Search

We investigate whether randomized first contact opportunities among first-year undergraduate students explain similarity in academic outcomes one year later.

We will examine perceived information and misinformation networks of a whole community of students participating in the Swiss StudentLife Study during the 2020 COVID-19 pandemic. We will relate these to other interpersonal networks, perceived norms on measures slowing the virus' spread, and attitudes towards multiple life and policy changes, all within the community

Since the 1950s, social scientists have been studying the impact of meeting opportunities in educational settings. They have confirmed through experimental research designs that randomized seating, housing of students, and overlap in their courses can have a strong impact on the development of friendships between them [1, 2, 3]. At the same time, scholars have argued for the importance of peer effects in educational settings and its impact on academic outcomes and well-being [4, 5]. If both mechanisms operate simultaneously, a network intervention in an educational setting can potentially induce social ties through which peer effects occur. When the network intervention follows a strict experimental design, it can provide causal evidence for peer effects in education. [1] L. Festinger, S. Schachter, and K. Back, Social pressures in informal groups; a study of human factors in housing. Harper, 1950. [2] R. L. Moreland and S. R. Beach, “Exposure effects in the classroom: The development of affinity among students,” Journal of Experimental Social Psychology, pp. 108–118, 1992. [3] J. Brockner and W. C. Swap, “Effects of repeated exposure and attitudinal similarity on self-disclosure and interpersonal attraction.,” Journal of Personality and Social Psychology, vol. 33, no. 5, p. 531, 1976. [4] A. Lomi, T. A. Snijders, C. E. Steglich, and V. J. Torl´o, “Why are some more peer than others? Evidence from a longitudinal study of social networks and individual academic performance,” Social Science Research, vol. 40, no. 6, pp. 1506–1520, 2011. [5] B. Sacerdote, “Peer effects with random assignment: Results for dartmouth roommates,” The Quarterly journal of economics, vol. 116, no. 2, pp. 681–704, 2001.

Analyses of the Swiss StudentLife data during parts of the COVID-19 crisis

Social influence, which we consider as the “change in an individual’s thoughts, feelings, attitudes, or behaviors that results from interacting with another individual or a group” (Rashotte, 2007, p. 4426), underpins many of the greatest problems we face in modern society and our ability to deal with them. Spread of misinformation and conspiracy theories, as well as increasingly polarized political parties, narratives, and elections, have impacted our ability to cope with global crises such as the Covid pandemic and climate change, and compromised trust in institutions. Understanding how and why these beliefs and behaviors spread through the process of social influence is a key step in addressing these problems. There exists a long tradition of literature analyzing the effect of network structure on social influence (sometimes also referred to as diffusion, contagion, spreading dynamics or network flow; Granovetter, 1973, Albert & Barabasi, 2002, Watts & Strogatz, 1998, Valente, 1996, Centola, 2018). Such work provides insights into why certain network structures are more suitable for diffusion, while others impede the adoption process. But structure alone cannot determine the pattern of behavior- or information-flow in networks. If it could, we would expect to witness identical cascade patterns for different pieces of information or behaviors within a given network. Here we argue that before we can understand the variable ways in which influence dynamics unfold across various networks, we must first understand the variable ways in which influence dynamics can unfold within a given network of individuals. Social Influence Processes In this direction, there have already been some key contributions to the literature, aiming to better describe processes of social influence, and enable modelling of different diffusion patterns within a given network. The introduction of a theoretical distinction between simple and complex contagions (Centola, 2018), and more broadly, the development of threshold models (Granovetter, 1978; Valente, 1996) have provided important pointers in understanding how we can analyze and think about why different outcomes may spread differently within the same network. Complex contagion theory introduces the idea that while some diffusion processes are “simple” (spread from a singular contact with an “infected ” other), others are “complex” (requiring some degree of social reinforcement from multiple “infected” contacts simultaneously before adoption). A classic example (Centola, 2018) is the different spreading patterns observed for sexually transmitted diseases (which require sexual contact with only one infected person) vs those of preventative measures to combat the spread of exactly these diseases (which, perhaps due to social stigma etc., may require reinforcement from multiple members of one’s network before adoption). More broadly, threshold models introduce the notion of “thresholds of activation” (minimum number or fraction of people in your network who must adopt in order for you to also adopt). These models are convenient because their abstraction allows us to introduce threshold distributions across the network (i.e., different thresholds of activation for different nodes), which likely vary depending on the specific nature of the topic, the focal individual, and the specific roles that those influencing us in our network play. These varying distributions (even across a fixed set of actors) can conveniently account for different diffusion processes unfolding differently even over the same set of actors. However, such models also pose challenges in an empirical setting and from a more sociological perspective. Firstly, it is not clear how threshold data can be collected meaningfully in the first place. Moreover, the model tells us nothing about the social processes and dynamics that lead to a specific threshold distribution across the network, nor the social dynamics that lead to varying distributions depending on contextual factors. On the other hand, there exists a body of empirical research which has aimed to estimate models that are specific to various contexts. Within this framework, effects can vary in direction and magnitude depending on the specific context to which models are estimated. Such work often aims to model social influence processes by analyzing the co-evolution of an outcome variable with a network variable. For example, Snijders et al (2007) studied alcohol consumption behavior (outcome variable) and how it is diffused on a friendship network (network variable). After estimating a model, they found significant evidence for an alcohol similarity effect, as well as a peer influence effect. They noted no significant positive tendency to drinking behavior in general. Meanwhile, Knecht et al. (2010) conducted a similar study on delinquency behavior (outcome variable) within a friendship network (network variable). They found evidence for a delinquency similarity effect, but no significant peer influence effect. Moreover, they observed an overall negative tendency towards delinquency, in contrast to the Snijder et al. study. Both studies also found evidence for common endogenous network effects, such as transitivity and reciprocity. But within other network settings (e.g., not in a school friendship network), studies often observe other patterns of endogenous effects. For example, in networks of advice within professional settings, we often also see a negative three-cycle effect (i.e., a tendency against three-cycle structures; e.g., Agneessens & Wittak, 2012), which can be a sign of hierarchical structure (Snijders et al., 2010). Such studies provide key insights into micro-level processes that occur within social networks, and allow researchers to capture the varying nature of different contexts. However, they still leave open certain questions. For instance, such work is often focused on modelling social influence through the lens of single issues within a given social context. This “single-lens” view makes it difficult to extract meaningful insights about why different diffusion cascades might occur within the same group of people. At the same time, a lot of research has focused on influence processes in friendship networks (e.g., Snijders et al. (2007); Knecht et al, (2010); Mercken et al (2010)). However, we argue that assuming that the social influence process of a particular outcome is best described by a general network measure, such as friendship, may be a limited view on people’s complex patterns of social relationships and the influence processes operating on these relationships. Indeed, it is well-documented in literature that we often fail to find evidence of social influence processes (Stadtfeld et al, 2010). One reason for this is a mismatch in amount of data we have related to ties vs amount of data we have related to the specific nodal attribute (~N2 vs ~N observations). But perhaps another factor may be that we are looking for influence in the wrong places by focusing on these general networks (such as friendship), which may be too broad and non-specific for a given diffusion process. For instance, we may expect to find stronger evidence of influence on drinking behavior occurring on a “drinking / socializing” network rather than on a generic friendship network. To address this, we make use of the multiplexity of social relationships in our study, using it as a proxy for outcome-related interactions (1). We argue that a more outcome-related type of network measure would be superior to a general network measure, such as friendship. Specifically, we argue that outcome-related interactions mediate the social influence processes that are observed on general networks (e.g., friendship), by facilitating discussions and activities related to the outcome itself. For example, we expect that someone’s studying hours (outcome measure) are more likely to be influenced by those interaction partners with whom they regularly study (outcome-related network measure) than, for instance, by those with whom they are friends with, or play sports or socialize (non-outcome-related network measures). In so doing, we implicitly assume that there is a tendency to discuss, for example, study-related topics with those with whom we study, and that these discussions facilitate social influence processes related to studying. Furthermore, we contribute to existing literature by studying social influence through a multi-outcome lens, rather than restricting to a single outcome. We believe that such an “outcome-specific lens" will help us draw more meaningful and general insights into the social processes that facilitate social influence. Hence, the primary focus of this article is to challenge the assumption that social influence processes occur within general networks (such as friendships), and ask whether we instead find evidence that these processes occur mostly on specific types of networks that are capturing with whom one has the most outcome-related interactions? Homophilic Selection Processes So far, we have only discussed social influence processes. However, when studying social influence, a closely related process requires (equally) careful consideration: the social selection process. Previous literature has highlighted the need to distinguish between these two distinct processes, which can both lead to an overrepresentation of similar behavior among those who share ties (Shalizi & Thomas, 2011; Steglich, Snijders & Pearson, 2010). Selection is the tendency to modify one’s network according to certain specifications (e.g., to become friends with those who are popular, to become friends with those who are similar along some dimension etc.). A particularly well-documented challenge in social network analysis literature is that of distinguishing effects of social influence on a certain outcome from effects of homophilic selection on the same outcome (Steglich et al., 2010), where homophilic selection is defined as the tendency to modify one’s network such that one’s behavior is more similar to those with whom one is tied (e.g., by dissolving ties with those who are less similar, and creating ties with those who are more similar on a given dimension). Because homophilic selection and influence are so deeply intertwined, another focus of this study is to uncover patterns of homophilic selection occurring on portions of the network in which the relevant outcome is more salient, and not in other portions of the network where it is less salient. This would point to the importance of carefully considering the specific social context in which these social processes are likely to be occurring. The Current Study To test for homophilic selection and influence processes, we will first cross-sectionally analyze networks for over-representation of dyads with similar outcomes. As a second step, we will conduct a temporal analysis which will uncover whether such processes can explain any over-representation of dyads with similar outcomes in the relevant networks. In addition to exploratory and descriptive analyses, we will use a stochastic actor-oriented approach to model the co-evolution of the relevant networks (studying, leisure, social support) with the relevant outcome variables (studying hours, leisure activities and time, mental health & psychotherapeutic treatment). Since we expect these networks to have a relatively high degree of overlap with each other, as well as with the friendship network, we will conduct further analyses to understand whether there is evidence for effects on alternative networks (in particular, friendship, which is widely used as a general network variable in many studies). If so, we will analyze how these compare in strength vs the hypothesized networks, and whether these can be explained through, for example, a mediation analysis. We will draw from the unique longitudinal dataset of the Swiss StudentLife study (Stadtfeld et al., 2019; Vörös et al., 2021), collected across three cohorts of students (Cohort I, N1 = 226; Cohort II, N2 = 244; Cohort III, N3 = 652). We will test for homophilic selection and influence processes on three sets of key outcomes which are among the most relevant within a student setting (studying outcomes, leisure activity outcomes, mental health outcomes), testing for homophilic selection and social influence where we expect these outcomes to be highly jointly salient for those who share a specific type of tie (e.g., within studying network, leisure activity network, social support network respectively), and not otherwise. To test for mediation, we will also test for influence processes on friendship networks. We expect to find evidence of selection and influence only within networks that are closely contextually related to the specific outcome (e.g., studying hours within networks of study collaboration). We argue that these results would be in line with literature on selective disclosure (Cowan and Baldassarri, 2018) and theory related to social foci / social circles (Feld, 1981; Simmel, 1955; Block, 2018) (2). The study aims to contribute to social theory and a body of empirical insights, by challenging the assumption that uniplex networks such as friendships are sufficient to understand selection and influence across outcomes, arguing instead that attention should be paid to the specific social contexts in which selection and influence are likely to occur. (1) Multiplexity is the notion that there can be different types of ties, and that the same person can be tied to us via more than one of these types of ties (e.g., someone can be a colleague and a friend, which is distinct to them just being a friend or just a colleague; Gondal, 2022). (2) Selective disclosure is the tendency of people to selectively disclose their opinions and thoughts on certain topics to others, for instance in anticipation of potential conflict (e.g., perceived disagreement on political topics) or simply because the topic is not relevant to the context in which a relationship is embedded (e.g., latest developments in one’s industry may not be relevant to a close friend who is not working in the same industry). Social foci literature theorizes that social interactions are arranged around certain “foci”, which could be shared interests, activities, institutions etc., which provide the opportunity for social ties to form (Feld, 1981; Small & Adler, 2019; Rivera et al., 2010).

Remembering and recalling social actors is a crucial cognitive function. The importance of the awareness and recall of the own social world is shown in different contexts, such as decision-making through social sampling (Galesic et al., 2018) or social status perception (Krackhardt, 1990). Despite its key role, the underlying memory structure and its search and recall processes are still unclear. It is assumed that the social memory as a mental representation of the individual’s highly complex social environment is not activated randomly but follows its own structure based on rules and patterns (Brashears, 2013; Smith et al., 2020). These rules and patterns are intended to allow access to the social environment in a simplified and efficient manner. Although it has an essential meaning for daily social life, the specification of these rules and patterns people use to search and recall their social environment remains unclear. Reviewing the literature on this question reveals different approaches, mainly focusing on the general recall of persons the participants know without any constraints. This research has indeed shown that specific memory patterns exist, for example, that individuals may recall alters following various aspects such as frequency of contact, closeness, and social groups (see Brewer, 1995; Brewer, 2000; Brewer et al., 2005, Hills & Pachur, 2012; Brasheas, 2013; Brashears & Quintaine, 2015, Marin, 2004). The most critical structural characteristics influencing the search and recall of the social memory seem to be social groups/contexts (Todd & Hills, 2020; Töpfer & Hollstein, 2021), frequency of contact, and social proximity (Hills & Pachur, 2012; Brewer, 2000). Nevertheless, additional aspects seem important for different contexts and research designs. The current study aims to provide an overview of various recall patterns discussed in the literature and test the extent to which they are relevant for recalling friendship relations. Friendship relations have been argued to be emotionally significant and role-related relations. As one of the most important relations for humans, friendship has an important role in the social environment of an individual (Kitts & Leal, 2021), and therefore its representation in social memory is highly relevant. The setting of the study is a complete network data set, which enables the possibility to understand better how the recall and search through social memory work in a more functional context in which specific relations are the recall aim, and the recall is constrained to the complete network. Moreover, it opens the possibility of including the perceptions of others as variables. Furthermore, we want to evaluate to what extent the patterns shown are consistent with existing cognitive theories about the mental organization of social relations. On the one hand, the findings may help to understand what shapes the structure, search, and recall in the social memory in the case of a specific recall task. On the other hand, these patterns provide evidence for the fundamental social processes that affect individuals` cognitive organization and awareness of the social networks they are embedded in.

This study investigates change in students’ social networks and mental health at the time of the COVID-19 crisis in April 2020. We surveyed multiple dimensions of social networks (pleasant interaction, friendship, social support, co-studying) and mental health indicators (depression, anxiety, stress, loneliness) before and during the crisis among Swiss undergraduate students (N=212). We find that interaction and co-studying networks had become sparser, and more students were studying alone. Furthermore, students’ levels of stress, anxiety, loneliness, and depressive symptoms got worse. Stressors shifted from fears of missing out on social life to worries about health, family, friends, and their future. Exploratory analyses suggest that COVID-19 specific worries, isolation in social networks, lack of interaction and emotional support, and physical isolation were associated with negative mental health trajectories. The results offer starting points to identify and support students at higher risk of social isolation and negative psychological effects during the COVID-19 pandemic.

Supplemental material, DS_10.1177_0038040718824095 for The Social Pipeline by Isabel J. Raabe, Zsófia Boda and Christoph Stadtfeld in Sociology of Education