10. General comparison ENRD and FNRD
Advantage: FNRD
• Can answer context questions
– how fragmented is the network as
a whole?
– How many links separate egos from
each other along the shortest
path?*
• Have incoming ties as well as out
– ENRD can ask ego who likes him,
but is still ego naming the alters
• Non‐ties are meaningful
– Can model tie/non‐tie for each
dyad as outcome of decision‐
making process
– in ENRD can ask ‘who do you not
like?’, but not ‘who do you have no
tie to?’
Advantage: ENRD
• Can employ standard
sampling techniques
– And so standard statistical
methods
• Cheaper & easier to deploy
– Can collect richer data – more
ties
• Fewer privacy/ethical issues
– May improve validity of data
*how quickly can something flowing through the network reach this node?
12. What can you do with ego net data?
• (if you only have) Ties to alters
– Network size for each kind of tie (e.g., number of
friends)
• (if you also have) Alter attributes
– Network composition (e.g., number of friends who
are top‐level managers)
• Testing homophily
• (if you also have) Ties among alters
– Structural holes
– All group‐level network measures (e.g., density of ties
among friends; avg distance; no. of components)
13. Number of ties
• Basically, this is network size
– Can be calculated different size for each type of tie, or all
ties combined
• Very well studied variable; has been very productive
– Health, power, satisfaction
• And there is still more to study
– Number of negative ties is understudied
• how many enemies, rivals, competitors, energy‐drains do you
have?
– Multiplex ties
• Suppose most of your friends are also co‐workers
– Most relationships A—B consist of both the friend and co‐worker tie
• What are the consequences for ego? Less freedom? More strain?
16. Network Composition
Property of network: Categorical Attributes Continuous Attributes
Summary of kind of alters ego
tends has, based on a given
attribute (e.g., wealth)
Example: Does ego have
mostly rich or mostly poor
friends? How many of each?
Example: What is the average
wealth of ego’s friends?
Measures: frequencies,
proportions
Measures: mean, median
Variability in the kinds of alter
an ego has, based on given
attribute
Example: whether ego has
equal number of rich, middle,
and poor friends, or mostly
one kind
Example: variance in wealth of
person’s friends
Measures: Blau/Herfindahl
heterogeneity; Agresti IQV
Measures: std deviation,
variance
Similarity of ego to alters with
respect to given attribute
Example: Prop. of ego’s friends
who are same wealth class as
ego
Example: Similarity between
ego’s wealth and friends’
wealth
Measures: E‐I index; PBSC;
Yules Q; Q modularity
Measures: avg euclidean
distance; identity coef.
19. Who do you discuss important
matters with?
Male Female
Male 1245 748
Female 970 1515
Age < 30 30-39 40-49 50-59 60+
< 30 567 186 183 155 56
30 - 39 191 501 171 128 106
40 - 49 88 170 246 84 70
50 - 59 84 100 121 210 108
60 + 34 127 138 212 387
White Black Hisp Other
White 3806 29 30 20
Black 40 283 4 3
Hisp 66 6 120 1
Other 21 5 3 34
Source:
Marsden, P.V. 1988. Homogeneity in confiding
relations. Social Networks 10: 57‐76.
General Social Survey 1985. Ego network study of 1500 Americans
• Rows are egos
• Columns are alters
• Cells are no. of ties
from type of ego to
type of alter
25. Homophily: preference vs opportunity
• With ENRDs we have information on ties but not non‐ties
– We can measure homophily as outcome, but not homophily as
choice
• Adequacy of homophily in ENRD depends on research
question
– If am American and 95% of my friends are American, this clearly
has certain effects on me
• even if this is only because 95% of people in my world are American
• So ENRD is ok
– But if I am trying to measure nationalistic tendencies, I need to
know whether 95% is more or less than expected if a person
were making choices without regard for nationality
• If 95% of my non‐ties are also American, we know that I am not
showing any preference for Americans – low nationalism score
26. Comparing individuals
• With ENRD, can we at least
compare egos to each other?
– Some ego’s have higher E‐I
index than others. Is this
interpretable as preference?
• In principle, yes
– if egos are drawn from the
same population, then …
– … significantly higher
homophily score indicates
greater preference for own
kind
• In practice, not clear what
“same population” means
– People live in segregated
worlds due to choices made by
others
• Example: Are male or female
students here at UAB more
homophilous with respect to
ethnic background?
• For each person, we measure
homophily using %H or E‐I
– Run t‐test/anova to compare
genders
• If all students face same ethnic
environment, then significant
difference in avg homophily is
meaningful as difference in
preference
29. Perspectives of action in SNA
Structuralist
In the social production of their existence,
men inevitably enter into definite relations,
which are independent of their will, namely
relations of production appropriate to a given
stage in the development of their material
forces of production. The totality of
these relations of production constitutes the
economic structure of society, the real
foundation, on which arises a legal and
political superstructure and to which
correspond definite forms of social conscious‐
ness. The mode of production of material life
conditions the general process of social,
political and intellectual life. It is not the
consciousness of men that determines their
existence, but their social existence that
determines their consciousness.
– Marx 1859 Preface to A Contribution to the
Critique of Political Economy
Cognitivist
“If men define situations as real, they are
real in their consequences”
– W.I. Thomas
Success
information
Actual no. of ties
confidence
Perceived no. of ties
33. Perception and ENRD
• With ENRD, all ties are perceived by ego
• Therefore, ENRD works well when …
• Predicting ego’s own behavior
• Predicting ego outcomes based on ego’s behavior
• Predicting ego outcomes AND we can assume ego is
accurate in perceiving ties
• Hard to use ENRD when the topic of interest is
understanding perceptual accuracy
– Can use hybrid designs where the alters are
interviewed about ties with ego
47. Slopes and intercepts
• Intercept is general
tendency to name others as
friends
– Gregariousness
• Slope is increase in friends
over time
• Can model via HLM
– Time is L1 unit
– Person is L2 unit
• L2 regression models slope
& intercept as function of
ego characteristics
– Optimism
– Social ability
person intercept slope
1 1.676 0.382
2 1.743 0.732
3 4.362 0.146
4 2.848 0.461
5 3.000 0.475
6 1.133 0.400
7 3.914 0.111
8 0.095 0.471
9 2.800 0.500
10 ‐1.029 0.329
11 3.276 0.307
12 1.933 0.325
13 2.638 0.379
14 2.581 0.286
15 2.524 ‐0.132
16 2.248 0.261
17 2.086 0.439
high increase
low increase
decline
49. T1Size T1 ties 3
T2Size T2 ties 3
NewTies Ties added at T2 2
LostTies Ties lost 2
KeptTies Ties present both time periods 1
AbsentTies Ties ABSENT both time periods 12
Changes for node RUSS
Changes within ego networks
T1
T2
How many ties that each node
add/drop between time points?
50. Egonet changes
T1
Size
T2
Size
New
Ties
Lost
Ties
Kept
Ties
Abse
nt
Ties
HOLLY 3 3 2 2 1 12
BRAZEY 3 3 2 2 1 12
CAROL 3 3 1 1 2 13
PAM 3 3 1 1 2 13
PAT 3 3 2 2 1 12
JENNIE 3 3 0 0 3 14
PAULINE 3 3 1 1 2 13
ANN 3 3 1 1 2 13
MICHAEL 3 3 0 0 3 14
BILL 3 3 1 1 2 13
LEE 3 3 1 1 2 13
DON 3 3 0 0 3 14
JOHN 3 3 1 1 2 13
HARRY 3 3 1 1 2 13
GERY 3 3 1 1 2 13
STEVE 3 3 0 0 3 14
BERT 3 3 1 1 2 13
RUSS 3 3 2 2 1 12
Women Men
------ ------
1 Mean 1.750 2.200
2 Std Dev 0.661 0.600
3 Sum 14.000 22.000
4 Variance 0.438 0.360
5 SSQ 28.000 52.000
6 MCSSQ 3.500 3.600
7 Euc Norm 5.292 7.211
8 Minimum 1.000 1.000
9 Maximum 3.000 3.000
10 N of Obs 8.000 10.000
Difference Sig
========== =====
-0.450 0.157
Number of ties KEPT
Significance for t‐test obtained via
randomization method
WomenMen
52. Modeling change as a function of
group membership
‐1 0 1
0 7 151 2
1 14 112 20
Chi‐Square 22.25 p = 0.001
Pearson Corr 0.10 P = 0.029
‐1 0 1 Odds Odds Ratio
0 0.044 0.944 0.013 0.013
12.540
1 0.096 0.767 0.137 0.159
Whether
alter is same
group as ego
Relationship improved (1),
worsened (‐1) or stayed same
P‐value constructed via QAP permutation test
57. Summary effects vs underlying
tendencies
• Measurements of network size, homophily,
propinquity etc can be used in two ways
– Summary of ego’s exposure to what flows
• Function of opportunities provided by environment
– Indication of ego’s strategies in tie formation
• Choices being made by the ego
• Examples
– Network size vs ability to make friends
– Observed exogamy vs preference for out marriage
ENRD FNRD
Overall effects Underlying tendencies
Consequences of homophily Reveal cognitive characteristics
58. Structuralist vs cognitivist mechanisms
• Some theoretical
mechanisms imply that
perceptions of the
network don’t matter
– Information benefits of
central position
• Others depend crucially
on perceptions
– My behavior is based on
my perceptions
• Outcomes vs behavior
• In pure ENRDs, all ties
are perceived
– Lack of true incoming
ties
– Very strong for
understanding behavior
– For understanding
outcomes, we need
additional assumption of
accuracy of perception
– People vary in
perception accuracy