Abstract
A quantitative approach to social network analysis involves the application of mathematical and statistical techniques and graphical presentation of results. Nonetheless—as with all sciences—subjectivity is an integral aspect of network analysis, manifested in the selection of measures to describe connection patterns and actors’ positions (e.g., choosing a centrality indicator), in the visualization of social structure in graphs, and in translating numbers into words (telling the story). Here, we use network research as an example to illustrate how quantitative and qualitative approaches, techniques, and data are mixed along a continuum of fusion between quantitative and qualitative realms.
Mixed methods research, which is described as combining qualitative and quantitative approaches to gain a richer and deeper understanding of a problem, has become increasingly popular (Creswell & Plano Clark, 2007). In the most typical manifestation of a mixed methods approach, qualitative and quantitative data can be used together to explain, explore, and/or triangulate findings, while individual quantitative or qualitative components remain fully distinguishable.
Some researchers, however, propose a more intimate communication between quantitative and qualitative approaches. Bazeley (2012), for example, proposes a continuum of integration, with one end representing the partial mixing of separated quantitative and qualitative components, and other end representing an “inherently mixed” study, in which the same data source provides both quantitative and qualitative information. Some researchers take a more radical approach and suggest that the quantitative–qualitative dichotomy is arbitrary, artificial, and inaccurate (Bernard, 2013; Gorard, 2010; Onwuegbuzie, 2002). Gorard (2010) criticized the current norm of defining quantitative research as involving numbers, and qualitative as involving everything else (e.g., text and discourse, images, observations, recordings, etc.) (p. 238). He proposed that the q-word dichotomy (quantitative vs. qualitative) does not specify the underlying logic of a study (i.e., paradigms) nor the technique of data collection (numbers or words), and therefore, is not necessary or helpful (p. 249). Gobo (2016) proposed “merged” methods to represent designs that entail a higher integration than “mixed methods” which still assume an inherent quantitative–qualitative separation (that he identified as a genetic bug in science!). In this line of thinking, meanings and narrative interpretations are closely knitted within the numerical presentation of quantitative results, and the traditional inductive–deductive distinction does not distinguish qualitative and quantitative research.
The real-world presentations of this greater intimacy between quantitative and qualitative research is a matter of debate, and some mixed methods scholars consider this to be an oversimplification, as despite blurry boundaries, quantitative and qualitative research have distinctive tendencies which should not be ignored (Morgan, 2018; Sale, Lohfeld, & Brazil, 2002). Others consider exploring these blurry boundaries as potential sites for innovation and development. But what is common between “inherently mixed” (Bazeley, 2012), “fully mixed” (Leech & Onwuegbuzie, 2009), and “merged methods” (Gobo, 2016) studies is that the quantitative and qualitative boundaries at data collection, analysis, and interpretation are blurred.
Bazeley (2012) has described mixed methods analysis as a “messy and still largely experimental” field with room for innovation (p. 825). Responding to Gobo’s (2015) call for “merged methods,”Bazeley (2016) provided examples of existing approaches within the realm of mixed methods, that could meet this definition of merged or hybrid research. Among Bazeley’s (2016) examples, she included social network analysis (SNA) as an example of an inherently mixed or fused mixed methods analysis. She later classified SNA as a presentation of “hybrid strategies” that “inherently combine both qualitative and quantitative elements to create a single source or set of data that is then, typically, further examined using iterative quantitative and qualitative strategies” (Bazeley, 2017, p. 242).
In this article, we extend on the arguments of Bazeley (2016) and others that (sometimes inadvertent) fusion between quantitative and qualitative data and approaches happens and these fusions hold potential for expanding our understanding of the continuum of integration in mixed methods research. Following on the metaphorical classification by Bazeley and Kemp (2012), we will use the term fusion for instances of blurriness between qualitative and quantitative data and approaches, where the distinction of the boundaries between the two is difficult. We prefer the term fused to hybrid, since the former connotes the fuzziness of boundaries more specifically than the latter, while methodologically describing a similar process.
We will use SNA as an example to demonstrate the continuum of fusion between quantitative and qualitative data, analysis, and interpretation. Social networks are suitable subjects for both quantitative and qualitative studies, since they embody both the structure (that could be observed from an outsider’s viewpoint) as well as content of social relations (which could be understood through the insider’s viewpoints of network actors) (Coviello, 2005; Edwards, 2010). As structures and forms of social relations (social structures) never exist independently from the contents of social processes, the qualitative and quantitative insights to social networks belong together (Crossley, 2010). In addition, the exploratory nature of (quantitative) SNA resembles many features that are generally ascribed to qualitative data analysis. Nevertheless, as will be briefly explained later, we believe that these features are not unique qualities of social networks, and fusions can potentially happen in studying other complex phenomena and while using other exploratory analytic techniques. We hope this discussion will push forward our understanding of the dimensions and possibilities of integration in mixed methods research.
Fusion as a Continuum
The natural blending of qualitative and quantitative approaches to inquiry is manifested in the works of great scientists as far back as Aristotle (Teddlie & Johnson, 2009) and Galileo (Bernard, Wutich, & Ryan, 2017). The belief that dichotomizing research into mutually exclusive qualitative and quantitative domains is unnecessary is gaining momentum (Johnson & Onwuegbuzie, 2004; Morgan, 2007). The two labels refer to paradigmatically and methodologically overlapping traditions (Yu, 2003), which coexist and complement each other. Some scholars, alternatively, believe in distinguishable identities of the two traditions, while acknowledging the blurry overlap (Morgan, 2018).
Defining quantitative research as a deductive approach using numbers, and qualitative research as an inductive approach using words misses the complexity of the relationship between numbers and words. Quality and quantity are inseparable attributes of any phenomenon (Gorard, 2010). The merge/mix happens in any social research to some extent (Bazeley, 2016). Even though a randomized controlled trial seems considerably different from a phenomenological study, traces of the themes that identify qualitative inquiry (Patton, 1980) are detectable in even seemingly quantitative studies. A quantitative study may try to be naturalistic and contextually sensitive (e.g., see the debate between pragmatic and explanatory clinical trials, Treweek & Zwarenstein, 2009). Simple random sampling is not the most acceptable sampling strategy in many population-based studies, and stratification may happen with the purpose of targeting theoretically important/interesting populations. A quantitative article may involve swinging between numbers/words and induction/deduction when interpreting and contextualizing the findings of the study in light of the theories/other literature in the field. As Onwuegbuzie and Leech (2005a) claim, qualitative and quantitative data are transformable. Meaning, such as that in constructs and assumptions, can lead to numbers. Conversely, numbers can be transformed to meanings when researchers interpret and verbalize quantitative findings. More formally, quantitative and qualitative data can be transformed through “qualitization” and “quantitization” processes (Teddlie & Tashakkori, 2009); such as labeling clusters in multidimensional scaling (Jaworska & Chupetlovska-Anastasova, 2009), and counting the themes in interview transcripts (Hsieh & Shannon, 2005). The repeated exploratory–confirmatory process of modifying and refining the models in structural equation modeling to find one that is “consistent with sound theory” (Mueller, 1997, p. 355) is a typical example of the spiral process of shifting between induction and deduction, and subjectivity and objectivity in a quantitative study. Though qualitative studies are usually inductive in nature, depending on the point at which the theory enters the study, they may apply deductive approaches, or alternate between induction and deduction in an iterative process (Sale & Thielke, 2018; Sandelowski, 1993).
Onwuegbuzie and Teddlie (2003) proposed the use of quantitative and qualitative techniques within the same framework and suggested an exploratory–confirmatory continuum instead of the conventional qualitative–quantitative dichotomy. On the one end of the continuum, an exploratory study aims to look for patterns and meanings, and on the other end, a confirmatory study tests predetermined hypotheses. Both quantitative and qualitative approaches and techniques can be used at various positions of this continuum (Onwuegbuzie & Leech, 2005b). Bazeley and Kemp (2012) used the metaphors of a conversation (a back-and-forth dialogue) and double helix DNA (spiral) to explain the process of generative integration, where the two approaches iteratively contribute to the gradual formation of meaning.
In summary, the spiral conversation and transformation of exploration–confirmation, induction–deduction, quality–quantity is a natural process of scientific enquiry, happens in various ways and to various extents, and forms a pattern that we call the continuum of fusion.
The Fusion Continuum in Network Research
SNA is studying the relations between actors in a network. Traditionally, SNA has been associated with quantitative methods (Freeman, 2004). Many network measures and patterns are informed by graph theory, a branch of mathematics focused on studying the connections among points. Consequently, construction of social networks in the form of graphs is an important aspect of SNA, both as a presentation technique and for the calculation of many network structural indicators. The reliance on mathematical and computational models and graphical imagery are now the building blocks defining modern SNA (Freeman, 2004).
When studying social networks, a researcher may focus on various aspects of the network structure (Marin & Wellman, 2011):
Studying the network structure as a whole (a bird’s eye or sociocentric view), by identification of overall structural features of the network, answering questions such as How densely are the network actors connected?, How evenly are the network characteristics distributed?, and How are the personal attributes distributed in the network?
Identification of prominent actors, answering questions such as Who are the most connected individuals?, Who are the actors who bridge structural holes and mediate communications?, Who are the actors with better access to others?, and How are the central actors connected to the rest of the network?
Composition of clusters, answering questions such as How are actors in a network clustered together?, Are some actors more connected to each other?, Who are the individuals with similar social positions?, and exploring whether personal characteristics can explain actors’ affinity to connect to each other or certain other actors.
The study of social networks is an appropriate field for mixed methods research (Bazeley, 2009; Edwards, 2010). Although quantitative network analysis provides important information regarding the patterns and shapes of social networks and the position of influential people, it may not be in-depth enough to uncover the subtle mechanisms through which social networks shape, emerge, and influence social and organizational processes (Dubini & Aldrich, 1991; Neergaard, Shaw, & Carter, 2005). However, despite the fact that qualitative approaches have been used in the context of network analysis as independent studies, combining qualitative and quantitative network analysis in the form of a mixed methods design is still an emerging field (Dominguez & Hollstein, 2014).
Martínez, Dimitriadis, Rubia, Gómez, and de la Fuentea (2003) proposed using a mixed methods approach to analyze social network data as a more efficient design than pure qualitative analysis, because the quantitative SNA findings are able to highlight critical issues and appropriate directions for subsequent qualitative enquiry. Quantitative and qualitative methods inform each other and consequently can help researchers gain more in-depth insight into social networks by gaining both outsider and insider views of the social structure (Edwards, 2010). Various scholars have examined different ways of mixing qualitative and quantitative approaches to social networks, ranging from fairly separate quantitative and qualitative analyses that complement each other in an overarching mixed methods study, to more intimate dialogue between quantitative and qualitative methods.
In the following sections, we will provide a brief overview of the continuum of fusion between quantitative and qualitative approaches in network research. Although there are various approaches to SNA research, in this article, we focus on the descriptive/exploratory analysis of a social network, in which the researcher tries to make sense of a complex structure using various features of the network (Wasserman & Faust, 1994). We will also provide examples of less intimate mixing between quantitative and qualitative approaches, including partial fusion (qualitative and quantitative studies on the same data source), and minimal fusion (separate quantitative and qualitative strands in a sequential or parallel mixed methods study). Figure 1 provides a schematic conceptual framework of the continuum of fusion in network research. The fusion can happen at the philosophical, methodological, and methods levels (Fetters & Molina-Azorín, 2017); however, in our discussion here we focus primarily on fusion along data collection, data analysis, and interpretation dimensions.
The continuum of fusion in network analysis.
Fusion in Exploratory SNA
Teddlie and Tashakkori (2009), in a chapter introducing mixed methods analysis identified SNA as an example of an “inherently mixed data analysis technique” (p. 273), wherein the numerical data are transformed into qualitative diagrams and narrative explanations of underlying processes and phenomena. In an exploratory SNA, interpretation is developed gradually through a dynamic back-and-forth between structural patterns and narrative meanings as manifested through (a) choice of indicators, (b) interpretation of graphs as qualitative data, and (c) translating numbers to meanings.
a) Choice of Indicators
Social networks are complex structures; and several concepts and measures have been developed that tap into different dimensions of the complexity. Usually there are several alternative choices of indicators for a single concept, and the relationship between the concept and the indicator is not very straightforward. There is no single gold standard.
The mere concept of prominence/centrality, as an example, could be studied from many perspectives, including but not limited to how popular each individual is (or how many people turn to her, also known as “in-degree”), how influential she is in connecting many unconnected pairs (which indicates the brokerage and bridging capabilities of actors; also known as “betweenness”), and how accessible she is by other network members (which implies that more people across the network have access to her if she agrees to advocate for the innovation, also known as “closeness”) (Freeman, 1978). Network analysts also have to decide whether they are interested in the structural advantage of an individual in relation to all other actors in the network, immediate neighbors, or other individuals who possess certain characteristics (e.g., other experts/central actors).
Freeman (1978) in his iconic article on the conceptual clarification of centrality, stated that “There is certainly no unanimity on exactly what centrality is or on its conceptual foundations, and there is little agreement on the proper procedure for its measurement.” (p. 217) The number of centrality measures at the time of Freeman’s article was far less than the number today, as the measures currently fill the whole periodic table of elements! (Scoch, 2016). So the “embarrassment of intellectual riches” (Freeman, 1978, p. 237) dealing with the multitude of choices of centrality concepts and techniques still applies today (Kolaczyk, 2010), and we still see efforts to create a universal theoretical framework for centrality (Boldi & Vigna, 2014).
The choice of indicators from the long list of possible candidates is not straightforward. The researcher should have a clear understanding of the meaning of the concept behind the mathematical indicator, possible social mechanisms that are represented by the concept (Borgatti, 2005), and deep understanding of the research context. Researchers may start by identifying the actors with the largest number of incoming and outgoing ties (i.e., in-degree and out-degree centrality). Identifying actors who mediate more connections (betweenness centrality), or actors with shortest distance from others (closeness centrality) will show other dimensions of prominence. Researchers may then compare the lists of central actors obtained by different techniques to see how prominent actors according to various definitions overlap. The analysis may continue with assessing the association between various centrality scores and demographic and behavioral characteristics. The analysis may also include revisiting the thresholds of centrality by which the most central actors were chosen. Hence, the question “who are the most prominent actors in the network?” has more than one simple answer, and the choice of indicators is a combination of the researcher’s assumptions about social behaviors and dynamics in the network and also the observed network structure. The researcher’s judgment is involved in choosing the most relevant concepts, indicators, and thresholds of centrality; and the final decision is made through assessing theoretical relevance, practical feasibility, and contextual meaningfulness of various centrality measures. These judgments are made based on observations and in-depth knowledge of the research site as well as inchoate impressions of the data.
As an example of exploratory analysis of centrality in networks, Leydesdorff (2007) studied how various measures of centrality (in-degree, closeness, betweenness) of journals in citation networks can indicate the multidisciplinarity of the journals. He concluded that betweenness centrality could be a potential indicator, and reflected that “the finding of betweenness centrality as a possible indicator of interdisciplinarity was originally a serendipitous result of my work on journal mapping.” (Leydesdorff, 2007, p. 1316)
b) Graphs as Qualitative Data
Graphical presentation of quantitative data is a common practice. In SNA, this often involves visual presentation of the network as dots connected by lines, called a network graph or sociogram. One interesting difference between a social network graph and a summary chart (such as a bar chart showing means or proportions) is that the network graph is still a representation of the raw data, and not statistical summaries. So, making sense of a network graph could still be considered data analysis, and not just interpretation of some quantitative findings.
There is no single accepted approach to graphical depiction of social networks. The researcher chooses the most satisfactory presentation through a subjective and repetitive process of trying a few layout algorithms (which refer to the methods of distributing nodes in the map), adjusting the sizes, colors, and shapes of the nodes according to the most informative/distinctive personal attributes, adjusting the thickness and the length of ties (to represent the strength of relations and the distance between groups), and highlighting social clusters.
Several algorithms have been developed to distribute the nodes in more meaningful ways. For example, force-directed algorithms position the nodes in a two-dimensional space by simulating nodes as little magnets, where nodes repulse and ties attract (Jacomy, Venturini, Heymann, & Bastian, 2014). These algorithms are useful in providing a meaningful and visually appealing picture of the network. The researcher may use a few techniques to find the one that provides a more satisfactory presentation. This exploration may include the reflective process of choosing a different algorithm, playing with the specification of one algorithm (such as adjusting repulsion and attraction forces in the model), and changing the colors, sizes, and shapes of the nodes and ties, to a point that researcher is happy with how the graph tells his or her intended story most vividly.
Drawing meaning from a graph is the result of several involuntary and voluntary processes (Bennett, Ryall, Spalteholz, & Gooch, 2007) operating at various levels: visceral/perceptive (such as tie detection and perception of symmetry), behavioral (which involve mostly subconscious reactions to perceptual input, such as usability and function), and reflective (conscious efforts to extract meaning). As a result of this complex process, observers of a graph may interpret it in ways that were not intended by the researchers, since the “observers bring a rich vocabulary of graphical idioms and conventions to the table when they interpret the visualization” (McGrath & Blythe, 2004, para. 3). Neither the researchers’ perception nor the readers’ would be an exhaustive understanding of the underlying social patterns. Interpretation of a network graph is not only influenced by the structure of the data and analytical choices but is also developed in light of the background knowledge of the observer about the social dynamics and patterns being depicted (Blythe, McGrath, & Krackhardt, 1995; McGrath, Blythe, & Krackhardt, 2014). That is why a rich description should be provided of the underlying social processes and contextual forces and constraints, to help the readers reach a more informed interpretation.
c) Translating Numbers Into Meanings
In an exploratory network analysis, the researcher labels actors and their groups based on their characteristics and their relationships with others, to make sense of patterns and differences. Certain actors are labeled as “central” or “peripheral” based on the centrality of their positions in the network (Freeman, 1978), or “bridging” or “constrained” based on their roles in connecting unconnected clusters (Burt, 2000). Similarly, the network ties are labeled as “weak” or “strong” based on various qualities of the relationship or the peripheral/central positions of the nodes (Granovetter, 1973); social clusters are labeled “cohesive” based on the level of connectedness, the extent of mutual connections, and the strength of the relations (Alba, 1973). Even though several mathematical techniques have been developed to assist with this labeling, interpretation often involves classification based on arbitrary cut-points on a numerical scale and subjective description of groups, as a result of a balance between sociological assumptions about underlying social dynamics and observed composition of network.
We would argue that this interpretation therefore makes use of qualitization or qualitative profiling process. Teddlie and Tashakkori (2009) suggested the terms qualitization and quantitization for the process of transformation of qualitative and quantitative data into one another. They defined qualitization as the process of forming narrative profiles from numbers and suggested various methods of labeling groups of study units according to their numerical characteristics. They classified these qualitization techniques into five categories: modal profiling (verbal description of the most frequent attributes), average profiling (around the mean), comparative profiling (based on the comparison of units), normative profiling (based on the comparison with a standard), and holistic profiling (verbal description based on overall impressions of the researcher). Here, we provide a few examples of how a combination of various qualitative profiling techniques could be applied to network analysis.
Comparative profiling
As mentioned earlier, identifying a central actor involves setting a threshold in the centrality distribution. This process of choosing a group of actors based on their centrality and labeling them as “central actors,”“brokers,” and so on aligns with comparative profiling in Teddlie and Tashakkori’s (2009) terminology.
As an example of labeling network actors based on comparative values of their network indicators, Kratzer and Lettl (2009) studied the association between centrality measures and the “lead user” (i.e., early users of innovations) and “opinion leader” roles (i.e., who can influence others to use innovations) in a social network of school children. They suggested that the “opinion leadership” construct could best be determined by in-degree centrality. Similarly, Iyengar, Van den Bulte, and Valente (2011) in a study of adoption of new products by physicians, defined actors with highest in-degree in referral and discussion networks as “social connectors” and “experts,” and concluded that metrically defined “opinion leaders” (actors with highest in-degree) showed a higher tendency to adopt early.
Modal profiling
This category of qualitization involves labeling a group of actors based on the characteristics of the majority of its members (Teddlie & Tashakkori, 2009). Sometimes researchers look for common characteristics of actors who are in previously defined groups in a network, such as units of an organization or different professional roles. Some analysis may include identifying cohesive groups that are more connected compared with the rest of the network, such as identifying cliques, that include a group of people who are connected to (almost) every other person in the group (Alba, 1973). Block modeling techniques try to group the actors into structurally equivalent subgroups, based on their pattern of relationship with others (Ziberna, 2007). Many of these techniques try to find the best approximate solution, as the theoretically perfect pattern is very rare. Labeling the resulting groups involves a high-level interpretation from the researcher’s perspective.
As an example, Lewis (2006) used block modeling to group the actors in a network of influential individuals in health policy. She qualitatively labeled the resulting groups based on her evaluation of what the majority of individuals in each group had in common. Examples of labels she used to describe the resulting groups include “core influential,”“consumer and legal,” and “peripheral but connected.” This type of labeling—based on the characteristics of the majority of group members—is analogous to modal profiling, as described by Teddlie and Tashakkori (2009).
Holistic profiling
Sometimes a network researcher labels a network or its subgroups based on some overall characteristics or use of multiple sources of information. This aligns with “holistic profiling” which is labeling based on overall impression (Teddlie & Tashakkori, 2009). Examples include labeling a network as a star network (with one actor at the center to which all other actors are exclusively connected), or resembling a “small world” structure, where the actors are located in several loosely connected dense local clusters (Watts, 1999).
As an example of labeling a network based on the overall characteristics, West, Barron, Dowsett, and Newton (1999) compared the degree centralization measures (an indicator of the variation in the number of network ties among different actors) in the networks of clinical directors and directors of nursing, and concluded that the latter network was more “centralized” and “hierarchical.”
Partial Fusion: Qualitative and Quantitative Studies on the Same Network Data Source
In addition to the inherent mixing of quantitative and qualitative insights in seemingly quantitative SNA, there are potentials for more intentional mixing of approaches in network research (Figure 1). Following Bazeley and Kemp (2012), we define “partial fusion” as types of mixing in which the identity of each component is still distinguishable, while the boundaries are blurred. The most prominent example is taking quantitative and qualitative approaches to study the same data (Bazeley, 2012). Here we will describe two common examples in network research.
Mixed Data Collection
The process of data collection in social networks generally involves online or paper-based surveys. “Name generator” surveys elicit the list of respondents’ network members using free recall from memory (Marsden, 2011). The lists will be pooled together to develop whole network matrices and maps. An alternative technique is a network chart or “target,” a bull’s eye diagram that engages the respondent in drawing his or her own personal network on a graphical platform, rather than providing a list. The network charts are concentric circles with the respondent at the center, where she places the members of her personal network as dots with varying distance from center (indicating tie strength), and at different wedges of the circle (indicating various social groups such as peers, family and friends, etc.) (Hogan, Carrasco, & Wellman, 2007; von der Lippe & Gamper, 2017). This approach has the potential to be integrated into a semistructured interview in which the interviewer asks complementary questions about the members of personal network to contextualize the social relations and to provide examples of the relationships (Marsden, 1990). The data obtained in the interview could be analyzed both quantitatively (by calculation of whole and personal network indicators) and qualitatively (by the analysis of content and context of social communications). Their intuitive structure, their potential for engagement of respondents in conversation about their social networks, and their ease of translation into quantitative structural analysis make the network chart/target a popular data collection tool in mixed methods studies (Tubaro Ryan, & D’angelo, 2016). As an example, Cheong, Armour, & Bosnic-Anticevich (2013) interviewed patients with asthma about people with whom they discussed asthma-related matters. The interviewer then assisted the patients to distribute these people in a network chart depending on their importance to the patient. While doing so, the interviewer asked about their reasons and meanings for placing each individual in a certain ring.
The mixing of network data collection has also been tested in developing whole networks (rather than personal networks). For example, Coviello (2005) used qualitative in-depth interviews with the founders of a business firm, in a case study aiming to understand the chronology of the development of an entrepreneurial firm. She asked the interviewees about social capital ties among stakeholders throughout the life story of their firm. The researcher coded interviewees’ accounts on individuals and the content of social relations to develop network matrices and to calculate whole-network indicators. Because of the inductive and iterative nature of the data collection, the resulting network maps were also developed inductively and “revised as necessary until the optimal representation of the firm’s history was captured in that it was considered by the informants to be comprehensive [and] detailed” (Coviello, 2005, p. 45)
Mixed Analysis of Network Graphs
Earlier we explained how the process of drawing meaning from network graphs is an inherently fused process. Some researchers use more systematic mixed methods approaches, such as using qualitative interviews to interpret the network maps developed through quantitative network analysis. For example, D’Angelo used this approach to interpret the network maps of Kurdish community organizations in London (Ryan & D’Angelo, 2018). The maps were shown to a selection of network actors to comment on the social structure. The reflections by the respondents were used to reinterpret and revise the maps. In this approach, a whole network map is no longer considered a (positivistic) presentation of reality, but more a “method of exploration” (Ryan & D’Angelo, 2018, p. 153), which is subject to iterative revision and reinterpretation by the network actors.
Separate Quantitative and Qualitative Strands in Mixed Methods Network Studies
Studying social networks is also a fertile ground for traditional sequential and parallel mixed methods designs. In these kinds of mixing, each strand keeps its distinct personality. A sequential mixed methods network study may usually include using the findings of a quantitative network analysis to identify potential interviewees (e.g., central network actors) in a subsequent qualitative enquiry aiming to deepen the researcher’s understanding of the social processes and dynamics.
For example, Yousefi Nooraie, Lohfeld, Marin, Hanneman, and Dobbins (2017), in a sequential mixed methods study, assessed the process and outcomes of a targeted training intervention to promote evidence-informed decision making among the staff of public health organizations. A selection of central and peripheral network actors in a quantitative SNA were invited to participate in semistructured interviews focusing on the process of their engagement in the intervention, the trajectories of their social relations during the intervention, and consequences of the implementation. Even though the findings of the quantitative network analysis were shared with the interviewees to seek their feedback on the quantitative findings, the results of the qualitative strand did not change the structure of the network maps that were developed through quantitative SNA. The qualitative strand helped the researchers explain and contextualize the quantitative SNA results.
Quantitative and qualitative strands could also be mixed in parallel mixed methods studies, in which two strands usually help triangulate the findings. For example, Chiu and West (2007) used SNA to understand how neighborhood and personal networks affect health interventions implemented by community health educators in the United Kingdom, and to what extent they were aware of their embeddedness in personal networks and consciously used their knowledge to promote their professional activities. The researchers used both network survey and focus group discussions. The integration of quantitative and qualitative findings helped the researchers to cross-reference the results and produce richer conclusions about the perception of neighborhood and its role in community health educators’ practices.
Discussion
In this article, we argued that network analysis holds potential for a continuum of fusion between quantitative and qualitative approaches, where the researcher steps into an iterative mixing of quantitative and qualitative data collection, analysis, and interpretation. The fusion in exploratory SNA can involve subjective selection of concepts and techniques, qualitative interpretation of graphs and visuals, and translation of numbers to words and meanings. Human perception and computational algorithms are used complementarily to make sense of patterns (Onwuegbuzie, Johnson, & Collins, 2009); and meaning development is the result of inductive process of empirical fitting and deduction from underlying theories (Newman & Hitchcock, 2011). The final understanding is always a subjective inference, or as suggested by Onwuegbuzie and Leech (2005a) follows the equation of “subjectivity + objectivity = subjectivity” (p. 377).
This back and forth process of gathering and organizing the pieces of the puzzle by analyzing various aspects of the complex network structure, and gradual formation of meaning in researcher’s mind resembles the spiral process of mixed methods integration, as explained by Fetters and Molina-Azorín (2017). The process entails back and forth exchanges between methods and approaches, with decreasing circular diameters, until it reaches an apex that represents full integration.
In her commentary on the new challenges facing mixed methods research, Bazeley (2016) suggested that the next challenge is not so much the need for creation of new merged methods, but to recognize the inherent potential for the merge between quantitative and qualitative approaches. We tried to disentangle some aspects of the continuum of fusion in network research. However, we argue that these processes are also present in other seemingly quantitative approaches. Multivariate statistical models are usually developed through a repetitive process to find a model with a good balance of fitness and parsimony, which generally involves revisiting the models in light of how well they represent the conceptual framework of the study and observed variations in the data (Ratner, 2010).
One notable example is exploratory cluster analysis, in which clusters are identified and labeled in a manner similar to the way emerging themes are developed in some forms of qualitative thematic analysis, whereas here the data only involve numbers. The process begins with statistical matching/distribution, in which, similar to SNA, the choice of clustering technique is usually subjective (Jaworska & Chupetlovska-Anastasova, 2009). It continues with a dominantly qualitative interpretation of patterns and a delineation of clusters in a graphical map or a dendrogram, and may iterate a few times until the researcher is satisfied with the findings. The process usually ends with labeling clusters and axes based on their meaningfulness, which involves translating patterns into words. Unfortunately, these formative processes are likely to be underreported in the quantitative literature, since researchers tend to report the final endpoints of their analysis with limited reporting of the iterative/stepwise mechanism of gradual generation (Bazeley & Kemp, 2012).
Long before the term social network analysis was coined, anthropologists such as Henry Morgan (1818-1881) and psychologists such as John C. Almack (1883-1953), and most prominently Jacob Moreno (1889-1974) used systematic approaches to study human interactions, mostly through blending of qualitative interviews and fieldwork with graphical presentation of patterns (Freeman, 2004). However, SNA, consistent with the dominant trend, has been developed in a quantitatively oriented tradition. Network analysts have developed systematic mathematical and graphical methods for the analysis of social structures. Through decades of advancement in scientific methodology, SNA is now considered by its pioneers as a theoretically coherent perspective developed through cumulative and incremental contributions of researchers attending “to each other’s work” within a common paradigmatic field with no major split (Hummon & Carley, 1993, p. 73), resembling the pattern of scientific development labeled by Kuhn as “normal science” (Hummon & Carley, 1993, p. 103). However, during the formative years of SNA, qualitative studies on social networks were too limited to become part of the main path of the incremental progress of the dominant SNA tradition (Hummon & Carley, 1993, p. 103).
In parallel to this quantitative development, various traditions in social anthropology and sociology have examined social relations from figurational (Elias, 1978) and symbolic (Blumer, 1969) viewpoints. In recent years, more attention has been paid to mixing quantitative and qualitative approaches in the study of complex social phenomena such as SNA (Dominguez & Hollstein, 2014; Edwards, 2010). Despite this, quantitative and qualitative approaches to SNA are considered separate traditions with limited integration (Fuhse & Mützel, 2011). We argue that SNA represents Normal Science if the inherent conversation between objectivity and subjectivity, blending of qualitative and quantitative data, and the cumulative and cyclical process of the development of meaning are further recognized and acknowledged.
Conclusions
Bazeley (2016) portrayed two challenges facing mixed methods researchers: “to recognize the inherent joint qualitative and quantitative elements that already exist in the data and analyses they are already using,” and “to fully exploit the integrative potential of their data during the analysis process” (pp. 191-192). We explained three main manifestations of SNA which make it a fitting example of fused analysis: choice of analytical techniques and network indicators, qualitative interpretation of data in graphs, and transformation of numbers into words. Fusion itself can happen over a continuum ranging from mixing distinguishable quantitative and qualitative studies to an inherently fused data analysis and interpretation. These manifestations are not unique features of network analysis, and could be applied, with context-sensitive modifications, to many other seemingly quantitative analytic approaches.
Inspecting the potential of SNA to fuse qualitative and quantitative data collection, analysis, and interpretation helps us learn more about the possibilities of integration, no matter if we consider fusion to be an example against the quantitative–qualitative distinction (Gorard, 2010), or rather an example of research that lies on the blurry boundaries between the two traditions (Morgan, 2018). Recognizing different manifestations of fusion will help researchers appreciate and exploit the potential for integration during the analysis process and take a more conscious approach to the development of mixed interpretations.
Footnotes
Acknowledgements
We are grateful to Professor H. Russell Bernard (University of Florida) whose encouragement and insightful feedbacks were influential in formation of this article.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.