Social networks Flashcards
What are nodes & edges?
What are nodes?
* Entities aka actors
* When we’re talking about social networks, actors consist of persons (i.e. pupils, workers in a firm, etc)
What are edges?
* Ties between them
* Connections in form of i.e. friendships, collaboration, bullying, exploitation, etc.
But can be much more complex/mathematical: nodes could be words that are used in a language & edges the frequency of how often they co-occur
What are paths?
The different ways through a network that follow ties along certain nodes
Directed / indirected networks
- Directed = friendship (directed arrow connecting the nodes)
- Undirected = relatives (no sense to use direction)
Why study social networks?
- Empirically important: provide the base for a lot of different social processes, e.g. transmission of infection/disease or opinion dynamics (who is able to make his opinion the dominant one), polarization or cultural change - gender roles diffusing through population
- Socially fascinating: Immediately at the core of sociology, not about action theories of individuals but social structure right away (no identity issue, identity still important: norms, roles, etc) but with networks you’re already at the macro level + if studied over time, one can also look at the dynamics - how does network change
- Methodologically different: compared to standard regression models and normal data collection methods (i.e. Allbus survey), where data is collected on the individual level + also randomly selected (ideally no relation with each other) with network analysis it is tried relatively early on to retain social structure so it is examined much more direct manner
What kind of problems do we encounter when studying SN?
- Statistical problem later on as not randomly sampled - error is not IID (independently & identically) + a lot of confounders as it is very difficult to capture everything that goes on between the nodes one has to control for every relevant mechanism and maybe even look beyond bounded network, always the risk of some unobserved source of correlation that biases our estimate
- “Theory Gap” - Granovetter (1979):
- Rapidly expanding literature on “social networks” BUT where is theoretical underpinning? Most of them are constructed on theoretical vacuum -> network theories tend to be rather technical
- Flap & Völker 2013: base of network theory is just an orienting statement “the structure of social networks determines the action of network members”
- Some people even say it is only about methods, no place in sociological theory -> Kroneberg critique: wrong bc social mechanisms in networks analysis very important
Egocentric Network Analysis
Interest
Interested in collecting data on personal networks of individuals, i.e.support network of elderly people or new immigrants
Egocentric Network Analysis
theoretical framework
More generally, a version of social capital theory deals with ties that are of value to us/ help us to attain our goals (Flap & Völker 2013)
2 hypothesis which often guide egocentric network analysis as a theoretical framework
- Social resource hypothesis: People better equipped with social capital will be better able to attain their goals
- Investment hypothesis: Therefore, people are ready to invest in social capital according to its instrumental value in producing their ends
Egocentric Network Analysis
What do we do in ENA?
Try to reconstruct the network of particular respondent’s ego (can be random sampled)
(1) Name-Generator
- Respondents are asked to give names of people who are most important to them, who they know, know by sight, etc. -> hierarchical levels of acquaintanceship
- Not every survey has these kinds of questions bc they take a lot of time (Side fact: study of close ties in the US, evidence suggested that people had fewer ties, but one of the reasons was that the respondents did not want to engage in this time-intensive activity again so they just told them about less people)
- Typical size around 50 (close-and-active network) but sometimes also valuable to look even beyond personal network (150), especially if interested in social capital: it is those weak ties where you get interesting information, usually in support group information u get is redundant
(2) Position-Generator
- You ask: Think about lawyer, physician, data scientist, whatever - do you know someone in this position? much more effective, especially when it comes to Social Capital
Sociocentric Network Analysis
Interest
- Data on all ties within a bounded setting i.e. gossip ties in company or friendship ties in school class (egocentric approach cannot tell you who is friends with each other, only who you consider your friends or your view on these bonds)
- Egos are not sampled randomly, but one tries to capture the whole bounded network
Sociocentric Network Analysis
Problems
- Boundary problem: What is the relevant setting? Where does is end? (i.e. friendships only within one class or should we look at parallel classes as well? The one below? The one above? Only academic context or neighborhood?)
- Non-response: At least 80% of network needs to participate otherwise too much missing
Sociocentric Network Analysis
Why is complete/socio-network analysis so promising?
- Main advantage: information on not only who is selected as network partner by whom BUT ALSO could have a tie but don’t - the absence of ties, ties that did not form reconstruct choices that underlie the social network
- Example: no tie bc no lawyer present in the network or no tie although there are lawyers present - completely different explanations - egocentric approach could not tell
Basic mechanisms of tie formation
Stadtfeld/Amati (2021)
- Transitivity = Triadic closure: a third tie builds between the first and last node (i.e. sharing a friend)
- Popularity = likely to receive more ties when already has a lot of ties
- Activity vs Attraction (opposites)
- Homophily
* selection mechanism based on certain shared attributes i.e. both are boys, sociologists which makes them share a tie, they assume that it is more rewarding (“gleiches und gleiches gesellt sich gern”)
* of great interest bc related to social cohesion & segregation, strong homophily = people prefer similar people fragmentation of society
* not an iron law, also heterophily possible, more likely to connect with someone who is different (exchange relations, better if specialized in different areas, i.e. “classical family”, men = labor, women = children)
* sometimes difficult to tell whether homophily (friends bc we are both criminal) or social influence (friends before and then I became criminal bc you are) longitudinal data to tell apart
Basic mechanisms of tie formation
Gremmen et al. 2017
Why network evolution interesting?
- Intrinsic interest: i.e. Is social cohesion declining? Or Do networks become more segregated?
- Base for causal inference: “why does it look how it looks?” - understanding the underlying mechanisms of tie formation
How do we analyze network evolution?
- Most prominent approach = Stochastic actor-oriented models (SAOM) for dynamic networks (Snijders et al. 2010) aka longitudinal data
- Basic idea: observing a model at two different time points (i.e. school class friendships) - going there maybe every 3 months - how do we get from state A to state B?
- Goal: model the evolution of the network between these time points
- Assumptions:
1) Really requires longitudinal data
2) Time is modelles as continuous-time process, changes can happen at any time
3) Individuals (nodes) decide whether to form a tie or not (behavioral model)
4) Only 1 tie change at a time –> big changes are due to small changes
5) some changes will be random
SAOMs - Actor-oriented element:
- reconstructs the choices that actors make, every actor can make a choice if it’s his/her turn whether to
1. dissolve tie (i.e. unfriend a certain person)
2. create a new tie (i.e. new friend)
3. do nothing
(then another actor is chosen and so on)
SAOMs
why quantitative approach
- How attractive is it to reciprocate a friendship tie with someone else? Or to befriend a friend’s friend? –> then we infer from the data that there is a tendency for reciprocation, triadic closures, etc.
- Not qualitative by asking but quantitative by reconstructing the mechanisms
SAOMS - The objective (utility) function
- models the attractiveness of network states x to actor i that are reachable in one step from the current network
- modelled like multinomial logistic regression (not a binary choice but the choice to form/un-form ties with everyone in the network)
- Parameters ßk express whether a feature of i’s neighborhood is desired or adverted, i.e. I like to have many reciprocated friendship ties or I like to have close triades -> ßk will be positive; if you don’t like to form those -> ßk negative = a way to estimate these tie-formation tendencies from the data
also possible to capture homophily: No. of friends who behave violently -> If violent class mates are more likely to be nominated as friends (over time) -> seems that students befriend violent people (but could also be the other way around)
Example: group-specific attractiveness of violent peers as friends
why difficult
Sometimes questions like “Do boys prefer rough, violent ties?” arise
1) so, objective function has to account for whether alter (different classmate) engages in violence and whether they are desired/avoided as friends
2) And even if tie formed from stage 1 to 2:
attractive du eto violent behavior or different reason? i.e. blue node = common friend (blue node), homophily (both red)
- If interested as a researcher to answer this question -> statistically control for other tendencies (needs to include all the other relevant mechanisms)
- This means: Causal interpretation requires strong assumption of unconfoundedness -> evidence is correlational data!
How to differentiate: Selection (Homophily) or Influence?
- Method: Not only a model for stochastic actor orientation but also one for how behavior changes over time, like two regression equation, one is the objective one (do I connect with others as friend) but also another one (do I start smoking)
- If u observe a correlation between network and individual characteristics -> at least 2 competing mechanisms:
1. Social influence: Network leads to behavioral alignment
2. Homophily: Actors’ behavior leads to network alignment - great ambition of these models is to tell those apart –> Caution: Subscribe to a lot of assumptions with these models as validity of results obtained by modeling is conditional on having all the relevant variables inside model specification
Multilevel network analysis
interest & advantages
- Multiple complete networks, i.e. not only interested in one school but 1000 schools
- Advantage of multilevel:
1. Allows scholars to study how characteristics of contexts may affect the operation of particular tie-formation mechanisms
2. Students of different ethnic groups, when is ethnic segregation stronger, what are the contextual characteristics of schools, also implying, how can we change those characteristics so that there is less segregation - BUT sometimes, also important to just look at one network bc intrinsically extremely important (i.e. trade relationships of nations)
Theoretical background of study by Small & Sukhu
- How do people acquire social resources? How and when do persons invest in others? What is the exact mechanism underlying these investments? (might there be different ways than looking at it as of instrumental value?)
- Standard view: Perry & Pescosolido 2010: “Functional specificity” hypothesis:
“Individuals engage in selective and purposive activation of ties”, they “engage in problem- or task-specific activation of social network ties, evaluating who in their networks in most willing and able to fulfill a particular need for support or companionship” (Perry & Pescosolido 2010: 356) –> u have a broken washing machine, so u wonder is there someone in the neighborhood available to help u fix the problem - 3 step process:
1. Seek: I need support, rather then fixinf it on my own and making it perhaps worse, decision to seek support
2. Select: Who will be best and willing, selecting some alter in your network
3. Activate: Call or ask the person
Critism of Small and Sukhu
- Not always so deliberate of a choice, maybe just someone who was available/accessible (i.e. if the printer broke u don’t call the printer company but rather talk to the next best available maybe even just a random person in the room)
- Could also be that one decides to not do anything about it and just leave the room in a frustrated manner, but someone else just enters it and so u think hm ok maybe I just ask him/her so not a consecutive process but almost simultaneously, not distinct decisions but opportunity created the idea (if greater effort, probably less spontaneous)
- Also, iterative process, not a one-shot process, maybe need to ask a couple of people
- One-sided view from the perspective of support seeker, but could also be analyzed from perspective of support giver (i.e. self-initiation, etc)
Empirical study of Small & Sukhu 2016
Setup
- Survey of 2000 college students across US, which they asked how they sought help around three forms of everyday social support. It asked them consider the last time they sought help with a
1. Mathematics or economics problem
2. Paper they were writing
3. and with a roommate problem - Over the previous year, 71.2% had sought help with a mathematics or economics problem; 57.6%, with a paper; 42.1%, with a roommate. The subset of respondents who sought help on at least one of these problems constitutes our analytical sample.