Homophily: Bringing Surrounding

Contexts into Network Analysis

— Examine additional processes (to triadic closure) that
affect the formation of links in the network.

— Surrounding contexts: factors that exist outside the nodes

and edges of a network.

— Represent the contexts together with the network in a

common framework.
Homophily (i.e. “love of the same”)
— Homophily principle: we tend to have similar
characteristics with our friends. People love those
who are like
themselves.
Similarity begets
friendship.
“birds of a feather flock together”
— Expression appears in the 16th century, a literal
translation of Plato's Republic.
— People of similar character, background, or taste tend to
congregate or associate with one another.
Homophily in Networks
— Links in a social network tend to connect people who are
similar to one another.
— basic notions governing the structure of social networks

— Its role in modern sociology by influential work of

Lazarsfeld and Merton in the 1950s.
— Paul Lazarsfeld and Robert K. Merton. 1954. Friendship as a
social process: A substantive and methodological analysis.
In Freedom and Control in Modern Society, pages 18–66.
Example

Social network from a town’s middle school and high school (students of
different races drawn as differently colored circles).
2 divisions:
• one based on race and
• the other based on friendships in the middle and high schools
Homophily vs. Triadic Closure
for Link Formation
— With triadic closure:
— –a new link is added for reasons that are intrinsic to the
network (need not look beyond the network)
— –Ex: a friendship that forms because two people are
introduced through a common friend
— With homophily:
— a new link is added for reasons that are beyond the
network (at the contextual factors)
— Ex: a friendship that forms because two people attend the
same school or work for the same company
Homophily vs. Triadic Closure
for Link Formation
— Strong interactions between intrinsic and contextual effects
— Both operating concurrently
— Triadic closure (intrinsic mechanism):
— B and C have a common friend A
— B and C have increased opportunities to meet
— Homophily (contextual mechanism):
— B and C are each likely to be similar to A in a number of
dimensions
— also possibly similar to each other as well
— Most links arise from a combination of several mechanisms
— difficult to attribute any individual link to a single mechanism
 Measuring Homophily
Given a characteristic (like
race, or age), how to test if a
network exhibits homophily
according to it?

friendship network:
–Exhibits homophily by
gender?
–Boys tend to be friends with
boys, and girls tend to be
friendship network of a (hypothetical)
friends with girls. classroom: shaded nodes are girls and the six
–Cross-gender edges exist. unshaded nodes are boys
Measuring Homophily
— Q: what would it mean for
a network not to exhibit
homophily by gender?
— A number of cross-gender
edges not very different
from randomly assigning
each node a gender
— according to the gender
balance in the original
network
Measuring Homophily
— p, the probability (fraction) of males
— q = 1-p, the probability (fraction) of females
— For a given edge:
— Homophily:
— Prob(both ends male) = p*p
— Prob(both ends female) = q*q
— Cross gender:
— Prob(ends male and female) = 2*p*q
— Homophily Test: If the fraction of cross-gender edges is
significantly less than 2pq, then there is evidence for
homophily.
 Measuring Homophily
— p = 6/9 = 2/3
— q = 1/3
— 2pq = 4/9 = 8/18
— 5/18 cross-gender edges
— Test: 5/18 < 8/18 => some evidence of
homophily
— Need definition of “significantly less than”
— standard statistical significance
— What if cross-gender edges more than
2pq?
— inverse homophily or heterophily (Ex:
network of romantic relationships)
— How to extend to characteristics with
more than 2 states?
Mechanisms Underlying Homophily
— Homophily has two mechanisms for link creation
— Selection: select friends with similar characteristics
— Individual characteristics drive the formation of links
— Involves immutable characteristics (determined at birth)
— Social influence: modify behavior close to behaviors of
friends
— the reverse of selection
— involves mutable characteristics (behaviors, activities, interests,
beliefs, and opinions)
The Interplay of Selection and Social
Influence
— Q: When homophily is observed, is it a result of selection
or social influence?
— –Have people adapted their behaviors to become more like
their friends, or have they selected friends who were
already like them?

— Track the network and monitor the results of the two

mechanisms
The Interplay of Selection and Social
Influence
— Most of the times, both mechanisms apply and interact
with each other
— Studies show that teenage friends are similar to each
other in their behaviors, and both selection and social
influence apply:
— Teenagers seek social circles of people like them and peer
pressure causes conformation to behavioral patterns within
these circles
— Q: how the two mechanisms interact and whether one is
more strongly at work than the other?
Affiliation
— Story so far:
— Homophily groups together similar nodes
— Selection and social influence determine the formation of
links in a network
— Similarity of nodes based on characteristics
— How to model these characteristics?
— They represent surrounding contexts of networks
— They exist “outside” the network
— How to put these contexts into the network itself?
Affiliation
— Represent the set of activities a person takes part in (a
general view of “activity”)
— E.g. part of a particular company, organization, frequenting
a particular place, hobby

— Refer to activities as foci: “focal points” of social

interaction
 Affiliation Networks
— Affiliation network:
— bipartite graph (two-modes)
— nodes divided into 2 sets
— no edges joining a pair of nodes
that belong to the same set
— people affiliated with foci
— Example
— Anna participates in both of
the social foci on the right
— Daniel participates in only one
Co-Evolution of Social and Affiliation
Networks
— Social networks change over time
— New friendship links are formed
— Affiliation networks change over time
— People become associated with new foci
— Co-evolution reflects interplay between selection and
social influence.
— 2 people participate in a shared focus can become friends
— If 2 people are friends, they can share their foci
— How to represent co-evolution with a single network?
 Social-affiliation networks

— Social-affiliation
network contains:
— a social network on
the people and
— an affiliation network
on the people and
foci
 Social-affiliation networks
— In social-affiliation networks link
formation as a closure process
— Several options for “closing” B-C
— triadic closure: A, B, and C
represent a person (already
examined)
— focal closure: B and C people, A
focus
— selection: B links to similar C
(common focus)
— membership closure: A and B
people, C focus
— social influence: B links to C
influenced by A
Example
— Bob introduces
Anna to Claire.

— Karate “introduces”
Anna to Daniel.

— Anna introduces
Bob to Karate.
Edges with bold are the newly formed
Tracking Link Formation in On-Line Data
— What we’ve discussed so far: a set of mechanisms that
lead to the formation of links
— triadic closure
— focal closure
— membership closure

— Tracking these mechanisms in large populations

— Their accumulation observable in the aggregate
Tracking Triadic Closure
— Likelihood of link as a function of common friends?
1. Two snapshots of the network
2. For each k, find all pairs of nodes with k common friends
in the first snapshot, but not directly connected
3. T(k): fraction of these pairs connected in the second
snapshot
— empirical estimate of probability that a link will form between
two people with k common friends
4. Plot T(k) as a function of k
— T(0) is the rate of link formation when it does not close a triangle
Tracking Triadic Closure
— Kossinets and Watts computed T(k)
— full history of e-mail communication (“who-talks-to-whom”)
— a one-year period
— 22,000 students at a large U.S. university
— observations in each snapshot were one day apart (average
over multiple snapshots)

— Gueorgi Kossinets and DuncanWatts. Empirical analysis of an

evolving social network. Science, 311:88–90, 2006.
Tracking Triadic Closure
— Interpret the result compared to a baseline

— Assume that each common friend that 2 people have,

gives them an independent probability p of forming a link
— 2 people have k friends in common => the probability they
fail to form a link is (1-p)^k
— 2 people have k friends in common => probability that they
form a link is 1-(1-p)^k
Tracking Triadic Closure
Tracking Focal Closure
— Likelihood of link formation as a function of the number
of common foci?

— Kossinets and Watts supplemented their university e-mail

dataset with information about the class schedules.
— Each class became a focus.
— Students shared a focus if they had taken a class together.
Tracking Focal Closure
Tracking Membership Closure
— Blogging site
LiveJournal
— social network
(friendship links)
— foci correspond to
membership in
user-defined
communities

probability of joining a LiveJournal

community as a function of the number of
friends who are already members
Tracking Membership Closure
— Wikipedia editors
— link editors when
they communicated
(user talk page)
— each Wikipedia
article defines a
focus (editor
associated with the
articles he/she
edited)
probability of editing a Wikipedia articles as a
function of the number of friends who have
already done so
Quantifying the Interplay Between
Selection and Social Influence
— How selection and social influence work together to produce
homophily?
— How do similarities in behavior between two Wikipedia
editors relate to their pattern of social interaction over
time?
— Similarity between 2 Wikipedia editors A, B:

— Is homophily (similarity) due to editors connected (talk) with

those edited the same articles (selection), or because editors
are led to edit articles by those they talk to (social influence)?
Quantifying the Interplay Between
Selection and Social Influence

“tick” in time whenever either A or B performs an action (editing or

talking). Time 0 is the point at which they first talked
Friendship Network on Facebook
— Wimmer and Lewis (2010) studied friendship network of
an entire college cohort of 1,640 students.
— Three measures of friendship: facebook friends, picture
friends, housing groups.
— Exponential random graph modeling (ERGM) techniques
is used to disentangle the effects of the various tie-
generating mechanisms and identify the (multiple) levels
of ethnoracial categorization on which homophily actually
occurs.
Mechanisms leading to homogeneity
Friendship Network on Facebook
— Main findings
— Some racial categories matter for social network formation
(black) , but not others (Asian). Some ethnic categories (e.g.
South Asian, Jewish, Chinese, British) matter a lot, but not
others (e.g. Italian).
— The failure to take into account opportunity structure and
difference in sociality leads to an overestimation of
homophily rate of large groups and groups of more sociable
individuals.
— The ignorance of triadic closure leads to an overestimation
of the tendency towards homophily.
 A Spatial Model of Segregation
— One of the most strong
effects of homophily is
in the formation of
ethnically and racially
homogeneous
neighborhoods in cities
— a process with a
dynamic aspect
— what mechanisms?

In blocks colored yellow and orange the percentage of African-

Americans is below 25, while in blocks colored brown and black the
percentage is above 75
The Schelling Model
— How global patterns of spatial segregation can arise from
the effect of homophily operating at a local level (Thomas
Schelling, 1969)
— an intentionally simplified mechanism
— works even when no one individual explicitly wants a
segregated outcome
 The Schelling Model
— Model assumptions:
— Population of individuals called
agents
— Each agent of type X or type O
— The two types represent some
characteristic as basis for
homophily (race, ethnicity, country
of origin, or native language)
— Agents reside in cells of a grid
(simple model of a 2-D city map)
— Some cells contain agents while
others are unpopulated
— Cell’s neighbors: cells that touch it
(including diagonal contact)
The Schelling Model

Cells are the nodes and edges connect neighboring cells.

We will continue with the geometric grid rather than the graph.
The Schelling Model
— Local mechanism:
— Each agent wants to
have at least some t
other agents of its own
type as neighbors (t
the same for all)
— Unsatisfied agents
have fewer than t
neighbors of the same
type as itself and move
to a new cell
— E.g. t = 3
 The Dynamics of Movement
— Unsatisfied agents move in rounds
— consider unsatisfied agents in some
order
— random or row-sweep
— unsatisfied agents move to an
unoccupied cell where will be satisfied
— random or to nearest cell that
satisfies them
— may cause other agents to be
unsatisfied
— deadlocks may appear (no cell that
satisfies)
— stay or move randomly
— All variations have similar results
— E.g. t=3, one round, row-sweep, move to
nearest cell, stay when deadlocks
Larger examples

Two runs (50 rounds) of the Schelling model with unsatisfied agents moving
to a random location. Threshold t=3, 150-by-150 grid with 10, 000 agents.
Each cell of first type is red, of second type blue, or black if unoccupied.
 Interpretations of the Model
— Spatial segregation is taking
place even though no individual
agent is seeking it
— Agents just want to be near t
others like them
— When t=3, agents are satisfied
being minority among its
neighbors (5 neighbors of the
opposite type)
— See the figure on right:
— A checkerboard 4x4 pattern can
make all agent satisfied (even for
large grids).
— We don’t see this result in
simulations.
Some other simulation results

t=4, 150-by-150
grid, 10, 000
agents, varying
number of rounds
(steps), not shown
until the end
Interpretations of the Results
— More typically, agents form larger clusters
— Agents become unsatisfied and attach to larger clusters
(where higher probability to be satisfied)
— The overall effect:
— Local preferences of individual agents have produced a
global pattern that none of them necessarily intended
— The Schelling model is an example that, as homophily draws
people together along immutable characteristics (race or
ethnicity), it creates a natural tendency for mutable
characteristics (decision about where to live) to change in
accordance with the network structure
Summary
— Homophily
— Underlying Mechanisms: Selection and Influence
— Affiliation
— Tracking Network using On-Line Data
— Email
— Wikipedia editing
— A Spatial Model of Segregation
— Schelling Model