Personal or Ego Network Analysis

Jeffrey C Johnson
Department of Anthropology
University of Florida
Ego or Personal Social Networks
Ego networks consist of a
focal node ("ego") and the
nodes to whom ego is
directly connected to
(these are called "alters")
plus the ties, if any, among
the alters.

Advantage: Fits nicely into a

survey research enterprise
By: Barry
Wellman
 Two kinds of Social Network Analysis
Whole Network Analysis Personal Network Analysis
Sociocentric Egocentric

Focus on interaction within Focus on effects of network

a group. Bounded by the
social context or only those
actors important for the
study at hand
Involves a census of sorts
Collect data from members
of a group about their ties to
other group members
on individuals social world
that influence attitudes,
behaviors and conditions
Involves a sample
Collect data from respondent
(ego) about the characteristics
and interactions with
network members (alters)
NOTE: All the measures used in
whole network analysis can be
used in personal network analysis
(and vice versa).

A network is a network!

Whole Networks-1 or a few networks

Personal Networks-Many Networks
 Ego Network Function
Social Support
Influence and Social Control
Social and Financial Resources
Role Models
World View
Social Capital
Trust
 Ego Network Characteristics
Ego network homophily (alters are strong ties and
have attributes similar to ego such as social class,
age, religion, occupation, etc.)
Composition matters (Diverse networks are
hypothesized to get you more of what you need)
Strong ties to ego imply at least some degree of
weak linkage among alters
Weak ties are important
 Similarity breeds connection, the sociologists Miller
McPherson, Lynn Smith-Lovin and James Cook wrote in
their classic 2001 paper on the subject, Birds of a
Feather: Homophily in Social Networks, and the result is
that peoples personal networks are homogeneous. This
year, other academics have cited homophily in elucidating
everything from why teenagers choose friends who smoke
and drink the same amount that they do to the strong
isolation of lower-class blacks from the interracial-
marriage market. Researchers at M.I.T. even published
Homophily in Online Dating: When Do You Like Someone
Like Yourself? which showed that you like someone like
yourself most of the time, online or off. So much for
opposites attract.

(Homophily AARON RETICA Published: Dec.10, 2006)

 Examples of Types of Measures in Ego
homophily Networks (after Borgatti)
size
average strength of ties
heterogeneity
components
structural holes
density
composition (e.g., % women, %whites, etc.)
range:
i. substantively defined as potential access to social
resources
ii. often defined as diversity of alters
iii. based on weak ties argument, density is thought of as
inverse measure of range
iv. size and heterogeneity also seen as measures of range
Ego Data (Campbell and Lee 1991)
Characteristics of focal individuals-egos(e.g.,
their gender or ethnicity);

Characteristics of ties between focal individual

and network members. These may be relational
characteristics (e.g.,each network member's
frequency of contact), relational contents (e.g., the
extent to which a network member provides
emotional aid), or relational types (e.g., whether a
network member is a friend, neighbor, workmate
or relative of a focal individual).
Characteristics of the network members (alters) with
whom focal individuals have ties (e.g., their gender or
ethnicity);

Network composition, the aggregated characteristics of

the network members and ties in each ego-centered
network (e.g., the mean frequency of contact that the
members of a network have with a focal individual, the
proportion of network members providing emotional aid);

Network structural characteristics (e.g., the overall

density of ties in egos network, the number of clusters in
egos network).
 Ego Network Data Collection
Person-Based (work forward)
Relation/Role-Based (work forward)
Know-Based (work backwards)
Position-Based
Resource-Based
 Person/Relation/Role-Based
(Forward)
Typically open ended name generator
Usually limited in scope (kin, friends, people you
do drugs with, provides emotional support)
Following extensive name generation ask for
attributes of alters, types of relations, strength of
relations with alters, whether alters know one
another, etc. (this is often referred to as name
interpreters)
Can ask about multiple relations (combine any
number of person/role-based relations of interest)
Typical Name Generator
Elicitation
Example from Flapp et al.
 Know Based (Backwards)
Name n people you know
- Would you please give us the names of 45 people you
know and who know you, with whom you have had contact
with in the past two years (face-to-face, by phone, or by
internet), and whom you could still contact if needed?
Then ask about the characteristics of alters
and the types and strength of relations ego
has with each alter
Then alter-alter ties
 Social Capital-Based
Position Approach
Primary purpose is the measurement of individual
social capital (Nan Lin)
Generates positions (e.g., occupations) as opposed
to names
Gives an indication of the embeddedness of ego
with respect to social or political resources (allows
for an occupational prestige index)
Advantage that it is easier to administer
Example from Flapp et al.
 Resource Approach
Gets at individual social capital like the
position generator
More directly looks at egos access to
resources
Example from Flapp et al.
Will Return to Role and Resource
Based Approaches Shortly
 The Research Problem often
Determines whether you use a
Whole vs. a Personal Network
Design

What do you want to understand?

 Classic Approach in the Social Sciences

Most social science research is designed to

understand or to predict attitudes, behaviors or
conditions of people based on their characteristics

Social scientists ask questions or observe

characteristics about respondents, then use the
variability in those characteristics to explain the
variability in outcome variables

From McCarty
 Example of a study problem and
design (classic non-relational)
Age
Education Number of
Income cigarettes smoked
per day
Height
Weight
Independent Dependent variable
variables
A social scientist may collect data on a sample of 500
respondents and try to predict their smoking behavior
using variability in their age, education, income, height
and weight (Classic non-relational approach)
 Conclusion
Age
Education Number of
cigarettes smoked
Income
per day
Height
Weight
Independent
Dependent variable
variables
The researcher concludes that age, level of education
and income are good predictors of how many cigarettes
are smoked in a day while height and weight are not.
 Social influence
Social scientists think that some outcome variables
are influenced by social factors

For example, it is widely accepted that

experimentation with smoking among adolescents
is due to peer influences

Since peer influence cannot be directly observed,

social scientists craft questions that can be used as
proxies for observing peer influence
 Proxy questions
Do either of your parents smoke?
(PARENTS)

Do most of your friends smoke?

(FRIENDS)

Have any of your friends ever offered you a

cigarette? (OFFER)
 Questions
Would more detail about the social
influence surrounding a respondent provide
more explanatory power?

What questions can we ask the respondent

to gather this kind of detail?

We propose using Social Network Analysis

 Questions
Would more detail about the social
influence surrounding a respondent provide
more explanatory power?

What questions can we ask the respondent

to gather this kind of detail?

We propose using Social Network Analysis

Whole network approach to smoking and
social influence

Select a group of students in a class

Ask each student to rate on a scale of 1 to 5

how much they socialize with each of the
others

Ask each student whether they smoke

 Adjacency Matrix of students
David Faith Rosanna Antonio Napp Lem Jim Beth Mark Kent Amber Thomas
David 5 2 2 0 0 1 0 3 1 0 2 0
Faith 1 5 5 0 0 0 0 1 0 0 2 0
Rosanna 2 5 5 0 0 1 0 2 0 0 4 0
Antonio 0 1 1 5 0 0 0 0 0 0 0 0
Napp 0 0 0 0 5 0 0 0 0 0 0 0
Lem 2 0 2 0 0 5 5 2 0 0 2 0
Jim 0 0 1 0 0 5 5 5 0 0 2 0
Beth 4 3 1 0 0 1 5 5 0 0 3 0
Mark 1 0 0 1 0 0 0 0 5 0 1 0
Kent 0 0 0 0 0 0 0 0 0 5 0 3
Amber 2 3 3 0 0 1 2 2 1 0 5 0
Thomas 0 0 0 0 0 0 0 0 0 3 0 5

Ratings of each person can be used to make a matrix representing the

relations between members of the class

Intersecting cells represent their assessment

David says he socializes with Faith at a level 2

Faith says she socializes with David at a level 1

 Network visualization

We can use the matrix to visualize the structure of relations

There is a large group in the middle
Amber and Beth smoke
Napp does not socialize with anybody and Thomas and
Kent only socialize with each other
 Network visualization

We can calculate several measures of this structure

There are two network components
Beth is the most degree central
Amber is the most between central
Could look at influence of covariates (e.g., smoking) on
network structure (ERGM)
 Conclusion
We may conclude that those in the group with
Beth and Amber are more likely to experiment
with smoking

Napp, Kent and Thomas are not

Drawback: This analysis says nothing about

influences from outside of this group

To study social influences across groups we use

Personal Network Analysis
Tom has a Personal Network of 10 people
(name generator)
Tom met these people in three groups (name
interpreter)
CLUB

FAMILY
WORK
Within the groups everybody knows each other
(alter-alter ties)
CLUB

FAMILY
WORK
There are also some ties between the groups (alter-
alter ties)
CLUB

WORK

FAMILY
Sometimes personal networks can be
complex
Overview of Personal Network Data
Collection
1. Identify a population

2. Select a sample of respondents

3. Ask questions about respondent

4. Elicit network members

5. Ask questions about each network member

6. Ask respondent to evaluate ties between network

members
 Identify a population
Personal network analysis begins much like
any social science research study.

The first thing to do is clearly identify the

population of interest (smokers, the elderly,
migrants, the homeless, hurricane victims or
any other population where there is an
outcome variable we think is impacted by
their social environment).
 Select a sample of respondents
Selecting a sample of respondents is also
like any other social science research study.
Sometimes added to standard social surveys
(usually random samples).
However, personal network data collection
can be a long interviewing process that
sometimes requires special software. This
may mean a trade-off between the
representativeness of the sample and the
detail about their personal networks.
 Ask questions about respondent
Like any other social science research study, we
are interested in knowing something about the
respondent (ego).
We want to know about outcome variables of
interest we think may be impacted by social
influences (Do they smoke, are they depressed,
how many times have they migrated).
We also want to know about other possible
explanatory variables that are not related to social
influence (How old are they, how much education
do they have, what is their income).
 Elicit network members
This is where personal network data collection
diverges from other social science research.
We ask ego a set of questions (name generators)
that elicit the names of people they know (alters).
This could be a free-listing of people they have
had contact with in the past year, people they talk
to about important matters, or any other question
that brings names to mind.
This is important as it defines the sample of
network alters.
 Ask questions about each network member

We then want to ask ego a set of questions

about each alter (name interpreters). Keep
in mind that this is typically the longest part
of the interview. If you elicit 50 alters from
each respondent and want to know ten
things about each one (e.g. sex, age, do they
smoke, where do they live, etc.) then this
becomes 500 questions.
You must balance the number of alters and
the amount of information you want to
know about each one.
 Ask respondent to evaluate ties
between network members
Finally, we want to collect the structural
data to form an adjacency matrix. This
means that ego must evaluate all the
possible ties between their alters.
Fortunately we usually assume that the ties
are symmetric, meaning that we just want to
know if the two alters are tied.
The number of ties to evaluate grows
geometrically as alters are added.
Alter Relations can be Time Consuming
Respondent burden as alters are added
Respondent burden by number of alters

1400

1200
Alter pair evaluations

1000

800

600

400

200

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
Alters
 What kind of data do we get?
Data on network composition. These are
summaries of the attributes of network
alters.

Data on network structure. These are

summary measures of the pattern of
relations

Combinations of composition and structure

 Personal network composition
Name Closeness Relation Sex Age Race Where Live Year_Met
Joydip_K 5 14 1 25 1 1 1994
Shikha_K 4 12 0 34 1 1 2001
Candice_A 5 2 0 24 3 2 1990
Brian_N 2 3 1 23 3 2 2001
Barbara_A 3 3 0 42 3 1 1991
Matthew_A 2 3 1 20 3 2 1991
Kavita_G 2 3 0 22 1 3 1991
Ketki_G 3 3 0 54 1 1 1991
Kiran_G 1 3 1 23 1 1 1991
Kristin_K 4 2 0 24 3 1 1986
Keith_K 2 3 1 26 3 1 1995
Gail_C 4 3 0 33 3 1 1992
Allison_C 3 3 0 19 3 1 1992
Vicki_K 1 3 0 34 3 1 2002
Neha_G 4 2 0 24 1 2 1990
. . . . . . . .
. . . . . . . .
. . . . . . . .

This ego has told us some things about each alter. For
example, Joydip is a 25 year old male she met in 1994 that she
is very close to.
 Now we can create a set of
compositional variables
Average age of each alter (ALTAGE)
Proportion of alters that are women
(ALTWOMEN)
Proportion of alters that are family
(ALTFAMILY)
Average length of time ego has known each
alter (DURATION)
Proportion of alters that smoke
(ALTSMOKE)
 And we can add these to our model
Age
Education
Number of
Income cigarettes smoked
per day
Altage
Altsmoke
Duration
Independent
Dependent variable
variables
For each respondent these now become variables about their
social environment that can be used to predict outcome
variables. In this case we may believe that higher proportions
of smoking alters leads to smoking.
 Personal Network Structure
Joydip_K Shikha_K Candice_A Brian_N Barbara_A Matthew_A Kavita_G Ketki_G . . .

Joydip_K 1 1 1 1 0 0 0 0 . . .

Shikha_K 1 1 0 0 0 0 0 0 . . .

Candice_A 1 0 1 1 1 1 1 1 . . .

Brian_N 1 0 1 1 1 1 1 1 . . .

Barbara_A 0 0 1 1 1 1 0 0 . . .

Matthew_A 0 0 1 1 1 1 1 1 . . .

Kavita_G 0 0 1 1 0 1 1 1 . . .

Ketki_G 0 0 1 1 0 1 1 1 . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

The same Ego also evaluated the ties between their alters. We
end up with an adjacency matrix for each ego. We can use this
to calculate structural measures.
Now we can create a set of structural
variables
Number of components (COMP)
Average betweenness centrality
(BETWEEN)
Closeness centralization (CLOSCENT)
Number of alters in network core
(CORESIZE)
 And these can be added to the model
Age
Education
Number of
Income cigarettes smoked
per day
Altage
Altsmoke Dependent variable
Duration In this model we want to test whether the
Comp structure of the personal network impacts
smoking. For example, betweenness
Between centrality is a measure of bridging.
Coresize Bridging represents exposure to different
Independent groups which may not tolerate smoking.
variables
 Some measures of personal network structure
Degree Centrality An alter is highly degree-central to the extent he or she is
directly connected to many other alters.

Closeness Centrality An alter is highly close-central if he or she is connected by

short paths to many other alters.

Betweenness Centrality An alter is highly between-central to the extent he or she

lies on many geodesics (shortest paths) between alters.

Components A set of alters who are connected to one another directly or

indirectly.

Isolates A node unconnected to any other node.

Network-Degree Centralization A measure of the extent to which the network is

dominated by a single alter using degree centrality.

Network-Closeness Centralization A measure of the extent to which the network

is dominated by a single alter using closeness centrality.

Network-Betweenness Centralization A measure of the extent to which the

network is dominated by a single alter using betweenness centrality.
 Example

Recovery From Natural Disasters

and Personal Networks
Do relations matter and how does
this vary?
 Natural Disasters and Well -Being

Earlier disaster studies Work on the sociology

have clearly observed a
link between destruction,
disruption, displacement,
and psychological stress
and trauma including, for
example, increases in
suicide (Provenzo and
Fradd 1995; Fullilove
1996; Peacock, Morrow,
and Gladwin 1997).
of disasters has found that
the economic, social, and
psychological problems
faced by disaster victims,
both during and after an
event, varies by ethnicity,
class, and gender
(Peacock, Morrow, and
Gladwin 1996; Peacock
and Ragsdale 1997; Dash,
Peacock, and Morrow
1997).
We ask two important and related
questions.
What forms of social and other resources
aid in lessening the long term social and
psychological impacts of natural disasters?
 Less well understood--Social factors that help
mitigate and lessen psychological stresses stemming
from a disaster.

Individuals vary on the These resources in turn

amount of personal (e.g.,
personal savings), social (e.g.,
kin and friends), and
institutional (e.g., insurance)
resources available to them
(Morrow 1997).
help in reducing the social,
psychological, and economic
uncertainties and impacts
presented by loss and disruption
(Fullilove 1996). For example,
many victims have access to kin
or friends who provide
temporary shelter, clean-up
help, or possibly financial
assistance while yet others have
limited social support due to
such things as recent migration
to the region or lack of
language proficiency.
 Even less well understood--Social
factors that help mitigate and lessen
psychological stresses stemming from
a disaster over an extended period of
time.
 The overall objective of this study is to understand the role
various social, personal, and institutional resources play in
mitigating the impacts of a catastrophic natural disaster. The
more specific objectives are:

To understand the types of social, personal, and

institutional resources used by hurricane victims
up to a year following the event.
To understand how the kind and extent of access
to social, personal, and institutional resources
relates to degree of social disruption and
psychological trauma experienced by victims.
 Research Approach
The instrument consisted of an ego-centered
network inventory approach to help in
understanding the extent and quality of access to
social resources (e.g., help in evacuating
belongings, help in finding shelter, etc.)(Wellman,
Mueller and Marin 1999). Psychological well
being will be determined with the Profile of Mood
States instrument that reliably measures
depression, anxiety, and vigor (Johnson, Boster,
and Palinkas 1996).
 Research Approach (Continued)
A sample of victims still without housing up
to a year following the disaster. Involves
sampling from residents in FEMA trailer
parks. Thus, most of the respondents
experienced catastrophic loss of property.
The Disaster
The Aftermath
The Destruction
Institutional Support
Reactions
 Variables of interest
Dependent From POMS including
Tension-Anxiety, Depression, Anger,
Fatigue, and Confusion.
Independent (Demo & Social & Geographic
Proximity) Education, Age, Household Size, #
Helping, Median days of Help from Alters, Log % Alters
Immediate Family, Log % Alters Extended Family, Log %
Friends, Log % Alters Neighbors, Log % Alters Local, Log
% Alters Outside Immediate Area, Log % Alters Outside
State, and Number of Organizations Helping
[83a] (RELHELP_83A) % of immediate family
[83b] (RELHELP_83B) % of extended family
[83c] (RELHELP_83C) % of acquaintances/friends
[83d] (RELHELP_83D) % of coworkers, bosses(working relationship)
[83e] (RELHELP_83E) % of new/none
[83f] (RELHELP_83F) % of other
[84] (PRMALE_84) sex- % of males that helped
[85] (AVGAGE_85) age- average age of those who helped
[86] (PDOMETHN_86) ethnicity- % of dominant ethnicity that helped
[87a-e] top 5 types of assistance from those who helped(see sheet)
[87a] (TOPASST1_87) top type of assistance
[87b] (TOPASST2_87) second highest type of assistance
[87c] (TOPASST3_87) third highest type of assistance
[87d] (TOPASST4_87) fourth highest type of assistance
[87e] (TOPASST5_87) fifth highest type of assistance
[88] (MEDASST_88) median(in days) period of help received
[89a-d] where alter lives 1=neighbor 2=Greenville area
3=outside area in state 4=outside state
[89a] (ALTERLIV_89A) % neighbor
[89b] (ALTERLIV_89B) % Greenville area
[89c] (ALTERLIV_89C) % outside area
[89d] (ALTERLIV_89D) % outside state
 Focus
Social Proximity the type of social resources.
Geographic Proximity geographic distance from
social resources.
Does Size Really Matter? amount of social
resources.
Institutional Resources Support from formal
institutions (e.g., FEMA, Red Cross, Salvation
Army).
Ego Network Examples

Ego
Immediate
Extended
Friends
 Tension/ Depression Anger Vigor Fatigue Confusion
Anxiety

Immediate 0.339** 0.350** 0.358** -0.237* 0.457*** 0.415***

Family (0.023) (0.027) (0.016) (0.116) (0.002) (0.005)
Extended -0.205 -0.012 -0.276* -0.021 -0.122 -0.181
Family (0.177) (0.942) (0.067) (0.892) (0.431) (0.240)
Friends 0.040 -0.147 0.102 0.171 -0.154 -0.043
Acquaintances (0.796) (0.366) (0.505) (0.260) (0.319) (0.780)
Coworkers -0.204 -0.146 -0.035 0.011 -0.143 -0.219
(0.179) (0.370) (0.822) (0.942) (0.355) (0.153)
Neighbors -0.109 -0.327** -0.174 0.252* -0.249* -0.322**
(0.476) (0.039) (0.253) (0.096) (0.103) (0.033)
Local 0.003 -0.004 -0.033 -0.067 0.145 0.054
(0.966) (0.980) (0.827) (0.633) (0.349) (0.726)
Outside Area 0.248* 0.293* 0.206 -0.181 0.264* 0.391***
(0.100) (0.067) (0.175) (0.234) (0.084) (0.009)
Outside State -0.278* -0.083 -0.043 0.060 -0.289* -0.255*
(0.064) (0.611) (0.777) (0.694) (0.057) (0.095)
# Alters Helping -0.094 -0.118 -0.062 0.266 -0.119 0.013
(0.528) (0.458) (0.679) (0.071) (0.430) (0.931)
Organizations -0.076 -0.038 -0.148 0.166 -0.119 -0.120
(0.614) (0.813) (0.327) (0.269) (0.437) (0.431)
Models Including Social
Proximity Variables
 Regressions (Tension Anxiety)
Effect Std Coef t-value P(2 tail)
Education -0.137 -0.856 0.398
Age -0.069 -0.419 0.678
Household Size -0.168 0.976 0.336
# Helping -0.012 -0.069 0.946
# Organizations 0.031 0.177 0.860
Helping
Log % Alters 0.695 2.639 0.012***
Immediate Family
Log % Alters 0.441 1.209 0.235
Extended Family
Log % friends 0.518 1.491 0.145
Constant 0.000 0.769
Multiple R 0.235 N= 44
squared
 Regressions (Depression)
Effect Std Coef t-value P(2 tail)
Education -0.292 -1.585 0.123
Age 0.054 0.320 0.751
Household Size -0.142 -0.743 0.463
# Helping -0.061 -.352 0.727
# Organizations 0.029 0.16 0.874
Helping
Log % Alters 0.553 2.475 0.019**
Immediate Family
Log % Alters 0.285 0.855 0.399
Extended Family
Log % friends 0.243 0.745 0.462
Constant 1.375 0.179
Multiple R 0.259 N=39
squared
 Regressions (Anger)
Effect Std Coef t-value P(2 tail)
Education -0.119 -0.750 0.458
Age 0.013 0.082 0.935
Household Size -0.129 -0.757 0.454
# Helping 0.025 0.155 0.878
# Organizations -0.146 -0.834 0.410
Helping
Log % Alters 0.545 2.480 0.018**
Immediate Family
Log % Alters 0.197 0.621 0.539
Extended Family
Log % friends 0.454 1.442 0.158
Constant 3.016 0.292
Multiple R 0.253 N=44
squared
 Regressions (Fatigue)
Effect Std Coef t-value P(2 tail)
Education -0.252 -1.668 0.105
Age -0.074 -0.466 0.644
Household Size -0.188 -1.168 0.251
# Helping 0.005 0.032 0.974
# Organizations 0.024 0.143 0.887
Helping
Log % Alters 0.615 2.661 0.012***
Immediate Family
Log % Alters 0.209 0.689 0.496
Extended Family
Log % friends 0.161 0.552 0.584
Constant 2.258 0.031
Multiple R 0.329 N=43
squared
 Regressions (Confusion)
Effect Std Coef t-value P(2 tail)
Education -0.125 -0.825 0.415
Age -0.027 -0.167 0.868
Household Size -0.177 -1.056 0.299
# Helping 0.101 0.628 0.534
# Organizations -0.029 -0.170 0.866
Helping
Log % Alters 0.779 3.019 0.005***
Immediate Family
Log % Alters 0.448 1.250 0.220
Extended Family
Log % friends 0.465 1.370 0.180
Constant 0.348 0.730
Multiple R 0.302 N=43
squared
Models Including Geographic
Proximity (e.g., outside area,
outside state)
No Significant Relationships
 Summary
There are bivariate relationships between over dependence
on strong family ties and a range of psychological
problems during extended recovery.
There are bivariate relationships between number of alters
outside the area and a range of psychological problems
during extended recovery.
The relationship between lack of more moderate to weak
ties and psychological well-being holds when controlling
for a number of demographic variables while the
relationship with geographical proximity disappears.
Position and Resource Based
Approaches in Personal
Networks and the Study of
Social Capital
 Brief China and the Rule of Law
Example
Is social capital, as reflected in position
generators, related to the perception of the
importance of relations in court?
Study Sites

Nanzhuan
g
 Urban
Sample
Shijiazhuang is the capital of Hebei
province. Until the city became a
railroad junction in the early 1900s,
it was a small village. Today its
location is at the intersection of
north-south and east-west highways
and railroads. It has textile,
fertilizer, pharmaceutical,
automotive, building materials and
paper industries. The jurisdiction of
prefecture of Shijiazhuang includes
six districts, twelve counties and
five county level cities, a total of 9
million people.. The population of
the urban area is 2.1 million.
 Rural Sample
Nanzhuang is a village in Zhao
County which is one of the
counties in the prefecture
Shijiazhuang. The drive from the
center of Shijiazhuang to
Nanzhuang takes approximately
60 minutes by private car. The
village has approximately 4830
inhabitants living in 900
household. 85% of the villagers
share the same last name. The
village is widely know for its pear
production. After
decollectivization in 1983 farmers
changed almost all agricultural
space to pear production. In
addition, many farmers have
started agricultural sidelines
businesses such as large scale
freezers, paper mills, paper carton
factories, and long-distance trade
activities
 Interviews on how relations matter
(social capital) in court
Examples of reasoning for going to court among young urban
residents (age 22 to 35):
.
I think going to court is a normal way to protect your own rights.
Reasoning for not going to court among rural residents (age 36 to 56):

What you have to do is mix the mud (huo xini). It will be suitable
for all kinds of situations. There are only two persons, so it should be
solved between them. That is how disputes are solved, just like they do
it at court anyway.
 Relational Index
Based on PCA- Sum of the Responses to the
Following Statements
It is impossible to win a case if you have evidence but no
guanxi at court.

If both sides give presents to the judge, whoever gives the most
(highest amount) will win the case.

People without relationships cannot win a lawsuit.

Mediation works only when you give gifts.

If both sides have relationships the one side that has the
stronger relationship will win the case in court.
Position generator for this study
 Variables in Models
Standard GLM
Dependent Variables
Relational Index (importance of social relations)
Independent Variables
Age Group
Gender
Location (rural, urban)
Class
Position Generator as Social Capital (Court
Capitals, Intellectuals)
 Dependent Variable RELATION_INDEX
N 290
Multiple R 0.363
Squared Multiple R 0.132
Adjusted Squared Multiple R 0.114
Standard Error of Estimate 2.304
Regression Coefficients B = (X'X)-1X'Y

Effect Coefficient Standard Std. Tolerance t p-value

Error Coefficient
CONSTANT 4.429 0.500 0.000 . 8.849 0.000
location of -0.787 0.283 -0.161 0.915 -2.777 0.006
residence

gender 0.176 0.275 0.036 0.973 0.642 0.521

age group 0.431 0.172 0.141 0.959 2.498 0.013
socioeconomi -0.406 0.158 -0.154 0.854 -2.574 0.011
c class power'

COURT_CAP 0.033 0.033 0.062 0.813 1.011 0.313

Number of -0.092 0.034 -0.165 0.802 -2.667 0.008

professors
known
 Summary
Relations (in court and other aspects of life
in China) are seen as less important among
upper class, urban actors with connections
to the intellectual elite.
Court capital (lawyers, judges not
important).
Relations seen as more important to rural
older residents.
 How do we collect and analyze
personal network data?
Many researchers develop paper
instruments or computerized instruments
that let them collect these data
Compositional data are calculated using a
statistical package (e.g. SAS or SPSS)
Structural analyses are not typical and are
often limited to personal network density,
since it is an easy measure to program
Examples of Personal Network
Software
Need for personal network software
A standardized software package would offer
many advantages
It provides a computer interface that edits and
standardized data input with complex skip patterns
It automatically can calculate compositional and
structural measures and export them to a data set
compatible with a statistical package
It makes it possible to analyze individual cases
 From Wikipedia:
The program allows to create
questionnaires, collect data and
provide comprehensive measures
and arrays of data that can be
used for subsequent analysis by
other software.
 EgoNet design
EgoNet is Java based and runs on both
Windows and Mac platforms
There are two programs:
1. Administrator program to create a study and
assemble a questionnaire
2. Client program to collect data and analyze it
Deign a Study
 Example data file from EgoNet

EgoNet outputs data across all the respondents and assembles it into
one file. Notice that the data set has data about ego (sex, age),
compositional data (Proportion of females, average alter age), and
structural data (components, cliques). This data set would be
difficult to produce without this software.
 Egonet can also visualize the
personal network of a single Ego

This is the personal network of Merced, a 19-year-old second

generation West African migrant in Spain. The dots represent
her alters and the lines represent a connection between alters
based on her evaluation of the ties.
 We can label the dots (nodes) with
information we collected from Merced about
each alter, like where they are from
We can also size the nodes, in this case by
Merceds assessment of how close she is to
each alter
And we can color the nodes, in this case by race
Finally, we can shape the nodes, in this case
by whether they smoke (smokers are the
squares)
We now have a picture we can use to interview Merced about
her acculturation experience in Spain. See the potential
influence of white, Spanish smokers in the upper right from
her high school
We now have a picture we can use to interview Merced about
her acculturation experience in Spain. See the potential
influence of white, Spanish smokers in the upper right from
her high school
This is Vivian, a 36 year old
Moroccan woman
And this is Jose, a 46 year old
Dominican man
We can also use Egonet to visualize structural
measures. Here is Merceds network with nodes
colored by betweenness centrality.
Here Merceds network is colored by her relation type (blue
nodes are relatives). EgoNet has done a cluster analysis and
circled nodes and labeled them with numbers.
 Conclusion: EgoNet
EgoNet provides a way to design a personal
network study and share the results with
colleagues
EgoNet exports a data file that puts together
variables about ego, compositional variables and
structural variables
EgoNet provides a flexible visualization of
individual networks that can be used to interview
respondents about their personal networks
E-Net
Row-wise format in E-Net for the Floyd flood example showing the ego
and alter characteristics. Characteristics are Im-Family (immediate
family), Ex-Famil (extended family),etc.
 Row-wise format in E-Net for the Floyd flood example
showing continuation of data from Table 4 for alter-alter ties
and tie characteristics.
Ego network visualization output
from E-Net for respondent 45
E-Net output for composition of
ego networks in the Floyd study
Table 8. Regression models comparing the effects of the various composition network variables
on the dependent variable depression while controlling for demographic variables (included are
standardized coefficients with t-values in parentheses).

Effect Model 1 Model 2 Model 3 Model 4

Education -0.284* -0.299* -0.235 -0.224
(-1.86) (-1.64) (-1.25) (-1.41)
Organizations -0.037 -0.120 -0.089 -0.094
(-0.24) (-0.71) (-0.52) (-0.60)
Age 0.042 -0.030 -0.014 0.056
(0.24) (-0.18) (-0.09) (0.35)
Immediate 0.412*** - - -
(2.66)
Extended - -0.131 - -
(-0.71)
Friends - - -0.019 -
(-0.10)
Neighbors - - - -0.320*
(-2.00)
Constant 0.000 0.000 0.000 0.000
(2.58) (2.74) (2.67) (2.75)
Squared 0.226 0.079 0.066 0.163
Multiple R
P<0.1*, p<0.05**, p<0.01***
Example of column-wise E-Net
format for the GSS data
"From time to time, most people discuss important
matters with other people. Looking back over the last six
months, who are the people with whom you discussed
matters important to you? Just tell me their first names or
initials. "
Continuation of the column-wise
format in E-Net showing alter-
alter ties
E-Net output for the structural
holes analysis
 Hypothesis (Using GSS data): The networks ties of whites will
be less redundant than those of other groups in that we
hypothesize they will have greater individual level social capital
as opposed to group level social capital (remember this is just an
illustrative example).
ANOVA output for the test of the hypothesis concerning constraint and race/ethnicity.

Analysis of Variance
Source Type III SS df Mean Squares F-ratio p-value
RACE 0.146 2 0.073 4.252 0.014
Error 26.323 1,531 0.017

Tukey's Honestly-Significant-Difference Test

RACE(i) RACE(j) Difference p-value 95.0% Confidence Interval
Lower Upper
Black Other -0.026 0.488 -0.078 0.027
Black White -0.033 0.010 -0.059 -0.006
other White -0.007 0.935 -0.054 0.040
EgoWeb
Others: e.g., Vennmaker
 Questions?

Wellman and
Berkowitz
1988