1. Introduction
Mixed-Method Social Network Analysis (MMSNA) has been discussed as a promising approach to addressing the theoretical calls for better understanding the interactions between human agency and social networks (Crossley, Reference Crossley2010; Crossley & Edwards, Reference Crossley and Edwards2016; Emirbayer & Goodwin, Reference Emirbayer and Goodwin1994). The meso-level network structures influence human agency at a micro, actor-oriented level of individuals’ meanings and motivations, while members’ actions impact upon network structure. The interactions between the two underlie many questions across disciplines of social science. The disciplinary and theoretical traditions have approached the structure–agency debate from multiple angles, and scholars have increasingly recognized the importance of their mutually formative influences (Tan, Reference Tan2011). Network researchers describe the interaction in social network terms, suggesting that norms, conventions, roles, shared practices, feelings, and meanings, and therefore interpretations and actions, are influenced by network structures, but the role of these elements within social relations and network structures has been underexplored. They then call for more research on how individual actions and network structures coevolve (Tasselli et al., Reference Tasselli, Kilduff and Menges2015).
Given the ongoing significance of the structure–agency debate—some authors suggesting that it reflects the most pressing social problems of the human condition (Archer, Reference Archer1996; Tan, Reference Tan2011)—empirical studies of their relative impact are warranted across the disciplines of social science and beyond. While MMSNA has been recognized as a relevant methodology for examining the structure–agency dualism, the purpose, and operational design for mixing multiple methods are often underexplained in this literature (Froehlich et al., Reference Froehlich, Van Waes and Schäfer2020). This paper has three interrelated aims: (1) to propose a replicable method and stepwise procedure for designing and implementing an embedded MMSNA to empirically analyze the structure–agency interaction; (2) to illustrate this approach through a case study on teacher agency within school social networks; and (3) to offer a critical reflection on the methodological and practical lessons learned from the research process.
This educational research project aims at understanding how school staff members (teachers but also managers, administrative and specialist staff) interacted to mobilize and generate knowledge in response to the rise of migrant students. With this case, we illustrate how our design and implementation of MMSNA allowed us to explore the interaction between human agency and network structure in teachers’ and schools’ efforts to address the increasing diversity of student population within their schools. In this introduction, we first discuss the importance of considering the dynamic interplay between human agency and network structure. Next, we turn to the potential of MMSNA for illuminating this dynamic interplay and examine how social networks develop and operate through the agency of network members and vice versa. We use a concrete example of an educational project to illustrate how we designed, implemented and evaluated an embedded MMSNA in the case study of teachers’ support networks within their school communities. Finally, we reflect on the lessons learnt and discuss its potential applications in social network sciences.
Like in other fields, MMSNA in education has commonly involved designs in which qualitative and quantitative analysis precede and inform each other (Pantić et al., Reference Pantić, Brouwer, Thomas and Froehlich2023). In this paper, we present a case of simultaneous use of quantitative and qualitative approaches that embed qualitative data within the social network. In Section 2, we describe the steps of this embedded MMSNA before discussing their unique contribution in Section 3 to illustrate how this design helped unravel the dialectic between the network structures and the embedded actors’ agency by capturing the network properties in concert with the contents that ‘flow’ through them.
1.1. Structure–agency debate and social networks
The structure–agency debate in social sciences has largely been about the relative influence of social structures and human agency in shaping societies and individuals in diverse domains. On the one hand, structuralist researchers stress the primacy of powerful social structures, such as institutions, norms, and stratification systems, in shaping how people act and think (Tan, Reference Tan2011). On the other hand, individualist accounts emphasize agency, which refers to a capacity of socially embedded actors to interpret, adapt to, reproduce and, potentially, resist and transform social structures through their actions, in accordance with their beliefs, reflexivity, understanding, or will (Crossley, Reference Crossley and Abrutyn2016; Emirbayer & Mische, Reference Emirbayer and Mische1998). Actors’ motivations and intentionality are a critical part of human agency based on their beliefs that certain goals are worth pursuing (Bandura, Reference Bandura2001). Social theorists and researchers on both sides of the debate increasingly acknowledge the interdependencies between agency and structures, suggesting that they presuppose each other in ways that cannot be easily disentangled (Archer, Reference Archer1982; Elder-Vass, Reference Elder-Vass2010; Giddens, Reference Giddens1984; Tan, Reference Tan2011). In a recent conceptual essay, Crossley (Reference Crossley2022, p. 167) proposed a relational model of social structure in which agency is central and argued that social structure “is an always evolving network of interaction, interdependence and relations between reflexive social actors who are formed […] within those relations and interactions.” For Crossley (Reference Crossley2022), the network structure is therefore not an omnipotent and impervious system that determines individuals’ attitudes and behaviors. The opportunities and constraints generated by the structure, such as resources or cultural norms, are negotiated by the actors, who reflexively interpret the content and meaning of the relationships in their actions. This agency thus mediates the effects which this structure has upon network members and potentially transform this structure. In turn, critical and reflexive actors are formed within social relations and networks, making network structure and human agency irreducible and interdependent constituents of social processes. Structure and agency are commonly seen as the two sides of the same coin in reference to Giddens’s (Reference Giddens1984) structuration theory, although critics have also pointed to the need for separating the two for analytical purposes in empirical investigations (Archer, Reference Archer2000; Parker, Reference Parker2000). The debate no longer centers on whether agency and structure interact, but rather on how they do so, and which structures and forms of agency combine to shape human behavior (Tan, Reference Tan2011). Empirical studies aim to specify these dynamics across different domains.
The interactions between agency and structures underly, whether implicitly or explicitly, the designs of many social network studies, most notably those focusing on agency in network formation that assume actors’ intentionality and reflexive deliberations (Mische, Reference Mische, Scott and Carrington2011; Tan, Reference Tan2011; Tasselli & Kilduff, Reference Tasselli and Kilduff2021). MMSNA aims to capture structure–agency dualism in the structures of social relationships, either through an outsider view of the network, and the processes which generate these structures, or insider perspectives of the actors within the network (Edwards, Reference Edwards2010a). The former can be, for instance, by locating mechanisms of network formation within a particular context (e.g., historical or geographical context). For example, our study focused on how teachers’ exercises relational agency to mobilize support for migrant pupils within the institutional settings historically designed for more homogenous school populations. The latter can be about qualitatively examining cultural elements in the ‘making’ of the relationships by actors (e.g., rituals, meanings, identities, and understandings) that cannot be adequately captured through numbers and standardization required by a quantitative approach in SNA. For example, our study of ties around migrant support assessed the nature of ties in terms of their alignment with inclusive pedagogical approaches. Social network data and methods cannot easily be located on the continuum of qualitative to quantitative research methods, nor is it simply a combination of the two, since the networks represent structures, rather than quantities, although structural properties of networks such as their density or centralization can be quantified (Froehlich, Reference Froehlich, Van Waes and Schäfer2020; Hollstein & Straus, Reference Hollstein and Straus2006). Some relational methods (and the data that they use) may be described as more qualitative-oriented methods, and others as more quantitatively oriented ones (Froehlich, Reference Froehlich, Van Waes and Schäfer2020). Both kinds of designs are employed in the network studies that use structure–agency as a theoretical lens (Pantić et al., Reference Pantić, Brouwer, Thomas and Froehlich2023).
However, despite arguments that agency and network structure cannot be fully understood separately from each other, social network analysis has often been criticized for failing to show the dynamic interplay between the two, and has been characterized as rooted in a structural determinist perspective (Borgatti et al., Reference Borgatti, Everett and Johnson2013; Kilduff & Brass, Reference Kilduff and Brass2010). Research designs that give weight to agency and structure simultaneously are still scant in most research areas (Tasselli et al., Reference Tasselli, Kilduff and Menges2015; Tasselli & Kilduff, Reference Tasselli and Kilduff2021). In educational research, for example, it is widely recognized that both social structures and agency of actors, such as teachers, can powerfully determine educational processes and outcomes. Yet, most network studies have focused on a structural analysis of school-based relationships and networks and failed to give adequate attention to the ways in which actors experience, interpret, and adjust to those relationships and networks (Pantić et al., Reference Pantić, Brouwer, Thomas and Froehlich2023). In this context, our embedded MMSNA approach has been designed to capture agency within teachers’ support networks.
1.2. MMSNA and structure–agency interactions
This section considers the value of MMSNA for analyzing structure–agency interactions, situating the discussion within broader debates in the network literature and illustrating its relevance through examples from educational research on teacher agency. Network researchers have aimed to comprehend the patterns of social relationships and how these patterns affect human behaviors by focusing on the actual, meaningful relationships between actors rather than on relationships based on institutional categories such as belonging to the same group or organization. In the late 20th century social science, the notion of social networks was used precisely as an alternative to predefined categories and social roles, such as social class or professional communities. In education, for example, this approach contributed to go beyond the conventional assumption that students or teachers form distinct and homogeneous groups. In network studies of that time, attention predominantly focused on formalist explanations and the structure formed by the links between the various elements of a network (Burt, Reference Burt1982; Wasserman & Faust, Reference Wasserman and Faust1994). Particularly in quantitative SNA, this contributed to abstracting network structures from the particular contexts from which they developed and the unique history of the relationships and networks developed by actors who gave them meaning.
More recent debates in the network literature have emphasized the importance of incorporating cultural elements (Crossley, Reference Crossley2010; Emirbayer & Goodwin, Reference Emirbayer and Goodwin1994; McFarland et al., Reference McFarland, Moody, Diehl, Smith and Thomas2014). Social relationships and networks have meaning and are constructed and interpreted by actors through processes of meaning making in specific cultural and institutional contexts. Informal interactions also proved to be strongly shaped by formal membership to constituted groups and institutions, so that social divisions revealed by SNA often reflect ‘standard’ role structures (Eve, Reference Eve2002). More fundamentally, social networks co-evolve with the system of social institutions in which networks are embedded and that have their own histories (Kadushin & Kotler-Berkowitz, Reference Kadushin and Kotler-Berkowitz2006). Social networks are not only about interconnectedness. They also constitute a context of shared meanings, identities, symbols, practices, etc. that must be situated in their institutional and cultural context (Crossley, Reference Crossley2010).
Scholarship on school reforms is a good example of this evolution in educational research (Coppe, Reference Coppe2024). Until the early 2000s, research on school reform mainly considered teacher networks as a given structure that facilitates or threatens the implementation of external change initiatives (e.g., curriculum reforms, see: Coburn, Reference Coburn2001). Within this structuralist perspective, studies have shown how the structure of teacher networks makes teachers more or less committed to adopting new instructional practices (e.g., Coburn & Stein, Reference Coburn and Stein2006). The approach to teacher networks in schools has evolved to recognize that the network structure is particular and contextual to each school and is developed and dynamically reshaped by teachers themselves according to their needs, beliefs and will—as they exercise their agency (Woodland et al., Reference Woodland, Douglas and Matuszczak2021).
By mixing methods, in particular by adopting qualitative and quantitative approaches in social network analysis, MMSNA offers valuable analytical tools for understanding how the institutional and network structures are intertwined with each other, and how agency is constituted in social relationships and networks, and act back upon and shape those relationships. Qualitative analysis enables an in-depth examination of the unique history of specific networks and mechanisms of network formation and transformation, including the complex and nuanced actors’ sense-making processes, identities, dispositions, and critical responses to relationships within networks (Crossley, Reference Crossley2010).
Using this approach, the day-to-day interactions can be seen as building blocks through which actors build more stable relational structures over time, which in turn enable or constrain their individual agency (Wubbels et al., Reference Moolenaar, Daly, Sleegers, Wubbels, den Brok, van Tartwijk and Levy2012). Through abstraction and standardization, a quantitative approach can be used to systematically map where ties exist and where they do not. But qualitative analysis is necessary to capture all the complex and nuanced reasons why actors choose to develop social relationships, how they navigate those structural opportunities and constraints in orienting their lives and actions or the role played by emotions, given that social relations are infused with affect (Tasselli et al., Reference Tasselli, Kilduff and Menges2015). In education, teacher agency has mainly been analyzed in the context of educational reforms (Coburn, Reference Coburn2001, Reference Coburn2004). These MMSNA studies have shown how teachers bring to bear their experience, motivation, cognition, and sense-making in implementing reforms (e.g., Coburn, Reference Coburn2001; Coburn et al., Reference Coburn, Russell, Kaufman and Stein2012) and how their intentionality is influenced by their network position (Moolenaar et al., Reference Moolenaar, Daly, Cornelissen, Liou, Caillier, Riordan and Cohen2014). Less common are studies of bottom-up agentic responses such as teachers’ mobilizing support for the increasing presence of migrant students in schools.
Building a deep understanding of the content and stories of network ties in addition to structural properties (Baker-Doyle, Reference Baker-Doyle2015) is critical for key issues in social sciences that involve social change. Qualitatively, network studies can examine the strategies individuals and group employ to navigate structural constraints (e.g., resisting norms) and use resources accessed through relationships and networks to challenge, reinterpret or transform existing network structures. In teacher-related research, teacher agency has been understood as a capacity to critically respond to problematic situations that is shaped by actors’ underlying beliefs about their professional roles and embedded in multi-layered social contexts (Biesta et al., Reference Biesta, Priestley and Robinson2015; Biesta & Tedder, Reference Biesta and Tedder2007; Pantić, Reference Pantić2015; Villegas & Lucas, Reference Villegas and Lucas2002). Research has used the concept of teacher agency to highlight the role of teachers as proactive actors in responding to reform but also when facing complex situations (Lukacs, Reference Lukacs2015). Examples of such situations include responses to the recent COVID-19 pandemic (Ehren et al., Reference Ehren, Madrid, Romiti, Armstrong, Fisher and McWhorter2021), the urge for climate change education (Andrzejewski, Reference Andrzejewski and Winograd2016), or the education of refugees (Rose, Reference Rose2019), where they displayed a particular form of relational agency in working flexibly with other actors to meet diverse student needs. MMSNA has also been used in educational studies that examine sociocultural contexts that constitute agency and shape what actors see as possible within given environments (Baker-Doyle, Reference Baker-Doyle2015; Eteläpelto et al., Reference Eteläpelto, Vähäsantanen, Hökkä and Paloniemi2013; Froehlich et al., Reference Froehlich, Rehm and Rienties2019; Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024). These studies have shown how social networks are both shaped by and shape agency, as both a consequence and antecedent of networks (Moolenaar et al., Reference Moolenaar, Daly, Sleegers, Wubbels, den Brok, van Tartwijk and Levy2012). For example, a given teacher network structure affects teachers opportunities for professional development, but teachers purposefully (re)design the network for specific purposes (Coppe et al., Reference Coppe, März and Raemdonck2023; Thomas et al., Reference Thomas, Tuytens, Devos, Kelchtermans and Vanderlinde2019), for example, as they exchange informal knowledge.
2. Designing an MMSNA study: Lessons from the TEAMS project
This section outlines the steps and tools involved in the embedded MMSNA procedure, using an illustrative case to describe how we designed an MMSNA study that enabled us to analyze structure–agency interaction in an educational project: Teaching that Matters for Migrant Students (TEAMS project for short). TEAMS examined the dynamic interplay between teacher agency and schools’ social structures across seven schools in Scotland, Finland, and Sweden, asking 1) how teachers use the support structures available to them to support the integration of migrant students in schools, and 2) how they exercise relational agency—that is, purposeful interactions with others (Edwards, Reference Edwards2010a; Pantić & Florian, Reference Pantić and Florian2015)—to build such social networks in the first place, reflecting the two sides of the structure–agency coin.
This section walks the reader through the stepwise procedure of designing and conducting an MMSNA study, using the TEAMS project to illustrate key aspects of the design process and the rationale behind specific choice of methods and analytical procedures. We first describe how the project objectives informed the overall approach, followed by specific data collection methods and related analytical procedures. Then, we present an integrated analysis as the final and critical step in embedded MMSNA.
2.1. From the research objective to the choice of an MMSNA
As in all research, the choices of methods and analytic procedures of embedded MMSNA are informed by the theoretical foundation and research purposes that vary across disciplines. Our MMSNA design was guided by the core objective of the TEAMS project to simultaneously capture the structure of teachers’ migrant support networks, what flows through the interactions in which they are built, and why teachers engage in these interactions (Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024).
We drew on the two fundamental assumptions rooted in works in educational sciences discussing how educational institutions and actors interact (e.g., Coburn, Reference Coburn2005; Coburn et al., Reference Coburn2016). On the one hand, in the school environment, teachers’ relationships with students and their families are subject to social and cultural norms; support for migrant students in schools is responsive to the wider immigration policies and discourses, and collaboration among educators vary by school- and country-specific legislation, for example, when these collaborations are formally encouraged through school meetings and training. These different aspects of agency–structure interaction are captured with the different methods. Social network survey collected data on the structural features of school networks that display hierarchies (with a central core and more peripheral actors). Power relations are also exogenous to networks, particularly in relation to how the authority is distributed in the school organization (bottom-up or top-down school management, for example), which reflected in both policy and fieldwork data. On the other hand, since individual teachers do not possess all the necessary expertise regarding migrant students, they and other school actors tend to develop social relationships with colleagues based on the specific resources or advice they seek, thereby shaping their support network. They may choose to connect with colleagues depending on their willingness to proactively share information or knowledge, or to engage in shaping collegial dynamics that support the integration of migrant students. Conversely, some may choose to minimize interaction when they do not feel personally concerned by the need to support these students. These choices based on how actors perceived the need for interaction were captured with the log data embedded in the network data as described below.
These assumptions interact in shaping the school social fabric and underscore the importance of considering both structure and agency in pursuing the study objectives. Naturally, what contextual factors matter—and how—varies by discipline and research question, shaping the design of an appropriate MMSNA. Because teachers’ work is regulated at the state (or regional) level but enacted daily within schools and classrooms, studying their agency requires attention to these multiple contextual layers (Ball, Reference Ball1993). When studying other actors situated in different institutional settings, these layers should first be made explicit before deciding which kinds of data can best illuminate the structure–agency dynamic. For instance, policy documents may provide insights into formal structures and regulations; social network data can reveal how resources and norms circulate; and interviews, observations, or qualitative logs may capture actors’ interpretations, strategies, and in situ actions. The relevance and integration of these data sources depend on the specific phenomenon and context under investigation, as we illustrate in the following section through our case study.
2.2. Deciding on the methods, data collection tools and related analysis
These contextual aspects and the objectives of the research inform the selection of specific tools and analysis relevant for addressing them. TEAMS employed four main data sources to examine the research questions with data from seven schools and three countries (three in Scotland, two in Finland, and two in Sweden) with data collection tools and strategies as follows.
Longitudinal social network survey:
Socio-centric approaches are commonly used to collect information on the social fabric of the social network (Jack, Reference Jack2010), in our case school networks. This whole-network approach, allowed us to map, visualize, and study the interactions among all the actors included in the network (Borgatti et al., Reference Borgatti, Everett and Johnson2013)—here, school staff members.
In each school, all staff members were invited to fill out an online social network questionnaire to collect information on their relationships within the school at three time point. These three waves were necessary to model the dynamics of network change over time using SAOMs (see Section 2.3). In each school, the online survey displayed a roster or list of staff who worked in the school at the time and asked staff members to nominate colleagues with whom they had interacted around migrant support in recent months. Staff could also nominate people outside the school that they interacted with. The name generator question specifically asked if they “turned to this person for support in matters concerning students from migrant backgrounds.” The responses were used to construct the whole-school networks, where a tie existed between staff member i and staff member j in migrant support networks if they interacted in the described ways. We retained information on the directionality of ties when constructing the whole-school networks (e.g., i → j) as our name generator question referred to support requested from someone. This approach aligns with common methodological practices in measuring support networks (e.g., Coppe et al., Reference Coppe, Sarazin, März, Dupriez and Raemdonck2022; Thomas et al., Reference Thomas, Tuytens, Devos, Kelchtermans and Vanderlinde2019). The survey also asked for demographic information (gender, migrant background) and staff roles (professional role, seniority). The response rates in the three time points (n = 359 T1; 416 T2; 341 T3) ranged from 51,5% to 80.7 % in Scotland, 39,3 % to 79% in Sweden, and 42,3 % to 77,5 % in Finland (See Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024 for details on the sample in each wave and school).
The social network survey was analyzed in two ways. First, we created visualizations of the general collaboration and migrant student supports networks using Gephi (Bastian et al., Reference Bastian, Heymann and Jacomy2009) and the ForceAtlas 2 layout algorithm (Jacomy et al., Reference Jacomy, Venturini, Heymann and Bastian2014) for each of the three waves and the seven schools. Then we estimated Stochastic actor-oriented models (SAOM) using the RSiena package in R (Ripley et al., Reference Ripley, Snijders, Boda, Andras and Preciado2023; Snijders et al., Reference Snijders, Van de Bunt and Steglich2010) to analyze the mechanisms of tie formation and network evolution across the three waves for each school. This allowed the researchers to examine the extent to which school staff attributes (such as their role) were associated with their interactions with colleagues when supporting migrant students, and how networks shaped these interactions. More information about the SAOM routine is reported in the article that presented study findings (Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024), including the predictors included in the model. Missing attribute data were retained as a separate category (“missing”) to preserve cases rather than exclude them. Teachers who did not participate in a given survey wave could still be nominated by others, although by definition they could not nominate teachers themselves. No data imputation was performed, as the extent of missingness was limited and imputation could introduce additional bias into the SAOMs.
In general, across research contexts, network visualizations are valuable tools for exploring the structural logic of entire networks, for example, identifying clusters or centralization (Jacomy et al., Reference Jacomy, Venturini, Heymann and Bastian2014). SAOMs, in turn, offer a longitudinal perspective on network evolution, enabling the analysis of the mechanisms driving structural change, such as selection processes based on actors’ attributes (Snijders et al., Reference Snijders, Van de Bunt and Steglich2010).
Online logs:
Second, researchers typically employ qualitative data to take account of the contexts and circumstances of interactions (Jack, Reference Jack2010). In the embedded MMSNA approach, we used an online log for Teacher Reflection on their Agency for Change (TRAC) to collect data on relational agency around migrant student support among school staff members (Pantić, Reference Pantić2021). TRAC asked staff to describe in detail a “time (over the past 6 months) when they reached out to someone to support or help a migrant student” in three sections that reflect aspects of the relational agency as follows: 1-WHAT was the purpose of an interaction the problem or situation that actors sought to address; 2-WHO they reached out to seek support including the role of the actors they interacted with; 3-WHY they reached out to them and how they were supported, including a reflection on the outcome. In each school all staff were invited to fill out the log at least once in each time point of data collection. The logs enabled us to examine the contents and contexts of the interactions on a subsample of responses within the network (See Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024 for rates across waves and schools).
The log data (a total subsample of 144 ranged from 30 to 40 % of SNA responses across waves) were coded deductively for the aspects of relational agency conceptualized as flexible ways of inter-professional working focused on solving problems together (Edwards, Reference Edwards2009)—that underlie the log sections: professional role/beliefs; nature of the interactions, and the perceived barriers/enablers for supporting migrant students. The data were coded using a scheme that noted instances of staff working proactively and flexibly with others to support migrant students coded as ‘Agents of Change’ (AoC), such as when a teacher approached “many different persons to get help with the student’s issues,” and those coded as ‘Role-implementers’ (RI) when staff described instances of implementing existing policies and procedures, e.g. “Home were notified but students behavior remained the same and declined.” We were particularly interested in actions that were coded as ‘Agents of change’ that reflected the principles of inclusive pedagogy—seeking support to remove barriers for learning, such as linguistic barriers (Pantić & Florian, Reference Pantić and Florian2015). Importantly, acting as agents of change is attributed to particular situations rather that to particular actors, because the same teacher might act as an agent of change in some situations and contexts and not in others in line with the situational nature of relational agency. Embedding these coded responses in the SNA networks enabled both visualizations and statistical analysis (see Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024).
Qualitative interviews:
Third, interviews are common ways of collecting qualitative data, both about the actors’ motives and contextual data from the sites. In our study, qualitative interviews (n = 141) were conducted with a subsample of staff in each school, who indicated their wish to be interviewed when filling out the log and survey. Staff interviews covered collaboration practices and migrant students support arrangements in each setting. The participants in each site included diverse range of roles, gender, ages, years of experience and, in some sites, migrant backgrounds, as detailed in Pantić et al. (Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024).
The interview transcriptions were coded deductively and analyzed for the aspects of relational agency, using the same theoretical scheme for the aspects of teacher agency, as well as those that characterize the nature of practices in terms of their alignment to the principles of inclusive pedagogy (Pantić & Florian, Reference Pantić and Florian2015) focusing on the practices that aim to include migrant students and recognize diversity as a resource for the school.
Policy documents:
Fourth, policy documents are used as sources of information about the national and international contexts that can help interpret the differences and similarities across different network structures. Because school education is regulated at the state (or regional) level, teachers’ work must be understood within the boundaries set by these regulations and the autonomy they exercise in implementing and/or negotiating them. Our policy analysis in each country focused on the support systems for migrant students, and teacher collaboration, for example, with services for language support. In total, 19 documents were collected based on the following criteria: (1) core legal and educational documents; (2) documents that provide guidance to schools; (3) documents presenting comparable types of policy intentions across the three countries; and (4) documents that were in force at the time of the project.
Policy documents were used to better understand the institutional arrangements that could be seen as enablers or barriers, which helped interpret particular interactions as implementing or adapting existing procedures. More details about the selection and analysis of the documents are presented in Tarnanen et al. (Reference Tarnanen, Oral, Niklasson, de Riba-Mayoral, Vähäsantanen, Pantić and Manninen2024).
2.3. Embedded MMSNA
The analysis integrated insights from multiple data sources through an iterative process within and across layers. We explain how the embedded MMSNA combined evidence from the social network survey, logs, interviews, and policy documents at different stages of the analysis. The integration process and corresponding steps are illustrated in Figure 1.
Figure 1. Data material and integration. Key: AoC: agent of change; RI: role implementation.
First, we combined the log data with the social network data. In each school, for the wave with the highest response rate, we mapped the interactions reported and analyzed with the logs (i.e., agent of change and role implementation) and quotes from the log onto the social network visualization to understand the content of some of the interactions, and the specific reasons why teachers were interacting. Then, in addition to this “mapping,” the descriptive network statistics were combined with the log analysis. The descriptive network statistics were compared for staff members whose outgoing interactions were coded as Agents of Change; those that were not coded as Agents of Change; and those who responded to the survey but not to the log, according to the number and diversity of staff members they declared turning to for migrant student support; and the number of actors outside of their school they reached out to (see Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024 for more details).
The content and contexts of interactions captured with the log and staff social networks were illustrated using the interview data to gain a deeper understanding of the organizational norms in each site, as well as wider institutional contexts, together with the information obtained from the policy documents about migrant integration and student support systems more generally. This integration constitutes a key step in understanding how the multiple layers interact to shape teachers’ practices and facilitate or constrain their agency, while acknowledging that contextual specificities exist, not only across countries, but also between schools.
Finally, results from the SAOMs were integrated with the insights from the previous steps (i.e., combination of logs, descriptive network statistics, interviews, and policy documents) to examine the interactions between social networks and relational agency. The content that actors reported in the log interactions was inspected in relation to their network position and whom they reached out to in the migrant support networks. Staff roles were examined for whether they had formal responsibility in supporting migrant students; their degree in the migrant support network; and tendency to interact with the same colleagues in this network—as to understand more precisely patterns visible in the SAOMs results. The researchers engaged in iterative analysis of quantitative and qualitative data for corroborating or not the evidence of network features in the qualitative data. The preliminary findings and analysis were extensively discussed in regular research team meetings to ensure the analytic procedures were aligned to generate comparable results across the three countries (see Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024 for more details).
Combining the analysis of social network and log data allowed the researchers to examine the interactions that indicated the purposes of relational agency against the backdrop of the social networks through which teachers accessed resources and expert advice for supporting migrant students and to compare the similarities and differences across the schools. Interview data provided complementary contextual information that were used to interpret the organizational structures and underlying norms around migrant integration. The full results and visualizations can be seen in Pantić et al. (Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024).
In the discussion section we will illustrate the insights that emerged from our analysis and discuss how the embedded MMSNA enabled us to generate unique insights of significance in discussion of agency–structure interaction.
3. Discussion
This paper illustrated an MMSNA approach that aimed at advancing our understanding of the interaction between structures and agency that is embedded in social networks.
Social networks are more than a set of nodes and links between them (Crossley, Reference Crossley2010; Sweet, Reference Sweet2016). Individual attributes involved beliefs, attitudes, motives, and other aspects of agency that inform individual actions and interactions with others. A quantitative approach and statistical analysis modeling social selection can only go so far in illuminating aspects of agency. For example, they do not tell us much about the particular purposes or contexts of interactions that generate relationships and networks. Embedding qualitative data in the network data facilitates a fuller understanding of how individuals engage with the social environments that shape their beliefs or behaviors in particular situations. In our case study, analyzing teachers’ relational agency within the school collaboration networks across different pedagogical, school and national contexts unraveled the intricacies of actors’ agency and institutional contexts of network structures.
In this final section we use some of the insights our study generated to demonstrate the potential but also challenges of the embedded MMSNA approach and discuss how it enabled us to generate insights into actors’ agency that was influenced by the structural features of their networks on the one hand, and how network structures were elaborated through the micro-foundations of teachers’ agency in turn.
3.1. Patterns in network structures
The Stochastic Actor-oriented models (SAOMs) enabled us to examine the patterns of network structures and the effect of individual attributes on tie formation and dissolution (Edwards, Reference Edwards2010b, Jack, Reference Jack2010; Sweet, Reference Sweet2016). This analysis helped examine changes over time in network structures, which were relatively small in our case study. SAOM results showed that teachers tended to reach out to colleagues in designated roles or with specific expertise for supporting migrant students, as was also found through the analysis of log data and interviews.
The log interactions coded as ‘agent of change’ (hereafter AoC) tended to document reaching out to a wider variety of actors, irrespective of their formal roles, such as colleagues who share a student’s language or in other ways proactively seeking relevant knowledge and support. SAOMs also showed that staff members having themselves a migrant background tended to be reached out by their colleagues in Scotland and Sweden, while each school in the Finnish sample had only 2 staff members with migrant background. Thus SAOM enabled us to examine the pattern of interactions, while the embedded log data enabled us to examine the qualitative nature of those interactions.
The logs coded as AoC reported interactions that had greater outdegrees, suggesting staff tended to reach out to more colleagues in the migrant student support network. In four schools, the logs coded as AoC also showed a slightly higher number of situations in which staff reached out to the actors outside school (See Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024, for the descriptive network statistics of log ties according to how logs were coded).
SAOMs focus on general patterns while qualitative data was used to better understand and investigate specifically how and why individual actors are forming or breaking social ties within networks (Baker-Doyle, Reference Baker-Doyle2015). The complementary qualitative methods collected contextual data that were analyzed with thematic approaches (McCrudden & Marchand, Reference McCrudden and Marchand2020). Coding the interactions as AoC enabled us to illustrate how teachers exercised relational agency to support migrant students relevant to their position in the network. In our study, the examples included situations of seeking to enable access and overcome language barriers for students, or to understand the cultural background, dealing with issues around different expectations and norms, home situations and sometimes behavior or attendance issues, or helping the newly arrived students to navigate an unfamiliar system. Thus, the embedded qualitative data enabled us to examine the nature of support relationships within the networks relative to the wider institutional norms and contexts.
3.2. Agency embedded in the networks
The embedded MMSNA procedures made it possible to inspect the nature of actors’ relational agency as an interplay between subjects and their social network environment from which action is generated (Tasselli & Kilduff, Reference Tasselli and Kilduff2021). Combining SNA modeling with qualitative approaches based on log and interview data enabled us to examine the meaning of ties, the actors’ motivations and rationale for actions, and the processes which generate network structure, as well as the contexts (e.g., policy, cultural, institutional) in which networks are embedded (Crossley, Reference Crossley2010; Crossley et al., Reference Crossley2015; Hollstein, Reference Hollstein, Scott and Carrington2011). By combining log data with the social networks, we found that interactions coded as AoC were more likely to be situated within groups where migrant support collaboration was more common in the social network. In our example, such interactions were often directed toward an English as Additional Language (hereafter EAL) teacher and the Social Studies department middle leaders, among whom information and advice sharing was prevalent with key positions in the migrant support network. These logs indicated that a habit of collaboration and information sharing (e.g., within a subject department or a group of teachers) was instrumental in supporting migrant students. They pointed to the importance of availability of colleagues for help that would not have been asked for if it was not usual for staff to reach out to colleagues in this way. The embedded MMSNA enabled us to simultaneously examine the contents of these logs, which indicated that habits of information sharing and collaboration facilitated interactions coded as AoC. In contrast, logs that were coded as role implementation tended to occur in departments where there was a relative dearth of migrant student support ties among staff members. These logs indicated that support mainly involved a superficial use of existing procedures (such as filling-in a well-being concern form) but without follow-up actions. The embedded MMSNA enabled us to inspect the contents of these interactions relative to the actors’ position in the network.
Complementary fieldwork and policy data helped understand that collaboration was a common feature of schools in Sweden and Finland compared to schools in Scotland. For example, teachers in the Swedish schools had regular meetings for discussing student needs as part of the students support routines that enabled discussions of migrant student needs among others. Qualitative data also helped us understand how support differed in relation to the different migrant pupils’ needs, even within the similar institutional settings. Importantly, presence or absence of a relationship did not qualify the interaction in terms of the nature of support for migrant students. Complementary qualitative data was used to exemplify the interactions between teachers and specialist support that were aligned to the principles of inclusive pedagogy, for example, when teachers used such support to enrich their own knowledge rather than delegate responsibility for migrant students to designated specialists.
MMSNA offered insights that combine the actors’ perception of how network dynamics work within their structural environments that may not be perceived by the actors involved in networks, such as the different levels of centrality of the EAL specialist. Comparison of AoC-coded interactions across sites showed that these interactions shared an essential quality of collaboration for knowledge sharing about students and creating a safe atmosphere, while reflecting the different roles of their institutional settings. Comparisons of the schools’ social networks uncovered how the same support roles were used differently. For example, the same EAL specialist who supports the three Scottish schools in the sample, was sought after differently across the migrant support networks in the three schools. In one school’s migrant student support network, she was by far the most central actor, in the second school, she was relatively central, but not among the most central actors in the school, while in the third school she was one among several very central actors (see Pantić et al., Reference Pantić, Sarazin, Coppe, Oral, Manninen, Silvennoinen, Lund, Päivi, Vähäsantanen and Li2024). These structural positions reflected in her interview data, which enabled us to triangulate the findings from the network and actor’s perspective.
Embedding log data in the social networks enabled us to understand agency that reflected in the actors’ attitudes and beliefs as they play out in particular situations, which is different from the common ways of studying personal or demographic attributes in networks, and critically missing for understanding how agency and structures co-evolve in the micro-foundations of organizational networks (Tasselli et al., Reference Tasselli, Kilduff and Menges2015). In particular, given the definition of relational agency as flexible problem-solving behavior, the study assessed to what extent teachers’ interactions ‘followed’ or worked around the formal roles and procedures. Here, qualitative data collected about both structures and agency, for example, in fieldwork and policy analysis was critical to understand how actors’ embeddedness in their organizational culture and norms influenced their relational practices, while statistical models with network data were critical for identifying and making sense of the blueprint of the networks (Crossley, Reference Crossley2010).
3.3. Uncovering mutually formative influences of agency and structure
The presented MMSNA has enabled analysis of network agency as a unit of analysis that involves ‘an embedded process of social engagement’ which is relevant in many disciplines such as social psychology, sociology or anthropology (see Tasselli & Kilduff, Reference Tasselli and Kilduff2021 for a review). Our study demonstrates the potential of network agency to explain how similar structures (e.g., EAL support) can be used differently in key aspects of teachers’ and school practices.
The embedded MMSNA has helped illuminate the role of agency in social networks within the structures of educational institutions (Baker-Doyle, Reference Baker-Doyle2015; Spillane et al., Reference Spillane, Healey, Kim and Daly2010, Reference Spillane, Hopkins and Sweet2018). Critically, the presented MMSNA approach, enabled examining the interactions of structure and agency simultaneously and in concert with each other to uncover which structures and interactions (regular collaboration) matter for particular purposes (coded as ‘agentic’) and how they matter across different contexts. In our study the relationships that matter for supporting migrant students were the ones aligned to the principles of inclusive pedagogy, which the researchers were able to explore by theoretical coding of the qualitative data, for example, the content of the logs that described specific interactions. The embedded MMSNA approach has helped us understand the nature of collaboration of interest, in this case whether teachers used specialists as a resource for themselves to better understand and support all students, or to delegate responsibility for some (Pantić & Florian, Reference Pantić and Florian2015). In this sense the nature of exchanges in the relationships were as relevant as their absence or presence.
The practices of interest will differ in different disciplines, but the potential of embedded MMSNA for understanding the formative power of agency relevant to the wider systemic structures will be applicable in diverse areas of research that involve actors exercising agency within social networks to change and/or sustain practices in their domain. For example, in the field of network agency in organizational research (Tasselli & Kilduff, Reference Tasselli and Kilduff2021), the embedded MMSNA approach can be used to understand how the norms and cultures that characterize institutions and organizations are both reproduced and changed as a result of situated, dynamic, intentional interactions.
This approach is relevant for addressing research questions that strive for understanding of mutually formative influence of human agency and social structures in concert with each other. The embedded MMSNA can be used for explaining how social networks and human agency interact in change processes, for example, to investigate their impact on outcomes, or the impact of particular policies mediated by the actors’ sense-making processes (Coburn, Reference Coburn2016).
While we have discussed the added value of using an MMSNA approach, it should be noted that it also has its limitations in that combining the different methods is time- and resource-intensive, and that it is difficult in practice to collect all of the data in each time point. In future studies, longitudinal designs that aim for simultaneous collection of quantitative and qualitative data at different points in time could integrate elements of qualitative tools such as logs into the network survey. This could enable researchers to systematically examine the changes in both social networks and agentic interactions over time, for example, to understand how relationships are reconfigured and reproduced through the moment-to-moment interactions as mechanisms for spreading innovation or, conversely, solidifying norms.
Competing interests
None.
Funding statement
This work was supported by the NordForsk (grant number 94935).
The data that support the findings of this study are openly available in Edinburgh Research Explorer https://www.research.ed.ac.uk/en/datasets/teaching-that-matters-for-migrant-students-teams-understanding-le
Open access
