Spreading: fundamental dynamics
- behavior
- belief
- information
- rumors
- innovations
Social psychology: How people influence each other. How much do we influence each other?
Asch's conformity experiment
shows how strong social influence can be
group conformity
3 or more people
people will deny what they see, and conform to group behavior (pressure)
informational conformity - convinced the group is right
normative conformity - apprehension that the group will disapprove to deviancy
Controlled experiments
- empirical studies
- (many studies)
- randomized assignment
- carefully study how we influence each other
It is very difficult to conclude that something is spreading on social network based on observational studies (Cosma Shalizi).
Examples
- Music Lab
- Election
Causes of Similarity and clustering
- induction
- homophily (selection)
- contagion (influence)
- confounding
How can you determine a person's behavior in a social network? Is it due to social influence, homophily or something else? Is it even possible to determine? (The general argument is: probably not.)
- social contagion/influence
- manifest homophily (observable characteristics)
- latent homophily (unobservable characteristics)
The challenge of social research is to determine which mechanism is responsible for a person's behavior.
It is not possible in an observational study.
- social contagion/influence
- biological contagion
- manifest homophily (on the observed characteristic of interest)
- secondary homophily (on a different yet observed characteristic)
- latent homophily (on an unobserved characteristic)
- common external causation
some are really difficult to distinguish social contagion from other reasons or other phenomena
some can be controlled using statistical models
if you cannot control, then it is very hard to get to causation
just having observational data makes it very difficult to draw any conclusion on causal effects
How to draw causality?
- you have a system (or person)
- there is treatment (some intervention)
- observe a state transition occurs in the system
Did the treatment cause the system state change or not? To test this, you need to go back in time and do another treatment (or absence of treatment) and see what happens.
Randomized Experiment
- basic idea
- you have many similar systems (or people)
- randomly assign to groups
- apply a different treatment to each group
- identify whether we get change we expect from certain treatment
- if yes, then it is safe to conclude that "a" influenced the change of the system state and "b" did not
Example:
- clinical trial for drugs
- randomly assign patients into two groups
- one group is given the actual drug
- the other group is given a placebo
- measure how the two groups differ from each other
Key:
- how random is the group assignment
- whether everything else is controlled, making it an equal setting
- in clinical trials, double blind testing is common in order to prevent any small leakage of information and assignment to the patient groups
The same approach is applied for measuring social influence or other network phenomena. We want to do randomized experiments in order to really draw causal conclusion.
homophily
from Ancient Greek ὁμοῦ (homou, "together") and Greek φιλία (philia, "friendship") is the tendency of individuals to associate and bond with similar others, as in the proverb "birds of a feather flock together".
contagion |kənˈtājən|
noun
the communication of disease (spreading) from one person to another by close contact: the rooms held no risk of contagion.
- a disease spread by the close contact of one person to another.
- the spreading of a harmful idea or practice: the contagion of disgrace.
causation |kôˈzāSHən|
noun
the action of causing something: investigating the role of nitrate in the causation of cancer.
- the relationship between cause and effect; causality.
confound |kənˈfound|
verb [ with obj. ]
- cause surprise or confusion in (someone), especially by acting against their expectations: the inflation figure confounded economic analysts.
- prove (a theory, expectation, or prediction) wrong: the rise in prices confounded expectations.
- defeat (a plan, aim, or hope): we will confound these tactics by the pressure groups.
- archaic overthrow (an enemy).
- (often be confounded with) mix up (something) with something else so that the individual elements become difficult to distinguish: 'nuke' is now a cooking technique, as microwave radiation is confounded with nuclear radiation.
manifest 1 |ˈmanəˌfest|
adjective
clear or obvious to the eye or mind: the system's manifest failings.
verb [ with obj. ]
display or show (a quality or feeling) by one's acts or appearance; demonstrate: Ray manifested signs of severe depression.
latent |ˈlātnt|
adjective
(of a quality or state) existing but not yet developed or manifest; hidden; concealed: discovering her latent talent for diplomacy.
-philous
Word Origin
- a combining form meaning “liking,” “having an affinity for” that specified by the initial element: dendrophilous.
| Question | Answer |
|---|---|
| In the obesity study, Geographical distance rather than social distance seemed to play a more important role in the spread of obesity. | False |
| In the obesity study, clusters of obese persons were found on the network, but these clusters didn't appear to be formed just from obese people making new social connections with other obese people. | True |
| The main take-away from the Shalizi paper was that teasing apart homophily and contagion is difficult in observational studies. | True |
| From the Shalizi paper, it was established that in a social network, a homophilous trait that is not a cause of a behavior won't be able to predict the behavioral outcome. | False |
| From the music lab paper, they found that social influence actually increased the unpredictability of song success. | True |
| In the Facebook voting study, people who saw their friend's faces in the message were more likely to self-report voting. | True |
| In the Facebook voting study, weak-ties rather than strong ties had more influence in determining whether friends voted. | False |
| What was the name of the famous social psychology experiment introduced in the last week? This experiment tests whether a subject can provide a (obviously) correct answer when there are several other participants deliberately telling a wrong answer. | Asch's conformity experiment |
| The experiment mentioned above demonstrated that | Correct: people's opinion can be strongly influenced by other's opinions Incorrect: it is our innate nature to conform to the majority. -> We don't know from the experiment whether it's innate or cultural. Also claiming this way is way too general. |
| “If your friend Joey jumped off a bridge, would you jump too?” Ian answered "yes". So you observed a social phenomenon that looks like a social contagion. Among the followings, which is NOT the possible reason behind your observation? |
None of these: 1. Biological contagion (parasites that induces similar behaviors) 2. Manifest homophily ("they are friends because they are similar") 3. Random fluctuation -> Any result can come from random fluctuation (noise) in the data. You should eliminate this possibility through statistical tests (which is still not perfect in many cases). 4. Social contagion - something is actually spreading |
| What is the main benefit of randomized controlled experiments? | It allows us to identify and infer causal relationships, which is almost impossible in observational studies. |
| Which factor among the followings makes it really difficult to identify social contagion? | Latent homophily -> Latent homophily is "latent", which means you cannot see it and thus cannot control for it in your model. |