Professional Documents
Culture Documents
1
MARCH,1949
1. Introduction
I n a n u m b e r o f b r a n c h e s of t h e social sciences one e n c o u n t e r s
p r o b l e m s of t h e a n a l y s i s o f r e l a t i o n s h i p s b e t w e e n t h e e l e m e n t s o f a
grouP. F r e q u e n t l y t h e r e s u l t s of t h e s e i n v e s t i g a t i o n s m a y b e p r e -
sented in d i a g r a m m a t i c f o r m as s o c i o g r a m s , o r g a n i z a t i o n c h a r t s ,
flow c h a r t s , a n d t h e like. W h e n t h e d a t a to b e a n a l y z e d a r e s u c h t h a t
a d i a g T a m of t h i s t y p e m a y be d r a w n , t h e a n a l y s i s a n d p r e s e n t a t i o n
of t h e r e s u l t s m a y b e g r e a t l y e x p e d i t e d b y u s i n g m a t r i x a l g e b r a , T h i s
p a p e r p r e s e n t s s o m e of t h e r e s u l t s o f a n i n v e s t i g a t i o n of t k i s a p p l i -
cation o f m a t r i c e s . I n i t i a l t r i a l s in t h e d e t e r m i n a t i o n of g r o u p s t r u c -
t u r e s i n d i c a t e t h a t t h e m a t r i x m e t h o d is n o t o n l y f a s t e r b u t a l s o less
p r o n e to e r r o r t h a n m a n u a l i n v e s t i g a t i o n . *
T h e second s e c t i o n of t h i s p a p e r p r e s e n t s c e r t a i n c o n c e p t s u s e d
in t h e a n a l y s i s a n d a s s o c i a t e s m a t r i c e s w i t h t h e g r o u p in q u e s t i o m
The third states the results obtained and the fourth gives illustra-
tions o f t h e i r a p p l i c a t i o n . F i n a l l y , s e c t i o n five c o n t a i n s a m a t h e m a t i -
c a l f o r m u l a t i o n o f t h e t h e o r y a n d d e r i v a t i o n of t h e r e s u l t s p r e s e n t e d
in section t h r e e .
2. Definitions
2.01. T h e t y p e s o f r e l a t i o n s h i p s w h i c h t h i s m e t h o d will h a n d l e
a r e : m a n a chooses m a n b as a f r i e n d , m a n a c o m m a n d s m a n b, a s e n d s
m e s s a g e s to b, a n d so f o r t h . Since in a g i v e n p r o b l e m w e c o n c e r n
*Some of these examples have been worked out by the Research Center for
Group Dynamics, Massachusetts Institute of Technology, in conjunction with
some of its research.
95
96 PSYCHOMETRIKA
b 1 0 0 1 ~
c 1 1 0 0
d 1 1 1 0
2.05. F r o m the X m a t r i x we e x t r a c t a symmetric m a t r i x S hav-
ing entries s~j determined by s~- = ss~ - - 1 if x~i ~---xj~ -~ 1, and other-
wise s~j ~ sj~ ~-~ 0 . All the s~mmetrie---s in t h e group are indicated in
the m a t r l x S . The S m a t r i x associated with t h e above X m a t r i x is:
0 1 0 0 7
1 0 0 1 i
0 0 0 0 |
0 1 0 0 J
*In the course of the present work it was brought to our attention that in
"A matrix approach tc the analysis of sociometric data," Socio,metry, 19~6, 9,
340-347, Elaine Forsyth and Leo Katz have used matrices o represent socio-
metric reIations. They considered a three-valued logic r a t h e r than the present
two-valued one, and the operations on the matrices are different from the ones
discussed in this paper.
R. D U N C A N LUCE AND ALBERT D. PERI%Y 97
set of m o r e than two people form a clique if they are all mutual
friends of one another. In addition, the definition specifies that sub-
sets of cliques are not cliques, so that in a clique of five friends we
shall not say that any three form a clique. Although the word "clique"
immediately suggests friendship, the definition is useful in the study
of other relationships.
2.10. This definition of clique has two possible weaknesses:
first, if each element of the group is related by ~- > to no more than
c other elements of the group, then we can detect only cliques with
at most c + 1 members; and second, there may exist ~Sthin the group
certain tightly knit subgroups which by the omission of a few s>nn-
metrics fail to satisfy the definition of a clique but which nonetheless
would be termed, non-technically, "cliques." It may be possible to
alleviate these difficulties by the introduction of so called "n-cliques"
which comprise the set of n elements which form two distinct n-chains
from each element of the set to itself. This requires that the n-chains
be redundant with the only recurring element being the end-point
and also that all the relations in the n-chains be symmetr/c.
This definition means that the four elements a,b,c, and d form
a 4-clique if the 4-chains (for example) a,b,c,d,a and a,d,c,b,a, both
exist. These by the definition of n-chain require that the relations
a<=> b, b < - - > c , c<--> d, d < - - > a
exist, but nothing is said about the relations between a and c, and
b and d . The original definition requires, in addition, that
a<=>e and b<=>d
for a,b,e, and d to form a clique of four members. Thus we see that
the definition of n-clique considers "circles" of symmetries, but it fails
to consider the symmetric "cross" terms that exist between the mem-
bers of the n-clique. These cross terms will be investigated, however,
by determining whether any m of these n-elements form an m-clique.
The usefulness of the definition of n-clique can be judged only
a f t e r experience has been gained in its application. This is not con-
veniently possible at present, unfortunately, because the problem of
the general determination of redundant n-chains has not been solved
(see 5.09).
The most general definition of a elique-like structure including
antimetries will not be discussed, for it is believed that this will not
be amenable to a concise mathematical formulation.
3. Statement of Results
3.01. In X ~ the entry x~j(~) ~ c if and only if there are c dis-
R. DUNCAN LUCE AND ALBERT D. PERRY 99
s,~9(~) = [ (4 - - 2) (4 - - 1) - - (0 - - 2) (0 - - 1)]
+ [(3--2) (3--1)-- (1--2) (1--1)]
+ [(5--2)(5--1)--(3--2)(3--1)] +2
-- 18 .
In the evaluation of this formula it is immaterial how the cliques are
numbered initially; however, it is essential once the n u m b e r i n g is
chosen t h a t we be consistent.
3.06. The r e d u n d a n t 2-chains of a m a t r i x X are the main diago-
nal e n t r i e s of X ~ (5.09). Thus f o r a m a t r i x
O
0 10 01 11 00 1
I 1 0
1
1
0
1
0
0
0
0
0
0
1
1
0
I
2I 00 01 01 21 1
X~__= 1 2 0 1 0 ,
1 2 0 1 0
0 1 1 2 0
the matrix of redundant 2-chains is
1 0 0 0 0
I
0 0 0 0 0
0 0 0 0 0
0 0 0 1 0
0 0 0 0 0
To obtain the m a t r i x of r e d u n d a n t 3-chains we compute the fol-
lowing matrix, in which the symbol R(:~) stands for the matrix of re-
d u n d a n t 2-chains:
X R c:2~ + R c 2 ~ X ~ S .
Deleting in this sum the main diagonal and replacing it by the main
diagonal of X 3 gives the m a t r i x of r e d u n d a n t 3-chains (5.09). If
t h e main diagonal of X R ~ + R ( = ~ X - - S is denoted b y y(3) and the
main diagonal of X C3) by Z (3), then let E ~3) = Z TM - - Y(a) and thus
the m a t r i x of r e d u n d a n t 3-chains, R (3), is given b y
R ~3~ = X R ' ~ + R c 2 ~ X + E TM - - S .
4. Examples
4.01. As the first example, let us compare and analyze the friend-
ship structure in the t w o following hypothetical groups. The matrices
are (where a blank entry indicates a zero) :
l,--i
~s
b~
~b
r j
~. ~ ~,-L~ ~..~l..~i,~
d~ ~b b~
c~ d)
b~
g,A9 ~"
,.,.i -q -.I
t~
oo O0
e~ ~J
b~
p~ t~
~ ~,~.
b
I .... I
I I 0
I
l l I -- I
i-.Ll-~l.~
b~
I-~
~--~ i~ . i-~
-',-1 ,..I
O0
b~ ~.~
b
I
J f -- I ~w . ;J
R. DUNCAN LUCE AND ALBERT D. PERRY 103
I II
1 2 3 4 56 7 8 9 10 12345678910
1 1415 14 1412 12 2 564 11 4
3
1514 14 1412 12
lr
2
3 5 241
2
11 1
4
5
1414 10
1
13 8 10
5 4
421
11
41
1 I
6
7
8
1414 13
1
1010 8 j
, 6 2
2
2 I
1212 8 10 6 7 1 2
9 9 1 141 2
10 1212 10 8 7 6 10 4 11 1 J
Since the corresponding m a i n diagonal t e r m s are non-zero, men 1, 2,
4, 7, 8, and 10 of group I are in cliques (3.03). These, w i t h 3 and 9
which have no symmetries in the group and 5 and 6 which are m u t u a l
friends, account for all members of the group. The t e r m s ss8(~ - -
S~.o.~.0(3~ - - 6 suggest a clique of f o u r m e m b e r s ; however, the-6xistence
of other mai--n diagonal terms makes it impossible to apply the for-
mula (t--2) ( t - - l ) (3.04). I n v e s t i g a t i n g in S first the elements 1,.
2, and 4 because t h e i r columns have the largest values in the t e n t h
row, we find t h a t elements 1, 2, 4, and 10 f o r m a clique of f o u r mere-
bers. In t h e eighth row the largest entries are in columns 1, 2, and
7, and an investigation reveals t h a t 1, 2, 7, and 8 f o r m a cl.ique of
four men, which then overlaps the first clique by the men 1 and 2. I n
row f o u r t h e largest entries are found in columns 1, 2, a n d 7. We
then find t h a t 1, 2, 4, and 7 f o r m a clique of f o u r elements which
104 PSYCHOMETRIKA
overlaps the previous two. All the men contained in cliques have been
accounted f o r at least once, and a check either with the formula f o r
main diagonal entries (3.05) or directly in the S m a t r i x indicates
that all the cliques have been discovered. This, coupled w i t h w h a t
we discovered in S ~, completely determines the symmetric s t r u c t u r e
of the first group.
F o r purposes of qualitative judgznent and a guide to carrying
out analysis, we note that the first t w o rows of S 3 present an inter-
esting s u m m a r y of the clique structure. The entries sl~ (3) and s2~(3)
have the largest values, next largest are in columns f o u r and seven,
and then finally in columns eight and ten. Men 1 and 2 a r e contained
in all three cliques, 4 and 7 are each contained in t w o cliques, and
finally men 8 and 10 are each in only one clique. This indicates that
the magnitude of the off-diagonal t e r m s determines to some extent
the amount and structural position of the overlap of cliques.
h group II there are only three elements with non-zero main-
diagonal entries, all with the value 2. This fits the formula (t--2)
( t - - l ) with t ------ 3 (3.04). Thus the men 1, 3, and 4 form a clique
of three members. R e t u r n i n g to S ~, we see t h a t t h e r e remains one
unaccounted s y m m e t r y each for men 4 and 9, hence 4 < - - > 9 .
In group I, the off-diagonal terms a r e large in magn.itude and
are quite dense in the array, w i t h some rows completely e m p t y or
with single entries in the S 3 matrix. This indicates a closely knit
g r o u p wit,h certain men definitely excluded. The S ~ m a t r i x f o r the
second group has f e w e r entries of a smaller value indicating a tess
tightly knit structure, b u t it has no e m p t y rows and only one row
with a single e n t r y ; t h a t is, it has f e w e r people t h a n gToup I who
are not accepted b y t h e group or w h o do not accept it.
A consideration of the mat~;ix X ~ S will give all t h e antimetries
in the groups and complete the analysis of the structures.
It. is clear t h a t this procedure gains s t r e n g t h as the complexity
of the problem increases, for the analysis of a twenty-element group
is little more difficult than t h a t of a ten-element group.
4.02. The second example is a communication system compris-
ing t w o - w a y links between seven stations such as might occur in a
telephone or telegraph circuit. The n u m b e r of channels of a given
number of steps (i.e., n-chains in the general theory) between any
t w o points and the minimum n u m b e r of steps required to complete
contact between t w o stations will be determined. Suppose the matrix
of one-step contacts is:
R. D U N C A N LUCE AND ALBERT D. PERRY 105
1234567
_
11 1-
2 1 1
3 11 11
4 1 111
5 11 11
6 11 1
7 1 111
which in this case is also the S matrix. Then two-step connections
are given by X~:
1 234567
1 -3 112210 t
2 1 211101
3 1 141123
4 2 114322]
5 2 113422j
6 1 022232
7 0 132224
and the three-step ones by X3:
12 3 4 56 7
24 8 4 44 8
42 5 3 33 3
85 41010 5 5
4 310 8 9 911
4 310 9 8911
43 5 9 96 8
8 3 511118 6
(2) (:2)
F r o m the former, the two connections 1 < ~- > 7 and 2 < ~ > 6 can-
not be realized because x17~) ~ x712) ~- 0 and x2+(~> ~ x+~<2) - - 0
(3.01). The contacts a r e possible i n - ~ r e e steps, however, since~ X 3
is completely filled. Thus two steps a r e sufficien~ for most contacts
and three steps f o r all.
In dete~Tnining the n u m b e r of paths between t w o points it is de-
sirable to eliminate r e d u n d a n t paths: F o r two-step communication
this is done by dele*ing the m a i n dia~onal of X ~. The r e m a i n i n g t e r m s
represent the n u m b e r of two-step paths between the stations indi-
cated. The m a t r i x of redundancies f o r three-step communication is
given by R ~+~ "-- X R ~2) ~+ R ( ~ > X + E ~> - - S (3.06), which works out
to be:
106 PSYCHOMETRIKA
1234567
1 246 6
2 425
3 65477
4 78767
5
6
7 6
77867
6656
7766 1
The matrix of non-redundant three-step communication paths is
X 3 ~ R(~):
1234567
1
2 2 43434323J
3 2 3355
4 433 234
5 4332 34
6 43533 2J
7 235442
We notice that the three-step paths between I and 2 and 2 and 3 are
all redundant but that there are two-step paths for these combin~-
tions. All other combinations have at least two t~ree-step paths join-
ing them.
5. Mathematical Theory
5.01 To c a r r y out the following mathematicM formulation and
the proofs of theorems it is convenient to use some of the symbolism
and nomenclature of point set theory. As there is some diversity in
the literature, the symbols used are:
Sets are either defined by enumeration or by properties of the
elements of the set in the form: symbol for the set [symbols used for
elements of the set I defining properties of these elements]. When
a single element i is treated as a set it will be denoted by (i), other-
wise sets will be denoted by upper case Greek letters.
The intersection of (elements common to) two sets P and
is denoted by F .
The union of two sets F and (elements contained in either or
both) is denoted by P + ~ . The context will make it clear whether
the symbol i+ refers to addition, m a t r i x addition, or union.
The inclusion of a set P in another set (all elements of P are
elements of ) is denoted by P < . The negation is P <* .
If < 2 , then the complement of with respect to 2 , % is de,
R. DUNCAN LUCE AND ALBERT D. PERRY 107
and if the indices were not so selected then a t least one x~, ~-- 0 . Thus
n-chains contribute 1 to the sum and other ordered sequences con-
t r i b u t e 0 . Since the indices take on each possible combination of
values j u s t once; every distinct n-chain is represented j u s t once. If
there are c such n-chains, then t h e r e a r e a total of c ones in the sum-
mation.
5.05. Theorem 2: An element of ~ has a m a i n diagonal value
of c in X ~ i f a n d only i f it is s y m m e t r i c ~'ith c elements of 2 .
P r o o f : Let ~ be the set of ]'s for which i < ~- > ]. B y definition
x . (~) = E x~jxj~ + E x ~ x j ~ = ~ ~+ ~,~.
je~ je"
~]~ " - c by theorem 1 (5.04) and N~ - - 0 because i and ] are not sym-
metric f o r ] s ~', so either x~j = 0 or xj~ ~---0 or both. Thus i f i is sym-
metric with c elements of 2 , x~j (~) : c .
R. DUNCAN LUCE AND. ALBERT D. PERRY 109
(/)+(k)<~
Proof: By definition
- - (i)+(k)<Y~
The set of all the pairs ] , k is the union of the following three mu-
tually exclusive sets:
T~ [ ] , } I t h e r e exists v such t h a t (]) + (k) < 0 , ; there does not
exist ,a such t h a t (]) :+ (k) < 0 ~ , (]) + (k) <" A~]
110 PSYCHOMETRIKA
= Z Z Z s~jsi~s~ .
V=l (j)+(k)<@ v
(j)+ (k)<*A v
Similarly
z={o
' (d'-
o 2). I) 11since AI--O.
Combining these,
~a
[
(tv - - 2 ) (tv - - 1 ) - -
(iv - - 2 ) ( t , - - 1)
(dv--2)(d~--l),~>l
, v -- 1.
R. DUNCAN LUCE AND ALBERT D. PERRY Iii
S u m m i n g o v e r ~, gives
+ (tl - - 2) (tl - - 1)
Since the entries szj(3) are uniquely determined from the entries
of S by the laws of matrix multiplication, all valid methods of cal-
culating s~ (a) will give the same result. Specifically, in the above
formula the numbering of the cliques is immaterial.
Similar formulas to that just deduced may be given for the off-
diagonal terms of S 3, but they are considerably more complex, and,
to date, they have not been found useful in applications.
5.08. T h e o r e m 5: I f 1) 0 is a set of t m e m b e r s w i t h t > 3 ,
2) s~/~) ~ ( t - - 2 ) ( t - - l ) f o r i c o n t a i n e d in O , a n d 3) sj/~) = 0 f o r ]
c o n t a i n e d in O', t h e n O is a clique of t m e m b e r s .
P r o o f : T h e r e a r e t w o cases:
S~/~) = Z Z s~jsjks~
U)+(~)<o
+ Z E s~jsj~sk~.
(j)+(k)<*O
S u (~) - - Y. Y~ si~sj~s~i - - 2 = ( t - - 2 ) ( t - - l ) - - 2
(j)+(k)<~
2. The B, class: There exists one and only one redundant pair
(k,m) and it has the property:
(k,m) (i,]) = i .
3. The C~ class: There exists one and only one redundant pair
(k,m) and it has the property:
( k , m ) (i,]) = ] .
4. The D,~ class: There exist two and only two redundant pairs
(k,m) and (p,q) and they have the properties:
(k,m) (i,D = i
( p , q ) (i,]) = ] .
5. The E, class: There exists one and only one redundant pair
(k,m) and it has the property:
(k,m) (i,j) = (i,i).
R ( z ) r-- 0
A (8) - - 0 .
(k,m) (i,D = i
(p,q) (i,]) -" ].
(p)+(q)<,~
or
A (~) = X R ( ~ ' 2 ) X .
114 PSYCHOMETRIKA
D -- T.. (3)
Substituting (3) in (2)
D-- U'DU
or
UD "- DU.
or
Since the dkj: are real and n is odd, equation (4) implies
u ~ [ ( G ~ ) v~ - - ( d . ) ""] = 0
or
B = u'Bu. (5)
By (1) and (3)
T~:D:B ,~ ,
6. A c k n o w l e d g e m e n t