Below you will find essays written for my tutorial in Social Network Analysis, hosted at the Oxford Internet Institute with Dr. Bernie Hogan as my tutor. I attended Oxford University during Trinity Term of 2019.


Construction of the Network in Social Network Analysis

Social networks are comprised of individual actors, or “nodes”, and relations between them, or “edges”. Analysts of social networks study different properties of and on these networks. Preceding the study of these networks, however, is the question of where these networks come from. More specifically: do social network analysts create or discover these networks? What constitutes an edge or relation? In this paper, I argue that social network analysts create, rather than discover, the networks they study. I further argue that these social networks only function as a rough approximation of reality, primarily because of (1) the static and low-dimensional nature of the model, and (2) the emphasis on the individual actor as the fundamental unit of Being rather than the relation.

The rendering of a social system into a social network requires some transformation of the relations that exist in real life into edges. This quickly becomes complicated, as “social network researchers grapple with where to draw lines across relational webs possessing no clearcut, natural boundaries” (Emirbayer, 1997). Hogan similarly notes: “In social network analysis, this linking [between nodes] is done by the analyst who would suggest some sufficient criteria. For example, a researcher might indicate links between two people if they have sent any direct communication in the last six months. This is obviously an artificial boundary, but a workable one” (Hogan, 2018). Other possible criteria: a friendship, a marriage, a sibling, or some other status. Importantly, however, “such a status does not necessitate a specific relationship that would happen between [them]” (Hogan, 2018). Thus, we see that an edge’s existence is only indicative of whether the nodes’ interactions meet the given criteria the researcher has set. In this sense, a unidimensional network is created, rather than discovered, as according to some dictated criteria.

Relations are also inherently dynamic, as “with every action. . . the individual is sending signals about their relationships that can then be used to update some strength of connections” (Hogan, 2018). The dimensional reduction of a dynamic process down to a binary variable inherently loses information. Furthermore, the reduction of multiple relationships to a binary variable implicitly equates different, dynamic relationships to the same static object. It is nearly, if not wholly, impossible to construct a metric or a question that will standardize to the same type of relationship as defining an edge. Bearman et al. support this in “Cloning Headless Frogs and Other Important Matters”, where they criticize the metric of asking “With whom do you discuss important matters with?”. They find that, since people have different ideas of what “important matters” constitute, this creates a network that is not necessarily solely composed of close relationships. Children can speak to their parents about “important matters” like finances without being close to them. Men can speak with casual acquaintances about “important matters” in the news without being close to them. Consider that we instead ask them more pointed questions to avoid this difference, for example, “with whom have you discussed your relationship, or your finances?”. We then run into a comparable problem, where we are solely measuring conversational content rather than the strength of a relationship. Neither question creates networks with edges that are equally representative of the same strength or type of relationship, and thus the network is not a true representation of reality or “discovery” of what is already there.

Arguably, individuals do something similar cognitively. As the brain tries to encode experiences and representations, we are prone to some of this oversimplification. At any singular moment, we are selectively choosing which parts of the relationship to characterize it by, and are thus similarly dimensionally reducing something of high or infinite dimension to a manageable amount (considerably smaller). Social networks, however, seem to take the static models of our social world in our head to an even further extent. Importantly: we see that, cognitively, encoding can interfere with our understanding of the phenomenon itself, as tangentially shown via experiments about verbal overshadowing (Hogan, 2018). This argument can extend to the construction of networks—in creating these models, especially without purposeful attention, it is possible for our simplistic encoding to obstruct our understanding of what is actually happening.

Another fundamental part of social network construction is the implicit assumption of “the elementary unit of social life [as] the individual human action” (Emirbayer, 1997). Social network analysis seems to rest on the substantialist view of self-subsistent entities (in this case, nodes, or agents) that “inherently, and hence necessarily, possess Being. . . come `preformed’, only after which we consider the dynamic flows in which they subsequently involve themselves” (Emirbayer, 1997). However, counter to this is the relationalist perspective that focuses on the relation as the elementary unit of social life, wherefrom “units involved in a transaction derive their meaning, significance, and identity from the (changing) functional roles they play within this transaction” (Emirbayer, 1997). A simple example that supports the relationalist view: the structure of the bonds in a chemical compound dictate the characteristics rather than the molecules themselves. I claim that this philosophical shift from a nodecentric view to an edge-centric view could help fuel social network models that are less static in nature and, moreover, don’t fall as prey to Hogan’s cited “reduction of processes to static conditions” as we optimize for the representation of these processes, i.e. the relationships (Hogan, 2018).

That being said, the psychological implications of switching to this perspective make this difficult and unnatural. By demoting actors to “individual persons. . . [that] are inseparable from the transactional contexts within which they are embedded”, we detract agency from the individual (Emirbayer, 1997). Suddenly the personality is not a set of “fixed traits or dispositions that endure across times and contexts but . . . a stable configuration of distinctive if-then situation-behavior transactions” (Emirbayer, 1997). This is naturally psychologically unsettling, as it undermines individual agency in favor of deterministic roles as a function of network position. In social network analysis, it is commonly accepted that the importance of a node by can be measured by their centrality on the network, as dependent on the type of flow on the network (Borgatti, 2005). As an extension of this argument, individuals are no longer bestowed with agency as a given but instead conditionally have the capacity to exercise it dependent upon their location within the network. This relationalist perspective, then, flips a lot of comfortable assumptions on their head. As ‘ego’s in our own network and lives, considering ourselves to be solely or more so products of our externalities is not a perspective that comes naturally to us, even as scientists.

In conclusion: when we construct networks and encode relationships (inevitably incorrectly), we are doing just that—constructing it. . These models are inherently incorrect as they attempt to offer static representations of dynamic processes. In operationalizing these dynamic processes—necessary for any sort of modeling or science—it is then important to do so purposefully and with open-eyed recognition of the assumptions we have made and criteria we have dictated. Although these social network representations are flawed (as all models are), they provide some approximation of reality that may prove useful. However, a shift from the substantialist view to the relational, though unnatural, may help analysts get closer to “discovering” rather than constructing the social networks underlying our interactions and societies.

References

Hogan, B (2018). Break-ups and the limits of encoding love. In Papacharissi, Z., ed. “A Networked Self and Love”. New York, NY: Routledge. Pp. 113-128.

Bearman, P., & Parigi, P. (2004). Cloning Headless Frogs and Other Important Matters: Conversation Topics and Network Structure. Social Forces, 83(4), 535–557.

Borgatti, S. P. (2005). Centrality and Network Flow. Social Networks, 27(1), 55–71.

Emirbayer, M. (1997). Manifesto for a relational sociology. American Journal of Sociology, 103(2), 281– 317.


Investigating the Generalizability of Network Intervention Studies

Through the study of social networks, we are able to gain insight into the impact our social ties have on our behaviors and habits. We then naturally wonder: in a world with an abundance of problems, can we harness this influence to our advantage, i.e. to help solve these problems? More specifically: is it possible for us to intervene on a social network and obtain a desired result?

The study of network interventions allows us to find answers to these questions in different contexts. For the purposes of this paper, I use Valente’s definition of network interventions as “the process of using social network data to accelerate behavior change or improve organizational performance”, noting that with these interventions, “social networks can be leveraged to accelerate behavior change, improve organizational efficiency, enhance social change, and improve dissemination and diffusion of innovations” (Valente, 2012). As we conduct these interventions, we gain knowledge about whether or not our efforts were successful. But, as all good science does, we ultimately want these studies to generalize past one-off yes/no answers. So to what extent are these network studies generalizable? Can we extend the bits of knowledge we gain from them into wider wisdom, or rules, that we can use to guide future interventions?

Without an underlying science of network interventions, I argue no. As network intervention studies stand, the field is more of an engineering than a science. In lieu of theory, scientists tinker to see what works and what does not, informed by feedback loops. As Valente states, “Network interventions are quite effective . . . yet, the science of how networks can be used to accelerate behavior change and improve organizational performance is still in its infancy” (Valente, 2012). Without inquiring into the mechanisms by which this behavior is spreading, it is hard to generalize these studies.

In this paper, I claim that there are multiple criteria that dictate the extent to which these findings are generalizable, a non-exhaustive list of which includes: the culture and values of the agents, how the network is constructed, and what the intervention requires of an individual. I will explicate each of these four criteria by looking at case studies of network interventions. I further claim that, without an underlying science, the only strategy we then have for understanding generalizability is strategic experimentation along dimensions. This consists of performing multiple experiments over which certain dimensions are held constant (e.g. behavior propagated, what an edge represents, etc.) to determine whether we can generalize along that dimension. This process of experimentation gives us insight into the underlying mechanisms by which a network intervention is successful or unsuccessful.

The first study I inspect is Valente et al.’s “Effects of a social-network method for group assignment strategies on peer-led tobacco prevention programs in schools”; here how peer leaders should be assigned to groups in tobacco prevention programs is evaluated. Three experimental conditions are created: “(1) random: leaders defined as those who received the most nominations by students, and groups created by randomly assigning students to leaders; (2) teacher: leaders and groups created by teachers; and (3) network: leaders defined as those who received the most nominations by students and groups created by assigning students to the leaders they nominated” (Valente et al., 2003). The network strategy is ultimately found to be the most effective way to structure the program.

This conclusion is then prone to the generalization that we should always select peer leaders via nominations and pair them up with their nominators. However, there are specific conditions this experiment is under that could be responsible for this strategy’s success. Valente proposes a few explanations for the network strategy’s success. He hypothesizes that perhaps “students learn to practice resistance skills with their near peers who probably will be present in situations where smoking will occur”, or “students may learn more if they are in a comfortable social setting with their friends” (Valente et al., 2003). Let us take the first hypothesis and suppose that being with near peers who “probably will be present ins situations where smoking will occur” is what makes the intervention more effective. This mechanism depends upon a culture that values social relationships and conformity. In other words, if we were to conduct this intervention in a culture where this is not valued as much, we may not get the same results. Furthermore, they note that “the majority [of students] had at least 1 foreign-born parent (74.4%) . . . A majority of the sample was Hispanic or Latino (54.4%), and a significant percentage was Asian American (23.4%)” (Valente et al., 2003). This makes us further consider: if these students are primarily first-generation immigrants or minorities, there is the possibility that they value cultural assimilation even more, especially at this young age. This cultural factor and its impact on what the agents implicitly value places conditions upon the generalizability of this study’s findings.

We can look at the Valente et. al study to also see how the construction of the network plays an important part in potential generalization. Nominations are collected by asking students to “think about the 5 people in the class who would make the best leaders for working on group projects” (Valente et al. 2003). This criteria is vague and does not necessarily translate to what is being tested, i.e. whether these students would be influential enough leaders to impact smoking behaviors and attitudes. Valente notes this shortcoming: “students chose leaders in our study on the basis of ‘working on group projects’. . . results may have been different if leaders were chosen for their status as role models, lifestyle trendsetters, or other attributes connected to tobacco use decisions” (Valente et al., 2003). As such, even Valente notes the lack of robustness in the results—or rather, its heavy dependence on a simple question—that limit its generalizability.

Centola’s 2010 study “The spread of behavior in an online social network experiment” further helps us explicate this claim about network construction. In this study, participants make decisions about whether or not to adopt a health behavior based on adoption patterns of their “health buddies” online. He varies the high-level network structure between clustered-lattice and random network conditions. He concludes that “whereas locally clustered ties may be redundant for simple contagions, like information or disease, they can be highly efficient for promoting behavioral diffusion”, i.e. clustered-lattice graphs were more effective than random networks for spreading this health behavior (Centola, 2010). Notably, however, Centola “blind[s] the identifiers that people used”, stripping any information about ‘health buddies’ that an agent can see aside from whether or not they adopted the behavior (Centola, 2010). This is an understated choice that has significant implications. Whereas the bottom line of Centola’s study seems to be that clustered lattice networks are more effective for behavior propagation, this is only shown to be true under this blind condition. It is possible that behavioral diffusion could spread effectively over a less clustered network if the strength of the ties and identifiers of agents are taken into account, which is arguably a more realistic and generalizable situation than Centola’s.

Thus, through both Valente’s and Centola’s studies, we see that the manner in which the network is constructed is incredibly influential on their ultimate intervention results. This is the same in network science in general: the way in which we operationalize relationships into constructed social network models can have a significant impact on our findings and analysis. Bearman and Parigi (2004) have explicated this claim in “Cloning Headless Frogs and Other Important Matters: Conversation Topics and Network Structure” by showing the differences in networks we create when we ask different questions, and how this impacts our analysis of the findings. Just as networks are subjective, then, similarly we see that “network interventions are not agnostic or impartial, but depend on the goals and objectives that initiate the intervention” (Valente, 2012).

Lastly, we look at Kim et al.’s 2015 study “Social network targeting to maximise population behaviour change: a cluster randomised controlled trial”, where two interventions—chlorine for water purification and multivitamins for micronutrient deficiencies— are introduced to the population through different strategies. They are ultimately able to create an effective nutritional intervention via networknominated targeting, but not significantly so with the chlorine intervention. By having two different types of health-related interventions, the generalizability of each is under closer speculation. Thus, where one could be tempted to say that “in these types of villages, with this type of strategy, a health-related intervention will work”, they are able to provide greater nuance to their claim. They hypothesize that their differing results are because “the chlorine intervention demanded the acquisition of new knowledge and a change in the use of a familiar product . . . The multivitamin intervention, then, demanded substantially less behavioral and ideational change” (Kim et al., 2015). In other words: although these interventions are both health related, and the populations are the same, and the strategies of intervention are the same, an important difference is in what the intervention requires of the individual in order to activate or change its state—a shift in mindset, behavior, etc. This is also, then, an important consideration for generalizability.

We can cobble together insights into generalizability via literature reviews. Wang et al. attempt to do this in a review of network interventions for condom usage (Wang et al., 2011). They correctly note that there are “large differences in how social network members were identified and involved in the interventions” they study (Wang et al., 2011). These range from “ethnographic observations. . . used to identify leaders of ‘social circles’” to “nominated peer-leaders (self-nomination allowed)” (Wang et al., 2011). Similarly, the construction of the network varies: in one study, participants were asked to “make a list of risk network members who might be eligible for the study”, while in another, they were “interviewed and asked to provide the interviewer with the first names of the [their] closest friends” (Wang et al., 2011). Not only is the network constructed in a different way, then, but the edges within them represent very different types of relationships, considerably restricting generalizability. The studies also vary by location, with “five studies conducted in high income nations, five studies in middle income nations, and one study in a low-income nation” (Wang et al., 2011). This allows us to understand the impact of culture and values on intervention success. Thus, by comparing each of these studies as though along different dimensions where certain variables are held constant (e.g. culture) and others are varied (e.g. what an edge represents), we can begin to understand along which dimensions we are able to generalize.

In conclusion, although many network intervention studies have proven effective, we need a more strategic method of experimentation to uncover an underlying science behind these interventions to be able to generalize findings from them. Only then will we be able to most effectively “use the power of human interaction to improve the human condition” (Valente, 2012).

References

Bearman, P., & Parigi, P. (2004). Cloning Headless Frogs and Other Important Matters: Conversation Topics and Network Structure. Social Forces, 83(4), 535–557.

Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996), 1194-1197.

Kim, D. A., Hwong, A. R., Stafford, D., Hughes, D. A., O'Malley, A. J., Fowler, J. H., & Christakis, N.A. (2015). Social network targeting to maximise population behaviour change: a cluster randomised controlled trial. The Lancet, 386(9989), 145-153.

Valente, T. W., Hoffman, B. R., Ritt-Olson, A., Lichtman, K., & Johnson, C. A. (2003). Effects of a social-network method for group assignment strategies on peer-led tobacco prevention programs in schools. American journal of public health, 93(11), 1837-1843. Valente, T. W. (2012). Network interventions. Science, 337(6090), 49-53.

Wang, K., Brown, K., Shen, SY. et al. AIDS Behav (2011) 15: 1298. https://doi.org/10.1007/s10461-011-0020-1.


Online Identity and Personal Online Networks

With the rising prevalence of social media and online interaction, we have begun to inhabit what Davis and Jurgenson refer to as a “networked era”. This “networked era” is defined as a “historical moment characterized by pervasive and increasing digital and social connectivity” (Davis and Jurgenson, 2014). Furthermore, they note that “within a networked era, social media platforms act as key hubs of interaction and identity negotiations” (Davis and Jurgenson, 2014). “Identity negotiations” reference the main topic of this paper: the explicit representation of the self that is required of individuals when participating in these platforms. Agents are then associated with not only an online personal network but an online identity. As such, we wonder: what can personal online networks tell us about the online identity of an agent?

I first argue that the most important insight personal networks can give us in this application is as empirical evidence for context collapse, which leads us to conceive of the online identity of an agent as distinct from authentic identity. I further explicate this claim by referencing studies that act as instantiations of Davis and Jurgenson’s (2014) classification of context collapse into (1) context collision and (2) context collusion. I then argue that social media platforms as biased “curators” (Hogan, 2010) accentuate the incongruence of online and authentic identity. Somewhat counterintuitively, I also argue that a defining feature of social media sites—the unknown audience—may actually incite online identity to be less distinct from authentic identity. Still, I ultimately conclude that heavy curation via the self and social media platforms severely limit the insight we can gain into the online and authentic identity of an agent through personal online networks.

It is widely accepted that the self is not constant but, rather, changes in different contexts. Context, then, is defined “in terms of role identities and their related networks . . . The self is made up of multiple identities, each of which exists within a network of others who hold particular expectations about who the actor is. These expectations inform appropriate— and inappropriate— lines of action and identity performance. In these terms, [context] collapse refers to the overlapping of role identities through the intermingling of distinct networks” (Davis and Jurgenson, 2014).

Through mapping online social networks and clustering, we are able to algorithmically find distinct groups of people that participants are able to name and identify as separate, as in McConnell et al. (2018), such as friends, family, and co-workers. As many social media sites only have one category of tie, “diverse Generalized Others converge into a single mass, requiring the actor to simultaneously engage with family, colleagues, and drinking buddies, each of whom harbours different views of who their actor is, and different interactional and performative expectations” (Davis and Jurgenson, 2014). This classification of qualitatively different types of relationships into binary ties is further evidence of the “limits of encoding” as described in Hogan (2018). Furthermore, the overlap of members from different contexts with different expectations of the agent constitutes a context collapse as defined above. Thus, by using personal online networks we are able to observe the structural underpinnings of context collapse. Context collapse can be further categorized into (1) context collusion, and (2) context collision, which are differentiated by intentionality (Davis and Jurgenson, 2014). I next present two studies that support the claim that each of these types of context collapse lead to the creation of an online identity distinct from authentic identity.

First, context collusion. Context collusion refers to “the process whereby social actors intentionally collapse, blur, and flatten contexts, especially using various social media” (Davis and Jurgenson, 2014). This collusion is inherent in the design of the early 2000’s social media site, Friendster. Friendster’s primary goal was to act as a dating website—but with a more credible touch through Testimonials by friends. As such, an agent’s network on Friendster consisted of both their friends and potential romantic interests. This produced an intentional blurring of contexts, and by choosing to participate in the Friendster platform, members willingly collapsed these contexts. In this case, it was observed that by trying to appeal to both friends and romantic interests, agents developed a set of behaviors to present the self (Boyd and Heer, 2006). Thus the “Testimonial, [although] technically between the author and the receiver, [became] equally intended for third parties” (Boyd and Heer, 2006). Similarly, “including event photos [was] simultaneously a signal of friendship structure to outsiders and an expression of appreciation to friends” (Boyd and Heer, 2006). Moreover, “profiles [were] interpreted as conversational anchors similar to clothing, providing valuable cues about the individual such as potentially shared interests” (Boyd and Heer, 2006). This inclusion of selective content and information constitutes a contrived self separate from the authentic self, especially because this self exists in a largely unnatural, context-collapsed space.

On the other hand, context collision is defined as “different social environments unintentionally and unexpectedly com[ing] crashing into each other” (Davis and Jurgenson, 2014). We can observe context collision most clearly through the study of those who have a large stake in it. McConnell et al. (2018) studied LGBTQ individuals with varying degrees of out-ness to various groups of people. More specifically, they analyzed the effect of Facebook’s context collapsing on the creation of the online self in these LGBTQ individuals. They ultimately found that “to negotiate these disclosure-related challenges, LGBTQ young people engage in a variety of identity management strategies, including monitoring their online self-expression, using privacy and security controls, strategically managing their friendship networks, creating multiple accounts, curating and editing personal photographs, and restricted LGBTQrelated content to other, more anonymous online contexts” (McConnell et al., 2018). This is a clear example of a situation in which we could analyze a network and not get accurate insights about the identity of an agent, as they are strategically managing their networks and content. Here again, then, we see the creation of an online self, even more purposefully distinct from the authentic self as agents aim to hide aspects of their identity.

Thus we see that in platforms that incite context collapse, the agent is challenged to produce content that is palatable to all cross-contextual members. This generalizes to Hogan’s “lowest common denominator” theory as a strategy to negotiate unknown and diverse audiences. Hogan theorizes that, when submitting content, an agent considers only two groups: “those for whom [they] seek to present an idealized front and those who may find this front problematic. . . One might not be posting for one’s parents (or children or students) on Facebook, but again, one is posting in light of the fact that these individuals may have access; these individuals define the lowest common denominator of what is normatively acceptable” (Hogan, 2010). Thus, by analyzing one’s online identity, we are only accessing what they feel comfortable sharing with the lowest common denominator they can think of. This leads to a flattened out version of their personality and opinions in all dimensions, with only expression of conventionally-agreed upon opinions and values. As an aside, it would be interesting to investigate the impact on individuality these context collapsed networks have if agents are constantly catering to the lowest common denominator.

Hogan expands his theory to a conception of social media websites as “exhibits” and platforms as “curators”, wherein “curators mediate our experience of social information . . . [via] filtering, ordering and searching” (Hogan, 2010). Curators, then, are ultimately responsible for choosing the audiences that see which content. It follows that curators theoretically have the power to shield from context collapse and provide content to the “correct” audiences, if given enough insight. However, I argue that the platforms that act as curators are not impartial. They often affect not only user behavior but the structure of the online networks we are able to observe via platform effects (Malik and Pfeffer, 2016). Furthermore, differences in online identity are noticed and accentuated by the curators in order to achieve their own goals. An example of this is on Facebook, where the “People You May Know” feature allowed for triadic closure to occur at higher rates than it would have otherwise (Malik and Pfeffer, 2016)—similarly, we could imagine social media platforms working to shape e.g. individuals who exhibit some bridging properties into highly bridging individuals. As such, we are even more restricted by what we can deduce about the online individual via their personal networks, as curators uncharacteristically play a part in shaping the exhibits they curate.

However, some factors do also play a part in making online identity more congruent with actual identity. I argue that cognitive limitations do so significantly. Davis and Jurgenson rightly note that “although users often act as though their audiences are bounded, they are in fact, potentially limitless” (Davis and Jurgenson, 2014). A significant characteristic of social media platforms is an unknown audience. Since artifacts online are persistent over time, “even if one can evaluate the audience at a given time, it is impossible to gauge potential audience” (Boyd and Heer, 2006). Thus, though “content may be produced and submitted with a specific audience in mind, those who view and react to this content may be different from those for whom it was intended” (Hogan, 2010). The lack of innate understanding of the timeinsensitive nature of our content and its implications allows us to be more authentic than we would if we did have this natural understanding. Arguably, this is a result of this technology being recent—our collective innate understanding may grow over time as generations acclimate to it as the default and see the repercussions of others’ past mistakes. Furthermore, the size of the audience, even in real-time, is largely incomprehensible to us. Agents often do “not consider everyone when submitting content but only two groups: those for whom we seek to present an idealized front and those who may find this front problematic… Their behavior is in reference to specific salient individuals, who are small enough in number to be coherent” (Hogan, 2010). Thus, due to cognitive limitations, agents submit content appropriate only for a group of “salient individuals” and thus reflect a more authentic self than if they were considering a larger audience (by virtue of the lowest common denominator strategy). Our cognitive incapacity to consider the scale of our content’s audience then ultimately lends authenticity to the online self.

Furthermore, there is still a rich amount of information we can gain from analyzing online networks about online identity and, in some cases, authentic identity. Far from denying this to be true, the arguments made in this paper simply serve as caveats to consider when attempting to gain insight from findings about online identity via online networks. With consideration of these factors, insightful analysis can be conducted. As examples: one can predict perceived social capital of an individual through analyzing their online network (Brooks et al., 2014). Through this network, one can also predict whether an individual is part of an individualistic or collectivist culture (Rui, 2013).

Thus, although there are some characteristics of online and authentic identity that we can learn about from online networks, we must engage in this study cautiously, as both online identity and networks are curated and purposefully constructed. Because of context collapse and the lowest common denominator approach, online identity is first heavily curated by the self. Online identity and networks are further curated by unbiased social media platforms. The combination of these factors limits the insight we can gain about online and authentic identity through the analysis of online networks.

References

Boyd, D., & Heer, J. (2006, January). Profiles as conversation: Networked identity performance on Friendster. In Proceedings of the 39th annual Hawaii international conference on system sciences (HICSS'06) (Vol. 3, pp. 59c-59c). IEEE.

Davis, J. L., & Jurgenson, N. (2014). Context collapse: Theorizing context collusions and collisions. Information, communication & society, 17(4), 476-485.

Hogan, B (2018). Break-ups and the limits of encoding love. In Papacharissi, Z., ed. “A Networked Self and Love”. New York, NY: Routledge. Pp. 113-128.

Hogan, B. 2010. “The Presentation of Self in the Age of Social Media: Distinguishing Performances and Exhibitions Online”. Bulletin of Science Technology and Society. 30(6): 377-386. https://doi.org/10.1177%2F0270467610385893 Malik, M.M., & Pfeffer, J. (2016). Identifying Platform Effects in Social Media Data. ICWSM.

McConnell, E., Néray, B., Hogan, B., Korpak, A., Clifford, A., Birkett, M. 2018. “Everybody puts their whole life on Facebook”: Identity management and the online social networks of LGBTQ youth. International Journal of Environmental Research and Public Health, 15(6), 1–19. https://doi.org/10.3390/ijerph15061078

Rui, J.R., & Stefanone, M. (2013). Strategic self-presentation online: A cross-cultural study. Computers in Human Behavior, 29, 110-118.


Self-monitoring and Social Networks

As we reduce agents to nodes and dynamic relationships to edges in social network analysis, we are prone to losing sight of the individual. In other words: when we study groups of people in this structure, we tend to forget the person. It is tempting to do so, especially with the justification of social networks as complex systems with emergent properties, where the whole system cannot be fully explained by constituent parts. However, the study of even complex systems can be informed by an understanding of the individual. As such, there is a need to bridge the gap between individual psychology and aggregate sociology. As Mehra et al. (2001) says, “rather than treat[ing] individual attributes and social attributes as separate realms of enquiry, we [can] seek to understand how social networks… are shaped by the attributes of interacting individuals”. In this paper, I aim to do this by examining the psychological behavior of self-monitoring and its impact on social networks. I begin by defining self-monitoring and its observed impacts on the individual and their network structure. I then discuss the impact that social media and participation online has on self-monitoring and its structural correlates. I argue that social media lowers both (1) selfmonitoring behavior because of context collapse (boyd and Heer, 2006) and the lowest common denominator approach (Hogan, 2010), and (2) structural “bridges” because of unbiased curation (Hogan, 2010) on behalf social media platforms.

According to Synder’s self-monitoring theory, people differ “in the extent to which they are willing and able to monitor and control their self-expressions in social situations” (Mehra et al., 2001). Individuals with high self-monitoring “present the right image to the right audience”, while those with low self-monitoring “insist on being themselves, no matter how incongruent their self-expression may be with the requirements of the social situation” (Mehra et al., 2001). Furthermore, “high self-monitors are better at scanning the social world for information about people and their intentions” (Mehra et al., 2001). This orientation towards high or low self-monitoring is thought to be a “distinctive aspect of each individual’s personality… accumulating evidence suggests that self-monitoring is a stable personality trait throughout one’s lifespan” (Mehra et al., 2001).

An individual’s self-monitoring orientation has been shown to have large impact. First, on their network structure. Burt (2012) claims that self-monitoring is a “psychological analogue to bridging structural holes”. This is because as individuals self-monitor, they are able to appeal to diverse sets of people in different contexts and thus bridge otherwise distinct clusters of people. High self-monitors can then function as network brokers who are able to “act as go-betweens… between disconnected others, facilitate resources flows and knowledge sharing” (Mehra et al., 2001). They further “tend to gain both information and control benefits… Actors whose social ties are limited to one clique are less likely to receive diverse information than are actors whose ties span cliques because information that circulates within a clique of highly connected [individuals] is likely to be redundant” (Mehra et al., 2001). As a result, these network brokers are often found to be “paid more, receive more positive evaluations and recognition, and get promoted more quickly to senior positions” (Burt, 2012). Lastly, high self-monitors are better able to take advantage of cultural resources, as Erickson (1996) states that “the most useful overall cultural resource is variety plus a well-honed understanding of which genre to use in which context”—the latter of which constitutes high self-monitoring and the former of which is generated through self-monitoring behaviors.

Given that self-monitoring orientation has an observed impact on networks, we then wonder what impact social media may have on self-monitoring behaviors. Intuition may first lead us to presume that the advent of social media would lead to the diversification of ties and information. This is because access to others’ content via social media, especially that of our “weak ties”, may provide us with increased exposure to their information or opinions than we would have had otherwise. And perhaps then we can reap all the benefits of cultural resources that come with that diverse network.

However, I argue that this is not the case because of (1) context collapse and (2) platforms as biased “curators”. First, context collapse. Social media provides an unknown and potentially incomprehensible audience (boyd and Heer, 2006; Hogan, 2010). Furthermore, through the “intermingling of distinct networks”, social media leads to context collapse, as constituted by “the overlapping of role identities” (Davis and Jurgenson, 2014). This blurring of audiences and role identities is hypothesized to lead to behavior such as the lowest common denominator approach (Hogan, 2010), where individuals create content as though for the lowest common denominator of what is normatively acceptable. This is supported by Merton’s role-identity theories, which claim that “when roles are played to the same people, that is, when the constituents for multiple roles overlap, there is pressure to behave consistently across the roles” (Burt, 2012). Individuals’ content, then, must be consistent with their behaviors in all their circles/audiences. We can see how this is can function as a natural antithesis to self-monitoring, as individuals are no longer able to customize their behaviors and opinions as according to context. As such, it is not easy for individuals to inhabit structural roles or, moreover, “expand their networks to reach new structural holes”, as self-monitors do (Burt, 1012).

These structural bridges are potentially further lowered by curation via social media platforms. I use Hogan’s framework of the platform as the “curator”, and further argue that platforms function as biased curators. Social media platforms have their own goals, which often accentuate differences in agents’ structural positions—an example of this is on Facebook, where the “People You May Know” feature allowed for triadic closure to occur at higher rates than it would have otherwise (Malik and Pfeffer, 2016). As algorithms decide which content who gets to see, it is possible that these algorithms are instead inducing “echo chambers” where weak ties are made weaker and strong ties are made stronger. Individuals then have less of an opportunity to build their weak ties—major sources of cultural diversification and information— and function in those structural holes.

Overall, I hypothesize that participation online lowers self-monitoring behavior and thus the number of structural “bridges” that are typically found in social networks. Further examination would be required to determine exactly how these structural effects impact community-wide culture. However, given that self-monitoring behavior has such a large effects on social capital (Erikson, 1996) and achievement in different spheres (Burt, 2012; Mehra et al., 2001), one could imagine that the impact of this would be high, and most likely negative. This paper has provided an example of melding the fields of individual-level psychology and aggregate-level sociology by inspecting individual self-monitoring behaviors and structural correlates in social networks.

References

Burt, R. S. (2012). Network-related personality and the agency question: Multirole evidence from a virtual world. American Journal of Sociology, 118(3), 543-591.

Boyd, D., & Heer, J. (2006, January). Profiles as conversation: Networked identity performance on Friendster. In Proceedings of the 39th annual Hawaii international conference on system sciences (HICSS'06) (Vol. 3, pp. 59c-59c). IEEE.

Davis, J. L., & Jurgenson, N. (2014). Context collapse: Theorizing context collusions and collisions. Information, communication & society, 17(4), 476-485.

Hogan, B. 2010. “The Presentation of Self in the Age of Social Media: Distinguishing Performances and Exhibitions Online”. Bulletin of Science Technology and Society. 30(6): 377-386. https://doi.org/10.1177%2F0270467610385893 Malik, M.M., & Pfeffer, J. (2016). Identifying Platform Effects in Social Media Data. ICWSM.

Erickson, B. H. (1996). Culture, class, and connections. American Journal of Sociology, 102(1), 217–251.

Mehra, A., Kilduff, M., & Brass, D. J. (2001). The social networks of high and low self-monitors: Implications for workplace performance. Administrative Science Quarterly, 46(1), 121–146.


What’s New About The “New” Science of Networks

Network Science vs. Social Network Analysis

Networks have recently seen the emergence of network science, also dubbed the “new” science of networks, that overlaps with but is distinct from social network analysis. In this paper, I explore the distinction between network science and social network analysis as fields of study. I argue that the “new” science of networks is considered “new” because of (1) the focus on generalized network models, and (2) the availability of data and computational power. I classify this field of study as a branch of social physics, as per Bernard’s 1979 criteria. Lastly, I outline the problems each field faces, and specify how network science and social network analysis can work together most effectively to combat their individual weaknesses.

The field of network science was (and is) driven by interest by different disciplines in network structure and problems in their respective fields, notable for its “unprecedented degree of synthesis. . . across many disciplines” (Watts, 2004). Protein-protein interaction networks, food webs, transportation networks, firm co-ownership networks, collaboration networks, and the Internet interest biologists, computer scientists, physicists, economists, sociologists, and more (Watts, 2004). As a result of this multidisciplinary stake, network science focuses on generalized models of networks that can apply to instantiations of them across disciplines. This allows for the study of networks in one discipline to inform and further its study in another. Furthermore, note that networks are classified as complex systems which have emergent phenomena (Martin, 2010). It is thus often not possible to predict the collective behavior of a network from our knowledge of the individuals. It then follows that a nuanced understanding of each actor in a network (e.g. a neuron in biology) may not highly inform macro-level behavior, further backing this generalized approach.

The rise of big data and computational power has also had a large stake in network science’s trajectory. Finding network data—of transactions, or relationships—is easier with interactions being recorded on technology. The combination of this data and increased computational power allows for network scientists to uncover more patterns and to introduce more mathematical complexity to their models. Being able to analyze data from different sources both on a larger scale and more thoroughly also allows for stronger generalized models. Note here, however, that this data comes with a caveat: platform effects, where “the design and technical features of a given platform constrain, distort, and shape user behavior on that platform” (Malik, 2016). As Malik states, this poses “a serious methodological concern… observed behavior could be artifacts of platform design” (Malik, 2016). Though this data is readily available, then, scientists are pressed to question what exactly this data is representative of, what it is being skewed by, and its potential for generalizability beyond the given platform of study. Furthermore, “variation in technology adoption creates online social networks that differ systematically from the underlying social network” (Malik, 2016). This also poses difficulties for generalizability, as we must be careful to claim that online social networks are representative of the real world.

The field of network science speaks well to Bernard’s 1979 paper, “Why Are There No Social Physics?”. He argues here for a positivist social science, where one “formulate[s] laws which govern the evolution and formation of human groups” (Bernard, 1979). These rules are meant to reduce complex phenomena down to a simple set of laws, just as physics does. Furthermore, these rules are temporally constant, as “if we deduce a law which holds for an array of present cultures, then it is legitimate to assume that the law holds for an array of cultures through time” (Bernard, 1979). Network science does just that: with a focus on quantitative patterns, aims to generalize findings to rules for the structure. Furthermore, the “failure” to date of social physics that Bernard notes is primarily driven by the following cause: “much of what we see in our daily lives is plausibly describable as turbulence. How much of what anthropologists observe is, in fact, turbulence. . . The trouble is that, by definition, what we observe in field research is detail, not mean behavior” (Bernard, 1979). Forty years later, big data and computational power now enable network science to study the behavior of means and avoid this barrier.

The focus of network science on the generalizable and the analysis of aggregate data, then, leads the field to feel more mathematical, impersonal, and robust. Conversely, in social network analysis lives a greater focus on the individual and the qualitative specifics, dating back to sociograms created via personal interviews. This can give the impression of network science as more “scientific” than social network analysis, but I claim that this is not the case. In both fields, the same scientific method is being applied to better understand the object of study. Furthermore, via tuning of social science methodology, we have learned about how people think and “turn[ed] that knowledge into technologies” such as polls, marketing, and public health (Bernard, 2012). Both fields are technically scientific—it is the stereotypes about what we perceive science as being—namely, mathematical and impersonal— that skew some into thinking network science is more so.

I claim that social network analysis and network science can work complementarily. This is not an entirely alien idea: sociology, in fact, has a history of incorporating physics as network science now does, with early sociologists holding that “societies could be seen as systems of forces and energies that could be analysed in terms of their specific equilibrium conditions. . . This was understood as a system of interpersonal forces, such as pressure, influence and constraint, which are at work within social groups . . [They] used concepts of attraction and repulsion in social situations to model the networks of social relations built through interactions” (Scott, 2011). Network science helps pull social network analysis out of the hole of potentially observing only turbulence and noise as it moves to analyzing the aggregate. It also lends more of a positivist perspective to the structure of networks, providing general patterns and laws that govern structures regardless of their instantiation. This gives different disciplines the freedom to interpret generalized findings in the language of their own discipline. On the other hand, we can gain new insight by studying specific dynamics of certain systems rather than applying a general model—this is where social network analysis can help by expanding on the models that network scientists provide. Furthermore, social network analysists can ground network scientists’ mathematical abstraction in real thought processes and behaviors. For example: What does it really mean when we observe a high clustering coefficient? What does this tell us about the actors in the system? As Watts states, “physicists may be marvelous technicians, but they are mediocre sociologists. . . [sociology] must offer guidance in, for example, the interpretation of theoretical findings, particularly in the context of policy applications” (Watts, 2004). By definition, the impersonal nature of network science is necessary: when one is aiming to create generalized laws about networks, one cannot focus on the identity of the nodes, their agency, and reasons why we see the structural properties we do. However, if we ultimately want to apply our knowledge about networks we gain in network science to intervening upon and understanding the real world, social network analysis is where we will get the answers to questions that are salient for doing so.

References

Bernard, H. R., & Killworth, P. D. (1979). Why Are There No Social Physics? Journal of the Steward Anthropological Society 11.1: 33-58.

Bernard, H. Russell. (2012). "The science in social science." Proceedings of the National Academy of Sciences 109.51: 20796-20799.

Malik, M.M., & Pfeffer, J. (2016). Identifying Platform Effects in Social Media Data. ICWSM.

Martin, J. L. (2010). Life’s a beach but you’re an ant, and other unwelcome news for the sociology of culture. Poetics, 38, 228–243. http://doi.org/10.1016/j.poetic.2009.11.004

Scott, J. (2011). Social Physics and Social Networks. In The SAGE Handbook for Social Network Analysis, 55-64.

Watts, D. J. (2004). The “New” Science of Networks”. Annual Review of Sociology 30:243-270. https://doi.org/10.1146/annurev.soc.30.020404.104342.


Social Relationships as Comprising Networks

Social networks are often used as models for understanding and analyzing social systems on a large scale. However, our way of learning about social systems is often via narratives and individual lives. A natural question then arises: how do these macrostructures relate to microstructures that we experience on a personal basis? Namely, to what extent can we understand social relationships as comprising networks? In this paper, I argue that viewing networks through the lens of social relationships is a necessary, but not sufficient, condition for understanding and analyzing social networks. I first establish why this is necessary, and to what extent this perspective is helpful. I then note some limits to this worldview that support the insufficient argument, touching on theories of performativity and complexity science.

As social networks abstract people away to nodes and relationships away to edges, it is tempting to depersonalize these entities. A focus on social relationships as comprising social networks, then, can lend us a more personal and relational account of an impersonal graph, grounding this abstraction in emotion and observable interactions. Pachucki and Breiger (2010) make an argument for this shift towards the personal: “Narrative can serve to describe, construct, and transform a web of relationships. . . a shift toward narrative thinking involving actors, actions, and temporally ordered events departs from more traditional structural analysis methods”. They further argue that “the difference is not merely methodological, but epistemological and intimately connected with human agency. . . A view of social reality fundamentally based on narrative data shifts sociologists’ concerns away from variables to actors, away from regression-based statistical models to networks, and away from a variable-based conception of causality to narrative sequences” (Pachucki and Brieger, 2010). Thus we see that if we hold the view that social relationships comprise networks, we naturally inquire into the individual motives and behaviors that lead to edge formation, the dynamics of the creation and flux of the network over time. Namely, structural properties we observe in the network suddenly have the potential to be causally connected to individual action. If we understand these causal links, we are better able to induce global change on the network via individual or local intervention: one of the main purposes of studying these social networks in the fields of healthcare, social policy, and more.

Furthermore, by modeling the social world as networks comprised of social relationships, we circumvent some key weaknesses in typical models. Network models are able to account for synchronic contingencies, which general linear models (GLMs) cannot (Abbott, 1988). Furthermore, through virtue of disaggregation, a given attribute can have different effects on an actor depending on that actor’s attributes (including, for example, certain characteristics of the agent or edge properties) in a network model. This then allows for “multiple meanings [which are] central for sociological methodology” (Abbott, 1988). Lastly, by viewing social relationships as comprising networks, we are then able to incorporate sequential narratives as dynamics that unfold on the network, congruous with narrative history and qualitative sociology. One could argue that implicit in this model is its assumption of fixed entities (actors/nodes) and attributes (edge existence) that is one of the shortfalls of GLMs (Abbott, 1988). By regarding entities as fixed and only attributes as changing, the model is unable to account for entity change—and an assumption that these actors are static is not particularly grounded in reality. However, I postulate that we could account for this by ascribing predicted dynamic qualities of the actors to additional attributes rather than inherent to the entity.

The performativity thesis, found in economics and finance, provides a potential argument against this network model view. The performativity thesis is the claim that, once the model is “carried out into the world. . . [it] reformats and reorganizes the phenomena the models purport to describe” (Healy, 2015). Explicatively, “before its public appearance, the system did not behave in accordance with the model’s predictions, whereas subsequently it does” (Healy, 2015). The mechanism for this shift is via “tools implementing formal models of action-- “calculative devices” [which are] put in the hands of social agents by the model-builders or their representatives. These devices act as ‘cognitive prostheses’ that enable actors to accomplish calculative tasks previously beyond their reach, but which are required by the theoretical models. . . they allow users to do things they were unable to do before, such as easily see three or four degrees out of their social network. . .” (Healy, 2015). This is cited as an objection to the proposed view because it postulates that, if we view social systems in this way (i.e. social relationships as comprising networks), simply by virtue of viewing it in this way and disseminating tools that similarly act on these assumptions, our models become self-fulfilling prophecies rather than representations of the world as it naturally exists. Some evidence for this happening exists from Facebook networks and more (Healy, 2015).

I do not view this as a particularly strong objection for the following reason. The fact that we view the system converging to our model (in strong performativity) and/or acting in expected ways (in weak performativity) shows that our underlying assumptions of the model are correct and thus not artificially created rather than discovered. The incongruent part with reality, then, is access to these ‘cognitive prostheses’ which give access to information that actors may not have had before. Without our model’s integration into popular culture and the tools provided, people act upon their perceptions of their network rather than the real network—all we are doing, then, is aligning their perception with the real network. In my view, this does not undermine the model’s integrity. Furthermore: inevitably, awareness of the system by the actors in the system will change the behavior of the system. Arguably, change towards what we predicted from our models is a testament to its strength.

The criticism I view as strongest comes from complexity science, which cites social networks as examples of complex systems. These complex systems “exhibit novel behaviors at the system level that are very different from anything exhibited by any components, a phenomenon called emergence” (Bernabeu Auban et al., 2012). If we then view social networks as solely comprised by social actors and social relationships, we run the risk of distilling emergent behavior down to an individual level where it cannot always be explained, just as “our bodies, which can carry out system-level behaviors that cannot be observed in any of its individual cells” (Bernabeu Auban et al., 2012). Thus we see that social relationships are necessary constituents of social networks, but are not sufficient, i.e. explanations that distill emergent phenomena down to social relationships may not have full explanatory power. Martin argues against this complexity point by stating that “complexity . . . is not an attribute of environments or objects, but rather a relationship between minds and objects. . . I propose, then, that our well-founded belief that the riddles of the cultural and social world are too much for us to unravel points to our fundamental cognitive poverty”, presumably rather than the inherent complexity of the phenomenon (Martin, 2010). I counter this by stating that it does not metaphysically matter whether the system itself is complex; what matters is that, as we are able to understand the world (cognitive limitations and all), its constituent parts are not sufficient to describe the behavior of the whole system. Adhering to Martin’s definition: if complexity is a relationship between minds and objects, and our minds are constant, it is reasonable to regard this phenomenon as complex for all practical purposes.

I conclude that we cannot predict and understand network behavior solely through the lens of social relationships, but we cannot predict and understand it without them. The explication of the network model down to its micro-level instantiations is important, as “models. . . come to influence our actual construing of social reality” (Abbott, 1988). Thus, as we use models aim to operationalize dynamic social processes and inevitably simplify them, it is important that we fully understand the choices we are making when doing so and their implications.

References

Abbott, A. (1988). Transcending general linear reality. Sociological Theory, 169–186.

Bernabeu Auban, J., Moreno Martín, A., & Barton, C. M. (2012). Complex systems,social networks and the evolution of social complexity. In M. Berrocal, L. GarcíaSanjuán, & A. Gilman (Eds.), The Prehistory of Iberia: Debating Early Social Stratification and the State (pp. 23–37). New York: Routledge.

Healy, K. (2015). The performativity of networks. European Journal of Sociology, 56(2), 175–205.

Martin, J. L. (2010). Life’s a beach but you’re an ant, and other unwelcome news for the sociology of culture. Poetics, 38, 228–243. http://doi.org/10.1016/j.poetic.2009.11.004

Pachucki, M. A., & Breiger, R. L. (2010). Cultural Holes: Beyond Relationality in Social Networks and Culture. Annual Review of Sociology, 36(1), 205–224. http://doi.org/10.1146/annurev.soc.012809.102615


Personality and Triadic Closure

In social network analysis, the triad is often referred to as the “fundamental social unit that needs to be studied”, as per Georg Simmel, an early sociologist and structural theorist (Krackhardt and Handcock, 2006). The triad is defined as a set of three nodes, or agents, who have relationships to one another as represented by an edge or the lack thereof. This is somewhat counterintuitive, as we might think of the dyad—the relationship between two individuals—as being the most basic. However, Simmel (1902) argues that “it is almost irrelevant… what defines a relationship… Rather, [intimacy] is based on the structure, the panoply of demands and social dynamics that impinge on that dyad. And those demands are best understood by locating that dyad within its larger context, by finding the groups of people (of at least three persons) that the dyadic members belong to” (Krackhardt and Handcock, 2006). Furthermore, the triad constitutes a group, which “develop[s] an identity, a ‘super-individual unit’… In contrast, dyads by themselves do not reflect this transition to a larger-than-self unit” (Krackhardt and Handcock, 2006). Thus we see that the triad is the smallest “group” or structure that can give us insight into social networks.

Studies have also found that “the presence of local triadic configurations has implications for global network structure. Triangles of ties suggest redundancy in that any two actors in the triad are reachable by both a one-path and a two-path; they also suggest ‘closure’ in that the paths ‘close in’ on one another” (Kalish and Robins, 2006)—note also that this introduces the concept of “triadic closure”, wherein relationships come to exist between all three agents in a triad. Thus, by studying the frequency of closed triads in a social network, we can understand the overall connectivity of a network. This has important implications for the characteristics of the network and its behavior as a whole. For example, Granovetter (1973) postulates that “the more local bridges in a community [less triadic closure] and the greater their degree, the more cohesive the community and more capable of acting in concert”, ultimately acting as an example for “why some communities organize for common goals easily and effectively whereas others seem unable to mobilize resources, even against dire threats”.

So far we see that local triads can offer us insight into more global and macro structures in the social network. It then follows that, if we can understand what individual processes incite agents to form these triads in certain ways, we can effectively bridge the individual and the aggregate. As Granovetter (1973) states, “the analysis of processes in interpersonal networks provides the most fruitful micro-macro bridge”. In this paper, I attempt to do so by first discussing how individual traits of agents—namely, personality— have been shown to shape triadic structure. I then claim that there are three factors that restrict an individual’s agency over their structural position. Lastly, I propose a course of action for explicating the impact of these factors and touch on potential applications of having a robust micro-level theory of the formation of closed triads.

As we inquire into how triads are constructed, we must study the most basic level of a social relationship between two individuals: the social interaction. From first-hand knowledge, we can imagine many individual differences which would have an impact on what the ego searches from this interaction, as well as from her network as a whole. This leads us to postulate that psychological traits would play a role in a theoretical account of social networks.

Accordingly, multiple studies have found effects of personality traits on propensity towards triadic closure. Kalish and Robins (2006) find that an external locus of control (the quality of seeing oneself as “vulnerable to external forces”) is positively associated with an increase in weak network closure. They further find that individualists (those who “focus on being different from others, including others in their own ‘social group’”), people with an internal locus of control, and neurotics tend to inhabit networks with structural holes, i.e. unclosed triads. Fang et al. (2015) find that self-monitoring is predictive of indegree centrality and brokerage (the ego’s propensity to act as a bridge, i.e. have structural holes rather than closed triangles in their network).

Furthermore, Heider’s (1946) balance theory provides a possible psychological process which motivates triadic closure to occur. According to Heider, people are motivated to correct imbalances in order to avoid “dissonance-like outcomes” (Kalish and Robins, 2006). Put in the context of triadic closure, the theory then claims that “If strong ties A-B and A-C exist, and if B and C are aware of one

another, anything short of a positive tie would introduce a ‘psychological strain’” (Granovetter, 1973).

These studies paint the picture of the individual as having agency in whether or not they facilitate triadic closure in their networks, and, moreover, over which structural positions they tend to inhabit. However: in both Kalish and Robins (2006) and Fang et al. (2015), though the traits studied are statistically significant, their effect is marginal. As an example, in Fang et al., “the total amount of variance explained in these network outcomes [by personality] ranged between 3% and 5%”. Furthermore, Heider’s balance theory has also proven to not be as empirically supported as Simmel’s theory, which effectively claims that closed triads exist and continue to exist because they are structurally strong rather than because of any special motivation on behalf of the agents to create them (Krackhardt and Handcock, 2006). This undermines the psychological process behind the closing of triads and, by extension, the individual’s agency. Instead it supports a theory that essentially says that triads happen upon us and, because of their properties, stick around (though, do note the extremely small sample size of 17 in this study).

What, then, is restricting the individual’s agency to shape their network? I postulate three main factors: (1) cultural and social norms, (2) personality interactions, and (3) complicated webs of causality. First, many of these network studies make the same assumption as in Kalish and Robins (2006): that “individuals are in a sufficiently unconstrained social environment so that it is possible to choose or reject social partners”. I claim that this is an unrealistic assumption to have, as there may be both social norms and cultural customs about the extent to which one is allowed to reject a tie, to whom it is appropriate to offer one, etc. An individual is never truly unconstrained to choose and reject social partners when the formation of a tie is dependent upon the mutual sentiment of an alter. Second, I note that the psychological traits of the alter also play a role in how she perceives the ego. Thus it is a mix of the two’s traits that enables a tie to form or not, and thus a triangle to close or not. As Kalish and Robins (2006) aptly state, “In advancing claims about individual predispositions and social ties, one needs to be careful about reducing what is a relational process (at the least, dyadic) into a solely individualist explanation, viewed from the perspective of one partner only”. Thus we must note that the personality traits that affect an ego’s network do so only through interaction with alters’ personality traits. Lastly, the web of causality between personality traits and one’s social network is quite complex. Kalish and Robins (2006) touch on this by saying that they intend to focus on “several psychological traits that a considerable body of theoretical and empirical evidence suggests may be enduring and relatively stable from infancy onwards” and thus argue that they can “treat these predispositions as antecedents to network structure”. However, I argue that the feedback structure of networks and one’s personality traits is underestimated here, even with respect to theoretically “stable” personality traits (to name a few they study: neuroticism, self-monitoring, external vs. internal loci of control). If an agent shapes their social network in a certain way because of a slight trait (and its interactions with others’ traits, as per point (2)), the social pressures to conform to this role may reinforce that trait and thereafter enable the agent to inhabit a greater number of similar structural positions to that one than would have occurred otherwise.

To explicate this web of causality and fully understand the extent of an individual’s agency over their structural positions in social networks, what is needed is longitudinal data. The collection of periodic social network data alongside measured personality data would also give us insight into (1) the cultural and social norms mentioned, as we see the constraints every agent acts under despite their individual differences. It would also give us insight into (2) personality interactions, as we see how relationships form or do not form in different interactions rather than as a given via ego’s brute force personality traits.

As we improve our understanding of the micro-level, psychological processes that drive the formation of macro-level phenomena, we can create robust theories that can help us in many applications. For example, as Granovetter (1973) mentioned: the capacity of a body of people to mobilize together can potentially be traced back to their constituent’s tendency towards triadic closure. With a theory, we can understand why different bodies of people have different tendencies towards triadic closure, and thus howto intervene on a system to make a culture more cohesive. Answering this question would have applications in public health as well. For example, structural holes in American girl adolescents increase the odds of thinking about suicide (Bearman and Moody, 2002). Again, with a robust theory, we can more effectively intervene on a micro-level after seeing this macro-level phenomenon.

References

Fang, R., Landis, B., Zhang, Z., Anderson, M. H., Shaw, J. D., & Kilduff, M. (2015). Integrating Personality and Social Networks: A Meta-Analysis of Personality, Network Position, and Work Outcomes in Organizations. Organization Science, 26(4), 1243–1260. https://doi.org/10.1287/orsc.2015.0972

Granovetter, M. (1973). The Strength of Weak Ties. American Journal of Sociology, 78, 1360–1380.

Kalish, Y., & Robins, G. (2006). Psychological predispositions and network structure: The relationship between individual predispositions, structural holes and network closure. Social Networks, 28, 56–84.

Krackhardt, D., & Handcock, M. S. (2006). Heider vs Simmel: Emergent features in dynamic structures. In Network Analysis: Models, Issues, and New(pp. 14–27). Berlin: Springer-Verlag. Retrieved from http://www.springerlink.com/index/Q778536584LG31L1.pdf

Bearman, Peter S. and Moody, James, “Suicide and Friendships Among American Adolescents”, American Journal of Public Health 94, no. 1 (January 1, 2004): pp. 89-95.

Previous
Previous

Mathematical Contest in Modeling (HiMCM)