Doreian, Batagelj, Ferligoj - Generalized Blockmodeling

'
&
$
%
Photo: Vladimir Batagelj, UNI-LJ
Blockmodeling
Anu ska Ferligoj
University of Ljubljana
Budapest, 16 September 2009
A. Ferligoj: Cluster Analysis 1
'
&
$
%
Outline
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Cluster, clustering, blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
4 Equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
9 Establishing blockmodels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
11 Indirect approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
19 Direct approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
24 Generalized blockmodeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
26 Pre-specied blockmodeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
31 Symmetric-acyclic blockmodel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
34 Blockmodeling of vauled networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
35 Blockmodeling of two-mode network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
39 Blockmodeling of three-way networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
42 Application: Clustering of Slovenian sociologists . . . . . . . . . . . . . . . . . . . . . . . 42
48 Open problems in blockmodeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Budapest, September 16, 2009
LL
S
L G L
S
L L
'
&
$
%
Introduction
The goal of blockmodeling is to reduce a
large, potentially incoherent network to a
smaller comprehensible structure that can be
interpreted more readily. Blockmodeling,
as an empirical procedure, is based on the
idea that units in a network can be grouped
according to the extent to which they are
equivalent, according to some meaningful
denition of equivalence (structural (Lorrain
and White 1971), regular (White and Re-
itz 1983), generalized (Doreian, Batagelj,
Ferligoj 2005).
LL
S
L G L
S
L L
'
&
$
%
Cluster, clustering, blocks
One of the main procedural goals of blockmodeling is to identify, in a given network
N = (U, R), R U U, clusters (classes) of units that share structural charac-
teristics dened in terms of R. The units within a cluster have the same or similar
connection patterns to other units. They form a clustering C = {C
1
, C
2
, . . . , C
k
}
which is a partition of the set U. Each partition determines an equivalence relation
(and vice versa). Let us denote by the relation determined by partition C.
A clustering C partitions also the relation R into blocks
R(C
i
, C
j
) = R C
i
C
j
Each such block consists of units belonging to clusters C
i
and C
j
and all arcs leading
from cluster C
i
to cluster C
j
. If i = j, a block R(C
i
, C
i
) is called a diagonal block.
LL
S
L G L
S
L L
'
&
$
%
The Everett network
a b c d e f g h i j
a 0 1 1 1 0 0 0 0 0 0
b 1 0 1 0 1 0 0 0 0 0
c 1 1 0 1 0 0 0 0 0 0
d 1 0 1 0 1 0 0 0 0 0
e 0 1 0 1 0 1 0 0 0 0
f 0 0 0 0 1 0 1 0 1 0
g 0 0 0 0 0 1 0 1 0 1
h 0 0 0 0 0 0 1 0 1 1
i 0 0 0 0 0 1 0 1 0 1
j 0 0 0 0 0 0 1 1 1 0
a c h j b d g i e f
a 0 1 0 0 1 1 0 0 0 0
c 1 0 0 0 1 1 0 0 0 0
h 0 0 0 1 0 0 1 1 0 0
j 0 0 1 0 0 0 1 1 0 0
b 1 1 0 0 0 0 0 0 1 0
d 1 1 0 0 0 0 0 0 1 0
g 0 0 1 1 0 0 0 0 0 1
i 0 0 1 1 0 0 0 0 0 1
e 0 0 0 0 1 1 0 0 0 1
f 0 0 0 0 0 0 1 1 1 0
A B C
A 1 1 0
B 1 0 1
C 0 1 1
LL
S
L G L
S
L L
'
&
$
%
Equivalences
Regardless of the denition of equivalence used, there are two basic approaches to the
equivalence of units in a given network (compare Faust, 1988):
the equivalent units have the same connection pattern to the same neighbors;
the equivalent units have the same or similar connection pattern to (possibly)
different neighbors.
The rst type of equivalence is formalized by the notion of structural equivalence and
the second by the notion of regular equivalence with the latter a generalization of the
former.
LL
S
L G L
S
L L
'
&
$
%
Structural equivalence
Units are equivalent if they are connected to the rest of the network in identical ways
(Lorrain and White, 1971). Such units are said to be structurally equivalent.
In other words, X and Y are structurally equivalent iff:
s1. XRY YRX s3. Z U\ {X, Y} : (XRZ YRZ)
s2. XRX YRY s4. Z U\ {X, Y} : (ZRX ZRY)
LL
S
L G L
S
L L
'
&
$
%
. . . Structural equivalence
The blocks for structural equivalence are null or complete with variations on diagonal
in diagonal blocks.
LL
S
L G L
S
L L
'
&
$
%
Regular equivalence
Integral to all attempts to generalize structural equivalence is the idea that units are
equivalent if they link in equivalent ways to other units that are also equivalent.
White and Reitz (1983): The equivalence relation on Uis a regular equivalence on
network N = (U, R) if and only if for all X, Y, Z U, X Y implies both
R1. XRZ W U : (YRW W Z)
R2. ZRX W U : (WRY W Z)
LL
S
L G L
S
L L
'
&
$
%
. . . Regular equivalence
Theorem 1.1 (Batagelj, Doreian, Ferligoj, 1992) Let C = {C
i
} be a partition
corresponding to a regular equivalence on the network N = (U, R). Then each
block R(C
u
, C
v
) is either null or it has the property that there is at least one 1 in
each of its rows and in each of its columns. Conversely, if for a given clustering C,
each block has this property then the corresponding equivalence relation is a regular
equivalence.
The blocks for regular equivalence are null or 1-covered blocks.
LL
S
L G L
S
L L
'
&
$
%
Establishing blockmodels
The problem of establishing a partition of units in a network in terms of a selected
type of equivalence is a special case of clustering problem that can be formulated as
an optimization problem (, P) as follows:
Determine the clustering C
for which
P(C
) = min
C
P(C)
where is the set of feasible clusterings and P is a criterion function.
LL
S
L G L
S
L L
'
&
$
%
Criterion function
Criterion functions can be constructed
indirectly as a function of a compatible (dis)similarity measure between pairs of
units, or
directly as a function measuring the t of a clustering to an ideal one with perfect
relations within each cluster and between clusters according to the considered
types of connections (equivalence).
LL
S
L G L
S
L L
'
&
$
%
Indirect approach
R
Q
D
hierarchical algorithms,
relocation algorithm, leader algorithm, etc.
RELATION
DESCRIPTIONS
OF UNITS
original relation
path matrix
triads
orbits
DISSIMILARITY
MATRIX
STANDARD
CLUSTERING
ALGORITHMS
LL
S
L G L
S
L L
'
&
$
%
Dissimilarities
The dissimilarity measure d is compatible with a considered equivalence if for each
pair of units holds
X
i
X
j
d(X
i
, X
j
) = 0
Not all dissimilarity measures typically used are compatible with structural equiva-
lence. For example, the corrected Euclidean-like dissimilarity
d(X
i
, X
j
) =
(r
ii
r
jj
)
2
+ (r
ij
r
ji
)
2
+
n
s=1
s=i,j
((r
is
r
js
)
2
+ (r
si
r
sj
)
2
)
is compatible with structural equivalence.
The indirect clustering approach does not seem suitable for establishing clusterings
in terms of regular equivalence since there is no evident way how to construct a
compatible (dis)similarity measure.
LL
S
L G L
S
L L
'
&
$
%
Example: Support network among informatics students
The analyzed network consists of social support exchange relation among fteen
students of the Social Science Informatics fourth year class (2002/2003) at the Faculty
of Social Sciences, University of Ljubljana. Interviews were conducted in October
2002.
Support relation among students was identied by the following question:
Introduction: You have done several exams since you are in the second class
now. Students usually borrow studying material from their colleagues.
Enumerate (list) the names of your colleagues that you have most often
borrowed studying material from. (The number of listed persons is not
limited.)
LL
S
L G L
S
L L
'
&
$
%
Class network - graph
b02
b03
g07
g09
g10
g12
g22
g24
g28
g42
b51
g63
b85
b89
b96
class.net
Vertices represent students
in the class; circles girls,
squares boys. Recipro-
cated arcs are represented by
edges.
LL
S
L G L
S
L L
'
&
$
%
Class network matrix
Pajek - shadow [0.00,1.00]
b02
b03
g07
g09
g10
g12
g22
g24
g28
g42
b51
g63
b85
b89
b96
b
0
2
b
0
3
g
0
7
g
0
9
g
1
0
g
1
2
g
2
2
g
2
4
g
2
8
g
4
2
b
5
1
g
6
3
b
8
5
b
8
9
b
9
6
LL
S
L G L
S
L L
'
&
$
%
Indirect approach
Pajek - Ward [0.00,7.81]
b51
b89
b02
b96
b03
b85
g10
g24
g09
g63
g12
g07
g28
g22
g42
Using Corrected Euclidean-like
dissimilarity and Ward clustering
method we obtain the following
dendrogram.
From it we can determine the num-
ber of clusters: Natural cluster-
ings correspond to clear jumps in
the dendrogram.
If we select 3 clusters we get the
partition C.
C = {{b51, b89, b02, b96, b03, b85, g10, g24},
{g09, g63, g12}, {g07, g28, g22, g42}}
LL
S
L G L
S
L L
'
&
$
%
Partition into three clusters (Indirect approach)
b02
b03
g07
g09
g10
g12
g22
g24
g28
g42
b51
g63
b85
b89
b96
On the picture, ver-
tices in the same
cluster are of the
same color.
LL
S
L G L
S
L L
'
&
$
%
Matrix
b02
b03
g10
g24 C1
b51
b85
b89
b96
g07
g22 C2
g28
g42
g09
g12 C3
g63
b
0
2
b
0
3
g
1
0
g
2
4

C
1
b
5
1
b
8
5
b
8
9
b
9
6
g
0
7
g
2
2

C
2
g
2
8
g
4
2
g
0
9
g
1
2

C
3
g
6
3
The partition can be used
also to reorder rows and
columns of the matrix repre-
senting the network. Clus-
ters are divided using blue
vertical and horizontal lines.
LL
S
L G L
S
L L
'
&
$
%
Direct approach
The second possibility for solving the blockmodeling problem is to construct an
appropriate criterion function directly and then use a local optimization algorithm to
obtain a good clustering solution.
Criterion function P(C) has to be sensitive to considered equivalence:
P(C) = 0 C denes considered equivalence.
LL
S
L G L
S
L L
'
&
$
%
Criterion function
One of the possible ways of constructing a criterion function that directly reects
the considered equivalence is to measure the t of a clustering to an ideal one with
perfect relations within each cluster and between clusters according to the considered
equivalence.
Given a clustering C = {C
1
, C
2
, . . . , C
k
}, let B(C
u
, C
v
) denote the set of all ideal
blocks corresponding to block R(C
u
, C
v
). Then the global error of clustering C can
be expressed as
P(C) =
C
u
,C
v
C
min
BB(C
u
,C
v
)
d(R(C
u
, C
v
), B)
where the term d(R(C
u
, C
v
), B) measures the difference (error) between the block
R(C
u
, C
v
) and the ideal block B. d is constructed on the basis of characterizations
of types of blocks. The function d has to be compatible with the selected type of
equivalence.
LL
S
L G L
S
L L
'
&
$
%
Empirical blocks
a b c d e f g
a 0 1 1 0 1 0 0
b 1 0 1 0 0 0 0
c 1 1 0 0 0 0 0
d 1 1 1 0 0 0 0
e 1 1 1 0 0 0 0
f 1 1 1 0 1 0 1
g 0 1 1 0 0 0 0
Ideal blocks
a b c d e f g
a 0 1 1 0 0 0 0
b 1 0 1 0 0 0 0
c 1 1 0 0 0 0 0
d 1 1 1 0 0 0 0
e 1 1 1 0 0 0 0
f 1 1 1 0 0 0 0
g 1 1 1 0 0 0 0
Number of
inconsistencies
for each block
A B
A 0 1
B 1 2
The value of the criterion function is the sum of all incon-
sistencies P = 4.
LL
S
L G L
S
L L
'
&
$
%
Local optimization
For solving the blockmodeling problem we use the relocation algorithm:
Determine the initial clustering C;
repeat:
if in the neighborhood of the current clustering C
there exists a clustering C
such that P(C
) < P(C)
then move to clustering C
.
The neighborhood in this local optimization procedure is determined by the following
two transformations:
moving a unit X
k
from cluster C
p
to cluster C
q
(transition);
interchanging units X
u
and X
v
from different clusters C
p
and C
q
(transposition).
LL
S
L G L
S
L L
'
&
$
%
Partition into three clusters: Direct solution (unique)
b02
b03
g07
g09
g10
g12
g22
g24
g28
g42
b51
g63
b85
b89
b96
This is the same par-
tition and has the
number of inconsis-
tencies.
LL
S
L G L
S
L L
'
&
$
%
Generalized blockmodeling
1 1 1 1 1 1 0 0
1 1 1 1 0 1 0 1
1 1 1 1 0 0 1 0
1 1 1 1 1 0 0 0
0 0 0 0 0 1 1 1
0 0 0 0 1 0 1 1
0 0 0 0 1 1 0 1
0 0 0 0 1 1 1 0
C
1
C
2
C
1
complete regular
C
2
null complete
LL
S
L G L
S
L L
'
&
$
%
Generalized equivalence / block types
Y
1 1 1 1 1
X 1 1 1 1 1
1 1 1 1 1
1 1 1 1 1
complete
Y
0 1 0 0 0
X 1 1 1 1 1
0 0 0 0 0
0 0 0 1 0
row-dominant
Y
0 0 1 0 0
X 0 0 1 1 0
1 1 1 0 0
0 0 1 0 1
col-dominant
Y
0 1 0 0 0
X 1 0 1 1 0
0 0 1 0 1
1 1 0 0 0
regular
Y
0 1 0 0 0
X 0 1 1 0 0
1 0 1 0 0
0 1 0 0 1
row-regular
Y
0 1 0 1 0
X 1 0 1 0 0
1 1 0 1 1
0 0 0 0 0
col-regular
Y
0 0 0 0 0
X 0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
null
Y
0 0 0 1 0
X 0 0 1 0 0
1 0 0 0 0
0 0 0 1 0
row-functional
Y
1 0 0 0
0 1 0 0
X 0 0 1 0
0 0 0 0
0 0 0 1
col-functional
LL
S
L G L
S
L L
'
&
$
%
Pre-specied blockmodeling
In the previous slides the inductive approaches for establishing blockmodels for a
set of social relations dened over a set of units were discussed. Some form of
equivalence is specied and clusterings are sought that are consistent with a specied
equivalence.
Another view of blockmodeling is deductive in the sense of starting with a blockmodel
that is specied in terms of substance prior to an analysis.
In this case given a network, set of types of ideal blocks, and a reduced model, a
solution (a clustering) can be determined which minimizes the criterion function.
LL
S
L G L
S
L L
'
&
$
%
Types of pre-specied blockmodels
The pre-specied blockmodeling starts with a blockmodel specied, in terms of
substance, prior to an analysis. Given a network, a set of ideal blocks is selected, a
family of reduced models is formulated, and partitions are established by minimizing
the criterion function.
The basic types of models are:
* * *
* 0 0
* 0 0
* 0 0
* * 0
? * *
* 0 0
0 * 0
0 0 *
core - hierarchy clustering
periphery
LL
S
L G L
S
L L
'
&
$
%
Pre-specied blockmodeling example
We expect that core-periphery model exists in the network: some students having
good studying material, some not.
Prespecied blockmodel: (com/complete, reg/regular, -/null block)
1 2
1 [com reg] -
2 [com reg] -
Using local optimization we get the partition:
C = {{b02, b03, b51, b85, b89, b96, g09},
{g07, g10, g12, g22, g24, g28, g42, g63}}
LL
S
L G L
S
L L
'
&
$
%
2 Clusters Solution
b02
b03
g07
g09
g10
g12
g22
g24
g28
g42
b51
g63
b85
b89
b96
LL
S
L G L
S
L L
'
&
$
%
Model
g07
g10
g12
g22
g24
g28
g42
g63
b85
b02
b03
g09
b51
b89
b96
g
0
7
g
1
0
g
1
2
g
2
2
g
2
4
g
2
8
g
4
2
g
6
3
b
8
5
b
0
2
b
0
3
g
0
9
b
5
1
b
8
9
b
9
6
Image and Error Matrices:
1 2
1 reg -
2 reg -
1 2
1 0 3
2 0 2
Total error = 5
center-periphery
LL
S
L G L
S
L L
'
&
$
%
Symmetric-acyclic blockmodel
Let us introduce a new type of permitted diagonal block and a new type of blockmodel,
respectively labeled the symmetric block and the acyclic blockmodel. The idea
behind these two objects comes from a consideration of the idea of a ranked clusters
model from Davis and Leinhardt (1972). We examine the form of these models
and approach them with the perspective of pre-specied blockmodeling (Doreian,
Batagelj, Ferligoj 2000, 2005).
We say that a clustering C = {C
1
, . . . , C
k
} over the relation R is a symmetric-acyclic
clustering iff
all subgraphs induced by clusters from C (diagonal blocks) contain only
bidirectional arcs (edges); and
the remaining graph (without these subgraphs) is acyclic (hierarchical).
LL
S
L G L
S
L L
'
&
$
%
An example of symmetric-acyclic blockmodel
The pre-specied blockmodel is acyclic with symmetric diagonal blocks. An example
of a clustering into 3 clusters symmetric-acyclic type of models can be specied in the
following way:
sym nul nul
nul,one sym nul
nul,one nul,one sym
The criterion function for a symmetric-acyclic blockmodel is simply a count of the
number of (1) asymmetric ties in diagonal blocks and (2) non-zero ties above the
diagonal.
LL
S
L G L
S
L L
'
&
$
%
Table 1: An Ideal symmetric-acyclic network
a . 1 1 1 . . . . . . . . . . . . . . . . . .
b 1 . 1 1 1 . . . . . . . . . . . . . . . . .
c 1 1 . 1 1 . . . . . . . . . . . . . . . . .
d 1 1 1 . 1 . . . . . . . . . . . . . . . . .
e . 1 1 1 . . . . . . . . . . . . . . . . . .
f 1 1 . 1 1 . 1 1 . . . . . . . . . . . . . .
g . 1 1 1 . 1 . 1 . . . . . . . . . . . . . .
h 1 1 . 1 1 1 1 . . . . . . . . . . . . . . .
i 1 . 1 1 1 . . . . 1 1 1 . . . . . . . . . .
j 1 1 1 1 1 . . . 1 . . . . . . . . . . . . .
k 1 1 1 . 1 . . . 1 . . 1 . . . . . . . . . .
l . 1 1 1 . . . . 1 . 1 . . . . . . . . . . .
m . 1 1 1 . . 1 . . . . . . 1 1 1 1 . . . . .
n 1 1 . . 1 1 . . . . . . 1 . 1 1 1 . . . . .
o 1 . 1 1 . 1 . 1 . . . . 1 1 . 1 1 . . . . .
p . 1 1 . 1 . 1 . . . . . 1 1 1 . 1 . . . . .
q . 1 . 1 1 . 1 . . . . . 1 1 1 1 . . . . . .
r . 1 . . . . . . 1 1 1 . . . . . . . 1 1 . 1
s 1 1 . . 1 . . . 1 1 1 1 . . . . . 1 . . 1 1
t 1 . 1 1 . . . . 1 . 1 . . . . . . 1 . . 1 1
u 1 1 1 . 1 . . . 1 1 . 1 . . . . . . 1 1 . 1
v 1 1 . 1 1 . . . . 1 1 1 . . . . . 1 1 1 1 .
LL
S
L G L
S
L L
'
&
$
%
Blockmodeling of vauled networks
Another interesting problem is the development of blockmodeling of valued networks (Batagelj,
Ferligoj 2000,

Ziberna 2007).
120 A.

Ziberna / Social Networks 29 (2007) 105126
Fig. 4. Optimal partition for valued blockmodeling obtained using null and max-regular blocks with m equal to 5 or using
null and sum-regular blocks with m equal to 10.
6.3. Binary blockmodeling
In this case binary blockmodeling does not produce as good results as the previous approaches.
It was explored on the matrix on Fig. 1 sliced at 1, 2, 3, 5, and 10 (values of ties equal or
grater to this values were recoded into ones). These values were suggested by the barplot on
Fig. 5.
Fig. 5. Histogram of cell values for the matrix on Fig. 1.
Ziberna (2007) proposed several approaches to

generalized blockmodeling of valued networks,
where values of the ties are assumed to be mea-
sured on at least interval scale.
The rst approach is a straightforward gen-
eralization of the generalized blockmodeling
of binary networks (Doreian, P., Batagelj, V.,
Ferligoj, A., 2005) to valued blockmodeling.
The second approach is homogeneity blockmod-
eling. The basic idea of homogeneity blockmod-
eling is that the inconsistency of an empirical
block with its ideal block can be measured by
within block variability of appropriate values.
LL
S
L G L
S
L L
'
&
$
%
Blockmodeling of two-mode network
It is also possible to formulate a generalized blockmodeling problem where the
network is dened by two sets of units and ties between them.
Therefore, two partitions row-partition and column-partition have to be determined.
Two-mode generalized blockmodeling can be done.
LL
S
L G L
S
L L
'
&
$
%
Supreme Court voting for twenty-six important decisions
Issue Label Br Gi So St OC Ke Re Sc Th
Presidential Election PE - - - - + + + + +
Criminal Law Cases
Illegal Search 1 CL1 + + + + + + - - -
Illegal Search 2 CL2 + + + + + + - - -
Illegal Search 3 CL3 + + + - - - - + +
Seat Belts CL4 - - + - - + + + +
Stay of Execution CL5 + + + + + + - - -
Federal Authority Cases
Federalism FA1 - - - - + + + + +
Clean Air Action FA2 + + + + + + + + +
Clean Water FA3 - - - - + + + + +
Cannabis for Health FA4 0 + + + + + + + +
United Foods FA5 - - + + - + + + +
NY Times Copyrights FA6 - + + - + + + + +
Civil Rights Cases
Voting Rights CR1 + + + + + - - - -
Title VI Disabilities CR2 - - - - + + + + +
PGA v. Handicapped Player CR3 + + + + + + + - -
Immigration Law Cases
Immigration Jurisdiction Im1 + + + + - + - - -
Deporting Criminal Aliens Im2 + + + + + - - - -
Detaining Criminal Aliens Im3 + + + + - + - - -
Citizenship Im4 - - - + - + + + +
Speech and Press Cases
Legal Aid for Poor SP1 + + + + - + - - -
Privacy SP2 + + + + + + - - -
Free Speech SP3 + - - - + + + + +
Campaign Finance SP4 + + + + + - - - -
Tobacco Ads SP5 - - - - + + + + +
Labor and Property Rights Cases
Labor Rights LPR1 - - - - + + + + +
Property Rights LPR2 - - - - + + + + +
The Supreme Court Justices and
their votes on a set of 26 impor-
tant decisions made during the
2000-2001 term, Doreian and Fu-
jimoto (2002).
The Justices (in the order in
which they joined the Supreme
Court) are: Rehnquist (1972),
Stevens (1975), OConner (1981),
Scalia (1982), Kennedy (1988),
Souter (1990), Ginsburg (1993)
and Breyer (1994).
LL
S
L G L
S
L L
'
&
$
%
. . . Supreme Court voting / a (4,7) partition
P
a
j
e
k

-

s
h
a
d
o
w

[
0
.
0
0
,
1
.
0
0
]
C
a
m
p
a
i
g
n
F
i
V
o
t
i
n
g
R
i
g
h
D
e
p
o
r
t
A
l
i
e
I
l
l
e
g
a
l
S
.
1
I
l
l
e
g
a
l
S
.
2
S
t
a
y
E
x
e
c
u
t
P
G
A
v
P
l
a
y
e
r
P
r
i
v
a
c
y
I
m
m
i
g
r
a
t
i
o
L
e
g
a
l
A
i
d
P
o
C
r
i
m
i
n
a
l
A
l
I
l
l
e
g
a
l
S
.
3
C
l
e
a
n
A
i
r
A
c
C
a
n
n
a
b
i
s
N
Y
T
i
m
e
s
S
e
a
t
B
e
l
t
U
n
i
t
e
d
F
o
o
d
C
i
t
i
z
e
n
s
h
i
P
r
e
s
E
l
e
c
t
i
F
e
d
e
r
a
l
i
s
m
F
r
e
e
S
p
e
e
c
h
C
l
e
a
n
W
a
t
e
r
T
i
t
l
e
I
V
T
o
b
a
c
c
o
L
a
b
o
r
I
s
s
u
e
P
r
o
p
e
r
t
y
Rehnquest
Thomas
Scalia
Kennedy
OConner
Breyer
Ginsburg
Souter
Stevens
upper conservative / lower liberal
LL
S
L G L
S
L L
'
&
$
%
Blockmodeling of three-way networks
The indirect approach to structural equivalence blockmodeling in 3-mode networks.
Indirect means embedding the notion of equivalence in a dissimilarity and deter-
mining it using clustering. A 3-mode network N over the basic sets X, Y and Z is
determined by a ternary relation R XY Z. The notion of structural equivalence
depends on which of the sets X, Y and Z are (considered) the same. There are three
basic cases:
all three sets are different
two sets are the same
all three sets are the same
LL
S
L G L
S
L L
'
&
$
%
Example: Articial dataset
Randomly generated ideal structure rndTest(c(5,6,4),c(35,35,35)):
LL
S
L G L
S
L L
'
&
$
%
Example: Solutions
19bfiko34dejl
muvwqrs257acgtxz68hnpy
0
5
0
0
1
0
0
0
1
5
0
0
2
0
0
0
Dendrogram of agnes(x = dist3m(t, 0, 1), method = "ward")
Agglomerative Coefficient = 1
dist3m(t, 0, 1)
H
e
ig
h
t
1349f
my27adr8gjnwx5bptzceilosv6hkqu
0
5
0
0
1
0
0
0
1
5
0
0
2
0
0
0
dist3m(t, 0, 2)
H
e
ig
h
t
145678abcqru39gjklosvx2defhmpyzintw
0
5
0
0
1
0
0
0
1
5
0
0
2
0
0
0
dist3m(t, 0, 3)
H
e
ig
h
t
LL
S
L G L
S
L L
'
&
$
%
Application: Clustering of Slovenian sociologists
The described blockmodeling approaches are applied to the collaboration network of
Slovenian sociologists.
The collaboration between sociologists is operationalized by coauthorship of publi-
cations. A tie between two researchers is measured by coauthorship of an original
article, a chapter of a monograph, or a scientic monograph in the years from 1996 to
2007.
The measured coauthorship network measures coauthorship only between Slovenian
researchers inside the eld of sociology. Nevertheless we know that they publish
together also with researchers from the other elds in Slovenia and also with the
researchers outside Slovenia.
The network consists of 95 units and 224 ties. Practically the whole network is one
component which means that all except 6 units are connected. These 6 units are in
three dyads and were excluded from further analysis. All analyses presented here are
for the 89 units in the single large component.
LL
S
L G L
S
L L
'
&
$
%
Indirect blockmodeling approach
First the indirect approach for structural equivalence for coauthorship network was
performed. The Ward dendrogram where the corrected euclidean-like dissimilarity
was used is presented in the dendrogram.
LL
S
L G L
S
L L
'
&
$
%
Pre-specied blockmodeling
In the case of the coauthorship network we can assume a core-periphery structure: one
or several clusters of strongly connected scientists that strongly collaborate among
themselves. The last cluster consists of scientists that do not collaborate among
themselves and also do not with the scientists of the other clusters. They publish
by themselves or with the researchers from the other scientic disciplines or with
researchers outside of Slovenia.
We performed the core-periphery pre-specied blockmodels (structural equivalence)
into two to ten clusters. The highest drop of the criterion function was obtained for
the blockmodels into three, ve, and eight clusters. As eight clusters have been shown
also by the dendrogram this clustering is further analyzed.
LL
S
L G L
S
L L
'
&
$
%
Core-periphery structure of the coauthorship network
LL
S
L G L
S
L L
'
&
$
%
If we permute rows and columns in the relational matrix in such a way that rst the
units of the rst cluster are given, than the units of the second cluster and so on, and
if we cut the relational matrix by clusters we obtain the table in the next slide. Here
a black square is drawn if the two researchers have at least one joint publication or a
white square if they have not publish any joint publication.
In the table the obtained blockmodel into eight clusters is presented (seven core
clusters and one peripherical one). The seven core clusters are rather small ones and
the peripherical large one.
Only cluster 4 is connected also to the rst and second clusters, all other core clusters
show cohesivness inside the cluster and nonconnectivity outside the cluster.
The rst two diagonal blocks are complete without any inconsistency compared to
the ideal complete blocks. All white squares in complete blocks and black squares in
zero blocks are inconsistencies. The number of all inconsistencies is the value of the
criterion function. For the blockmodel the value of the criterion function is 310.
LL
S
L G L
S
L L
'
&
$
%
Main results
There are several small core groups of the Slovenian sociologists that publish
among themselves and much less or not at all with the members of the other
groups.
The core groups overlap with the organizational structure: sociologists of each of
these core groups are members of a research center.
The periphery group composed by the sociologists that are mostly not coauthors
with the other Slovenian sociologists is very large (more than 56 %). Most of
them publish only as a single author. Probably some of them publish with the
other Slovenian nonsociologists or with the researchers outside of Slovenia.
The coauthorship network is quite a sparse one which tells us, that the preferred
publication culture by Slovenian sociologists is to publish as a single author.
More detailed analysis of the periphery group should be further studied.
LL
S
L G L
S
L L
'
&
$
%
Open problems in blockmodeling
The current, local optimization based, programs for generalized blockmodeling can
deal only with networks with at most some hundreds of units. What to do with larger
networks is an open question. For some specialized problems also procedures for
(very) large networks can be developed (Doreian, Batagelj, Ferligoj 2000; Zaver snik,
Batagelj 2004).
LL
S
L G L
S
L L

Doreian, Batagelj, Ferligoj - Generalized Blockmodeling

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Doreian, Batagelj, Ferligoj - Generalized Blockmodeling

Uploaded by

Copyright:

Available Formats

'

such that P(C

Ziberna (2007) proposed several approaches to

You might also like