Social Networks 43 (2015) 136148

Contents lists available at ScienceDirect

Social Networks
journal homepage: www.elsevier.com/locate/socnet

Reconstructing Granovetters network theory

Andreas Tutic a, , Harald Wiese b,1
a
b

Universitt Leipzig, Institut fr Soziologie, Beethovenstrae 15, 04107 Leipzig, Germany

Universitt Leipzig, Wirtschaftswissenschaftliche Fakultt, Grimmaische Strae 12, 04109 Leipzig, Germany

a r t i c l e

i n f o

Keywords:
Granovetters thesis
Weak ties
Graph theory
Cooperative game theory
Gloves game

a b s t r a c t
We employ concepts from graph theory and cooperative game theory to reconstruct Granovetters famous
thesis concerning the strength of weak ties. In contrast to existing formal models related to this thesis,
our approach captures the mechanisms Granovetter invokes in the derivation of his thesis. Notably, our
model allows for an analytical distinction between the strength of ties and the value of ties a distinction
empirical research on the labor market has shown to be of great importance. We use our model to test
the theoretical validity of Granovetters thesis and to evaluate its robustness if implicit assumptions in
Granovetters argumentation are dropped.
2015 Published by Elsevier B.V.

1. Introduction
One of the most important and most widely cited articles in
sociology is Granovetter (1973). He is concerned with the ties or
links between agents. According to Granovetter (1973, p. 1366)
these links may have different strengths which stand for e.g. the
amount of time invested in a relationship, the emotional intensity governing the relationship, or the velocity of transmission of
information.
Granovetter bases his work on two postulates. First, he relates
ties between two agents to common ties with other agents.
Granovetter (1973, p. 1362) argues that a stronger tie between
agents 1 and 2 leads to a higher proportion of third agents to
whom both are tied. We call this the common-friends postulate.
Granovetter nds arguments from various elds in support of this
claim.
Granovetters second postulate is known as the forbidden triad.
In a typical network the strong ties an agent 1 has with 2 and 3 imply
a tie between agents 2 and 3. For example, 2 and 3 are friends with
1 and hence friends with each other (e.g. having met at parties
organized by 1). Therefore, Granovetter (1973, p. 1363) wants to
rule out the situation where 2 and 3 (both being friends with 1) do
not have a link between themselves. Triads that are not forbidden
are called balanced and so are networks without forbidden triads.

Corresponding author. Tel.: +49 0341 9735651.

E-mail addresses: andreas.tutic@sozio.uni-leipzig.de (A. Tutic),

wiese@wifa.uni-leipzig.de (H. Wiese).
1
Tel.: +49 0341 9733771.
http://dx.doi.org/10.1016/j.socnet.2015.05.002
0378-8733/ 2015 Published by Elsevier B.V.

With some cautious wording, Granovetter claims that the

common-friends postulate implies the triad postulate. Since Granovetter argues in a non-formal way, he cannot substantiate this
alleged implication to which we will come back later.
A typical network, in Granovetters view, consists of cliques
of friends (called a densely knit clump of social structure in
Granovetter, 1983, p. 202) with strong ties inside each clique
(intra-clique) and weak ties (called bridges) between cliques (interclique). (Of course, other network structures exist also marriages
concern people of different sexes and clubs often try to attract
people with different occupations.) Since the bridges open up communication or business opportunities to agents outside ones own
clique, they are more important than strong ties between friends.
Granovetter (1973, p. 1366) states his thesis as follows:
The contention here is that removal of the average weak tie
would do more damage to transmission probabilities than
would that of the average strong one.
Hence the title of Granovetters contribution: The strength of
weak ties.

1.1. Empirical tests of Granovetters thesis

Granovetters thesis provoked a lot of both empirical and theoretical work in the social sciences. Interestingly, most empirical
studies provide a rather skeptic impression on the validity of
Granovetters thesis, while scholars have been eager to construct
formal models that corroborate the thesis with little qualication.
In the following, we briey summarize the most relevant literature
to motivate our own work.

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

From the beginning, the thesis of the strength of weak ties was
applied to the labor market. Granovetter (1974) provides empirical support for the claim that workers use contacts in addition
to formal means if they search for a new job. Further, for reasons outlined above, scholars conjectured that weak ties should
be more important in the search for a new job, i.e., more jobs
should be found through the use of weak ties, these jobs should be
paid better and should have higher occupational prestige. Empirical studies show that about 4050% of all job matches are due to
social contacts (e.g. Franzen and Hangartner, 2006; Granovetter,
1974), conrming his rst claim. However, the second claim
weak ties are more important for job matches than strong ties
is highly controversial. Some studies provide clear negative evidence (e.g. Bridges and Villemez, 1986; Forse, 1997; Marsden and
Hurlbert, 1988), while other studies provide rather modest support
(e.g. Biang and Ang, 1997; Sprengers et al., 1988). More recently,
Tassier (2006, p. 706), whose own study validates the claim, comments on the state of the art as follows: In summary, despite the
intuitive appeal of the notion that using weak ties to nd a job may
increase income, for the most part, past effort to show a clear empirical link between weak ties and income have failed (Mouw, 2003).
Granovetter (2005, p. 37) himself comes to a somewhat more positive assessment: Whether the use of weak or other ties in nding
jobs signicantly affects wages, wage growth, job satisfaction and
productivity has been debated but not resolved. Large aggregated
data sets sometimes do not show clear effects (as in Mouw, 2003),
but more focused and specialized samples often do.
1.2. Theoretical models related to Granovetters thesis
While these empirical studies show that Granovetters thesis is
too general and must be qualied, the formal models related to
Granovetters ideas either corroborate his thesis or use the thesis
as an assumption to derive additional implications. Following are
short discussions of the most relevant models with respect to Granovetters thesis, i.e., Boorman (1975), Montgomery (1992, 1994),
Fararo (1983) and Fararo and Skvoretz (1987).
Boorman (1975) constructs a game-theoretical model based on
the following ideas. Each of a set of agents has to distribute a xed
time budget on weak ties and strong ties. Strong ties are more
expensive to maintain than weak ties. In each time period there
is a certain probability that an agent loses his job and there is a certain probability that an agent gets the information about a vacant
position (this information comes from outside the model). Each
agent employs a priority rule: If he gets the information about a
vacancy and is unemployed, he takes the job himself. If he is not
unemployed, he offers the job to some of his unemployed strong
contacts. If he and all of his strong contacts already have a job, he
offers the job to some of his unemployed weak contacts. Assuming that each agent wants to minimize the probability of being
unemployed, Boorman (1975) shows that in a symmetric equilibrium agents invest all their time in weak contacts, provided that
the probability of losing a job is not too close to 1.
While Granovetter and Boorman come to a similar conclusion
weak ties are more important than strong ties for the transmission of information in networks the underlying mechanisms are
very different. Granovetters argument rests on the forbidden triad
which has the consequence that all bridges are weak ties. Boorman
(1975, p. 224) explicitly rejects this idea and in fact assumes that
there are no closed triads, i.e., he assumes that if a is connected
with b and b is connected with c, there is no connection between a
and c. Instead, Boormans model is driven by the assumption that
strong ties take more time to maintain than weak ties. While this
assumption certainly is in line with Granovetters reasoning, Boormans model does not capture the central mechanism Granovetter
invokes to establish his thesis.

137

Similar criticism must be directed at the two formalizations

of Granovetters thesis due to Montgomery (1992, 1994). Both
models deal with the application of the thesis to the labor market. Montgomery (1992) builds on economic job-search theory
(cf. Mortensen, 1986) to analyze the effects of network composition on wages. It is unnecessary to describe this model in detail,
because it does not derive Granovetters thesis from some underlying assumptions. Instead Montgomerys model rests on the thesis
to derive further implications regarding life earnings. More specifically, Montgomery (1992, p. 588) comes to the conclusion that if
[. . .] a given weak tie is more likely to produce new information
than a given strong tie the proportion of contacts via weak ties has
a positive inuence on the reservation wage and hence expected
life earnings. In a nutshell, this model is best interpreted as studying the consequences of Granovetters thesis on the labor market,
not as a derivation of Granovetters thesis.
Montgomery (1994) constructs a Markov model of employment
transitions. In this model, society is organized in dyads which are
connected via strong ties. At any point in time, each dyad is in
one of three states: Both individuals are employed, both individuals are unemployed, or exactly one individual is employed. At any
point in time, there is a certain probability that each individual is
either in contact with his dyad partner or with some other individual via a weak tie. Employed individuals lose their job at a xed
rate. Unemployed individuals get jobs via employed contacts and
through formal channels, the latter at a constant rate. Montgomery
assumes that unemployed contacts never know of any job, that
employed dyad partners know of a job with some xed probability,
and that employed contacts via weak contacts provide a job with
a certain probability. From this assumption, Montgomery derives
the transitions rates between the states of the dyads. Of course, in
equilibrium these rates are required to equal zero. It turns out that
for a plausible range of parameter settings, the employment rate is
increasing in the proportion of social interaction via weak ties.
As in Boormans approach, in this model there is little akin to
the postulate of the forbidden triad. Hence the underlying mechanism generating the benecial consequences of weak ties are very
different to Granovetters reasoning.
Thus, the approaches by Boorman and Montgomery partly
support Granovetters conclusion, but they do not capture the
underlying mechanism Granovetter describes in his original paper.
The model closest to Granovetters reasoning stems from a series
of papers by Thomas J. Fararo and John Skvoretz (e.g. Fararo, 1983
and Fararo and Skvoretz, 1987). They employ the theory of biased
random nets to explicate the strength of weak ties. A random net
consists of a set of nodes. Consider some subset of nodes; in the
rst stage (t = 1) of the so-called tracing procedure, each node from
the starting subset connects with a > 0 nodes, each node having
the same probability of being contacted. Some of the nodes contacted in the rst stage will not be from the starting set. Each of
these nodes contacts a > 0 other nodes in the second stage (t = 2) of
the tracing procedure, and so on (t = 3, . . .). Theoretically, this setup raises the following question: What is the expected number of
connected nodes in stage t? For the case of unbiased random nets,
i.e., if each node has the same probability of being contacted by
some other node, Rapoport (1979, p. 6) showed that if t , the
expected fraction of connected nodes, , is implicitly dened by
 = 1 ea  .
Numerical analysis shows that , called connectivity, is a positive function of a, called element density. Fararos formalization of
Granovetters thesis rests on this observation and the concept of a
biased random net. As already explained, in an unbiased net each
node has the same probability of being contacted by some other
node. In a biased random net, the probability that two nodes get in

138

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

contact is inuenced by properties of the already evolved network.

Specically, leaning on the work of Foster et al. (1963), Fararo posits
the equation
P (BA|CB and CA) =  + (1 ) P(BA),
in which A, B, and C denote nodes, X Y means that node X is
connected with node Y, and  [0, 1]. The equation states that
the conditional probability of a tie between A and B is generically
greater than the unconditional probability, where the condition
refers to some common friend C. This equation allows to model Granovetters forbidden triad. The bias parameter  should be greater if
the pairs C B and C A are connected via strong ties. There are different versions for the formalization of this idea (e.g. Fararo, 1983;
Fararo and Skvoretz, 1987, p. 1196). In any case, mathematical analysis combined with numerical studies show that the connectivity of
a biased random net increases if the strength of weak ties increases
(the strength of weak ties is measured by a variable that determines
the differential effect of a strong-tie conguration to a weak-tie
conguration on the bias parameter ).
While we strongly appreciate this model, because it is quite
close to Granovetters original reasoning, there is a subtle point
we want to improve upon in our own attempt. Granovetter argues
as follows: If the postulate of the forbidden triad holds, then weak
ties are more important for the benet of society than strong ties.
However, the delineated model offers the following conclusion: The
more satised the postulate of the forbidden triad, the greater the
benet for society.
1.3. Our approach
This paper is intended to develop a formalization of Granovetters thesis that is closer to Granovetters argumentation than all
of the approaches considered above. As in Granovetters argument,
we postulate some property of networks and check if, given this
property, the removal of a weak tie hurts the social aggregate more
than the removal of a strong tie. Our paper considers various related
models, all of which are based on the following three fundamental
ideas:
First, to our mind the strength of a tie lends itself to a probabilistic interpretation. Indeed, the probability that two agents interact,
communicate, or cooperate may be an expression of the amount
of time invested, the emotional intensity or the mutual conding, etc. While Granovetter does not bind himself to any such
interpretation (nor do we), he refers to probabilistic arguments
on several occasions. Granovetters qualitative discussion distinguishes between absent, weak, and strong ties, only. In contrast,
our formal model knows any strength between 0 (absent tie) and
1 (extremely strong tie).
Second, we argue that the triangular inequality dened and
defended below is a very appropriate translation of (and indeed
improvement upon) Granovetters postulate of forbidden triads.
Third, turning to Granovetters thesis, we argue that it is not sufcient to just consider networks but that we also need to theorize
about the value (not only the strength) of links. Our approach
allows to differentiate analytically between the strength of ties
and the value of ties by an integration of network theory and
cooperative game theory. That is, we will use the Myerson value
for probabilistic networks to evaluate the payoffs for the players,
which depend on the strengths of the links and on the economic
rents coalitions of players can achieve.
We will use this setup to formalize Granovetters reasoning and
test its theoretical validity. Additionally, our model allows to check
if Granovetters implicit assumptions are crucial for the derivation

of his hypothesis. In our view, Granovetter bases his theory on

the implicit assumptions of typical networks and typical benet
structures: First, Granovetter assumes homophily, i.e., strong links
typically exist with people similar to myself while I have weak or no
links to people that are very different. Second, Granovetter assumes
that bridging different cliques actually provides benets. In a rst
step, we will apply our formal model to a specic game (the gloves
game) and a specic network that satisfy these assumptions in a
pure form. In a second step, we will alter the network to check
whether the validity of Granovetters thesis actually depends on
these assumptions. Finally, we will relate our results to the aforementioned empirical literature on network effects in labor markets
and especially to Lins solution to the puzzle, why there is only
mixed evidence with respect to the strength of weak ties in status
attainment (cf. Lin, 1999).
The remainder of this paper is organized as follows: In Section 2, we introduce both undirected (nonprobabilistic) graphs and
undirected probabilistic graphs. Section 3 explains the cooperative game theory needed and also presents the Myerson value
and the probabilistic Myerson value. Section 4 discusses several
alternatives of how to formalize the common-friends postulate
and the forbidden triad. After all this preparatory work, we are
in a position to test Granovetters thesis for an instructive class of
models in Section 5 and relate our results to the empirical literature on applications of Granovetters network theory to the labor
market. Section 6 concludes the paper.
2. Graph theory
In this section, we present some important concepts of graph
theory together with concepts found in the Granovetter paper.
Granovetters discussion uses the threefold distinction between
absent, weak, and strong ties. This may be sufcient for qualitative discussions but is severely restricting for both theoretical and
empirical work.
Granovetter himself repeatedly refers to probabilistic arguments (e.g. Granovetter, 1973, p. 1362). Therefore, we put numbers
between 0 and 1 on every link in order to overcome the threefold
distinction between absent, weak, and strong ties. Formally, we
are dealing with probabilistic networks. For every verbal denition due to Granovetter, we present two formalizations. The rst
one is couched in terms of (deterministic) graphs (N, E) with vertices N and edges E (Section 2.1) while the second one makes use of
probabilistic networks (Section 2.2).
2.1. Graphs
Leaning on Bollobas (1998) we introduce the denition of a
graph. Fig. 1 is an example where you are invited to ignore the
different strengths of lines.
Denition 1 (undirected graph). An undirected graph G is a tuple
(N, E) where N is a set of so-called vertices (or players, or agents)

Fig. 1. Cliques and bridges.

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

and E, the set of edges or ties or links, is a subset of N(2) := {{i, j} : i,

j N, i =
/ j}. The set of such sets of links is denoted by E(N). Links
{i, j} are also written as ij or ji. We assume that both N and E are
nite as well as |N| = n. By N(G) and E(G) we mean the vertices or
the edges of graph G. i and j are called endvertices of the edge ij.
If ij is contained in E(G), i and j are called adjacent or neighboring
vertices of G. We also say that i and j are directly linked. G = (N , E )
is a subgraph of G = (N, E), G G, if N N and E E.
Every nonempty subset K N gives rise to a subgraph of (N, E)
denoted by (K, E(K)) where E(K) := {{i, j} E : i, j K}. By G ij we
mean the graph (N(G), E(G) \ ij). For i, j N(G), we write G + ij for
(N(G), E(G) ij).
Since we are interested in connections between agents, paths
are a central object:
Denition 2 (path in an undirected graph). A path
P(ij) = (i = i0 , . . ., j = ik )
in G from i to j (a i j path) can be identied with the subgraph
given by
N(P(ij)) = {i = i0 , . . ., j = ik } and
E(P(ij)) = {i0 i1 , . . ., ik1 ik },
where i = i0 and j = ik . A link j is said to be contained in P(ij) if j
E(P(ij)) holds. We also write P(ij) = (i = i0 , . . ., j = ik ). The set of such
paths is denoted by P(ij). Players i and j are called connected or
linked if an i j path exists or if i = j holds. i and j are called connected
within a subset K N if they are connected by an i j path obeying
N(P(ij)) K or if i = j holds. A graph (N, E) is called connected if every
pair of players is connected.
A coalition structure P on N (sometimes written as (N, P))
is a partition ofN into components C1 , . . ., Cm , i.e., P =
m
{C1 , . . ., Cm }.Thus, j=1 Cj = N, Cj Ck = for all j, k {1, . . ., m}, j =
/ k
/ for all j {1, . . ., m}. The set of all partitions on N is
and Cj =
denoted by P(N). For any player i N, P(i) denotes the component
containing i. Partitions on nonempty subsets K N are denoted by
PK . The set of such partitions is denoted by P(K). It is well-known,
that the equivalence classes of an equivalence relation on a set form
a partition of that set.
Denition 3. Consider the graph (N, E) and the equivalence relation on a nonempty coalition K N given by

139

Note that the number of players and thus the number of paths
are nite.
As mentioned in the introduction, Granovetters typical
networks consist of cliques and bridges. A simple example is given
in Fig. 1 where
N = {1, . . ., 8} and
E = {12, 13, 14, 23, 24, 34, 26, 35, 56, 57, 67, 78}.
The players 1 through 4 form a clique and so do the players 5
through 7. Player 8 is a clique on his own. Thus, these cliques
partition N = {1, . . ., 8} into three components. The partition is
P = {{1, 2, 3, 4}, {5, 6, 7}, {8}}. Links 26, 35 and 78 are bridges.
Preparing the following denition, note that 26 is a bridge of
degree 3 because in the absence of this link players 2 and 6 would
need to be linked by the path (2, 3, 5, 6) with ((2, 3, 5, 6)) = 4 1 = 3.
Intuitively, Granovetters denition of the degree of a bridge refers
to the opportunity cost (in terms of path length) of not using a
specic bridge. Note, also, that this degree does not depend on the
strengths of the ties, but only on the length of the path.
Denition 5 (undirected graph, Granovetter). A Granovetter graph
G is a tuple (N, Eweak , Estrong ) where N is a set of vertices (or players,
or agents), Eweak (Estrong ) is the set of weak (strong) links such that
Eweak Estrong = .
Denition 6 (cliques and bridges, Granovetter). Assume a Granovetter graph (N, Eweak , Estrong ) and the existence of a partition PENstrong
(according to Denition 3). Components of this partition are also
called cliques. Every weak link between agents from different
cliques C1 and C2 is called a (local) bridge (Granovetter, 1973, p.
1364) between C1 and C2 . If there is only one such link, the bridge
is called global. The degree of a bridge ij is dened by
bridge (ij) :=

min
P(ij) P(ij),

(P(ij)).

ij not contained inP(ij)

A bridge ij between different cliques C1 and C2 is a best bridge for
c1 C1 and c2 C2 if ij is contained in a best path between c1 and c2 .
Note that our denitions differ slightly from Granovetter (1973,
p. 1364): Our global bridge is simply called bridge in the original
paper while a local bridge that is not a global one, is denoted a local
bridge by Granovetter. It is a weak tie linking up cliques with a
degree of 2 or more.

ij : i is connected to j within K.
The resulting partition of equivalence classes is denoted by PEK
P(K).
Consider, for example, N = {1, 2, 3} and E = {12, 13}. We obtain
the partitions
PEN = {N}, PE{1,2} = {{1, 2}}, PE{1,3} = {{1, 3}} and

2.2. Probabilistic networks

We now generalize Granovetter graphs by putting numbers on
every link where zero indicates the absence of a link. If there are
n players, we can think of a probabilistic network as a symmetric
n n matrix

PE{2,3} = {{2}, {3}}.

Denition 4 (social distance, Granovetter). Consider players i and j,
/ j, connected by a path P(ij) = (i = i0 , . . ., j = ik ). We say that i and j
i =
are connected with degree (P(ij)) := |N(P(ij))| 1 = k along this path.
The social distance between i and j (Granovetter, 1973, p. 1336,
footnote 10) is the minimal degree found by considering all paths
linking players i and j:
(ij) := min (P(ij)).
P(ij)P(ij)

We also let (ii) := 0. The set of best paths is given by

:= arg min (P(ij)).

P(ij)
P(ij)P(ij)

Denition 7 (probabilistic network). A probabilistic symmetric

weight function is a real valued function s on N(2) :
s : N (2) [0, 1].
Clearly, s can be identied with a n n matrix obeying s(i, j) = s(j, i)
for all i, j N (in line with Granovetter, 1973, p. 1361, footnote 10).
If s is considered a matrix, we also let s(i, i) = 1 for all i N.
These functions are also called (probabilistic) networks. We
often write s(ij) rather than s(i, j). s(ij) is called the probability of

140

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

We also let (ii) := 0. The set of best paths is given by

:= arg min (P(ij)).

P(ij)
P(ij)P(ij)

Probabilistic social closeness or proximity is dened by

cl(ij) := max m1
s(ij , ij+1 ).
j=0
P(ij)P(ij)

Agents ij are said to be directly connected if s(ij) > 0 holds and indirectly if cl(ij) > 0.
Note
Fig. 2. A probabilistic network.

(ij) N0 for Granovetters social distances and

(ij) [0, 1]for probabilistic social distances.

link ij or the strength of the link between i and j. By s(N) we mean

the set of all probabilistic networks on N. For any two probabilistic
networks s and s we write s s if s(ij) s (ij) for all i, j, i =
/ j. 1 s(N)
is the probabilistic network with 1(ij) = 1 for all i, j N. Of course, if
all link strengths are 0 or 1, we identify s with E := {{i, j} : s(i, j) = 1,
/ j}.
i =
Consider any subset K N together with E K(2) . By

The intuition is the same in both cases: The higher the social distance measured by (ij), the more difcult it is for agents i and j
to link up. Note also that the probabilistic concepts are related by
cl(ij) = 1 (ij).
We now want to introduce probabilistic cliques which is a group
of agents with social distances below a certain threshold.

sK (E) := eE s(e)eK (2) \E [1 s(e)]

Proposition 10.

we dene the probability that within K the set of links E is realized.

i(s,)
j : (ij)

Consider Fig. 2. The associated network is dened by N = {1, . . .,

4} and

denes a reexive and symmetric relation on N.

s(1, 2) =

1
,
4

s(1, 3) =

3
, s(1, 4) = 0,
4

s(2, 3) =

s(3, 4) = 0.

1
,
8

s(2, 4) =

1
,
5

Note also the probability for the link set 12, 13 on {1, 2, 3}:
s{1,2,3} (12, 13) =

1 3
1
1
4 4
8

21
128

and the probability for the link set on 1, 2, 3

s{1,2,3} () = 1


1
4


3
4


1
8

21
.
128

We now turn to probabilistic paths:

Denition 8 (probabilistic path). A probabilistic path (of length
m 1) between two players i, j N, is a sequence of players and
link strengths
P(ij) = (i = i0 , s(i0 , i1 ), i1 , s(i1 , i2 ), . . ., s(im1 , im ), im = j).
By P (ij), we denote the set of probabilistic paths linking players i
and j. A cycle (of length m) is a path P (ii) where m 3 and all players
i0 , i1 , . . ., im1 are distinct from each other.
We do not rule out i = j or link strengths zero. Of course, P (ij) =
(i = i0 , i1 , 1, . . ., 1, im = j) corresponds to the path in an undirected
graph P (ij) = (i = i0 , i1 , . . ., im = j).
Denition 9 (probabilistic social distance). ;Consider players i and
j, i =
/ j, and a path
P (ij) = (i = i0 , s (i0 , i1 ) , i1 , s (i1 , i2 ) , . . ., s (im1 , im ) , im = j) .
We say that i and j are socially apart with degree

(P (ij)) := 1 m1
s ij , ij+1
j=0

along this path. The probabilistic social distance between i and j is

the minimal degree found by considering all paths:
(ij) := min (P(ij)).
P(ij)P(ij)

For any [0, 1] and any i, j N,

The proof of this proposition is immediate and hence omitted.

Proofs for all other propositions can be found in the appendix.
Denition 11 (probabilistic cliques and bridges). Assume that (s,)

is a transitive (and hence an equivalence) relation. The resulting

partition of equivalence classes is denoted by P(s,)
and its compo

nents (s, )-cliques.

Consider two (s, )-cliques
C1 and C2 and two
players i1 , i2 with i1 C1 and i2 C2 . If s(i1 i2 ) > 0 holds, i1 i2 is called

between C1 and C2 . The degree of a bridge i1 i2 is

a (s, )-bridge
dened by
bridge (i1 i2 ) :=

min
P(ij) P(i1 i2 ),

(P(ij)).

i1 i2 contained in P(ij)
Consider c1 Ci and c2 C2 . i1 i2 is called a best bridge for c1 and c2
if i1 i2 is contained in a best path between c1 and c2 .

Note that the strength of a (s, )-bridge

is below by construc.
Note
also
that
we
do not differentiate
tion of the partition P(s,)

between local and global bridges. The reason will become clear once
we have introduced the triangular inequality.
3. Cooperative game theory
Our paper in using cooperative game theory to map interaction situations and network structures to payoff distributions is
in the tradition of sociological exchange theory (Bienenstock and
Bonacich, 1992, 1993; Braun and Gautschi, 2006; Tutic et al., 2011).
One may ask the question why we use cooperative rather than noncooperative game theory in our reconstruction of Granovetters
network theory. The main reason for this is that noncooperative game theory is [. . .] obsessed with procedural details [. . .]
(Aumann in van Damme, 1998, p. 198) concerning the rules and
structure of the game, the order of decision making, and the specic payoff functions of the players. Solutions to noncooperative
games are very sensitive to slight modications of the setup, which
restricts the applicability and the explanatory power of the models. In contrast, games in cooperative game theory are much more
abstract models of interaction situations the precise rules and

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

141

sequences of decision making are not captured by the model

and hence have no impact on the predicted outcomes. From our
perspective, the greater abstractness of cooperative models ts
Granovetters quite general argument, which is not restricted to
specic interaction situations.

concept for games with a coalition function that has an extension

to probabilistic networks.
The Myerson value (Myerson, 1977)  is a solution concept in
cooperative game theory that attributes a payoff vector to any game
v on N and any network (N, E). It is dened by

3.1. Basic denitions and the Shapley value

(v, E) = Sh(vE ),

A game (in coalition function form) is a pair (N, v) (often abbreviated by v) where N = {1, 2, . . ., n} is a nite set and v a function
2N R such that v() = 0. Elements of 2N are also called coalitions,
N the grand coalition and v a coalition function. The set of all games
on N is denoted by G(N).
The gloves game is one of the most popular market games in
cooperative game theory. It presupposes a player set N = L R where
L and R are disjunct sets of players holding one left or one right
glove, respectively. The coalition function for the gloves game is
given by

vL,R (K) = min(|L K|, |R K|).

Thus, the worth of coalition K is equal to the number of pairs of
gloves coalition K can assemble.
n
A payoff vector
x for N is an element of R or a function N R.
By xS we mean
iS xi .
Player i N is a null player, if

Two players i, j N are called symmetric if for all coalitions K obey/ K and j
/ K we have
ing i

v(K {i}) = v(K {j}).

R,

K  v|T (K) = v (K) .

A coalition function v( ) is superadditive if and only if for every
pair of coalitions S, R N, S R = the inequality v(R) + v(S) v(R
S) holds.
Permutations  on S are written as (1 , . . ., |S| ) where 1 is the
rst player in the order, 2 the second player etc. Formally, rank
orders on S are bijective functions {1, 2, . . ., |S|} S. The set of all
rank orders on S is denoted by RO(S). By bijectivity, for every i S
there exists a j(i) {1, 2, . . ., |S|} such that j(i) = i. Then, we dene
Ki () := {1 , . . ., j(i) }. Thus, Ki () is the set of players up to and
including player i (for a given rank order ).
The Shapley value (Shapley, 1953) and other related values
make heavy use of marginal contributions of players. For any coalition S N and any player i N we dene
MC Si (v) := v(S {i}) v(S\{i}).
The Shapley value Sh is given by
Shi (v) =

1
n!

MC Ki i () (v),

i N.

RO(N)

3.2. The Myerson value and the probabilistic Myerson value

Our reconstruction of Granovetters network theory relies on an
extension of the Shapley value to probabilistic networks. The choice
of this particular solution concept is motivated twofold. First, in
comparison to other point-solution concepts such as the nucleolus
(Schmeidler, 1969) the Shapley value is relatively easy to compute.
Second, as far as we know, the Shapley value is the only solution

v(C).

CPE

Consider, for example, N = {1, 2, 3} and E = {12, 13}. We have

vE ({1, 2}) = v({1, 2}) and vE ({1, 2, 3}) = v({1, 2, 3}),

but vE ({2, 3}) = v({2}) + v({3}).
The probabilistic Myerson value has been concocted by Calvo et al.
(1999). By a slight abuse of notation it will also be denoted by .
It attributes a payoff vector to any game v and any probabilistic
network s. It is dened by
(v, s) = Sh(vs ),
where vs is given by

sK (E)vE (K).

EK (2)

Consider N = {1, 2, 3} and the probabilistic graph s dened by

s(1, 2) =

The restriction of a coalition function v to a coalition T is the coalition function

v|T :

vE (K) =

vs (K) =

v(K {i}) = v(K) for allK N.

2N

where vE is a coalition function on N given by

1
3
1
, s(1, 3) = , and s(2, 3) = .
3
4
4

We have

vs ({1, 2, 3}) = sN (12, 13, 23)v{12,13,23} (N) + sN (12, 13)v{12,13} (N)

+sN (12, 23)v{12,23} (N) + sN (13, 23)v{13,23} (N)
+sN (12)v{12} (N) + sN (13)v{13} (N) + sN (23)v{23} (N)
+sN ()v (N)
=

1 3 1
1 3
1
v(N)
v (N) + 1
3 4 4
3 4
4 


1
3
1
1
3 1
+ 1
v(N) + 1
v(N)
3 
4 4
3
4
4


1
3
1
+ 1
1
[v({1, 2}) + v({3})]
3
4

4 3 
1
1
+ 1
1
[v({1, 3}) + v({2})]
3
4
4

 
 1
1
3
+ 1
1
[v({2, 3}) + v({1})]
3 
4 4



1
3
1
+ 1
1
1
[v({1}) + v({2})
3
4
4
+v({3})].

If s is to be identied with a given link set E (by s(ij) = 1 for ij E and

/ E), we have vs = vE and hence (v, s) = (v, E).
s(ij) = 0 for ij
4. The forbidden triad
4.1. Granovetters triad postulate
A central concept in Granovetters original paper is the forbidden triad. Granovetter wants to rule out the top-left graph in Fig. 3
where the ties between 1 and 2 on the one hand and 1 and 3 on
the other hand are strong and, at the same time, there is no tie
between 2 and 3. Note that Granovetter does not rule out that 1
and 2 as well as 1 and 3 are connected via strong ties, but 2 and 3
are only connected via a weak tie (cf. Easley and Kleinberg, 2010,
p. 49).

142

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

Denition 13 (triangular balancedness). A probabilistic graph s is

triangular-balanced iff the triangular inequality holds for any three
players.
Thus, if s(ik) and s(jk) are large (strong ties), s(ij) cannot be very
small, very much in line with Granovetters forbidden triad. Also,
Granovetters forbidden triads point to a possible violation for at
most one link, very much the same as the triangular inequality:
Proposition 14. Consider any three players 1, 2, and 3 in a network.
The triangular inequality is violated for at most one of the three links
between them.
The triangular inequality can be generalized in the obvious fashion.
Axiom 15 (cycle inequality). For any cycle
Fig. 3. Granovetter versus triangular balancedness.

While we sympathize with the general idea, Granovetters postulate has two drawbacks:
The triad postulate merely rules out an absent tie between players 2 and 3 in case of strong ties 12 and 13. However, it would
clearly be desirable to make the minimum strength of the tie 23
a monotonic function of the strengths of the ties 12 and 13.
It is unclear how to extend the forbidden triad to quadruples or
more involved networks. Consider Fig. 4 and the top-left network
where the links 12, 13 and 34 are strong. Would it be in the spirit
of the forbidden triad to demand a weak or strong link 24? Granovetters postulate implies links 14 and 23 which could be weak
or strong. If both are weak, the link 24 can be absent (top right);
if one of them or both are strong, the link 24 may not be absent
(middle right).
The following subsection tackles these problems.
4.2. Triangular inequality
We now offer a formalization of Granovetters postulate of forbidden triads. Consider the bottom left network in Fig. 3 with link
strengths s(12), s(13), and s(23). Players 2 and 3 may be linked
directly or indirectly (via player 1). If their direct link is very weak
they will prefer to link up with the help of player 1. Thus, we have a
lower bound on s(23). To put this idea in general terms, we suggest
the following postulates and denitions:
/ k =
/ i, we
Axiom 12 (triangular inequality). For all i, j and k N, j =
postulate s(ij) s(ik) s(jk).

P(ii) = (i = i0 , s(i0 , i1 ), i1 , s(i1 , i2 ), . . ., s(im1 , im ), im = i)

we postulate s(i0 , i1 ) m1
s(ij , ij+1 ).
j=1
Denition 16 (cycle balancedness). A probabilistic graph s is cyclebalanced iff the cycle inequality holds for every cycle.
In the spirit of forbidden triads, we need to rule out networks
(or parts of networks) where the strength between any two players
i and j is smaller than the strength along any other path connecting
i and j. Consider the lower part of Fig. 4. In order to balance the left
network we need to increase some links. The strength of link 24 is
given by max(0.8 0.63, 0.7 0.72) = max(0.504, 0.504) = 0.504.
We argue that the triangular and cycle balancedness neatly capture Granovetters idea of forbidden triads while addressing the
problems mentioned in the previous section:
In contrast to Granovetters approach, the minimum strength of
tie ij is a monotonic function of the strengths of the ties ik and
jk. Furthermore, this minimum strength can be defended with
reference to the detour via k open to players i and j.
If we are dealing with a quad etc., we can work with cycle balancedness rather than triangular balancedness. However, the
following proposition makes clear that triangular balancedness
is all we need.
Proposition 17. Triangular balancedness and cycle balancedness are
equivalent.
Denition 18. A probabilistic graph s is called balanced if it is
triangular-balanced. Otherwise, it is called unbalanced.
4.3. Formalizing the common-friends postulate
According to Granovetter, the forbidden triad results from
the common-friends postulate (Granovetter, 1973, p. 1362): The
stronger the tie between individuals i and j, the larger the proportion of other individuals tied to both i and j. In our mind, the ratio
mentioned by Granovetter is
CFR(ij) :=

|{k N\{i, j} : s(ik) > 0 and s(jk) > 0}|

,
|{k N\{i, j} : s(ik) > 0 or s(jk) > 0}|

the number of other agents k tied to both i and j over the number
of agents k tied to i or j (including i and j). Therefore, we suggest to
translate the common-friends postulate by the following axiom:
Axiom 19 (common-friends I). For any players i and j, i =
/ j, we have

CFR(ij)
0,
s(ij)
Fig. 4. Is the quad with three strong links forbidden?

where we let

0
0

= 1.

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

While this axiom is a good translation of Granovetters words,

it is rather crude in using the above ratio rather than cashing in
on our probabilistic model. In particular, the ratio is insensitive
to changes of various strengths unless these changes involve zero
links. The same criticism can be levied against a second version of
the common-friends axiom:
Axiom 20 (common-friends II). For any players i and j, i =
/ j, we
have

143

5.2. The gloves game

We now consider the Granovetter thesis for the specic example
of a gloves game (see Section 3). We consider the gloves game with
four players vL,R , where we have L = {l1 , l2 } and R = {r1 , r2 }. The leftglove owners form a clique and so do the right-glove owners. In
particular, we assume the probabilistic network

s(ij) CFR(ij),
0
0

where we let

= 1.

Consider the network consisting of players 1, 2, and 3 with

s(12) > 0 and s(13) > 0. Common friends II means
s(12) CFR(12).

where
min(cl , cr ) > max(b11 , b12 , b21 , b22 ) and

In case of s(23) = 0 (violating Granovetters triad in case of large s(12)

and s(13) and also the triangular inequality), we arrive at the contradiction s(12) 10 . Using Granovetters forbidden triad, a weak or
strong tie s(23) leads to the true inequality s(12) 11 . Similarly, the
triangular inequality postulates s(23) s(21) s(31) > 0.
Proposition 21. The triangular inequality implies common-friends
II but not vice versa.
In our minds, the triangular inequality is the best and simplest
way to express Granovetters postulate in a probabilistic world and
we will rely on it subsequently. Besides its advantages in terms of
sensitivity to changes in link strengths, the triangular inequality
also trumps common-friends with respect to the handy property
of continuity (see Proposition 28 in the appendix).
5. Are weak ties really strong?
5.1. What does strong mean in our framework?
We now want to reconsider the Granovetter thesis in our framework and propose two tests of its theoretical validity. The rst one
is closer to Granovetters words, the second one is an alternative
that uses the probabilistic structure.
Denition 22 (Granovetters thesis I). Assume a player set N, a coalition function v and a balanced probabilistic graph s. Assume also a

partition P(s,)
of (s, )-cliques. Granovetters thesis I is said to hold

if, for any two (s, )-cliques

C1 and C2 , any players c, d C1 with s(c,
d) < 1 and any two players i1 , i2 with i1 C1 and i2 C2 ,

min(b11 , b12 , b21 , b22 ) > 0.

Thus, we have the two (s, 1 min(cl , cr ))-cliques L and R, six links in
N, the two intra-clique ones, cl and cr , and the four inter-clique links
b11 , b12 , b21 and b22 . Note that this network may be unbalanced but
is certainly balanced in case of b11 = b12 = b21 = b22 and cl = cr .
We have chosen the gloves game and the specic class of probabilistic networks for several reasons. First, the gloves game models
a simple matching market (cf. Shapley and Shubik, 1969) value
is created by matching complementary gloves, i.e., a left glove
and a right glove. This matching problem resonates with labor
markets in which rms offering similar jobs compete for workers with similar qualications, et vice versa. For example, one can
think of the right-glove holders as factory owners and of the leftglove holders as workers. This interpretation of the gloves game
allows us to relate our ndings to the empirical literature on network effects in labor markets, one of the most extensively studied
applications of Granovetters theory (see Section 5.5). Second, the
specic network under consideration satises Granovetters two
implicit assumptions in a pure form. That is, Granovetter assumes
homophily, i.e., strong ties connect similar actors and he assumes
that bridges between cliques provide benets. After testing the theoretical validity of Granovetters thesis in this pure setting, we will
check, if these assumptions are actually crucial for his conclusions.
Proposition 24. The worth of the grand coalition for the coalition
function v{l1 ,l2 },{r1 ,r2 } and the above probabilistic network s is

vs{l

1 ,l2 },{r1 ,r2 }

(N) = b21 + b22 + b12 + b11 (b11 + b22 )(b12 + b21 )

+ b11 b12 b21 + b11 b12 b22 + b11 b21 b22 + b12 b21 b22

v(scd) (N) > v(si1 i2 ) (N),

+ b11 b12 cl + b21 b22 cl + b11 b21 cr + b12 b22 cr

where s cd (s i1 i2 ) is the (possibly unbalanced) network resulting from s if the link strength of cd (of i1 i2 ) is reduced to zero.

2b11 b12 b21 b22 (cl + cr )(b11 b12 b21 + b11 b12 b22
+ b11 b21 b22 + b12 b21 b22 ) 2cl cr (b11 b12 + b11 b21

Note that our denition refers to the worth of the grand coalition. Of course, other denitions are also defendable. For example,
one might compare the payoff increase for players inside a clique
with the payoff increase for players forming a bridge.
Denition 23 (Granovetters thesis II). Assume a player set N, a
coalition function v and a balanced probabilistic graph s. Assume

also a partition P
of (s, )-cliques.
Granovetters thesis II is said

(s,)

C1 and C2 , any players c, d C1

to hold if, for any two (s, )-cliques
with s(c, d) < 1 and any two players i1 , i2 with i1 C1 and i2 C2 ,
s

dv (N)
dv (N)
<
.
ds(cd)
ds(i1 i2 )

+ b11 b22 + b12 b21 + b12 b22 + b21 b22 )

+ 2b11 b12 b21 b22 (cl + cr ) + cl cr (3b11 b12 b21
+ 3b11 b12 b22 + 3b11 b21 b22 + 3b12 b21 b22 )
4b11 b12 b21 b22 cl cr + cl cr (b11 + b12 + b21 + b22 ).
In the special case of b := b11 = b12 = b21 = b22 and c := cl = cr , we have

vs{l

1 ,l2 },{r1 ,r2 }

(N) = 2b4 + 4b3 4b2 + 4b 4b4 c 2 + 4b4 c + 12b3 c 2

2

8b3 c + 4b2 c 12b2 c 2 + 4bc .

144

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

5.3. Testing Granovetters thesis

We now put Granovetters thesis to some tests.
Proposition 25. Assume the gloves game v{l1 ,l2 },{r1 ,r2 } and the above
probabilistic network s in the special case of b := b11 = b12 = b21 = b22
and c := cl = cr . Assume also c := 1. Granovetters thesis
I holds for sufciently large bridge strengths (in particular b > 12 3 12 0.37) and
Granovetters thesis II is true
for sufciently small bridge strengths
(more precisely b < 32 21 5 0.38). However, for c < 1, we often nd
that Granovetters thesis II holds for sufciently small or for sufciently
large bridge strengths.
As the proof in the appendix makes clear, we apply the two
testing denitions to one weak and one strong link, as it should be.
We make use of the identical link strengths afterwards.
Broadly speaking, Granovetters thesis says that weak links are
more important than strong ones. When using variant I, a weak
link (an inter-clique one) and a strong link (an intra-clique one) are
both eliminated. Of course, if the weak bridge is very close to zero,
the effect of its removal is close to zero, also. This explains why the
weak bridge must not be too weak.
Summarizing, we feel that the proposition can be interpreted
as a conrmation of Granovetters thesis. After all, the proposition
tells us that the specic social system under consideration is worse
off when losing a weak inter-clique link of sufciently high strength
than doing without a strong intra-clique link (variant I). According
to variant II, if all the bridging links are weak, the economy benets more (suffers less) from a small increase (small decrease) of
a bridging link than from a small increase (small decrease) of an
intra-clique link.
5.4. Testing Granovetters assumptions
As already pointed out, Granovetter uses two assumption
implicitly: He assumes homophily and he assumes that bridges
between cliques are benecial. Under these assumptions, we have
conrmed Granovetters thesis, cum grano salis. We now want to
test if the assumptions are necessary. Note that the gloves game
doesnt allow to test each of the assumptions by itself. However,
we can analyze a network in which strong ties connect heterogeneous players (a left-glove owner with a right-glove owner) and
weak ties connect holders of the same type of glove, respectively.
Consider the probabilistic network

with the two cliques {l1 , r1 } and {l2 , r2 } and the associated clique
assumptions
min(c1 , c2 ) > max(bl , br , b12 , b21 ) and
min(b11 , b12 , b21 , b22 ) > 0.
In testing Granovetters thesis I, we need two comparisons. If we
remove the strong link l1 r1 , we can remove either the weak link l1 l2
or the weak link l1 r2 .
Proposition 26. Assume the gloves game v and the above probabilistic network s in the special case of b := bl = br = b12 = b21 and c := c1 = c2 .
In case of c := 1, Granovetters thesis I does not hold for the link pair
l1 r1 and l1 l2 . For any c, Granovetters thesis I does not hold for the link
pair l1 r1 and l1 r2 . These results are strictly in disfavor of Granovetters
thesis with the notable exception of equality for the (symmetric!) link
pair l1 r1 and l1 r2 in case of b = c.

Similarly, Granovetters thesis II does not hold for the link pair l1 r1
and l1 l2 , but for the link pair l1 r1 and l1 r2 .
Thus, Granovetters implicit assumptions are vital for the validity of his thesis, if we look at our rst test (thesis I). Surprisingly,
these assumptions seem not to be too important, if we use an alternative test (thesis II). When interpreting this nding, keep in mind
that the alternative test is not as close as the rst test to Granovetters original argumentation.
5.5. Relationship to the empirical literature
Though our analysis primarily aims at a better understanding of the theoretical underpinnings of Granovetters thesis, our
results also connect nicely with some of the most interesting ndings regarding the workings of social capital (cf. Flap and Volker,
2013) in the labor market. As already indicated in the introduction,
empirical research on the use of informal means in job search provides only modest support for Granovetters thesis. While social
contacts are frequently used in getting a job, weak ties do not necessarily trump strong ties in terms of providing access to better
jobs.
Lin (1982, 1999; see also 2001) proposed the probably most
inuential theoretical solution to the puzzle, why there is only
mixed evidence with respect to the benecial effects of using weak
ties in job search. In terms of social capital, Lin argues that (1) the
higher the social status of the contact, the better the job outcome
for the seeker. He adds that (2) weak ties are not valuable per se, but
weak ties have a greater capability of bridging gaps between status groups. That is, especially for low-status actors weak ties might
provide the only connection to high-status actors who in turn could
boost their status attainment. Granovetter (1983) embraces these
ideas and essentially argues that (2) might hold to different degrees
depending on the actors occupational eld and status. This in turn
explains, why some studies nd positive effects of weak-tie connections on status attainment, while other studies nd negative
effects. In the meantime, Lins account has received considerable
empirical support (e.g. Biang and Ang, 1997; Ensel, 1979; Lai et al.,
1998; Lin et al., 1981; Lin and Dumin, 1986; Marsden and Hurlbert,
1988; Volker and Flap, 1999), although part (2) of his argument
appears to be a bit more shaky than part (1).
Our reconstruction backs Lins argument from a formal point
of view. Recall, part (1) of Lins argument states that ties to highstatus actors are valuable to job seekers. Though in a strict sense
status is absent in our model, it is captured indirectly via the productivity and hence the value of ties. Propositions 25 and 26 differ
with respect to the productivity of the dyads connected by weak
ties. That is, if weak ties connect complements as in the premise of
Proposition 25, part (2) of Lins argument is fullled, in the sense
that weak ties are valuable for both parties involved. Backing Lins
argument, we nd favorable conditions for the truthfulness of Granovetters thesis under these conditions. However, in case weak ties
connect substitutes as in the premise of Proposition 26, part (2) of
Lins argument is violated, because weak ties do not provide benets for the connected actors. In line with Lins reasoning, we nd
that Granovetters thesis generally lacks validity in this scenario. In
this sense our reconstruction of Granovetters network theory captures Lins reasoning on network effects on the labor market. Also,
to the extent that empirically networks of the type described by
Proposition 25 are more likely to occur than networks of the type
assumed in Proposition 26, our model predicts overall benecial
effects of weak ties on status attainment.
Generally speaking, the main lesson to learn from the empirical
literature on Granovetters network theory is the fact that weak ties
are not valuable per se but, depending on the application under consideration, under additional assumptions. As demonstrated with

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

respect to network effects on the labor market, our framework provides a powerful tool to theorize both the strength of ties and the
value of ties by combining probabilistic networks and cooperative
game theory.

145

Proof of Proposition 17. Assume a probabilistic network s that is

triangular-balanced and a cycle P(ii) = (i = i0 , s(i0 , i1 ), i1 , s(i1 , i2 ), . . .,
s(im1 , im ), im = i). Triangular balancedness implies
s(i0 , i1 )

s(i1 , i2 )s(i2 , i0 )
s(i1 , i2 )[s(i2 , i3 )s(i3 , i0 )]

6. Conclusion

...
While there are some formal models related to Granovetters thesis, this paper provides a formalization not only of his
thesis, but also of its derivation. This is accomplished by combining concepts from graph theory and cooperative game theory.
The usage of game theory allows us to take into account the
major lesson from empirical research on Granovetters thesis:
The analytical distinction between the strength of ties and the
value of ties. We use our model to put Granovetters thesis to a
test of its theoretical validity. It turns out that cum grano salis
Granovetters thesis holds, provided that Granovetters implicit
assumptions concerning homophily and benecial consequences
of bridges between distinct cliques hold. Also by analyzing a situation in which these assumptions are violated and comparing
the results to a situation in which they are satised, we nd support for Lins theory regarding the strength of weak ties in job
search.
In our view, a major drawback of Granovetters argument and
hence also of our reconstruction is the fact that a proper notion
of long run equilibrium is missing. That is, Granovetter starts
with balanced networks and removes links from these in deriving
his thesis. The removal of links from a balanced network typically leads to an unbalanced network, and hence the question
arises how the network restores balance in the long run. We feel
that future work on Granovetters network theory should explore
the question under what circumstances his thesis holds when
long term consequences of balancing tendencies are taken into
account.
Finally, we want to stress that cooperative game theory proved
to be a very powerful tool in our reconstruction of Granovetters network theory. With the very notable exception of Braun
and Gautschi (2006) little work has been put in this direction
recently. To us it seems that cooperative game theory has much
more potential to be useful for formalizing the ideas of sociologists like Granovetter or Emerson than is recognized at the
moment.

s(ij , ij+1 ),
m1
j=1
so that s is also cycle-balanced. The other direction of the proof
obtains for m = 3.
Proof of Proposition 21. The triangular inequality implies that
indirectly connected nodes are directly connected. Therefore, if the
triangular inequality holds and if s(ij) is strictly positive, we nd
|{k N\{i, j} : s(ik) > 0 and s(jk) > 0}|
= 1,
|{k N\{i, j} : s(ik) > 0 or s(jk) > 0}|
so that common-friends II is fullled for any s(ij) [0, 1]. Common
friends II does not imply the triangular inequality which can be seen
from the network consisting of three players 1, 2, and 3 with link
strengths s(12) = s(13) = 1, s(23) = 0.4 which satisfy common-friends
II but not the triangular inequality.
Proof of Proposition 24. The proof consists of straightforward
but very (!) tedious calculations. It is available from the authors
upon request.
Proof of Proposition 25. We rst check on variant I. Eliminating
the link between l1 and l2 leads to the value difference

vs{l

1 ,l2 },{r1 ,r2 }

1 l2 )
(N) v(sl
{l ,l },{r
1 2

1 ,r2 }

(N)

= b11 b12 cl + b21 b22 cl cl (b11 b12 b21 + b11 b12 b22 + b11 b21 b22
+ b12 b21 b22 ) 2cl cr (b11 b12 + b11 b21 + b11 b22 + b12 b21 + b12 b22
+ b21 b22 ) + 2b11 b12 b21 b22 cl + cl cr (3b11 b12 b21 + 3b11 b12 b22
+ 3b11 b21 b22 + 3b12 b21 b22 ) 4b11 b12 b21 b22 cl cr
+ cl cr (b11 + b12 + b21 + b22 ),
while the elimination of the link l1 r1 yields

vs{l

1 ,l2 },{r1 ,r2 }

1 r1 )
(N) v(sl
{l ,l },{r
1 2

1 ,r2 }

(N)

= b11 b11 (b12 + b21 ) + b11 b12 b21 + b11 b12 b22 + b11 b21 b22
Acknowledgements

+ b11 b12 cl + b11 b21 cr 2b11 b12 b21 b22 (cl + cr )(b11 b12 b21

We thank Patrick Doreian and a referee for helpful comments

regarding the empirical application of our model and the readability of the paper. Also, thanks to Hendrik Kohrs for his excellent
proofreading of the manuscript.

+ b11 b12 b22 + b11 b21 b22 ) 2cl cr (b11 b12 + b11 b21 + b11 b22 )
+ 2b11 b12 b21 b22 (cl + cr ) + cl cr (3b11 b12 b21 + 3b11 b12 b22
+ 3b11 b21 b22 ) 4b11 b12 b21 b22 cl cr + cl cr b11 .
Granovetters thesis I holds for our simple gloves game if

Appendix

Proof of Proposition 14. Assume s(12) < s(13)s(23) so that the

triangular inequality is violated. Then
s(13)

s(13)s(23)

v(sl1 l2 ) (N) > v(sl1 r1 ) (N) vs{l

1 ,l2 },{r1 ,r2 }

< vs{l

1 ,l2 },{r1 ,r2 }

(N) v(sl1 l2 ) (N)

(N) v(sl1 r1 ) (N)

which holds for

b21 b22 cl cl (b11 b12 b21 + b11 b12 b22 + b11 b21 b22 + b12 b21 b22 )

> s(12)

2cl cr (b12 b21 + b12 b22 + b21 b22 ) + cl cr (3b12 b21 b22 )

s(12)s(23)

+ cl cr (b12 + b21 + b22 ) b11 + b11 (b12 + b21 ) b11 b12 b21

so that the triangular inequality for s(13) (and symmetrically for

s(23)) holds.

b11 b12 b22 b11 b21 b22 b11 b21 cr + 2b11 b12 b21 b22
2b11 b12 b21 b22 cr < 0.

146

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

Considering the special case of identical strengths intra-clique and

inter-clique and assuming c := 1, Granovetters thesis I holds for

and using the Proof of Proposition 25, the test of Granovetters

thesis I boils down to

4b3 4b2 + 2b < 0,

v(sl1 r1 ) (N) > v(sl1 l2 ) (N)

i.e., for sufciently large bridge strengths (in particular, b >

1
0.366).
2
With respect to Granovetters thesis II, we nd
(s,l1 l2 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 l2 )

1
2

= b11 b12 + b21 b22 (b11 b12 b21 + b11 b12 b22

c1 b12 c2 c1 b21 c2 c1 b21 br + 2c1 b12 b21 c2

+ b11 b22 + b12 b21 + b12 b22 + b21 b22 )

2c1 b12 b21 c2 br > 0.

+ cr (b11 + b12 + b21 + b22 )

The special case (including c = 1) simplies this inequality to the

contradiction
b3 (b 6) > 1.
We now turn to the second comparison, the one between l1 r1 versus
l1 r2 . We obtain

and

ds(l1 r1 )

+ bl br (b12 + b21 + c2 ) c1 + c1 (b12 + b21 ) c1 b12 b21

+ b11 b21 b22 + b12 b21 b22 ) 2cr (b11 b12 + b11 b21

+ 3b11 b21 b22 + 3b12 b21 b22 ) 4b11 b12 b21 b22 cr

(s,l1 r1 )
(N)
1 ,l2 },{r1 ,r2 }

b21 c2 bl bl (c1 b12 b21 + c1 b12 c2 + c1 b21 c2 + b12 b21 c2 )

2bl br (b12 b21 + b12 c2 + b21 c2 ) + bl br (3b12 b21 c2 )

+ 2b11 b12 b21 b22 + cr (3b11 b12 b21 + 3b11 b12 b22

dv{l

and

= 1 (b12 + b21 ) + b12 b21 + b12 b22 + b21 b22

1 r1 )
v(sl
(N) = b21 + c2 + b12 c2 (b12 + b21 ) + b12 b21 c2
{l1 ,l2 },{r1 ,r2 }

+ b21 c2 bl + b12 c2 br (bl + br )b12 b21 c2

+ b12 cl + b21 cr 2b12 b21 b22 (cl + cr )(b12 b21

2bl br (b12 b21 + b12 c2 + b21 c2 ) + bl br (3b12 b21 c2 )

+ b12 b22 + b21 b22 ) 2cl cr (b12 + b21 + b22 )

+ 2b12 b21 b22 (cl + cr ) + cl cr (3b12 b21 + 3b12 b22
+ 3b21 b22 ) 4b12 b21 b22 cl cr + cl cr .
Turning again to the special case of identical intra-clique and interclique strengths, the two derivatives simplify:
(s,l1 l2 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 l2 )

+ bl br (b12 + b21 + c2 )
and
1 r2 )
v(sl
(N) = b21 + c2 + c1 (c1 + c2 )b21 + c1 b21 c2 + b21 c2 bl
{l1 ,l2 },{r1 ,r2 }

+ c1 b21 br (bl + br )c1 b21 c2 2bl br (c1 b21 + c1 c2

+ b21 c2 ) + bl br 3c1 b21 c2 + bl br (c1 + b21 + c2 ).

= 2b2 4b3 12cb + 2b4 + 12b3 c 4b4 c + 4bc

Equal inter-clique and intra-clique strengths (we do not need
c = 1) implies

and
(s,l1 r1 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 r1 )

= 1 2b + 3b2 + 2bc 2b3 6cb 6c 2 b + 4b3 c

2 2

3 2

+ 9b c 4b c + c .
Now, we have Granovetters thesis II, i.e., the inequality
(s,l1 l2 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 l2 )

(s,l1 r1 )

<

dv
(N)
,
ds(l1 r1 )

if
2

2b2 4b3 12cb + 2b4 + 12b3 c 4b4 c + 4bc (1 2b + 3b2

2

+ 2bc 2b3 6cb 6c 2 b + 4b3 c + 9b2 c 2 4b3 c 2 + c 2 ) < 0

or

v(sl1 r1 ) (N) > v(sl1 r2 ) (N) (c b)(1 b)

(b + 3b2 c bc 2b2 1) > 0
where c = b implies v(sl1 r1 ) (N) = v(sl1 r2 ) (N) and
b + 3b2 c bc 2b2 1 > 0
is obviously false.
With respect to Granovetters thesis II and for the comparison of l1 r1 versus l1 l2 , we nd (use substitution and the Proof of
Proposition 25)
(s,l1 r1 )
1 ,l2 },{r1 ,r2 }

dv{l

(N)

ds (l1 r1 )

= (b 1)

2bc 4b2 c + 3b2 + 1

and
2

b 5bc + 4b2 c 4b3 c + 4b2 c 2 2bc + 2b3 + c 2 + 1 > 0

holds (we always have b < 1). In the special case c = 1, we nd
2

b(2 b) < 1
and Granovetters thesis II is true
for sufciently small bridge
strengths (more precisely, b < 32 12 5 0.381 97). In the special
case of c = 0.8, however, Granovetters thesis II holds for b < 0.537
or for b > 0.716 (both gures are approximated).
Proof of Proposition 26.
versus l1 l2 . Substituting

We begin with the comparison of l1 r1

cl bl , cr br , b11 c1 , b22 c2

(s,l1 l2 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 l2 )

= 2b(b 1)(c 1)(b + 2c 2bc)

for the equal probabilities. We obtain a conrmation of Granovetters thesis II if

(s,l l )

(s,l r )
dv{l ,l1 2},{r ,r } (N)
dv 1 1 (N)
1 2
1 2
<
ds(l1 r1 )
ds(l1 l2 )

or
2

(1 b)(4b3 c + 3b3 + 4b2 c 2 b2 4bc + 2bc + b 1) > 0

holds,
which
is
not
true
for
c := 1
4b3 + 3b3 + 4b2 b2 4b + 2b + b 1 = (1 b)( 2b + b2 1).

by

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

Turning to Granovetters thesis II and the comparison of l1 r1

versus l1 r2 , we nd
(s,l1 r2 )
(N)
1 ,l2 },{r1 ,r2 }

dv{l

ds(l1 r2 )

Assume max( 0 , 1 ). Then

s(ij)

= s .

+ c1 bl + c2 br 2c1 b21 c2 (bl + br )(c1 b21

+ c1 c2 + b21 c2 ) 2bl br (c1 + b21 + c2 )
+ 2c1 b21 c2 (bl + br ) + bl br (3c1 b21 + 3c1 c2
+ 3b21 c2 ) 4c1 b21 c2 bl br + bl br .
dv

dv

Since

(s,l1 r1 )
(N)
{l1 ,l2 },{r1 ,r2 }

ds(l1 r1 )

(s,l1 r2 )
(N)
{l1 ,l2 },{r1 ,r2 }

ds(l1 r2 )

= 1.

1 holds for any b [0, 1], Granvotters thesis

holds with respect to link pairs l1 r1 versus l1 r2 .

Remark 27. The triangular inequality can be expressed by the setvalued function (the correspondence) Stri (ij) : s(N) 2[0,1] given by
tri

S (ij)(s) = {s(ij) [0, 1] : s(ij) s(ik) s(jk) forall k =

/ j and
/ i}, ij N
k=

(2)

Stri (ij)(s)

is the set of link strengths between players i and j allowed

by the triangular inequality.
Common friends II can be expressed by the correspondence ScfII (ij) : s(N) 2[0,1]
given by ScfII (ij)(s) = {s(ij) [0,
1] : s(ij) CFR(ij)}. We call Stri (ij) (ScfII (ij)) the triangular (commonfriends-II) correspondence.
Proposition 28. For all i, j N, the triangular correspondence is continuous while the common-friends-II correspondence is not.

Proof of Proposition 28. We have to show both upper hemicontinuity and lower hemicontinuity of Stri (ij). Note, rst of all,
Stri (ij)(s) = [s , 1] for s := max s(ik) s(jk). We now show upper
/ j,
k=
/ i
k=
hemicontinuity. According to de la Fuente, 2000, p. 112) it sufces to show that Stri (ij) is closed. Consider any sequence (s ) of
probabilistic networks that converges to some probabilistic network s. It gives rise to a sequence S tri (ij) (s ) of allowed sets of link
strengths. Consider the
(s(ij) ) of link strengths obeying
sequence

s(ij) S tri (ij) (s ) = s , 1 that converges to some t (ij). We have



shown the closedness of S tri (ij) if we can show t (ij) S tri (ij) s =
[s , 1].
/ S tri (ij)(s) = [s , 1]. Then there exists an > 0 such
Assume t(ij)
that t(ij) + < s . By s(ij) t(ij) there exists an 0 such that
s(ij) + < s for all 0 . By s s there exists an 1 such
that

max |s (jk) s(jk)| <  :=

j,k1,...,n

(1 + ) 1 > 0 for all 1 .

For these 1 we nd
s

:= max s(ik) s(jk)

/ j,
k=
k=
/ i
max (s (ik) + ) (s (jk) + )
/ j,
k=
k=
/ i

= max s (ik) s (jk) + s (ik) + s (jk) + 2

/ j,
k=
/ i
k=

max s (ik) s (jk) +  +  + 2

/ j,
k=
k=
/ i
= s + 2 + 2 .

< s
s + 2 + 2

= 1 + c1 (c1 + c2 ) + c1 b21 + c1 c2 + b21 c2

Assuming equal probabilities and c := 1, we have

147

Thus s(ij)
/ S tri (ij) (s ) = s , 1 for all max( 0 , 1 ) which
yields the desired contradiction.
We now turn to lower hemicontinuity. Following de la Fuente
(2000, p. 111) we consider a sequence (s ) of probabilistic networks
with s s. We also consider an arbitrary element t(ij) from
S tri (ij)(s) = [s , 1]. We need to show the existence of a sequence

(s(ij) ) of link strengths obeying s(ij) S tri (ij) (s ) = s , 1 and
that s(ij) t(ij).
Assume, rst, s = 1. Then t(ij) = 1. Dene s(ij) := 1 so that we
have s(ij) S tri (ij) (s ) and s(ij) t(ij)
Assume, now, s < 1. Let :=

t(ij)s
.
1s

Then 0 1. We dene

s(ij) := s + 1 s .
We have s(ij) S tri (ij) (s ) and, by s s, s s . Hence s(ij)
s + (1 s ) = t(ij). This completes the proof of Stri (ij)s continuity.
We now show that ScfII is not upper-hemicontinuous at
s = (s (12) , (13) , s (23)) = 0, 13 , 23 . Take the sequence (s ) =

 1

, 13 ,

N,



2
3

we

which

converges
S cfII

obtain

|{kN\{2,3}:s(k2)>0 and s(k3)>0}|

|{kN\{2,3}:s(k2)>0 or s(k3)>0}|

s (23) [0, 1] : s (23)

s = 0, 31 ,

to

(23) (s ) =

= [0, 1] ,

2
3

For

all

s (23) [0, 1] : s (23)

while

S (23) 0, 13 ,

|{kN\{2,3}:s(k2)>0 and s(k3)>0}|

|{kN\{2,3}:s(k2)>0 or s(k3)>0}|

the limit. Since any subsequence of the sequence

with elements 12 taken from (S (23) (s )) has limit
hemicontinuity is indeed violated.

1
2

0
1

1
2

2
3

=
at

, 12 , . . .

=
/ 0, upper

References
Biang, Y., Ang, S., 1997. Guanxi networks and job mobility in China and Singapore.
Soc. Forces 75, 9811005.
Bienenstock, E.J., Bonacich, P., 1992. The core as a solution concept to negatively
connected networks. Soc. Netw. 14, 231243.
Bienenstock, E.J., Bonacich, P., 1993. Game-theory models for exchange networks:
experimental results. Sociol. Perspect. 36, 117135.
Bollobas, B., 1998. Modern Graph Theory. Springer, New York.
Boorman, S.A., 1975. A combinatorial optimization model for transmission of job
information through contact networks. Bell J. Econ. 6, 216249.
Braun, N., Gautschi, T., 2006. A nash bargaining model for simple exchange networks.
Soc. Netw. 28, 123.
Bridges, W., Villemez, W., 1986. Informal hiring and income in the labor market. Am.
Sociol. Rev. 51, 574582.
Calvo, E., Lasaga, J., van den Nouweland, A., 1999. Values of games with probabilistic
graphs. Math. Soc. Sci. 37, 7995.
de la Fuente, A., 2000. Mathematical Methods and Models for Economists. Cambridge
University Press, Cambridge.
Easley, D., Kleinberg, J., 2010. Networks, Crowds, and Markets, Reasoning about a
Highly Connected World. Cambridge University Press, Cambridge.
Ensel, W.M., 1979. Sex, Social Ties, and Status Attainment. State University of New
York at Albany Press, Albany.
Fararo, T.J., 1983. Biased networks and the strength of weak ties. Soc. Netw. 5, 111.
Fararo, T.J., Skvoretz, 1987. Unication research programs: integrating two structural theories. Am. J. Sociol. 92, 11831209.
Forse, M., 1997. Capital social et emploi. LAnnee Sociol. 47, 27215.
Flap, H., Volker, B., 2013. Social capital. In: Wittek, R., Snijders, T., Nee, V. (Eds.), Handbook of Rational Choice Social Research. Stanford University Press, Stanford, pp.
220251.
Foster, C., Rapoport, A., Orwant, C., 1963. A study of a large sociogram II: elimination
of free parameters. Behav. Sci. 8, 5665.
Franzen, A., Hangartner, D., 2006. Social networks and labour market outcomes: the
non-monetary benets of social capital. Eur. Sociol. Rev. 22, 353368.
Granovetter, M.S., 1973. The strength of weak ties. Am. J. Sociol. 78, 13601380.
Granovetter, M.S., 1974. Getting a Job: A Study of Contacts and Careers. University
of Chicago Press, Chicago.

148

A. Tutic, H. Wiese / Social Networks 43 (2015) 136148

Granovetter, M.S., 1983. The strength of weak ties: a network theory revisited. Sociol.
Theory 1, 201233.
Granovetter, M.S., 2005. The impact of social structure on economic outcomes. J.
Econ. Perspect. 19, 3350.
Lai, G.W., Lin, N., Leung, S., 1998. Network resources, contact resources, and status
attainment. Soc. Netw. 20, 159178.
Lin, N., 1982. Social resources and instrumental action. In: Marsden, P., Lin,
N. (Eds.), Social Structure and Network Analysis. Sage, Beverly-Hills, pp.
131145.
Lin, N., 1999. Social networks and status attainment. Ann. Rev. Sociol. 25,
467487.
Lin, N., 2001. Social Capital: A Theory of Social Structure and Action. Cambridge
University Press, Cambridge.
Lin, N., Dumin, M., 1986. Access to occupations through social ties. Soc. Netw. 8,
365385.
Lin, N., Ensel, W.M., Vaughn, J.C., 1981. Social resources and strength of ties: structural factors in occupational status attainment. Am. Sociol. Rev. 46, 393405.
Marsden, P., Hurlbert, J., 1988. Social resources and mobility outcomes: a replication
and extension. Soc. Forces 66, 10381059.
Montgomery, J.D., 1992. Job search and network composition: implications of the
strength-of-weak-ties hypothesis. Am. Sociol. Rev. 57, 586596.
Montgomery, J.D., 1994. Weak ties, employment, and inequality: an equilibrium
analysis. Am. J. Sociol. 99, 12121236.

Mortensen, D.T., 1986. Job search and labor market analysis. In: Ashenfelter, O.,
Layard, R. (Eds.), Handbook of Labor Economics II. North-Holland, Amsterdam,
pp. 849919.
Mouw, T., 2003. Social capital and nding a job: do contacts matter? Am. Sociol. Rev.
68, 868898.
Myerson, R.B., 1977. Graphs and cooperation in games. Math. Oper. Res. 2, 225229.
Rapoport, A., 1979. A probabilistic approach to networks. Soc. Netw. 2, 118.
Schmeidler, D., 1969. The nucleolus of a characteristic function game. SIAM J. Appl.
Math. 17, 11631170.
Shapley, L.S., 1953. A value for N-person games. In: Kuhn, H., Tucker, A.W. (Eds.),
Contributions to the Theory of Games. Princeton University Press, Princeton,
pp. 307317.
Shapley, L.S., Shubik, M., 1969. Pure competition, coalitional power, and fair division.
Int. Econ. Rev. 10, 337362.
Sprengers, M., Tazelaar, F., Flap, H., 1988. Social resources, situational constraints,
and reemployment. Neth. J. Sociol. 24, 98116.
Tassier, T., 2006. Labor market implications of weak ties. South. Econ. J. 72, 704719.
A., Pfau, S., Casajus, A., 2011. Experiments on bilateral bargaining in markets.
Tutic,
Theory Decis. 70, 529546.
van Damme, E., 1998. On the state of the art in game theory: an interview with
Robert Aumann. Games Econ. Behav. 24, 181210.
Volker, B., Flap, H., 1999. Getting ahead in the GDR. Social capital and status attainment under communism. Acta Sociol. 42, 1734.