ordered triad) and independence (chaotic triad) behaviour. These compositions may help highlight how team members choose to collaborate. Further studies are required to test the assumption of interdependence and independence behaviours and their relationship with the grouped temporal motif compositions in a collaborative learning context, which could offer valuable insights for practitioners to enhance the learning experience.

Teams that assume specific roles, such as team members and leaders, analysing their motifs, such as triadic compositions, may be challenging if the role of team cognition varies among teams (Cooke et al., 2013). A team’s mental processes and structures for coordinating and collaborating may prioritise learning strategies where members focus on tasks outside the LMS while allocating one or two members to study the learning content within the LMS. Such scenarios may be weak in inferring task cohesion using temporal network motifs and may change the meaning behind the motifs’ development, frequency and composition. Therefore, efforts are needed to capture trace data related to various team cognitive behaviours, extending beyond LMS data collection to realise its full potential in collaborative learning contexts.

In this study, which focuses on task cohesion and perceptions, an alternative approach could involve targeting social cohesion to detect team diversity. By analysing motif compositions, frequency, and the nature of conversations, it may be possible to identify instances where teams with different perspectives engage in deliberative discussions. This approach could provide valuable insights into how diversity manifests in team interactions and how social cohesion plays a role in recognising and leveraging diverse viewpoints.

Overall, the findings give credence to existing theoretical assumptions that using techniques that capture the learners’ synchronous interactions with their tasks demonstrates that it is a potential proxy for cohesion (Forsyth, 2021; Van Swol & Kane, 2019). Future studies may need to pivot towards trace data exhibiting emergent state patterns that go beyond counts. One approach could be to focus on dynamic interactions that explore team processes using network methodologies to illustrate the reciprocal connections between team members (Mohammed et al., 2021; Santoro et al., 2015). Therefore, applying temporal methodological approaches in team learning contexts where trace data can be collected from LMS to identify the delicate behavioural patterns that indicate cohesion is one way to measure the complexities of cohesion and further research (Zamecnik et al., 2022).

Implications

This exploratory study brings several practical implications. First, it highlights the importance of promoting shared perceptions and high team cohesion to enhance performance and collaborative learning outcomes in educational contexts. Second, identifying diverse teams and understanding the reason behind their changing perceptions can aid educators in offering targeted support through feedback (Gabelica et al., 2012), enhancing team unity, and minimising negative impacts on collaboration (Kozlowski & Klein, 2000). Third, using temporal network motifs as a proxy for task cohesion offers a promising approach to examining the effectiveness of collaborative learning. It provides insights into team dynamics and can guide educators with interventions to enhance team cohesion and performance (Zamecnik et al., 2023). Educational providers facilitating team learning are encouraged to monitor and assess shared perceptions within teams closely. These observations could lead to more informed decision-making and tailored interventions to enhance team task cohesion. Fourth, the adoption of this analytical approach may indirectly foster teamwork skills. If issues related to cohesion are addressed early in the collaborative learning experience, team members may learn how to work effectively, leading to better learning outcomes. Grote and Kozlowski (2023) emphasise that tools adopted by educational providers that support teams may help prepare learners for challenges working in teams. Finally, the integration of the temporal motif approach into the methodological toolkit for learning analytics offers a unique opportunity to develop a nuanced understanding of learning processes over time. By considering the temporal dimension of learning, researchers can identify subtle patterns and trends that may be indicative of academic success or risk. This approach has the potential to provide valuable insight into how learners progress and evolve over time, enabling educators to intervene more effectively and tailor their support to individual needs.

Limitations and future work

The survey questionnaire was restricted to four items and did not utilise the full extent of the remaining questions typically seen in team cohesion studies of a similar context. The decision for the smaller number of questions was based on the educational provider’s restricted survey capability, given that they were hosting many other questions to examine different attributes of the teams. Another limitation of this study is that a 3-week, full-time project is not a typical study length. However, notable and relevant observations related to the team’s cohesion concur with the existing literature that indicates cohesion manifests relatively quickly (S. W. J. Kozlowski & Chao, 2012). The results give credence to future work in courses longer than 3 weeks. As such, more transition points can be observed to understand how team cohesion takes place over a longer duration. In future work, analytical techniques considering the dynamic processes of teams that identify temporal reciprocal relationships, rhythms and fluctuations are recommended to explore for collaborative learning instead of relying on simple statistical summarisation to measure a complex emergent phenomenon. While this study was exploratory and limited to WIL contexts, it has the potential to be applied to more collaborative learning environments.

Conclusion

In this study, we revealed the characteristics of team cohesion and their perceptions in an online WIL environment by extrapolating connections with their learning engagement, performance outcomes and group synchrony interrelations. We draw on these insights using temporal motif analysis that support existing theories regarding the importance of shared perceptions, high task cohesion for team performance and collaborative learning benefits. Using temporal network motifs provides an alternative approach to studying task cohesion using learner trace data and shows promise as an indicator of cohesion based on previous studies (Zamecnik et al., 2023). The research highlights that teams with different perceptions, based on their task cohesion responses, have distinguished performance and group synchrony characteristics. There is a need for future studies to explore the emergent patterns, dynamic interactions and alternative methodologies to capture the complexities of cohesion in team learning contexts. The research underscores the importance of fostering shared perceptions among teams to optimise their performance and collaborative learning experience.