PSYCH(~METRIKA~V0L. 14, No.

MARCH,1949

A METHOD OF MATRIX ANALYSIS

OF GROUP STRUCTURE

R. DUNCAN LUCE AND ALBERT D. PERRY

GRADUATE S T U D ~ T S ~ DEPARTME~qT OF MATHI~CIATICS
MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Matrix methods may be applied to the analysis of experimental

data concerning group structure when these data indicate relationships which can be depicted by line diagrams such as sociograms.
One may introduce two concepts, n-chain and clique, which have simple relationships to the powers of certain matrices. Using ,them it
is possible to determine the group structure by methods which are
both faster and more certain than less systematic methods. This
paper describes such a matrix method and applies it to the analysis
of practical examples. At several points some unsolved problems in
this field are indicated.

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 illustrations 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

ourselves with one sort of relation, no confusion arises f r o m ~eplacing the description of t h e relationship by a symbol : > . Thus, instead of " m a n i chooses m a n ] as a f r i e n d , " we w r i t e " i : > ]." If, on
t h e other hand, m a n i had not chosen m a n ], we would have w r i t t e n
" i if: > ] , " using t h e symbol :~" > to indicate the negation o f the relationship denoted by ~--- > .
2.02. Situations such as m u t u a l choice of f r i e n d s or two-way communication would thus be indicated by i ---~ > ] and ~
> i, or briefly,
i < - - > ]. We describe such situations by s a y i n g t h a t a s y m m e t r y
exists between i and ].
2,03. When the choice is not mutual, t h a t is i -~ > 3" or ] ~ > i
but not both, we say an a n t i m e t r y exists between i and ].
2.04.* The data to be analyzed are presented in a m a t r i x X as
follows: the i,] e n t r y (x~) has the value of 1 if i ~ > ] and the value
0 if i > ]. F o r convenience we place the re&in diagonal terms equal
to zero, i.e., x~ ~--- 0 for all i. This convention, i > i, does not restrict the applicability of the method, since there is little significance
in such statements as " J o n e s chooses himself as a f r i e n d . "
Suppose, f o r example, t h a t we had a gToup of f o u r members w i t h
the following relationships: a ~--- > b, b = > , b ~--~> d, d ~ > b,
c ~ > a, c ~ > b, d ~ > a, and d ~-~ > c. All other possible combinations of a,b,c, and d are related by the symbol ~ > . The X m a t r i x
associated with this group is:

b
c
d

1
1
1

0
1
1

0
0
1

1 ~
0
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 having entries s~j determined by s~- = ss~ - - 1 if x~i ~---xj~ -~ 1, and otherwise 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

1
0
0
1

0
0
0
0

0 7
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 sociometric 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

To ~ndicate the i,] entity of the m a t r i x X ~, which is the nth

power of X , we shall employ the symbol x~/~k Similarly, the i,] entry of S ~ is s~/~.
2.06. In the group considered above, we had a : > b, b ~- > d ,
and d : > c as three of the relations. I f the symbol___~ > indicates the
relationship "sends messages to," .it appears t h a t a can send a message to c in three steps, via b and d . We call this three-step path a
3-chain f r o m a to c. l~ather t h a n w r i t i n g out the above sequence of
relations, we m a y omit the symbol : > and simply write the 3-chain
as a,b,d,c.
In a group involving more elements one m i g h t have the 5-chains:
a,e,c,b,d,f and a,d,b,c,d,e. We notice t h a t the first sequence involves
five steps between six elements of the group. The second sequence
also involves five steps but only five elements of the group, since the
element d appears as both the second and fifth member of the sequence. Thus, although these two five-step sequences contain different numbers of elements of the group,, t h e y both have six members.
Using this concept of membership in a sequence, an n-step sequence
has ~ + 1 members.
These examples of 3-~hains and 5-chains suggest a general definition f o r a property within the group: an ordered sequence with n + l
members, i , a , b, . - . , p , q, ] , is an n-chain f r o m i to ] if and only if

i--- > a , a ~

> b ,...,p~

> q ,q-~ > ].

2.07. When two n-chains have the same elements in the same
order, i.e., the same members, then t h e y are said to be equal, and
otherwise t h e y are distinct. I t is impol'tant in this definition of equality t h a t it be recognized t h a t both .the elements of the group and t h e i r
order in the sequence are considered. The two chains i,],k,l,p and
i,p,k,],l are distinct though t h e y contain the same five elements.
2.08. When the same element occurs more t h a n once in an
n-chain, the n-chain is said to be redundant. (Thus, in a group of
elements any n-chain w i t h n g r e a t e r t h a n m is r e d u n d a n t ) . The chains
a,b,e,d,b,c and a,c,a,b,d,c,e are, for example, both redundant, for the
element b occurs t~dce in the f o r m e r and the elements a a n d c both
occur twice in the latter. An example of a non-redundant 5-chain is

a,d,p,b ,q,] .
2.09. A subset of .the group f o r m s a c,lique provided t h a i it consists of three or more members each in the symmetric relation to
each other member of the subset, and provided f u r t h e r t h a t there
can be found no element outside the subset t h a t is in the s y m m e t r i c
relation to each of t h e elements o f the subset. The application of this
definition to the concept of friendship is immediate: it states t h a t a

98

PSYCHOMETRIKA

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 subsets 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>nnmetrics 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 members 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 conveniently 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

tinct n-chains f r o m i to j ( f o r p r o o f see 5.04). Thus, if in t h e fifth

power of a m a t r i x of data X we find t h a t the n u m b e r 9 occurs in t h e
third r o w of the s e v e n t h column, we m a y conclude t h a t there a r e 9
distinct 5-chains f r o m element 3 to element 7.
3.02. In X ~ the i ~h main diagonal e n t r y has t h e value m if a n d
only if i is in the s y m m e t r i c relation w i t h m elements of t h e g r o u p
(5.05). Since b y the definition of a clique each element i in a
clique of t m e m b e r s m u s t be in the symmetric relation to each of the
t--1 other elements, it is necessary t h a t x~ (:2~ > t--1 for i to b e in a
clique of t members. We m a y not, however, conclude f r o m t h e f a c t t h a t
x~ (~ > t--1 t h a t i is necessarily contained in a clique of t members.
3.03. An element i is contained in a clique if and only if t h e i ~
entry of the main diagonal of S ~ is positive (5.06). The main diagonal t e r m s of S 3 will be either 0 or even positive n u m b e r s in all cases,
and when the value of the e n t r y is 0 the associated element is not in
a clique.
3.04. If, in S 3, t entries of the main diagonal have the value
(t--2) (~--1) and all o t h e r entries of the main diagonal a r e zero, then
these t elements f o r m a clique of t m e m b e r s (5.08). It also follows
f r o m the next ~tatement (3.05) t h a t if there is only one clique
of t m e m b e r s then these t elements will have a main diagonal value
in S s of (t--2) ( t - - l ) . The f o r m e r s t a t e m e n t is, however, t h e more
significant in analysis, for it is the aim to go f r o m the matrix representation to the g r o u p structure. There is no difficulty in going f r o m
the structure to the matrices.
3.05. Since b y s t a t e m e n t 3.03 the main diagonal values of S s
are dependent only on t h e clique s t r u c t u r e of the group, it is to be
expected t h a t a formula relating these values and the clique structure is p o s s i b l e . I f an element i is contained in m different cliques
each having t~ members, and if there are d~ elements common to the
k t~' clique and all the preceding ones, then
s ~ ~3~ - - Y~ { (iv - - 2 ) (iv - - 1) - - (d~ - - 2 ) (d~ - - I ) ) + 2
V=l

(5;07). Thus, if we have three cliques: (5,7,9,10), (1,4,9), and

(1,2,5,9,11), then d~ -- 0, for there are no preceding cliques; d___--~1,
for only element 9 is common to the second and first cliques; and
d~ -- 3, for clique three has the elements 1,5, and 9 common wi~h
t h e first t w o cliques. Substituting ~ ~ 4 , t~ - - 3 , ~ ---- 5 , d ~ - 0 ,
d - - 1, and d3 ---- 3 and evaluating t I ~ - f o r m u - ~ f o r element 9, which
is the only one common to all three cliques, w e obtain

100

PSYCHOMETRIKA
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 diagonal e n t r i e s of X ~ (5.09). Thus f o r a m a t r i x

X=

with the square

X~__=

O
0 10
1 0

01
0

11
0

00 1
1

1
1

0
0

0
0

1
0

I
I
I

0
1

2I
1
1
0

00
2
2
1

01
0
0
1

01
1
1
2

21 1
0
,
0
0

the matrix of redundant 2-chains is

1
0
0
0
0

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 following matrix, in which the symbol R(:~) stands for the matrix of red 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 .

It has not y e t been possible to develop formulas which will give

,the m a t r i x of r e d u n d a n t n-chains f o r n larger t h a n 3. W h a t w o r k
t h a t has been done in this direction is presented in 5.09.

1%. DUNCAN LUCE AND ALBERT D. PERRY

101

3.07. The several theorems on cliques give a method that to some

extent determines the clique structure independent of the rest of the
group structure. I t would be desirable to find a simple scheme t h a t
determines the clique structure directly. Since a certain amount of
knowledge in this direction can be obtained from S 3 , it is conjectured t h a t possibly there is a simple f o r m u l a relating clique structure to the n u m b e r s in S 3. As yet no such formula has been developed.
In a consideration of this problem, it was questioned w h e t h e r
certain aspects of the s t r u c t u r e would be lost in the multiplication,
which, i f true, might m a k e the discovel~y of the desired formula impossible. The following theorem shows t h a t neither t h e clique ~tructure n o r any of the properties of S are lost in the matrix S~: A n y
real s y m m e t r i c m a t r i x has one and only one real s y m m e t r i c n ~ root
if n is a positive odd integer (5.12). This theorem is somewhat m o r e
general than was required, since it does not restrict the entries in
the n t~ root to 0 and 1, and since it is true f o r a n y odd root r a t h e r
than j u s t the cube root. (In general the real symmetric even roots
are not unique.)
This theorem suggests a f u r t h e r problem to be solved: to find
a symmetric group structure which will insure the presence of certain prescribed minimum n-chain conditions f o r odd n . To c a r r y this
out it will probably prove necessary to discover a theorem that uses
not only the realness and s y m m e t r y of the S m a t r i x and i t s powers,
but in addition the fact that only the numbers 1 and 0 m a y be entries
in S .

4. Examples
4.01. As the first example, let us compare and analyze the friendship structure in the t w o following hypothetical groups. The matrices
are (where a blank entry indicates a zero) :

g,A9 ~"

l,--i

b~

....

p~

oo

,.,.i

c~

d~
d)
b~

b~
~b

~s

~.~

--

I-~

-',-1

e~

-q

t~

t~

~b

~.

r
~

~--~

~,-L~

--

~w

i~ .

i-.Ll-~l.~

i-~

~,~.

~..~l..~i,~

;J

O0

,..I

b~

O0

-.I

b~

~J
b~

R. DUNCAN LUCE AND ALBERT D. PERRY

103

(3.02). Thus, as f a r as symmetric relationships are concerned, these

men are isolated f r o m the group. In the same w a y we determine t h a t
5 and 6 each have j u s t one s y m m e t r i c friendship relation (s~5(~) - - s6J ~
-- 1, 3.02) w h i c h we determine to be 5 < - - > 6 f r o m the S matrix.
TKe remaining elements in S 2 f o r m a ra---fher dense set of quite large
numbers, which means, roughly, a t i g h t l y k n i t group.
In the second group, on the other hand, every m a n has a non-zero
main diagonal in S 2. The men 2, 5, 7, 8, and 10 each have a single
mutual fniend, which we determine to be: 2 < -~ > 6, 5 < - - > 1 ,
7 < -- > 6 , 8 < - - > 9 , and 10 < ~- > 1 . Then since s6J 2 ~ 2 a n d since
we-h-~ve j u s t cited 6's two mutua-l-friends, 6 need not b~-~onsidered
further. We note t h a t the off-diagonal areas of this S ~ m a t r i x are
not so completely filled as group I, indicating t h a t the group is not
so tightly bound,
The S 3 matrices indicate the differences in compactness of the
structures quite clearly:

1
3
4
5
6
7
8
9
10

I
1 2 3 4 56
1415
14
1514
14

7 8 9
1412
1412

10
12
12

1414

13

10

10

1
1414
1212
1212

13
8
10

lr

2
3

,
1010
10 6
8

5
6

II
12345678910
2
564
2
241
5
421
11
4

8 j
7
6

11

11 1
41
1

I
I

2
9
10

1
4

1
141
11

2
2
1

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 formula (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 merebers. 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 interesting 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 maindiagonal 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 comprising 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

1234567
11

2
3
4
5
6
7

1-

11

11
1
111
11
11
11
1
111

105

which in this case is also the S matrix. Then two-step connections

are given by X~:
1 234567
1
2
3
4
5
6

-3

112210
211101
141123
114322]
113422j
1 022232
0 132224

1
1
2
2

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 cannot 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 desirable 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 indicated. 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

1
2
3
4
5
6
7

1234567
246
6
425
65477
78767
77867
6656
6
7766

The matrix of non-redundant three-step communication paths is

X 3 ~

R(~):

1234567
1
2
3
4
5
6

2 43434323J
2
3355
433
234
4332
34
43533
2J
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 joining 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), otherwise 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

fined by + ' - - ~= and ' ----- 0 w h e r e 0 is the null set.

The inclusion of ~ single element i in a set is denoted by i ~ ~ .
For a n y two elements i and ] of a set 2 and a subset ~9 of 2:
(i) + (j) < ~2 i f and only if i s ~2 a n d j ~ ~ .
(i) + (]) <* t2 implies i e ;2' a n d / o r ] e ~2'.
The symbol 6i~ = l
--0

if
if

i=j
i=#]

5.02. Consider a finite set 3 of x elements denoted by 1 , 2 , . . . ,

i , . . . , ] , - . . , x f o r which ~here is defined a relationship - - > between
elements and its negation ~ > having the properties:
1.

E i t h e r i - - > ] or i ~ > ] f o r all i a n d ] e ~ .

2. i ~ = > i .
Let a n u m b e r x~j be associated w i t h i and 3"such t h a t
--1
----0

if
if

i-->]
i>].

A matrix X ---- [x~] is f o r m e d f r o m the numbers x~s. I t wilt be found

useful to denote the i,] e n t r y of the nth power of X , X ~, by x~/~.

A symmetry is said to exist between i and ] if a n d only i f i ---- > ]

and ] - - > i , in which case we m a y w r i t e i ' < = > ] . F o r the m a t r i x
X this--requires t h a t x~s ---- x~ ----- 1. If, however, e i t h e r _ / / = > ] and
] = > i or i > ] and ] = > i then an antimetry is said to exist between i and ] .
The symmetric matrix S associated w i t h the m a t r i x X is defined
by S ---- [s,] , where
--1
s~j -----sj, ~__0

if x ~ j = x j i = l ,
otherwise.

i.e.,

i<:>].

The i,] e n t r y of the n *h power of S is s~/~.

5.03. Definitions:
1. An ordered sequence w~th n + l members, i -- ;~, ?~, ---, 7 , ,
?**~ -----] , is an n-chain P f r o m i to ] if and only if
i = ~ ' - - > ~'~, ?~-- > ~s, " " , ? ~ = > ~',*l'~]
(n)

In brief, i - - > j indicates t h a t there exists a n n-chain f r o m i to ] ,

which may-EIso be enumerated as i -- 71, ?,2, --- , ~'~, ~'~,~=- ] , or,
when no a m b i g u i t y will arise, as i , k , l , . . . , p , q , ] with the ordering being indicated by the w r i t t e n order of the sequence.

108

PSYCHOMETRIKA

2. Two n - c h a i n s F and a r e e q u a l if and only if the r ta member of F equals the r th member of ~ , i.e., 7~ - - ~, f o r 1 < r < n + 1.
I f this is not true, t h e n / " and are d i s t i n c t .
3. Each pair of elements 7~ and 7~ of a n n-chain with 1 < k <
m < n + 1 a n d 7~ : 7.~ is said to be the r e d u n d a n t p a i r ( k , m ) . A n
n-chain is r e d u n d a n t if and only i f it contains a t least one r e d u n d a n t
pair.
4. The elements 1 , 2 , - . . , t (t > 3) f o r m a c l i q u e 0 o f t members i f a n d only if each element of O is s y m m e t r i c w i t h each other elem e n t of 0 , and .there is no element not in O s y m m e t r i c w i t h all elements of O .
This is equivalent .to
x i j = 1 - - &j f o r / , ] = 1 , 2 , . - . , t but not f o r i , ] =
t + 1 , whatever the (t + 1) ~ element.

1,2,.-.,

t,

5.04. Theorem 1: x ~ (') = c if and only if there exist c, distinct

n-chains f r o m i to ] .
Proof: By definition of m a t r i x multiplication
X~j (n) ~ : ~ .. ~ Xi~X~ .. XmXcj,

w i t h the summations over n--1 indices. Suppose t h a t t h e indices have

been selected such t h a t i , k , i, . . . , p , q , 3" is an n-chain f r o m i to ] .
Then by definition 1 (5.03)
~ik ~

Xkl = ' ' "

X p q - - - Xqi =

1 ,

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 cont 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 summation.
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 symmetric f o r ] s ~', so either x~j = 0 or xj~ ~---0 or both. Thus i f i is symmetric with c elements of 2 , x~j (~) : c .

R. DUNCAN LUCE AND. ALBERT D. PERRY

109

I f xi~ (2) ~ _ c , then b y t h e o r e m I t h e r e exist c distinct j's such t h a t

x~j----xs~ ~-__1, i.e., i < ---->__] f o r c ] ' s .
5.06. Theorem 3: An element i is contained in a clique if and
only if the i t~ e n t r y of the main diagonal of S ~ is positive.
Proof: Suppose t h a t i is contained in a clique O .
By definition
(/)+(k)<~

S e l e c t ] and k s u c h t h a t
(]) :+ (k) < O and such t h a t i # ] #
k
# i . Such elements exist b y t h e definition of a clique (definition 4,
5.03). I t is t r u e b y t h e definition o f a clique a n d of the m a t r i x S
that: s~s ---- sj~ = s~ - - s~j : s~ = s~ - - 1 f o r such ] and k . Thus this
choice of ] and k contributes 2 to--~e summation, and because s~j > 0
for all i and ] t h e r e are no negative contributions to the s u m ; therefore s~ (~) > 2 > 0 .
Suppose t h a t s~ (8) > 0 . Then there exists at least one p a i r of
elements of ] and k such t h a t s~i = sj~ ---- sk~ ~--- 1 a n d this implies
i <=>
3",3' < - - >
k , and k < = >
i . I f there a r e no other elements symmetric w i t h i , ] , and k then these t h r e e f o r m a clique. I f
there is another element s y m m e t r i c with these three, then consider
the set of f o u r f o r m e d b y adding it to the previous three. I f t h e r e ~s
no other element s y m m e t r i c with these four, they f o r m a clique. I f
there is, add it to the set and continue t h e process. Since the set 2
contains only a finite n u m b e r of elements, the process must terminate
giving a clique containing i .
5.07.

Theorem 4: I f 1) O~ are cliques of t~ members, 2) the sets

zl~ ~--- O~ (e~ :+ 0~ + -.. + O~_~) have d~ members, and 3) i is contained in the cliques @~, z - - 1 , 2 ,

--., m , then

s,,(~) = ~ ( (t~ - - 2) (t~ - - I ) - - (d~ - - 2) (d~ - - I ) } :+ 2.

Proof: By definition
- -

(i)+(k)<Y~

The set of all the pairs ] , k is the union of the following three mutually 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

N~ [ j , k I there does not exist a such t h a t (j) + (k) < O,]

T~ [ j , k i there exists a such t h a t (j) + (k) < 0~, (j) + (k) <* A~] .
1.

F o r T1 then either

a)

(j) + (k) <* O~ for all a .

This is not possible because

(]) + (k) < or;

b)
AI=0;

(j)

(k) < A~ f o r all a .

This is not possible because

ore)
(j) + (k) < O~ if and only if (j) + (k) < A ~ f o r a l l a .
This is not possible because A~ = 0 . Thus gJ~ is empty.

2. (j) + (k) < T2 impties'suss.~s~ = 0 for SuSjkSkl ~-- 1 implies t h a t i ; ] , and k are either a clique or a subset of a clique (by
t h e a r g u m e n t of theorem 3), but (j) + (k) < gz implies ] and k
are not contained in a n y clique.
3.

T~ gives t h a t
(j)+ (k) <q,~

= Z
V=l

s~jsi~s~

(j)+(k)<@ v
(j)+ (k)<*A v

We observe t h a t : 2~[j,k 1 (j) + (/~) < O,] = D 2 [ j , k

I (])+

(k) <

o.~, (y) + (k) < A,] + Q~[S,k i (]) + (k) < O,, (y) + (k) <* Av]
and since .% - 2~ = 0, it follows t h a t Y~ = Y~ + N or Y~ = N -- ~ .
~h

Ft2

fl~

~3

f~l

fte

t?, is the set of all ordered pairs (j) + (k) < O,,. I f i C j C k i ,
then s~ = s~k = s~ = 1 , othetavise one of the s~ - - 0 . Since every
O, contains t,, elements, there are %-aP~ ordered pairs s a t i s f y i n g these
conditions. Thus:

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
s.~ (~) = ~ ( (t, - - 2) (t~ - - 1) - - (d~ - - 2) ( & , - 1) }
Y=2

+ (tl - - 2) (tl - - 1)

~- ~ ( (t~ - - 2) (t~, - - 1) - - (d~ - - 2) (d, - - I ) } + 2.

V=l

Since the entries szj(3) are uniquely determined from the entries
of S by the laws of matrix multiplication, all valid methods of calculating 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 offdiagonal 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:
1. i < = > 2" f o r all i , 2" s O , t h e n 0 is a clique b y definition 4
(5.03) and t h e o r e m 3 (5.06), and it ,has t m e m b e r s b y p a r t 1 of
the hypothesis.
2. T h e r e e x i s t p a n d q s O such t h a t p a n d q a r e n o t s y m m e t r i c .
T h e n b y definition

S~/~) =

Z Z

s~jsjks~

U)+(~)<o

Z E

s~jsj~sk~.

(j)+(k)<*O

I f s~jsj~sk{ ~- 1 , t h e e l e m e n t s i , 2", a n d k a r e a clique o r a s u b s e t o f

a etique a n d t h u s b y h y p o t h e s i s (3) a n d t h e o r e m 3 (5.06) t h e y a r e
all c o n t a i n e d in O; t h e r e f o r e t h e second s u m ---~ 0 . I n t r o d u c e in 2
sufficient r e l a t i o n s h i p s p = > q to m a k e 0 a elique { of t m e m b e r s .
Sinee s{~ > 0 f o r all i a n d 2", t h e i n t r o d u c t i o n of t h e s e spq - - 1 m u s t
i n c r e a s e t h e s u m b y 2 o r m o r e , f o r a t l e a s t t w o a d d i t i o n a l 3-chains
a r e i n t r o d u c e d (i,p,q,i a n d i,q,p,i) ; h e n c e b y t h e o r e m 4 (5.07)

S u (~) - -

Y. Y~ si~sj~s~i - - 2 = ( t - - 2 ) ( t - - l ) - - 2
(j)+(k)<~

< (t--2) (t--l),

112

PSYCH0~ETRIKA

which is contrary to hypothesis (2). Therefore 0 is a clique of t

members.
5.09.

Redundancies:

By definition 3 (5.03) an n-chain is redundant if and only if it

contains at least one redundant pair (k,m), where a redundant pair
defines two members of the n-chain ?k and ?~ with ?~ - - ~,~ and k < m .
I f these ordered subscript pairs (k,m) and the end point pair (i,])
(the latter not necessarily a redundant pair) are considered as sets,
then five classes of mutually exclusive redundant n-chains may be defined which include all redundant n-chains:
1. The A~ class: There exists at least one redundant pair (k,~)
and it has the property:
(k,m)

(i,]) =

o.

2. The B, class: There exists one and only one redundant pair
(k,m) and it has the property:
(k,m)

3.

(i,]) = i .

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).

If there are t n-chains i ~ ' > ] of the class A , from i to ] , then

define a~s(") - - t . From these numbers the matrix A(") = [a~/~)] is
formed. Th-i~ is the matrix of redundant n-chains--~ the class A , .
If R (') is the matrix of redundant n-chains it follows, if analogous
definitions are made for matrices of the other four classes, that
R ( ' ) - " A (' + B ('*) '+ C ('~) + D ( ' ) + E ' .

It follows directly from the definitions and the limitations on n

that

113

R. D U N C A N L U C E A N D A L B E R T D. P E R R Y
R ( z ) r-- 0
A (8) - - 0 .

I t will now be p r o v e d t h a t D (3) = S . B y t h e definition of t h e

class D 3 , t h e r e exist t w o and only t w o r e d u n d a n t p a i r s ( l , m ) a n d
( p , q ) , and t h e y h a v e the p r o p e r t i e s :
(k,m)

(p,q)

(i,D

= i

(i,]) -" ].

These pairs m a y define in total either t h r e e or f o u r m e m b e r s of t h e

3-chain ( t h r e e m e m b e r s w h e n m ~- p , b u t no f e w e r f o r i f / c - - p and
m = q t h e n ( k , m ) ( i , ] ) = ( i , ] ) , which is c o n t r a r y to t h e definition
of D:~). Suppose m = p , t h e n either i ---- y~ - - ] o r i = y~ = ] , w h i c h
is impossible f o r i > i b y assumption. T h u s m p . W i t h f o u r
m e m b e r s t h e r e a r e two possibilities f o r a r e d u n d a n t 3-chain= e i t h e r
i ----- ?~, 7~ - - ] o r i ----- ~,~, ~,~ -~ ] . T h e f o r m e r is impossible by t h e
previous a r g u m e n t ; thus the only 3-chains o f t h e class D~ a r e o f t h e
form
i,r~,),~,]--i,],i,];

t h a t is,
--1
d~/8) : 0

if i < = > ]
otherwise,

T h e r e f o r e , b y the definition of S , we h a v e D (3) = S .

I f t h e m a t r i c e s of redundancies up t o and including R (~-2) a r e
known, t h e n we can find A (~') b y A (~) ----- X R ( ' ~ 2 ) X .
P r o o f : B y t h e definition o f the class A,~, a r e d u n d a n t n-chain o f this
class has t h e f o r m
(a)

~ = yl , y~ , ~ ,

(b)

y~ , - - ,

(c)

y,~ , - - ,

7~ , ~'~,l-=

wherea+b+c+5=n,k<m,andyk=y,~.
(~t-2)

It follows f r o m t h e definition t h a t p - n--2 chain, and each such distinct n--2

one distinct r e d u n d a n t n-chain f r o m i
d u n d a n t n-chains of t y p e A~ f r o m i to
tions p - ~2 and q -- y~ f o r t h e n u m b e r
p to q , t h a t is,

72
> ~ ----- q is a r e d u n d a n t
chain d e t e r m i n e s no m o r e t h a n
to ] . T h u s t h e n u m b e r of re] is t h e sum o v e r all combinaof r e d u n d a n t n : - 2 chains f r o m

(p)+(q)<,~

or

A (~) = X R ( ~ ' 2 ) X .

114

PSYCHOMETRIKA
I f the m a t r i x [e~/*)] is defined as
[e,/-)] = X R ( , ~ ) X

+ D oo

then the relations

A ('~) + B (~) ~+ D(~) = R ~ ) X
A (~) + C TM) + D('O = X R ( ~ x )

TM)

= [~j (x~fcr - - e#.~ ) ]

follow through an enumeration of cases and b y using similar p a t t e r n s

of p r o o f to t h a t j u s t given.
These various relations permit the specific conclusions:
R(2) = [&ixi/~)] = E(2)
R (3) = X R (2) + R ( 2 ) X + E (3) ~ S

and the general result

R(~) = X R ( ~ )

+ R(~-I)X _ XRO~2)X

+ EOO - - D(~) .

This latter expression is not useful in its present f o r m because D (")

has not been expressed in t e r m s of t h e matrices of redundancies up
to and including R (--t). This problem of t h e determination of the
m a t r i x of r e d u n d a n t n-chains is left as an unsolved problem of both
theoretical and practical interest.
5.10.

Uniqueness:

In certain applications it is desirable to know w h e t h e r a p o w e r

of a m a t r i x uniquely determines the matrix. This is not t r u e in general, f o r Sylvester's t h e o r e m gives a multiplicity of n t~ roots of a
matrix. The matrices being considered are r a t h e r specialized, however, and it is possible t h a t some degree of uniqueness m a y exist.
The following two t h e o r e m s indicate certain sufficient conditions
f o r uniqueness. Since these theorems do not utilize completely the
special characteristics of the matrices in this study, it is probable
t h a t m o r e appropriate theorems can be proved.
5.11. Theorem 6: I f p and q a r e positive integers, if t w o integers a a n d b can b e found such t h a t a p ~ b q - - 1 , and if X is a nonsingular matrix, then t h e p o w e r s Xp and X~ uniquely determine X .
P r o o f : Suppose t h a t t h e r e exist t w o non-singular matrices X and Y
such t h a t X~ = YP and Xq = Yq. Then X ~ = Y~ and X ~q = Y~q. Now,
f o r m X b q Y = y b q y = y~q~ - - y~p, since ap - - b q = 1. Similarly
X ~ X = X bq~= X ap. B u t since X ~p - - Y~ it follows t h a t X ~ q X - - X b q Y .

R. DUNCAN LUCE AND ALBERT D. PERRY

115

Since X is non-singular, IX~I 0 , and t h u s there exists a unique inverse of X % X -~q, such t h a t X ~ X bq = I ; therefore X = Y .
5.12. Theorem 7: I f n is a positive odd integer and S a real
symmetric matnix, then there is one and only one real symmetric n ~I'
root of S .
Proof: 1. There is one such n t~ root.
Since S is real and s~m~netric there exists a real orthogonat
m a t r i x P such t h a t P ' S P = D ( P ' is the transpose of P) is diagonal
with real entries d~ which are the characteristic roots of S .* Assume
P is so chosen t h a t dl~ ~< d~ < .-. < d ~ . L e t B be the diagonal matrix of the real n tj' roots of the elements of D , i.e., b~ - - real ( d ~ ) ~/" ,
so
B~=D.

Define R = P B P ' .

(1)

Then R " = S , f o r

R ~=

( P B P ' ) ~ ~-- P B ~ P ' = P D P ' =- S .

Since B ~s real and diagonal and P is real and orthogonal, R is real

and symmetric.
2.

There is only one such n th root.

Suppose there exists a real s y m m e t r i c m a t r i x / 7 1 not equal to R

such t h a t R~" = S . Then there exists an orthogonal m a t r i x Q such
t h a t Q ' R ~ Q - ~ T is diagonal in the characteristic roots of R1, and ordered as before. Consider the n th power of T:
T ' ~ : ( Q R ~t Q )
: Q'PDP'Q

n - -

- - Q R !~ Q ~ - Q ' S Q
= (P'Q)'D(P'Q)

(2)

T ~ -~- U ' D U ' 7 -

where U is the orthogonal m a t r i x P ' Q . Since U' = U -~, T" a n d D are

similar, and hence have the same characteristic roots.t Because t h e y
are diagonal in the characteristic roots, ordered in the same way,
they are equal:
D -- T..

Substituting (3) in

(3)

(2)
D--

U'DU

or
UD "- DU.

*MacDuffee, C. C. Vectors and matrices. Ithaca, N. Y.: The Mathematical

Association of America, 1943, pp. 16~-170.
tIbid., p. 113.

116

PSYCHOMETR!KA

By definition of matiix multiplication this means

u~jd~k = Y~ d . u j ~ .
J

Since D is diagonal, this reduces to

or

u~k (d,k - - d . )

= 0.

(4)

Since the dkj: are real and n is odd, equation (4) implies
u ~ [ ( G ~ ) v~ - - ( d . ) ""] = 0

where the (Ga)~/= are real. Thus by the definition of B ,

U B -~--B U
or

(5)

B = u'Bu.

By (1) and (3)

T~:D:B

,~ ,

but by construction T and B a r e both real diagonal matrices and n

is odd, so this implies
T=B.

This substituted in (5) gives

T -- U'BU--

Q'PBP'Q

or

QTQ' ~- PBP'.

But

Q T Q ' ---- R1

and

PBP'

= R

by definition ; therefore
RI ~ R .

6. A c k n o w l e d g e m e n t

We wish to acknowledge our indebtedness to Dr. Leon Festinger,

Assistant Professor of Psychology, Massachusetts Institute of Te&nology, for his kindness in directing this research to useful ends, encouraging the application of this method to practical problems, and
providing many constructive criticisms of the work.