doi:10.1016/j.socnet.2006.08.002
Social Networks 29 (2007) 173–191
An introduction to exponential random graph (p*)
models for social networks
Garry Robins ∗, Pip Pattison, Yuval Kalish, Dean Lusher
Department of Psychology, School of Behavioural Science, University of Melbourne, Vic. 3010, Australia
Abstract
This article provides an introductory summary to the formulation and application of exponential random
graph models for social networks. The possible ties among nodes of a network are regarded as random
variables, and assumptions about dependencies among these random tie variables determine the general
form of the exponential random graph model for the network. Examples of different dependence assumptions
and their associated models are given, including Bernoulli, dyad-independent and Markov random graph
models. The incorporation of actor attributes in social selection models is also reviewed. Newer, more
complex dependence assumptions are briefly outlined. Estimation procedures are discussed, including new
methods for Monte Carlo maximum likelihood estimation. We foreshadow the discussion taken up in other
papers in this special edition: that the homogeneous Markov random graph models of Frank and Strauss
[Frank, O., Strauss, D., 1986. Markov graphs. Journal of the American Statistical Association 81, 832–842]
are not appropriate for many observed networks, whereas the new model specifications of Snijders et al.
[Snijders, T.A.B., Pattison, P., Robins, G.L., Handock, M. New specifications for exponential random graph
models. Sociological Methodology, in press] offer substantial improvement.
© 2006 Elsevier B.V. All rights reserved.
Keywords: Exponential random graph models; Statistical models for social networks; p* models
In recent years, there has been growing interest in exponential random graph models for
social networks, commonly called the p* class of models (Frank and Strauss, 1986; Pattison and
Wasserman, 1999; Robins et al., 1999; Wasserman and Pattison, 1996). These probability models
for networks on a given set of actors allow generalization beyond the restrictive dyadic indepen-
dence assumption of the earlier p1 model class (Holland and Leinhardt, 1981). Accordingly, they
permit models to be built from a more realistic construal of the structural foundations of social
behavior. The usefulness of these models as vehicles for examining multilevel and multitheoretical
hypotheses has been emphasized (e.g., Contractor et al., 2006).
∗ Corresponding author.
E-mail address: garrylr@unimelb.edu.au (G. Robins).
0378-8733/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.socnet.2006.08.002
mailto:garrylr@unimelb.edu.au
dx.doi.org/10.1016/j.socnet.2006.08.002
174 G. Robins et al. / Social Networks 29 (2007) 173–191
There have been a number of major theoretical and technical developments since Anderson
et al. (1999) presented their well-known primer on p* models. We summarize these advances in
this paper. In particular, we consider it important to ground these models conceptually in their
derivation from dependence assumptions, as the underlying basis of a model is then made explicit
and more readily linked with hypotheses about (unobserved) social processes underlying network
formation. It is through such an approach that new models can be developed in a principled way,
including models that incorporate actor attributes. Recent developments in model specification
and estimation need to be noted, as do new technical steps regarding setting structures and partial
dependence assumptions that not only expand the class of models but have important conceptual
implications. In particular, we now have a much better understanding of the properties of Markov
random graphs, and promising new specifications have been proposed to overcome some of their
deficiencies.
This article describes the models and summarizes current methodological developments with
an extended conceptual exposition. (More technical recent summaries are given by Wasserman and
Robins, 2005; Robins and Pattison, 2005; Snijders et al., in press.) We begin by briefly describing
the rationale for analyzing social networks with statistical models (Section 1). We then provide
an overview of the underlying logic of exponential random graph models and outline our general
framework for model construction (Section 2). In Section 3, we discuss the important concept of
a dependence assumption at the heart of the modeling approach. In Section 4, we present a range
of different dependence assumptions and models. For model estimation (Section 5), we briefly
summarize the pseudo-likelihood estimation (PLE) approach, and review recent developments in
Monte Carlo Markov chain maximum likelihood estimation techniques. In Section 6, we present
a short example of fitting a model to network data. In conclusion, we note the importance of the
new model specifications that are the focus of attention in other papers in this special edition.
1. Why model social networks?
There are many well-known techniques that measure properties of a network, of the nodes,
or of subsets of nodes (e.g., density, centrality and cohesive subsets). These techniques serve
valuable purposes in describing and understanding network features that might bear on particular
research questions. Why, then, might we want to go beyond these techniques and search for a
well-fitting model of an observed social network, and in particular a statistical model? Reasons
for doing so include the following:
(1) Social behavior is complex, and stochastic models allow us to capture both the regularities
in the processes giving rise to network ties while at the same time recognizing that there
is variability that we are unlikely to be able to model in detail. Moreover, as Watts (1999)
has cogently demonstrated, “adding” a small amount of randomness to an otherwise regular
process can dramatically alter the properties of the possible outcomes of that process. It is
therefore important to allow for stochasticity if we believe that it best reflects the processes
we aim to model. Perhaps most importantly, a well-specified stochastic model allows us
to understand the uncertainty associated with observed outcomes: we can learn about the
distribution of possible outcomes for a given specification of a model, or we can estimate,
for given observed data, the parameters of the hypothesized model from which the data may
have been generated (and also obtain quantitative estimates of the uncertainty associated with
estimation).
G. Robins et al. / Social Networks 29 (2007) 173–191 175
(2) Statistical models also allow inferences about whether certain network substructures – often
represented in the model by one or a small number of parameters – are more commonly
observed in the network than might be expected by chance. We can then develop hypotheses
about the social processes that might produce these structural properties.
(3) Sometimes, different social processes may make similar qualitative predictions about network
structures and it is only through careful quantitative modeling that the differences in predic-
tions can be evaluated. For instance, clustering in networks might emerge from endogenous
(self-organizing) structural effects (e.g., structural balance), or through node-level effects
(e.g., homophily). To decide between the two alternatives requires a model that incorporates
both effects and then assesses the relative contribution of each.
(4) The more complex the network data structure, the more useful properly formulated models
can be in achieving efficient representation. It is notable that there are a variety of deterministic
approaches for analyzing single binary networks, but many of these are not appropriate, or are
too complex, for more complicated data. To understand network evolution (Snijders, 2001)
or multiple network structures (Lazega and Pattison, 1999), models can be of great value.
(5) Several longstanding questions in social network analysis relate to the puzzle of how localized
social processes and structures combine to form global network patterns, and of whether
such localized processes are sufficient to explain global network properties. It is difficult to
investigate such questions without a model, as in all except rather simple cases the global
outcomes resulting from the combinations of many small-scale structures are not immediately
obvious, even qualitatively. With good locally specified models for social networks, it may
be possible to traverse this micro-macro gap, often through simulation.
We particularly emphasize the value of developing plausible models that are estimable from data
and hence empirically grounded. There are many models in the network literature that are impor-
tant tools for simulation, hypothesis generation, and “thought experiments”. But our principal
goal is to estimate model parameters from data and then evaluate how adequately the model rep-
resents the data. These complementary approaches serve useful but different purposes, with the
distinctive value of the data-driven approach clearly being its capacity for empirical interrogation
of the assumptions underpinning model construction.
2. The logic behind p* models for social networks—an outline1
We describe as the observed network the network data the researcher has collected and is
interested in modeling. The observed network is regarded as one realization from a set of possible
networks with similar important characteristics (at the very least, the same number of actors), that
is, as the outcome of some (unknown) stochastic process. In other words, the observed network
is seen as one particular pattern of ties out of a large set of possible patterns. In general, we do
not know what stochastic process generated the observed network, and our goal in formulating a
model is to propose a plausible and theoretically principled hypothesis for this process.
For instance, one of our research questions may be whether in the observed network there are
significantly more, or less, structural characteristics of interest than expected by chance. We might
see these characteristics as the outcomes of local social processes. For example, we might ask –
as Moreno and Jennings (1938) did in one of the first applications of statistics to social networks –
1 For other introductions to the logic of p* modeling, see Monge and Contractor (2003) and Contractor et al. (2006).
176 G. Robins et al. / Social Networks 29 (2007) 173–191
whether the observed network shows a strong tendency for reciprocity, over and above the chance
appearance of a number of reciprocated ties if relationships occurred completely at random. In
other words, do actors in the observed network tend to reciprocate relationship choices? Here
the structural characteristic (reciprocated ties) is the outcome of a social process (individuals
choosing to reciprocate the choices of others). Thus, as a simple example, we might posit a
stochastic network model with two parameters, one that reflects the propensity for ties to occur
at random and one that reflects an additional propensity for reciprocation to occur.
In general, the structural characteristics in question help to shape the form of the model. An
assumption of a reciprocity process leads us to propose a model in which an index of the level
of reciprocity is a parameter. The assumption also reflects an expectation about what sort of
networks are more likely. A statistical model for a network on a given set of actors assigns a
probability to all possible networks on those actors. For instance, since reciprocity of ties is a
commonly observed feature in friendship networks, a good model is likely to imply that networks
with reciprocation are more common and networks without reciprocation are rather improbable.
As is usual, we represent networks as graphs of nodes and edges. For a given model, the
node set is regarded as fixed. The range of possible networks, and their probability of occurrence
under the model, is represented by a probability distribution on the set of all possible graphs
with this number of nodes. In this distribution of graphs, those graphs with substantial levels
of reciprocation are likely to have higher probability than graphs with little reciprocation, with
the precise probabilities depending on the value of relevant parameters, such as a reciprocity
parameter. Note that the observed network is a particular graph in this distribution and so it also
has a particular probability.
Of course, at the outset, we do not know which parameter values to use in assigning probabilities
to graphs in the distribution. Our goal, rather, is to find the best values (by estimating model
parameters) using the observed network as a guide. The essential maximum likelihood criterion
is to choose parameter values in such a way that the most probable degree of reciprocation is
that which occurs in the observed network. If the model has a reciprocity parameter (defined
to be zero when reciprocal ties occur by chance), and if there are many reciprocated ties in the
observed network, then a model that is a good fit to the data in terms of degree of reciprocation
will have a positive reciprocity parameter. If we estimate a reciprocity parameter for the observed
network, and if we can be confident that this parameter is positive, we may infer that there is more
reciprocity in the observed network than expected by chance.
Once we have defined a probability distribution on the set of all graphs with a fixed number
of nodes, we can also draw graphs at random from the distribution according to their assigned
probabilities, and we can compare the sampled graphs to the observed one on any other charac-
teristic of interest. If the model is a good one for the data, then the sampled graphs will resemble
the observed one in many different respects. In this ideal case, we might even hypothesize that
the modeled structural effects could explain the emergence of the network. And we can examine
the properties of the sampled graphs in order to understand the nature of networks that are likely
to emerge from these effects.
As an example, consider friendship in a school classroom. The observed network is the network
for which we have measured friendship relations. There are many possible networks that could
have been observed for that particular classroom. We examine the observed friendship structure in
the classroom in the context of all possible network structures for the classroom. Some structures
in the classroom may be quite likely and some very unlikely to happen, and the set of all possible
structures with some assumption about their associated probabilities is a probability distribution
of graphs. We are placing the observed network within this distribution, rather than comparing
G. Robins et al. / Social Networks 29 (2007) 173–191 177
the observed network to friendship networks in other classrooms. (Of course, our model for the
observed network may also be a good model for other classrooms but that is not the issue at this
point.)
Note that the assumption is that the network is generated by a stochastic process in which
relational ties come into being in ways that may be shaped by the presence or absence of other
ties (and possibly node-level attributes). In other words, the network is conceptualized as a self-
organizing system of relational ties. Substantively, the claim is that there are local social processes
that generate dyadic relations, and that these social processes may depend on the surrounding
social environment (i.e. on existing relations). For example, we can assume that actors with
similar attributes are more likely to form friendship ties (homophily), or that if two unconnected
actors were connected to a third actor, at some point they are likely to form a friendship tie between
them (transitivity). Note that in addition to the assumption of stochasticity, this description is also
implicitly temporal and dynamic.
2.1. A general framework for model construction
In positing an exponential random graph model for a social network, a researcher implicitly
follows five steps. While the focus of research is on the final step of parameter estimation and
interpretation, it is through all the five steps that a researcher makes explicit choices that connect
theoretical decisions to data analysis. And as shown below, it is through these earlier steps that we
can locate certain earlier network models within the rubric of exponential random graph models.
2.1.1. Step 1: each network tie is regarded as a random variable
This step implies a stochastic framework with a fixed node set. By assuming that a tie is a ran-
dom variable we do not imply that people form relations in an ad hoc fashion: some relationships
might be highly probable. Rather, we are simply stating that we do not know everything about
relationship formation, that our model is not going to make perfect deterministic predictions, and
that as a result there is going to be some statistical “noise”, or lack of regularity, that we cannot
successfully explain.
With possible network ties established as random variables, it is timely to review some basic
notation. For each i and j who are distinct members of a set N of n actors, we have a random
variable Yij where Yij = 1 if there is a network tie from actor i to actor j, and where Yij = 0 if there
is no tie. We specify yij as the observed value of the variable Yij and we let Y be the matrix of
all variables with y the matrix of observed ties, the observed network. Of course, y can also be
construed as a graph on the node set N, with the edge set specified by those pairs (i,j) for which
yij = 1. Y may be directed (in which case Yij is distinguished from Yji) or non-directed (where
Yij = Yji and the two variables are not distinguished). It is also possible for y to be valued, although
for this article we will restrict attention to binary ties.
2.1.2. Step 2: a dependence hypothesis is proposed, defining contingencies among the
network variables
This hypothesis embodies the local social processes that are assumed to generate the network
ties. For instance, ties may be assumed to be independent of each other, that is, people form
social connections independently of their other social ties. This is not usually a very realistic
assumption. In the example of the school classroom with reciprocity processes in place, if student
A likes student B, then student B will quite probably like student A implying some form of dyadic
dependence. Ties may also depend on node-level attributes (see Section 4.4 below), with for
178 G. Robins et al. / Social Networks 29 (2007) 173–191
instance possible homophily effects in the classroom. Notice that each of these processes can be
represented as a small-scale graph configuration: for instance, a reciprocated tie, or a tie between
two girls.
2.1.3. Step 3: the dependence hypothesis implies a particular form to the model
It can be proven that well-specified dependence assumptions imply a particular class of models
(the Hammersley–Clifford theorem, Besag, 1974). Each parameter corresponds to a configuration
in the network, that is, a small subset of possible network ties (and/or actor attributes—although
that is for later). These configurations are the structural characteristics of interest (e.g., recipro-
cated ties), referred to above. The model then represents a distribution of random graphs which
are assumed to be “built up” from the localized patterns represented by the configurations. For
instance, a single tie is a configuration, as may be a reciprocated tie (in a directed graph), a tran-
sitive triad and a two-star. Parameters related to the presence of each of these configurations in
the observed graph may be included in a model.
Dependence assumptions and the general form of the model are discussed in Section 3 below.
Particular dependence assumptions are presented in Section 4.
2.1.4. Step 4: simplification of parameters through homogeneity or other constraints
In order to define a model clearly, we need to reduce the number of parameters. This is
often done by imposing homogeneity constraints. In effect, we ask whether some parameters
should be equated or related in other ways. For instance, we usually propose one parameter for a
reciprocity effect across the entire network, by assuming that the reciprocity parameters for each
possible reciprocated tie are all equal. Parameter constraints for particular models are illustrated in
Section 4.
2.1.5. Step 5: estimate and interpret model parameters
Of course, estimation and interpretation are usually a focus of particular research applications,
but reaching this step implies that the other four have already been undertaken, even if only
implicitly. This step is complicated if the dependence structure is complex, as it probably needs
to be for any realistic model. Having obtained parameter estimates, as well as estimates of the
uncertainty of estimation, we may then take full advantage of having a statistical model for the
network that is constructed from specifiable dependence assumptions and that is estimated from
observed network data. For example, we can explore the range of network outcomes predicted
by the model, a step that can be very helpful in assessing how good the model is, and we can
make inferences about model parameters. For instance we can infer whether any model parameter
is significantly different from zero and so whether the corresponding configuration is present in
the observed graph to a greater or lesser extent than expected by chance, given other parameter
values. We discuss parameter estimation in Section 5.
3. The general form of the exponential random graph model: dependence assumptions
and parameter constraints
Exponential random graph models have the following form:
Pr(Y = y) =
(
1
κ
)
exp
{∑
A
ηAgA(y)
}
(1)
G. Robins et al. / Social Networks 29 (2007) 173–191 179
where (i) the summation is over all configurations A; (ii) ηA is the parameter corresponding to the
configuration A (and is non-zero only if all pairs of variables in A are assumed to be condition-
ally dependent);2 (iii) gA(y) =
∏
yij ∈ Ayij is the network statistic corresponding to configuration
A; gA(y) = 1 if the configuration is observed in the network y, and is 0 otherwise;3 (iv) κ is a
normalizing quantity which ensures that (1) is a proper probability distribution.4
All exponential random graph models are of the form of Eq. (1) which describes a general
probability distribution of graphs on n nodes. The probability of observing any particular graph y
in this distribution is given by the equation, and this probability is dependent both on the statistics
gA(y) in the network y and on the various non-zero parameters ηA for all configurations A in the
model. Configurations might include reciprocated ties, transitive triads and so on, so the model
enables us to examine a variety of possible structural regularities.
So why are dependence assumptions important here? Dependence assumptions have the con-
sequence of picking out different types of configurations as relevant to the model. Note from point
(ii) above, parameters are zero whenever variables in a configuration are conditionally indepen-
dent of each other. In other words, the only configurations that are relevant to the model are those
in which all possible ties in the configuration are mutually contingent on each other.5
It is worth noting that if a set of possible edges represents a configuration in the model, then
(1) implies that any subset of possible edges is also a configuration. Thus, single edges are always
configurations, as demonstrated in Section 4.
So the dependence assumption is crucial in constraining which configurations are possible in
the model. We will discuss particular examples in Section 4. A configuration A refers to a subset
of tie variables, and corresponds to a small network substructure. For instance, if for a directed
network we apply a dyadic dependence assumption (see Section 4) it will follow that reciprocity
parameters will be in the model. In this case, one configuration in the model is the set of variables
{Y12, Y21}, another is {Y13, Y31}, and so on, with every dyad providing its own configuration.
Obviously for any of these configurations, if both of the ties are present in the observed graph, we
see a reciprocated tie, so the configuration represents a type of network substructure that may be
observed in the graph y. We can think of this configuration diagrammatically as that substructure,
i.e. a reciprocated tie.
But of course there is no guarantee that all possible edges in a given configuration will be
present in a realized graph y, so we will observe some of these possible substructures but not
others. Some ties will be reciprocated, some will not. Configurations represent possibilities. The
graph statistic, gA(y), on the other hand, tells us whether the configuration A is in fact observed
in the network y. For a reciprocity configuration A, that statistic simply tells us whether there are
reciprocated ties between the relevant pair of nodes or not.
We can think of the graphs in the distribution as being generated by these potentially overlapping
configurations. For instance, suppose there is a reciprocity effect at work in the process generating
2 i.e. conditional on the rest of the graph.
3 We write gA(y), rather than gA, to remind ourselves that the statistics relate to the graph y.
4 It is possible to assert a model of the form of (1) by incorporating more general statistics than configuration and
subgraph counts (see Wasserman and Pattison, 1996). But then dependence assumptions may not be clear. Our preference
is for an explicit dependence structure in order to be able to link the model to interpretations regarding local social
processes.
5 More technically, the dependence assumptions may be represented in a dependence graph, first introduced into the
network literature by Frank and Strauss (1986), following the approach described by Besag (1974). The configurations
A are represented by the cliques of the dependence graph. Interested readers should consult Frank and Strauss (1986) for
further details; see also the review by Robins and Pattison (2005).
180 G. Robins et al. / Social Networks 29 (2007) 173–191
the network. If we could observe the evolution of the network, and if the network started with few
reciprocated ties, we might expect to see more reciprocated ties emerge over time. In thinking this
way, though, we need to bear in mind that as a particular tie emerges through an imagined process
of generation, its presence may affect other potential neighboring ties. So there is an implicitly
dynamic and self-organizing quality to this hypothetical construction process: as one tie emerges
or disappears, other neighboring ties are likely to emerge or disappear as well, and there may be no
natural endpoint to this ongoing stochastic process. Nonetheless, the strength and direction of any
particular parameter value will affect how frequently the corresponding configuration is observed.
If the parameter is large and positive, we expect to observe the corresponding configuration in
graphs in distribution (1) more frequently than if the parameter were zero. So if a reciprocity
parameter were large and positive, we would expect to see many reciprocated ties in the observed
network. Likewise, when a parameter is large and negative we expect to see the configuration
(e.g., reciprocated ties) relatively less frequently than if the parameter is zero.
Because (1) has an exponential term in the right hand side, such distributions have been referred
to as exponential random graph models. The Markov random graphs of Frank and Strauss (1986)
are one particular class of exponential random graph models. The network analytic community also
refers to the exponential random graph model class as p* models because they are a generalization
of dyadic independence models, of which p1 models (Holland and Leinhardt, 1981) were a popular
early example.
3.1. Constraints on parameters
Notice that Eq. (1) refers to different configurations for sets of different nodes. For instance, for
models with reciprocity there is a separate configuration for {Y12,Y21}, for {Y13,Y31}, and so on.
In this general form, then, the model implies many parameters. For instance, there are n(n − 1)/2
parameters relating to reciprocity alone.
This is simply too many parameters and the model cannot be estimated from a single network
observation. Some parameters need to be set to zero, equated or otherwise constrained. Following
Frank and Strauss (1986), we often impose a homogeneity assumption by equating parameters
when they refer to the same type of configuration. For instance, in considering reciprocity, Paul
may tend very strongly to reciprocate friendship offers from others, but Mary might be more
cautious. For the purpose of constructing a simpler model, however, we may assume that there
is a single tendency for reciprocity shared by both Mary and Paul. The resulting error is then
consumed into the model as statistical noise. This approach assumes that certain regularities are
the same for the entire network, for example, that there is a single tendency for reciprocity across
the network, irrespective of which nodes are involved. We term this homogeneity of isomorphic
network configurations, where parameters are equated if the configurations are the same when
we ignore the labels on the nodes (in which case the configurations are said to be isomorphic).
A less radical assumption is also possible: for instance, if we were able to measure whatever
characteristics of individuals incline them to reciprocate ties, we could allow the reciprocity
effect to depend on those node characteristics.
When we make this homogeneity assumption, we produce a model with the same form as Eq.
(1) but now the (isomorphic) configurations refer to generic effects (e.g., the overall reciprocity
effect). The statistics then become the counts of the corresponding configurations in the network
(e.g., the number of reciprocated ties).
But there are several other ways in which constraints on the parameters may be applied, and
different constraints result in different models. Another method of applying constraints may be to
G. Robins et al. / Social Networks 29 (2007) 173–191 181
equate parameters for isomorphic configurations involving similar types of actors. For example,
in the case of reciprocity in classroom friendship networks, we could propose one reciprocity
parameter for girl–girl configurations, one for girl–boy configurations and another for boy–boy
configurations.
Even with sensible homogeneity constraints in place the model may still have too many param-
eters to be estimable. In that case, we might consider limiting the number of configurations by
setting some parameters to zero (see Section 4.3), or by introducing hypothesized constraints on
the values of parameters associated with larger configurations (as proposed by Snijders et al., in
press; see Section 4.6).
4. Dependence assumptions and models
4.1. Bernoulli graphs: the simplest dependence assumption
Bernoulli random graph distributions are generated when we assume that edges are indepen-
dent, for instance if they occur randomly according to a fixed probability α (see Erdös and Renyi,
1959; Frank and Nowicki, 1993). The dependence assumption is simple in this case: all possible
distinct ties are independent of one another. We noted above that the only configurations relevant
to the model are those in which all possible ties in the configuration are conditionally dependent
on each other. When all possible ties are independent, the only possible configurations relate to
single edges {Yij}. So from (1) the general model is:
Pr(Y = y) =
(
1
κ
)
exp
⎛
⎝∑
i,j
ηijyij
⎞
⎠
Note that compared to (1) every set A comprising a single possible edge Yij is a configuration
in this model, and there is a parameter ηij for each of these configurations. The network statistic
gA(y) = gij(y) = yij tells us whether that configuration is observed or not. If we impose a homo-
geneity assumption so that the effect for each tie is identical we equate parameters such that ηij = θ
for all i and j, hence:
Pr(Y = y) =
(
1
κ
)
exp(θL(y)) (2)
where L(y) =
∑
i,jyij is the number of arcs in the graph y and the parameter θ is related to the
probability of a tie being observed.6 The parameter θ is called the edge or density parameter.
There are other possibilities for imposing homogeneity. Suppose we have actors in two a priori
blocks and we impose block homogeneity, so that ηij = θ11 if both i and j are in block 1, ηij = θ12
if i is in block 1 and j in block 2, and so on. Then it is simple to show that
Pr(Y = y) =
(
1
κ
)
exp(θ11L11(y) + θ12L12(y) + θ21L21(y) + θ22L22(y))
where L11(y) is the number of arcs within the first block and L12(y) is the number of arcs from
block 1 to block 2, and so on.
6 Specifically, α = exp θ/(1 + exp θ). The homogeneity assumption means that there is a fixed probability for all possible
edges across the graph, i.e. that there is a single α.
182 G. Robins et al. / Social Networks 29 (2007) 173–191
4.2. Dyadic models: the dyadic independence assumption
A somewhat more complicated (but not usually very realistic) assumption for directed networks
is that dyads, rather than edges, are independent of one another. With this dependence assumption
we have two types of configurations in the model, single edges and reciprocated edges. With
homogeneity imposed, the model then becomes:
Pr(Y = y) =
(
1
κ
)
exp
⎛
⎝θ∑
i,j
yij + ρ
∑
i,j
yijyji
⎞
⎠ =
(
1
κ
)
exp(θL(y) + ρM(y)) (3)
where L(y) is the number of ties in y and M(y) =
∑
i,jyijyji is the number of mutual ties in y. A
slightly more complex homogeneity assumption results in the p1 model of Holland and Leinhardt
(1981).
Related but more complex and realistic models include the p2 model (Lazega and van Duijn,
1997; Van Duijn et al., 2004) which assumes dyadic independence but conditional on node-level
attribute effects. The p2 model is appropriate when structure is expected to arise from attributes.
It is an extension of the p1 model with sender and receiver effects treated as random effects and
with actor and dyadic effects included. The more complex assumptions underpinning this model
make it more realistic for actual network data, especially when attribute effects are expected to
be strong. It differs from usual exponential random graph models in the incorporation of random
effects.Of course, in the case of non-directed networks, Bernoulli and dyad dependence models
are identical: for non-directed networks, the reciprocity parameter ρ in Eq. (3) is irrelevant and
the model reduces to that of Eq. (2).
4.3. Markov random graphs
Bernoulli and dyadic dependence structures are unrealistic assumptions in many circumstances,
both empirically and theoretically. Frank and Strauss (1986) introduced Markov dependence, in
which a possible tie from i to j is assumed to be contingent on any other possible tie involving
i or j, even if the status of all other ties in the network is known. In this case, the two ties are
said to be conditionally dependent, given the values of all other ties.7 Markov dependence can
be characterized as the assumption that two possible network ties are conditionally dependent
when they have a common actor. For instance, the relationship between Peter and Mary may well
be dependent on the presence or absence of a relationship between Mary and John (especially if
the relationship is a romantic one!) We can express this more formally by assuming conditional
dependence between the possible ties Ypm and Ymj. These two possible ties are conditionally
dependent because they share the node m (Mary).
If we also assume homogeneity, we obtain the Markov random graph model, with configu-
rations (and associated parameters) for directed and non-directed networks presented in Fig. 1.
These parameters relate to some well-known structural regularities in the network literature. For
directed networks, we have already seen the edge (τ15) and reciprocity (τ11) parameters from the
Bernoulli and dyadic independence models. There are various two-star effects: the two-out-star
parameter (τ12) can be thought of as relating to expansiveness, the two-mixed-star parameter
7 If two ties are conditionally dependent, then if the value of one tie changes, the probability of the other tie is affected,
even if all other ties in the network remain the same.
G. Robins et al. / Social Networks 29 (2007) 173–191 183
Fig. 1. Configurations and parameters for Markov random graph models.
(τ13) relates to two-paths, and the two-in-star parameter (τ14) relates to popularity. Note the
important transitivity and cyclic configurations (τ9 and τ10). The inclusion of these parameters
is a strength of these models because there is a paucity of network models that incorporate these
effects (Newman, 2003), and very few indeed that are estimable from data. The full parameter set
includes all possible higher order stars as well, although if all such stars are included there are too
many parameters for the model to be estimable. Although some early applications of the Markov
random graph model included only two-star effects, it is now known that it may be important to
184 G. Robins et al. / Social Networks 29 (2007) 173–191
include a non-zero parameter for at least the three-star effect in models for many social networks
(Robins et al., 2004, 2005). An alternative approach (see below) includes all higher-order star
parameters but imposes constraints on the relationships between higher-order star parameters and
lower-order ones.
For example, a Markov random graph model for a non-directed network with edge, two-star,
three-star and triangle effects is:
Pr(Y = y) =
(
1
κ
)
exp(θL(y) + σ2S2(y) + σ3S3(y) + τT (y)) (4)
where S2(y) and S3(y) are the numbers of two-stars and three-stars, respectively, in the network y
and T(y) is the number of triangles in y. Note that for Markov random graphs, it is also possible
to include parameters for stars of higher order than three (four-stars, five-stars, etc). The model in
Eq. (4) is an example of how we might set certain higher order parameters to zero (Section 3.1).
In this case, we are assuming that the distribution of stars (in effect, the degree distribution) can
be adequately explained by the two- and three-star effects.8
It should be noted that the statistics in the Markov model are often related to each other, in the
sense that some are higher-order to others. For instance, suppose there is a three-star in a non-
directed network centered on node i. Then it is also the case that there are three two-stars (and three
edges) also centered on i. This is analogous to higher order interactions in more familiar general
linear model procedures. This is an important feature of the model that assists interpretation. If,
for instance, a network has many two-stars present, then some will form triangles just by chance.
But if there is a substantial triangle effect in a Markov random graph model, this is over and
above any two-star effect, and we can infer that the level of triangulation did not occur simply
because of the chance overlapping of many two-stars (or indeed of many edges). In that case, we
would infer that triangulation was an important process in this network, independently of other
effects.
Several elaborations of this basic Markov random graph model have also been developed:
for multivariate networks (Pattison and Wasserman, 1999); for valued networks (Robins et
al., 1999); for affiliation networks (Skvoretz and Faust, 1999; see also Pattison and Robins,
2004).
4.4. Dependence structures with node-level variables
There are various ways of introducing node-level effects (actor attributes) into Markov and
other exponential random graph models. We assume a vector X of binary attribute variables with
Xi = 1 if actor i has the attribute and Xi = 0, otherwise. The vector x is then the set of observations on
X. It is possible to generalize to polytomous and continuous attribute measures but we will restrict
the current discussion to binary attributes. Here, as an example, we briefly describe social selection
models where attributes are assumed to be exogenous predictors of network ties (Robins et al.,
2001a).9 We can investigate a similarity or homophily hypothesis as a basis for social selection
– that social ties tend to develop between actors with the same attributes – by looking at the
8 To be confident about this, we could simulate a distribution of graphs from a fitted model and inspect the degree
distributions as compared to the observed network. Examples of such goodness of fit diagnostics are presented in other
papers in this special edition.
9 Another method to incorporate actor attributes is through social influence models, where network ties were taken as
exogenous predictors of attributes (Robins et al., 2001b).
G. Robins et al. / Social Networks 29 (2007) 173–191 185
Fig. 2. (A–H) Configurations for a Markov attribute—Markov graph social selection model.
distribution of ties given the distribution of attributes. In other words, as distinct from Eq. (1),
our interest is in the probability of the graph y given the observations of attributes x, that is,
Pr(Y = y|X = x).
A simple dependence assumption between the attribute and network variables is that the
attribute of i influences possible ties that involve i (i.e. Yij), referred to as a Markov attribute
assumption. For example, in an organizational setting, an actor’s seniority (say, senior manage-
ment versus junior management) may influence the possible ties of that actor. If we consider
Markov attributes along with Markov network dependencies, for a non-directed network the
model contains the configurations (up to three-stars) shown in Fig. 2, with a filled node repre-
senting an actor who has the attribute seniority (i.e. the actor is a senior manager), and an empty
node (with dotted line) just representing an actor, irrespective of whether junior or senior. In
other words, the configuration (A) represents tendencies for senior managers to have ties with
each other; whereas the configuration (B) represents the tendency for a senior manager to have
many ties, and so on. A large positive parameter estimate for configuration (A) is evidence for
homophily effects in the network.
It is notable from Fig. 2 that the only non-dyadic configurations with attributes involve two-
or three-stars, with the actor with the attribute at the centre of the star. To produce triangle
configurations with attribute variables requires additional dependence assumptions.
4.5. More complex dependence assumptions
Elaborations of exponential random graph models that go beyond Markov random graphs have
been developed. Pattison and Robins (2002) presented two innovations. With setting structures,
they confined dependencies within social settings. Drawing on Feld (1981), they suggested as
possible examples settings based on a spatiotemporal context, such as a group of people gathered
together at the same time and place; settings based on a more abstract sociocultural space, such as
pairs of persons linked by their political commitments; and settings that reflect external “design”
constraints, such as organizational structure.
An additional motivation to introduce settings is that Markov dependence seems unrealistic
for large networks, where individual actors may not even be aware of each other, and have no
means to come into contact, yet their possible tie still is taken to influence other possible ties. If
the setting structure hypothesis is well-founded, there are implications for the type of data that
186 G. Robins et al. / Social Networks 29 (2007) 173–191
needs to be collected for a full understanding of a social network. For further elaborations, see
also Schweinberger and Snijders (2003).
A second direction presented by Pattison and Robins (2002) was to propose non-Markov
dependencies among ties that did not share an actor but might be interdependent through third
party links. For instance, Yij may be conditionally dependent on Yrs for four distinct actors if there
is an observed tie between either i or j and either r or s. These realization-dependent models can be
developed through what Pattison and Robins (2002) described as partial dependence structures.
These models also permit the introduction of triangles involving attribute effects.
4.6. New model specifications
There is mounting evidence that homogeneous Markov random graph models are not good
models for many observed social networks (see Section 5.2 below), so these models are not
always useful in practical terms. Based on realization-dependence structures, Snijders et al. (in
press) developed new specifications for exponential random graph models that include new higher
order terms. These models introduce constraints on k-star parameters, as well as new higher-order
k-triangle configurations which allow for the measurement of highly clustered regions of the
network where two individuals may be connected to a large number of k others (a k-triangle). For
these models, many higher order star and triangle effects are included (rather than set to zero) but
they are constrained in the form of a weighted sum with alternating signs. The motivation behind
these innovations, and the success of these new model specifications, are discussed in other papers
in this special edition.
5. Estimation
Anderson et al. (1999) in their p* primer used pseudo-likelihood estimation introduced by
Strauss and Ikeda (1990) in order to estimate the parameters of Markov models. We now know
that, depending on the data, there may be serious problems with pseudo-likelihood estimates for
these models. But for Markov random graph models, standard maximum likelihood estimation
is not tractable for any but very small networks, because of the difficulties in calculating the
normalizing constant in Eq. (1). What this means is that standard statistical techniques cannot be
applied to these models. These problems have been overcome in recent times by the development
of new Monte Carlo maximum likelihood techniques. We begin by making some rather brief
comments about pseudo-likelihood and then introduce the new estimation approaches.
5.1. Pseudo-likelihood estimation: an approximate technique
The use of maximum pseudo-likelihood to estimate interactive models was first proposed by
Besag (1975), and was suggested for Markov random graph models by Strauss and Ikeda (1990).
In the general statistical community, pseudo-likelihood has given way to Monte Carlo techniques
where feasible, although it still has its adherents (see Wasserman and Robins, 2005, for some
of the literature). The advantage of pseudo-likelihood estimation in the context of exponential
random graph models is that it is relatively easy to fit even complicated models. The disadvantage
is that the properties of the estimator are not well understood and it is known that for many data
sets pseudo-likelihood estimates are not accurate.
Pseudo-likelihood estimation is best understood by transforming Eq. (1) – the joint form of
the model – into the following equivalent conditional form (see Strauss and Ikeda, 1990, for more
G. Robins et al. / Social Networks 29 (2007) 173–191 187
detail):
log
[
Pr(Yij = 1|yCij )
Pr(Yij = 0|yCij )
]
=
∑
A(Yij)
ηAdA(y) (5)
where (1) the sum is over all configurations A that contain Yij; (2) ηA is the parameter corresponding
to configuration A; (3) dA(y) is the change statistic; the change in the value of the network statistic
zA(y) when yij changes from 1 to 0; (4) y
C
ij is all the observations of ties in y except the observation
yij.
The calculation of the change statistic has been discussed extensively by a number of authors
(Anderson et al., 1999; Pattison and Robins, 2002; Wasserman and Pattison, 1996; Wasserman
and Robins, 2005), so we do not go into it further here. With the change statistics calculated,
to produce the pseudo-likelihood estimates, each possible tie Yij becomes a case in a standard
logistic regression procedure, with yij predicted from the set of change statistics (Anderson et al.,
1999).
This procedure looks like a logistic regression – or indeed, a loglinear model – but it is not.
Logistic regression assumes independent observations, an assumption we explicitly do not make
with Markov and higher order models. So the parameter estimates may be biased; and the standard
errors are approximate at best, and may be too small. One should not rely on the Wald statistic as
a means to decide whether a parameter is significant or not. As well, one cannot assume that the
pseudo-likelihood deviance is asymptotically distributed as Chi-squared (which would be the case
in normal logistic regression). When the dependence among observations is not so strong, it is
generally the case that PL estimates will be more accurate. Pseudo-likelihood estimation has been
used to date as a pragmatic convenience (given that alternatives have not hitherto been readily
available) and the method does not have a principled basis. Whenever possible, the preferred
option is to use Monte Carlo estimation procedures.
5.2. Markov chain Monte Carlo maximum likelihood estimation (MCMCMLE)
Important recent developments in Monte Carlo estimation techniques for exponential ran-
dom graph models have been presented and reviewed by a number of authors (see Snijders,
2002; Handcock et al., 2006; Snijders et al., in press; Wasserman and Robins, 2005), and are
further discussed in other articles in this special edition, so we include only a brief summary
here.
To begin, we note that simulation of these models can be implemented in a relatively straight-
forward way. Without going into details, simulation of the graph distribution for a given set of
parameter values can be achieved through a number of algorithms (e.g., algorithms well-known
in statistics more generally, such as the Metropolis algorithm). Simulation is at the heart of Monte
Carlo maximum likelihood estimation. Procedures for simulating exponential random graph dis-
tributions have been described by Strauss (1986), Snijders (2002) and Robins et al. (2005).
Although there are variations between different Monte Carlo estimation techniques (Snijders,
2002; Hunter and Handcock, 2006), they are based on the same central approach: simulation of a
distribution of random graphs from a starting set of parameter values, and subsequent refinement
of the parameter values by comparing the distribution of graphs against the observed graph, with
this process repeated until the parameter estimates stabilize. Recent software that implements
Monte Carlo maximum likelihood estimation for exponential random graph models is reviewed
in other papers in this special edition.
188 G. Robins et al. / Social Networks 29 (2007) 173–191
Both estimation and simulation studies have raised issues of model specification for Markov
random graphs. Handcock (2003) defined near degeneracy as occurring when a model implied
that only a few graphs had other than very low probability (often these were the full graph
or the empty graph). If a model implies only these rather uninteresting outcomes, it will not
be useful for modeling real networks. Simulation studies suggest that Markov graph mod-
els that contain at least non-zero three-star parameters tend to exhibit less near degeneracy
than those with two-stars as the highest order non-zero star parameter (Robins et al., 2005).
But the inclusion of three-star parameters often is not sufficient to remove near degeneracy
behavior in Markov graph models, particularly when attempting to find models that reproduce
the high levels of transitivity often observed in human social structures (there is an extended
discussion in Snijders et al., in press). The fact that these problems may not occur for pseudo-
likelihood estimation simply means that for near degenerate models, pseudo-likelihood estimates
may be particularly misleading. The primary problem in these cases is that the model is not
well-specified.
The bottom line is that various Monte Carlo estimation techniques are now available and,
wherever practicable, are to be preferred. These new approaches highlight certain inadequacies
in Markov random graph models when, for instance, transitivity effects are strong. If this does
happen for a given data set, researchers fitting Markov random graph models will notice that it
is impossible to obtain consistent parameter estimates with Monte Carlo maximum likelihood
estimation (technically, the estimation process does not converge). This means that the Markov
graph models are inappropriate for the data. It is for such reasons that Snijders et al. (in press)
introduced their new specifications for exponential random graph models, mentioned in Section
4.6, and discussed in other papers in this special edition.
6. A short example: a Markov random graph model for Medici business network
Other papers in this special edition provide examples of fitting exponential random graph
models to data, so here we present a very short example. We fit a Markov random graph model
for the well-known non-directed network of business connections among 16 Florentine families,
available in UCINET 5 (Borgatti et al., 1999). (For a fuller description of the context of the
data, see Padgett and Ansell, 1993.) The model includes edge, two-star, three-star and triangle
parameters as in Eq. (4). This model is not degenerate for this data set and parameter estimates
successfully converge. MCMCMLE parameter estimates are presented in Table 1. We see that the
density and triangle parameters are substantial in magnitude in comparison with their standard
errors.10 Interpretation is therefore relatively simple. The negative density parameter indicates
that edges occur relatively rarely, especially if they are not part of higher order structures such
as stars and triangles. The positive triangle parameter can be interpreted as providing evidence
that the business ties tend to occur in triangular structures and hence to cluster into clique-
like forms. The star effects are not significant, so perhaps do not merit interpretation. But the
parameter values suggest that there is a tendency for multiple network partners (the positive two-
star estimate) but with a ceiling on this tendency (the negative three-star parameter). So, while
there is tendency for network actors to have multiple partners, there are few actors with very many
partners.
10 The distribution of the statistic formed as the ratio of the estimate to its standard error is not known, but likely to
approximate a t distribution (Snijders, 2002). As a result, ratios exceeding two in absolute magnitude suggest non-zero
effects.
G. Robins et al. / Social Networks 29 (2007) 173–191 189
Table 1
Parameter estimates for Markov graph model: Florentine families business network (maximum likelihood estimates with
standard errors in brackets)
Parameter Configuration Estimate (standard error)
θ −4.27 (1.13)
σ2 1.09 (0.65)
σ3 −0.67 (0.41)
τ 1.32 (0.65)
7. Conclusion
This article provides an introductory exposition of the formulation and application of exponen-
tial random graph models for social networks. We have concentrated on presenting the underlying
logic and derivation of these models. Given the limitations of space, we have only given summary
attention to more recent developments which will be discussed in other papers in this special
edition.
Recent work on the Markov random graph models of Frank and Strauss (1986) shows that they
may be inadequate for many observed networks. In reviewing developments in these models to
this point, we have deliberately made no more than very summary comments on improved model
specification. The new specifications of Snijders et al. (in press) offer substantial improvement
in the practical use of exponential random graph models. They also indicate a way forward to
developing other innovative specifications. One of our aims in this paper has been to lay the
groundwork for an understanding of these new developments, which are given a fuller exposition
in other papers in this special edition.
Acknowledgements
We thank an anonymous reviewer for helpful comments in improving earlier versions of the
paper. This research was assisted by grants from the Australian Research Council.
References
Anderson, C.J., Wasserman, S., Crouch, B., 1999. A p* primer: logit models for social networks. Social Networks 21,
37–66.
Besag, J., 1974. Spatial interaction and the statistical analysis of lattice systems. Journal of the Royal Statistical Society,
Series B 36, 96–127.
Besag, J.E., 1975. Statistical analysis of non-lattice data. The Statistician 24, 179–195.
Borgatti, S., Everett, M., Freeman, L., 1999. UCINET 5 for Windows: Software for Social Network Analysis. Analytic
Technologies, Harvard, MA.
190 G. Robins et al. / Social Networks 29 (2007) 173–191
Contractor, N., Wasserman, S., Faust, K., 2006. Testing multi-theoretical multilevel hypotheses about organizational
networks: an analytic framework and empirical example. Academy of Management Journal 31, 681–703.
Erdös, P., Renyi, A., 1959. On random graphs. I. Publicationes Mathematicae (Debrecen) 6, 290–297.
Feld, S., 1981. The focused organization of social ties. American Journal of Sociology 86, 1015–1035.
Frank, O., Nowicki, K., 1993. Exploratory statistical analysis of networks. Annals of Discrete Mathematics 55, 349–
366.
Frank, O., Strauss, D., 1986. Markov graphs. Journal of the American Statistical Association 81, 832–842.
Handcock, M.S., 2003. Statistical models for social networks: degeneracy and inference. In: Breiger, R., Carley, K.,
Pattison, P. (Eds.), Dynamic Social Network Modeling and Analysis. National Academies Press, Washington, DC,
pp. 229–240.
Handcock, M., Hunter, D., Butts, C., Goodreau, S., Morris, M., 2006. Statnet: An R Package for the Statistical Analysis
and Simulation of Social Networks. Manual. University of Washington, http://www.csde.washington.edu/statnet.
Holland, P.W., Leinhardt, S., 1981. An exponential family of probability distributions for directed graphs (with discussion).
Journal of the American Statistical Association 76, 33–65.
Hunter, D., Handcock, M., 2006. Inference in curved exponential family models for networks. Journal of Computational
and Graphical Statistics 15, 565–583.
Lazega, E., Pattison, P.E., 1999. Multiplexity, generalized exchange and cooperation in organizations. Social Networks
21, 67–90.
Lazega, E., van Duijn, M., 1997. Position in formal structure, personal characteristics and choices of advisors in a law
firm: a logistic regression model for dyadic network data. Social Networks 19, 375–397.
Monge, P.R., Contractor, N.S., 2003. Theories of Communication Networks. Oxford University Press, NY.
Moreno, J.L., Jennings, H.H., 1938. Statistics of social configurations. Sociometry 1, 342–374.
Newman, M., 2003. The structure and function of complex networks. SIAM Review 45, 167–256.
Padgett, J.F., Ansell, C.K., 1993. Robust action and the rise of the Medici, 1400–1434. American Journal of Sociology
98, 1259–1319.
Pattison, P.E., Robins, G.L., 2002. Neighbourhood-based models for social networks. Sociological Methodology 32,
301–337.
Pattison, P.E., Robins, G.L., 2004. Building models for social space: neighborhood based models for social networks and
affiliation structures. Mathematiques des Science Humaines 168, 11–29.
Pattison, P.E., Wasserman, S., 1999. Logit models and logistic regressions for social networks. II. Multivariate relations.
British Journal of Mathematical and Statistical Psychology 52, 169–194.
Robins, G.L., Pattison, P.E., 2005. Interdependencies and social processes: generalized dependence structures. In: Car-
rington, P., Scott, J., Wasserman, S. (Eds.), Models and Methods in Social Network Analysis. Cambridge University
Press, New York, pp. 192–214.
Robins, G.L., Elliott, P., Pattison, P.E., 2001a. Network models for social selection processes. Social Networks 23,
1–30.
Robins, G.L., Pattison, P.E., Elliott, P., 2001b. Network models for social influence processes. Psychometrika 66, 161–190.
Robins, G.L., Pattison, P.E., Wasserman, S., 1999. Logit models and logistic regressions for social networks. III. Valued
relations. Psychometrika 64, 371–394.
Robins, G.L., Pattison, P.E., Woolcock, J., 2004. Models for social networks with missing data. Social Networks 26,
257–283.
Robins, G.L., Pattison, P.E., Woolcock, J., 2005. Social networks and small worlds. American Journal of Sociology 110,
894–936.
Schweinberger, M., Snijders, T.A.B., 2003. Settings in social networks: a measurement model. Sociological Methodology
33, 307–341.
Skvoretz, J., Faust, K., 1999. Logit models for affiliation networks. Sociological Methodology 29, 253–280.
Snijders, T.A.B., 2001. The statistical evaluation of social network dynamics. Sociological Methodology 31, 361–395.
Snijders, T.A.B., 2002. Markov chain Monte Carlo estimation of exponential random graph models. Journal of Social
Structure 3, 2.
Snijders, T.A.B., Pattison, P.E., Robins, G.L., Handock, M. New specifications for exponential random graph models.
Sociological Methodology, in press.
Strauss, D., 1986. On a general class of models for interaction. SIAM Review 28, 513–527.
Strauss, D., Ikeda, M., 1990. Pseudo-likelihood estimation for social networks. Journal of the American Statistical Asso-
ciation 85, 204–212.
Van Duijn, M.A.J., Snijders, T.A.B., Zijlstra, B.J.H., 2004. p2: a random effects model with covariates for directed graphs.
Statistica Neerlandica 58, 234–254.
G. Robins et al. / Social Networks 29 (2007) 173–191 191
Wasserman, S., Pattison, P.E., 1996. Logit models and logistic regressions for social networks. I. An introduction to
Markov graphs and p*. Psychometrika 61, 401–425.
Wasserman, S., Robins, G.L., 2005. An introduction to random graphs, dependence graphs, and p*. In: Carrington, P.,
Scott, J., Wasserman, S. (Eds.), Models and Methods in Social Network Analysis. Cambridge University Press, New
York, pp. 148–161.
Watts, D.J., 1999. Small Worlds: The Dynamics of Networks Between Order and Randomness. Princeton University
Press, Princeton, NJ.
An introduction to exponential random graph (p*) models for social networks
Why model social networks?
The logic behind p* models for social networks-an outline1
A general framework for model construction
Step 1: each network tie is regarded as a random variable
Step 2: a dependence hypothesis is proposed, defining contingencies among the network variables
Step 3: the dependence hypothesis implies a particular form to the model
Step 4: simplification of parameters through homogeneity or other constraints
Step 5: estimate and interpret model parameters
The general form of the exponential random graph model: dependence assumptions and parameter constraints
Constraints on parameters
Dependence assumptions and models
Bernoulli graphs: the simplest dependence assumption
Dyadic models: the dyadic independence assumption
Markov random graphs
Dependence structures with node-level variables
More complex dependence assumptions
New model specifications
Estimation
Pseudo-likelihood estimation: an approximate technique
Markov chain Monte Carlo maximum likelihood estimation (MCMCMLE)
A short example: a Markov random graph model for Medici business network
Conclusion
Acknowledgements
References