an introduction to exponential random graph models (ergm) johan koskinen the social statistics...
TRANSCRIPT
![Page 1: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/1.jpg)
An introduction to exponentialrandom graph models (ERGM)
Johan Koskinen
The Social Statistics Discipline Area, School of Social Sciences
Mitchell Centre for Network Analysis
Workshop: Monday, 29 August 2011
![Page 4: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/4.jpg)
Networks matter – ERGMS matter
6018 grade 6 children 1966 – 300 schools Stockholm
![Page 5: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/5.jpg)
Networks matter – ERGMS matter
6018 grade 6 children 1966 – 200 schools Stockholm
-10
-5
0
5
10
15
Density-4
-2
0
2
4
6
8
Mutuality-4
-3
-2
-1
0
1
2
Alt. in-stars-20
-15
-10
-5
0
5
10
Alt. out-stars
-1
0
1
2
3
Alt. trans trian-2
-1.5
-1
-0.5
0
0.5
1
1.5
Alt. indep 2-paths-2
0
2
4
6
Homophily girl-15
-10
-5
0
5
10
Main girl
CI
WLS
CIWLS
Koskinen and Stenberg (in press) JEBS
![Page 6: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/6.jpg)
Networks matter – ERGMS matter
6018 grade 6 children 1966 – 200 schools Stockholm
-10
-5
0
5
10
15
Density-4
-2
0
2
4
6
8
Mutuality-4
-3
-2
-1
0
1
2
Alt. in-stars-20
-15
-10
-5
0
5
10
Alt. out-stars
-1
0
1
2
3
Alt. trans trian-2
-1.5
-1
-0.5
0
0.5
1
1.5
Alt. indep 2-paths-2
0
2
4
6
Homophily girl-15
-10
-5
0
5
10
Main girl
CI
WLS
CIWLS
j
i
Koskinen and Stenberg (in press) JEBS
![Page 7: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/7.jpg)
Networks matter – ERGMS matter
6018 grade 6 children 1966 – 200 schools Stockholm
-10
-5
0
5
10
15
Density-4
-2
0
2
4
6
8
Mutuality-4
-3
-2
-1
0
1
2
Alt. in-stars-20
-15
-10
-5
0
5
10
Alt. out-stars
-1
0
1
2
3
Alt. trans trian-2
-1.5
-1
-0.5
0
0.5
1
1.5
Alt. indep 2-paths-2
0
2
4
6
Homophily girl-15
-10
-5
0
5
10
Main girl
CI
WLS
CIWLS
h1
j
i
h1
j
i
h2 h1
j
i
h2 h3
h1
j
i
h2
hk
h3
(a)
Koskinen and Stenberg (in press) JEBS
![Page 8: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/8.jpg)
Networks matter – ERGMS matter
6018 grade 6 children 1966 – 200 schools Stockholm
-10
-5
0
5
10
15
Density-4
-2
0
2
4
6
8
Mutuality-4
-3
-2
-1
0
1
2
Alt. in-stars-20
-15
-10
-5
0
5
10
Alt. out-stars
-1
0
1
2
3
Alt. trans trian-2
-1.5
-1
-0.5
0
0.5
1
1.5
Alt. indep 2-paths-2
0
2
4
6
Homophily girl-15
-10
-5
0
5
10
Main girl
CI
WLS
CIWLS
j i
i j
Koskinen and Stenberg (in press) JEBS
![Page 10: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/10.jpg)
Social Networks
• Notational preliminaries• Why and what is an ERGM• Dependencies• Estimation
Geyer-Thompson Robins-Monro Bayes (The issue, Moller et al, LISA, exchange algorithm)
• Interpretation of effects• Convergence and GOF• Further issues
![Page 11: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/11.jpg)
Networks matter – ERGMS matter
Numerous recent substantively driven studies• Gondal, The local and global structure of knowledge
production in an emergent research field, SOCNET 2011• Lomi and Palotti, Relational collaboration among spatial
multipoint competitors , SOCNET 2011• Wimmer & Lewis, Beyond and Below Racial Homophily,
AJS 2010• Lusher, Masculinity, educational achievement and social
status, GENDER & EDUCATION 2011• Rank et al. (2010). Structural logic of intra-organizational
networks, ORG SCI, 2010.
Book for applied researchers: Lusher, Koskinen, Robins ERGMs for SN, CUP, 2011
![Page 12: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/12.jpg)
The exponential random graph model (p*) framework
• An ERGM (p*) model is a statistical model for the ties in a network
• Independent (pairs of) ties (p1, Holland and Leinhardt, 1981; Fienberg and Wasserman, 1979, 1981)
• Markov graphs (Frank and Strauss, 1986)• Extensions (Pattison & Wasserman, 1999; Robins, Pattison
& Wasserman, 1999; Wasserman & Pattison, 1996)• New specifications (Snijders et al., 2006; Hunter &
Handcock, 2006)
![Page 14: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/14.jpg)
ERGMS – modelling graphs
• We want to model tie variables• But structure – overall pattern – is evident• What kind of structural elements can we
incorporate in the model for the tie variables?
![Page 15: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/15.jpg)
ERGMS – modelling graphs
If we believe that the frequency of interaction/density is an important aspect of the network
li
j k
Counts of the number of ties in our model
We should include
![Page 16: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/16.jpg)
ERGMS – modelling graphs
If we believe that the reciprocity is an important aspect of the (directed) network
li
j k
Counts of the number of mutual ties in our model
We should include
![Page 17: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/17.jpg)
ERGMS – modelling graphs
If we believe that an important aspect of the network is that
i
j
k
two edge indicators {i,j} and {i’,k} are
conditionally dependent if {i,j} {i’,k}
We should include counts of
degree distribution; preferential attachment,
etcfriends meet through friends; clustering; etc
![Page 18: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/18.jpg)
ERGMS – modelling graphs
If we believe that the attributes of the actors are important (selection effects, homophily, etc)
Heterophily/homophily
We should include counts of
Distance/similarity
![Page 19: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/19.jpg)
ERGMS – modelling graphs
If we believe that (Snijders, et al., 2006)
i
j
k
two edge indicators {i,k} and {l,j} are
conditionally dependent if {i,l} , {l,j} El
clustered regions
![Page 20: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/20.jpg)
ERGMS – modelling graphs
li
j k
m n
log
1 # +2 #
i kPr
i kPr
given the rest
given the rest
+ #+3 # +
li
j k
m n
li
j k
m n
li
k
i
j k
+ 2 =adding edge, e.g.:
![Page 21: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/21.jpg)
ERGMS – modelling graphs
logi kPr
i kPr
given the rest
given the rest
)()()()()Pr(log 2211 xzxzxzxX pp
The conditional formulation
)()()( 22
11 xxx p
ikpikik
May be ”aggregated” for all dyads so that the model for the entire adjacency matrix can be written
)()()( xzxzx rijrrik where Is the difference in counts of
structure type k
![Page 22: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/22.jpg)
ERGMS – modelling graphs
For a model with edges and triangles
The model for the adjacency matrix X is a weighted sum
The parameters 1 and weight the relative importance of
ties and triangles, respectively
- graphs with many triangles but not too dense are more
probable than dense graphs with few triangles
)()()()Pr(log 1 xTxLxX
where #)( xL #)( xT
![Page 23: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/23.jpg)
ERGMS – modelling graphs: example
Padgett’s Florentine families (Padgett and Ansell, 1993) network of 8 employees in a business company BusyNet <- as.matrix(read.table(
”PADGB.txt”,header=FALSE))
![Page 24: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/24.jpg)
ERGMS – modelling graphs: example
BusyNetNet <- network(BusyNet, directed=FALSE)plot(BusyNetNet)
Requires libraries
'sna’,'network’
![Page 25: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/25.jpg)
ERGMS – modelling graphs: example
Observed
15)( ji
ijxxL
and
1202
16
M
Density:
125.08
1
120
15)()(
M
xLxd
![Page 26: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/26.jpg)
ERGMS – modelling graphs: example
For a model with only edges
)()()Pr(log 11 xLxX
Equivalent with the model:
heads
tails
For each pair flip a p-coin
Where p is the probability coin comes up heads
![Page 27: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/27.jpg)
ERGMS – modelling graphs: example
i.e., an estimated one in 8 pairs establish a tie
The (Maximul likelihood) estimate of p is the density
here:
heads
tails
125.0)(ˆ xdpMLE
For an ERGM model with edges )()()Pr(log 11 xLxX
and hence )(
)(
1
1
1 xL
xL
e
ep
![Page 28: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/28.jpg)
ERGMS – modelling graphs: example
for 1, we have that the density parameter
Solving
heads
tails
)(
)(
1
1
1 xL
xL
e
ep
)1/1log(1 p
and for the MLE
)1ˆ/1log(ˆ ,1 MLEMLE p
9451)11/8log( .
125.0)(ˆ xdpMLE
Let’s check in stanet
![Page 29: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/29.jpg)
ERGMS – modelling graphs: example
Estim1 <- ergm(BusyNetNet ~ edges)summary(Estim1)
125.0)(ˆ xdpMLE
945.1ˆ ,1 MLE
MLE,1Parameter corresponding to #)( xL
approx. standard error of MLE of 1,
Success: )1ˆ/1log(ˆ ,1 MLEMLE p
![Page 30: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/30.jpg)
ERGMS – modelling graphs: example
Do we believe in the model:
heads
tails
For each pair flip a p-coin
Let’s fit a model that takes Markov dependencies into account
)()()()()()Pr(log 33221 xTxSxSxLxX
where #)( xL #)( xT#)(2 xS #)(3 xSstatnet
![Page 31: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/31.jpg)
ERGMS – modelling graphs: example
#)( xL #)( xT#)(2 xS #)(3 xS
MLE
approx. standard error of MLE of ,
Estim2 <- ergm(BusyNetNet ~ kstar(1:3) + triangles)summary(Estim2)
![Page 33: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/33.jpg)
Likelihood equations for exponential fam
)()()()()Pr(log 2211 xzxzxzxX pp
”Aggregated” to a joint model for entire adjacency matrix X
)()}({ˆ obsxzXzEMLE
Sum over all 2n(n-1)/2 graphs
The MLE solves the equation (cf. Lehmann, 1983):
![Page 34: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/34.jpg)
Likelihood equations for exponential fam
Solving )()}({ˆ obsxzXzEMLE
• Using the cumulant generating function (Corander, Dahmström, and Dahmström, 1998)
• Stochastic approximation (Snijders, 2002, based on Robbins-Monro, 1951)
• Importance sampling (Handcock, 2003; Hunter and Handcock, 2006, based on Geyer-Thompson 1992)
![Page 35: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/35.jpg)
)}()({ obs)(1
0)()1(
)( xzxzDa mr
mmm
Robbins-Monro algorithm
Solving )()}({ˆ obsxzXzEMLE
Snijders, 2002, algorithm
- Initialisation phase
- Main estimation
- convergence check and cal. of standard errors
MAIN:
Draw using MCMC
![Page 36: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/36.jpg)
Robbins-Monro algorithm
Parameter space Network space
We can simulate the distribution of graphs associated with any given set of parameters.
Observed data
A set of parameter values
In this case, the distribution of graphs from the parameter values is not consistent with the observed data
![Page 37: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/37.jpg)
Robbins-Monro algorithm
Parameter space Network space
We want to find the parameter estimates for which the observed graph is central in the distribution of graphs
![Page 38: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/38.jpg)
Robbins-Monro algorithm
Parameter space Network space
So: 1. compare distribution of graphs with observed data2. adjust parameter values slightly to get the distribution closer to the observed data3. keep doing this until the estimates converge to the right point
Observed data
Based on this discrepancy, adjust parameter values
![Page 39: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/39.jpg)
Robbins-Monro algorithm
Over many iterations we walk through the parameter space until we find the right estimates
Parameter space Network space
![Page 40: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/40.jpg)
Robbins-Monro algorithm
Phase 1, Initialisation phase
Find good values of the initial parameter state
And the scaling matrix
(use the score-based method, Schweinberger & Snijders, 2006)
![Page 41: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/41.jpg)
Robbins-Monro algorithm
Phase 2, Main estimation phase
Iteratively update θ
)}()({ obs)(1
0)()1(
)( xzxzDa mr
mmm
by drawing one realisation
from the model defined by the current θ(m)
Repeated in sub-phases with fixed
![Page 42: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/42.jpg)
200 400 600 800 1000 1200 1400-1.8
-1.6
-1.4
-1.2
-1ed
ge p
aram
eter
1(m
)
iterations
200 400 600 800 1000 1200 1400-0.05
0
0.05
0.1
0.15
0.2
alte
rnat
ing
tria
ngle
par
amet
er 2(m
)
iterations
200 400 600 800 1000 1200 1400-40
-30
-20
-10
0
10
20
z 1( x
(m) )
- z
1( x
obs )
iterations
200 400 600 800 1000 1200 1400
-20
0
20
40
z 2( x
(m) )
- z
2( x
obs )
iterations
![Page 43: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/43.jpg)
Robbins-Monro algorithm
Phase 2, Main estimation phase
Relies on us being able to draw one realisation
We can NOT do this directly
from the ERGM defined by the current θ
We have to simulate xMore specifically use Markov chain Monte Carlo
![Page 44: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/44.jpg)
Robbins-Monro algorithm
What do we need to know about MCMC?
Method:
Generate a sequence of graphs
for arbitrary x(0), using an updating rule… so that
as
![Page 45: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/45.jpg)
Robbins-Monro algorithm
What do we need to know about MCMC?
So if we generate an infinite number of graphs in the “right” way we have the ONE draw we need to update θ once?
Typically we can’t wait an infinite amount of time so we settle for
In Pnet very large ismultiplication factor
![Page 46: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/46.jpg)
Robbins-Monro algorithm
Phase 3, Convergence check and calculating standard errors
At the end of phase 2 we always get a value
But is it the MLE?
Does it satisfy
?
![Page 47: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/47.jpg)
Robbins-Monro algorithm
Phase 3, Convergence check and calculating standard errors
Phase 3 simulates a large number of graphs
And checks if
A minor discrepancy - due to numerical inaccuracy - is acceptable
Convergence statistics:
![Page 48: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/48.jpg)
)()()( obs1
)()(1)1()1()( xzxzwIM
m
mmggg
Geyer-Thompson
Solving )()}({ˆ obsxzXzEMLE
Handcock, 2003, approximate Fisher scoring
MAIN:
Approximated using importance sample from MCMC
![Page 49: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/49.jpg)
Bayes: dealing with likelihood
The normalising constant of the posterior not essential for Bayesian inference, all we need is:
y
p
k kk
p
k kk
yz
xzx
1
1
)}(exp{
)}(exp{);(
)();(d)();(
)();()|(
x
x
xx
… but
Sum over all 2n(n-1)/2 graphs
![Page 50: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/50.jpg)
Bayes: MCMC?
Consequently, in e.g. Metropolis-Hastings, acceptance probability of move to θ
y
p
k kk
y
p
k kk
yz
yz
1
*
1
)}(exp{
)}(exp{
)|(
)|(
)();(
*)();(min
)|(
)|(
)|(
)|(,1min
*
**
*
**
prop
prop
prop
prop
q
q
x
x
q
q
x
x
… which contains
![Page 51: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/51.jpg)
Bayes: Linked Importance Sampler Auxiliary Variable MCMC
LISA (Koskinen, 2008; Koskinen, Robins & Pattison, 2010): Based on
Møller et al. (2006), we define an auxiliary variable
And produce draws from the joint posterior
m
j
K KK1
,,1,,1 X
)(})(exp{
)(
})(exp{
})(exp{)|,( ,
yz
P
yz
xzx
kk
B
kk
obskkobs
using the proposal distributions
),(~| )()(* tt N and )}(exp{
)(~|
*
*
,** *
yz
P
kk
F
![Page 52: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/52.jpg)
Bayes: alternative auxiliary variable
LISA (Koskinen, 2008; Koskinen, Robins & Pattison, 2010): Based on
Møller et al. (2006), we define an auxiliary variable
Improvement: use exchange algorithm (Murray et al. 2006)
m
j
K KK1
,,1,,1 X
Many linked chains:- Computation time- storage (memory and time issues)
),(~| )()(* tt N and )ERGM(~| *** x
Accept θ* with log-probability: ))}(*)((*)(,0min{ obsT xzxz
Caimo & Friel, 2011
![Page 53: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/53.jpg)
Bayes: Implications of using alternative auxiliary variable
Improvement: use exchange algorithm (Murray et al. 2006)
),(~| )()(* tt N and )ERGM(~| *** x
Accept θ* with log-probability: ))}(*)((*)(,0min{ obsT xzxz
Caimo & Friel, 2011
• Storing only parameters• No pre tuning – no need for good initial values• Standard MCMC properties of sampler• Less sensitive to near degeneracy in estimation• Easier than anything else to implement
QUICK and ROBUST
![Page 55: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/55.jpg)
Problem with Markov models
Markov dependence assumption:
i
j
k
two edge indicators {i,j} and {i’,k} are
conditionally dependent if {i,j} {i’,k}
We have shown that the only effects are:
degree distribution; preferential attachment,
etcfriends meet through friends; clustering; etc
![Page 56: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/56.jpg)
Problem with Markov models
Often for Markov model
degree distribution; preferential attachment,
etcfriends meet through friends; clustering; etc
)()}({ˆ obsxzXzEMLE
Matching
hardImpossible!or
![Page 57: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/57.jpg)
Problem with Markov models
)()}({ˆ obsxzXzEMLE
Matching
hardImpossible!or
# triangles
# edges
or Some statistic k
![Page 58: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/58.jpg)
0 50 100 150 200 250 300 350 400number of triangles
Problem with Markov models
)()}({ˆ obsxzXzEMLE
Matching
hard
# triangles
The number of triangles for a Markov model with edges (-3), 2-stars (.5), 3-stars (-.2)
and triangles (1.1524)
![Page 59: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/59.jpg)
0 0.5 1 1.550
60
70
80
90
100
110
120
130
140
num
ber
of e
dges
triangle parameter0 0.5 1 1.5
0
50
100
150
200
250
300
350
400
num
ber
of t
riang
les
triangle parameter
Problem with Markov models
)()}({ˆ obsxzXzEMLE
Matching
hard
Exp. # triangles and edges for a Markov
model with edges, 2-stars, 3-stars and
triangles
![Page 60: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/60.jpg)
Problem with Markov models
)()}({ˆ obsxzXzEMLE
Matching Impossible!
# 2-stars # edges
If for some statistic k
Similarly for max (or conditional min/max)e.g. A graph on 7 vertices with 5 edges
Implies:
![Page 61: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/61.jpg)
Problem with Markov models
First solutions (Snijders et al., 2006)
Markov:
adding k-star adds (k-1) stars
alternating sign compensates but eventually zk(x)=0
Alternating stars:
Restriction on star parameters - Alternating sign
prevents explosion, and
models degree distribution
![Page 62: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/62.jpg)
Problem with Markov models
First solutions (Snijders et al., 2006)
Markov:
adding k-star adds (k-1) stars
alternating sign compensates but eventually zk(x)=0
#)( xL #)( xT#)(2 xS #)(3 xS
![Page 63: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/63.jpg)
Problem with Markov models
First solutions (Snijders et al., 2006)
Alternating stars:
Restriction on star parameters - Alternating sign
prevents explosion, and
models degree distribution
Include all stars but restrict parameter:
/23 /34 /45
);()()()( 113322 xAKSxSxSxS AKSnn new
![Page 64: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/64.jpg)
Problem with Markov models
First solutions (Snijders et al., 2006)
Include all stars but restrict parameters:
/23 /34 /45
);()()()( 113322 xAKSxSxSxS AKSnn new
Expressed in termsof degree distribution
![Page 65: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/65.jpg)
Problem with Markov models
First solutions (Snijders et al., 2006)
Include all stars but restrict parameters:
/23 /34 /45
);()()()( 113322 xAKSxSxSxS AKSnn
Interpretation:Positive parameter (λ ≥ 1) – graphs with some high degree nodes and larger degree variance more likely than graphs with more homogenous degree distribution Negative parameter (λ ≥ 1) – the converse…
![Page 66: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/66.jpg)
Problem with Markov models
Markov:
triangles evenly spread out
but one edge can add many triangles…
Alternating triangles:
• Restrictions on different order triangles – alternating sign
• Prevents explosion, and
• Models multiply clustered regions
• Social circuit dependence assumption
First solutions (Snijders et al., 2006)
![Page 67: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/67.jpg)
Problem with Markov models
Markov:
triangles evenly spread out
but one edge can add many triangles…
First solutions (Snijders et al., 2006)
k-triangles:
1-triangles: 2-triangles: 3-triangles:
![Page 68: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/68.jpg)
Problem with Markov models
Alternating triangles:
• Restrictions on different order triangles – alternating sign
• Prevents explosion, and
• Models multiply clustered regions
First solutions
/1 kk
);()()()( 223322 xAKTxTxTxT AKTnn
Weigh the k-triangles:
Where:
![Page 69: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/69.jpg)
Problem with Markov models
Alternating triangles:
• Restrictions on different order triangles – alternating sign
• Prevents explosion, and
• Models multiply clustered regions
First solutions
Underlying assumption:
i
j
k
two edge indicators {i,k} and {l,j} are
conditionally dependent if {i,l} , {l,j} E
l
Social circuit dependence assumption
![Page 70: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/70.jpg)
Problem with Markov models
Alternative interpretation
We may (geometrically) weight together :
i j
![Page 71: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/71.jpg)
Problem with Markov models
Alternative intrepretation
We may also define theEdgewise Shared Partner Statistic:
i j… and we can weigh together the ESP statisticsusing Geometrically decreasing weights: GWESP
![Page 73: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/73.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
Collaboration network among 36 lawyers in a
New England law firm (Lazega, 2001)
Boston office:
Hartford office:
Providence off.:
least senior:
most senior:
![Page 74: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/74.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
)()()()()Pr(log 2211 xzxzxzxX pp
Fit a model with ”new specifications” and covariates
ijx
)( jiij aax )( jiij bbx
)( jiij bbx 1
)( jiij ccx 1
)( jiij ddx 1
323
12
1
)()1(
)()(3
n
nn xtxtxt
Edges:
Seniority:
Practice:
Homophily
Sex:
Office:
GWESP:
Practice:
Main effect
![Page 75: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/75.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
Fit a model with ”new specifications” and covariates
PNet:
![Page 76: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/76.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
Fit a model with ”new specifications” and covariates
ijx
)( jiij aax )( jiij bbx
)( jiij bbx 1
)( jiij ccx 1
)( jiij ddx 1
323
12
1
)()1(
)()(3
n
nn xtxtxt
Edges:
Seniority:
Practice:
Homophily
Sex:
Office:
GWESP:
Practice:
Main effect
![Page 78: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/78.jpg)
Bayesian Data AugmentationFitting an ERGM in Pnet: a business communications network
office sales
officeRb for attribute
R for attribute
sales
For ”wwbusiness.txt” we have recorded wheather the employee works in the central office or is a traveling sales represenative
![Page 79: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/79.jpg)
Bayesian Data AugmentationFitting an ERGM in Pnet: a business communications network
office sales
officeRb for attribute
R for attribute
sales
Consider a dyad-independent model
)()()Pr(log 1 jiijRbjiijRijijij OFFOFFxOFFOFFxxxX
![Page 80: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/80.jpg)
Bayesian Data AugmentationFitting an ERGM in Pnet: a business communications network
office sales
officeRb for attribute
R for attribute
sales
With log odds
jiRbjiRij
ij OFFOFFOFFOFFX
X
)()0Pr(
)1Pr(log 1
1
R 1
1jOFF 0jOFF
1iOFF
0iOFF
)0Pr(
)1Pr(log
ij
ij
X
X
RbR 21
R 1
![Page 82: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/82.jpg)
Unpacking the alternating star effect
Alternating stars a way of- “fixing” the Markov problems (models all degrees)- Controlling for paths in clustering
);()()()( 113322 xAKSxSxSxS AKSnn
k-triangles measure clustering…
1-triangles: 2-triangles: 3-triangles:
![Page 83: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/83.jpg)
Unpacking the alternating star effect
Alternating stars a way of- “fixing” the Markov problems (models all degrees)- Controlling for paths in clustering
);()()()( 113322 xAKSxSxSxS AKSnn
Is it closure or an artefact of many stars/2-paths?
1-triangles: 2-triangles: 3-triangles:
![Page 84: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/84.jpg)
Unpacking the alternating star effect
Interpreting the alternating star parameter:
centralization – +
variance of degree measure centralization
![Page 85: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/85.jpg)
Unpacking the alternating star effect
Interpreting the alternating star parameter:
centralization – +
alternating star parameter – +
![Page 86: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/86.jpg)
Unpacking the alternating star effect
Statistics for graph (n = 16); fixed density; alt trian: 1.17
alternating star parameter – +
![Page 87: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/87.jpg)
Unpacking the alternating star effect
Statistics for graph (n = 16); fixed density; alt trian: 1.17
alternating star parameter – +
![Page 88: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/88.jpg)
Unpacking the alternating star effect
Graphs (n = 16); fixed density; alt trian: 1.17
alternating star parameter – +
![Page 89: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/89.jpg)
Problem with Markov models
Alt, stars in termsof degree distribution
Note also the influence of isolates:
![Page 90: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/90.jpg)
Problem with Markov models
Note also the influence of isolates:
alternating star parameter – +
![Page 91: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/91.jpg)
Problem with Markov models
Note also the influence of isolates:
This is because we impose a particular shape on the degree distribution
);()()()( 113322 xAKSxSxSxS AKSnn
![Page 92: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/92.jpg)
Problem with Markov models
The degree distribution for different values of:al
tern
atin
g st
ar p
aram
eter
–
+
![Page 93: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/93.jpg)
Unpacking the alternating triangle effect
Alternating triangles measure clustering
We may also define theEdgewise Shared Partner Statistic:
i j… and we can weigh together the ESP statisticsusing Geometrically decreasing weights: GWESP
![Page 94: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/94.jpg)
Unpacking the alternating triangle effect
Statistics for graph (n = 16); fixed density; alt star: 2.37
alternating Triangle parameter – +
![Page 95: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/95.jpg)
Unpacking the alternating triangle effect
Statistics for graph (n = 16); fixed density; alt star: 2.37
alternating Triangle parameter – +
![Page 96: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/96.jpg)
Unpacking the alternating triangle effect
Statistics for graph (n = 16); fixed density; alt star: 2.37
alternating Triangle parameter – +
clus
terin
g
+ closed
open
![Page 97: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/97.jpg)
Unpacking the alternating triangle effect
Statistics for graph (n = 16); fixed density; alt star: 2.37
alternating Triangle parameter – +
clus
terin
g
+ closed
open
![Page 98: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/98.jpg)
Unpacking the alternating triangle effect
Alternating triangles model multiply clustered areas
For multiply clustered areas triangles stick togetherWe model how many others tied actors have
Edgewise Shared Partner Statistic:
i j
![Page 99: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/99.jpg)
Unpacking the alternating triangle effect
Statistics for graph (n = 16); fixed density; alt star: 2.37
# edwise shared partners – +
Alte
rnat
ing
tria
ngle
par
a
+
![Page 101: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/101.jpg)
Revisiting the Florentine families
#)( xL #)( xT#)(2 xS #)(3 xS
PNet
statnet
Why difference?
![Page 102: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/102.jpg)
Revisiting the Florentine families
#)( xL #)( xT#)(2 xS #)(3 xS
Pnet checks convergence
in 3rd phase
)()}({ˆ obsxzXzEMLE
![Page 103: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/103.jpg)
Revisiting the Florentine families
#)( xL #)( xT#)(2 xS #)(3 xS
Lets check
For statnet
)()}({ˆ obsxzXzEMLE
![Page 104: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/104.jpg)
#)( xL
#)( xT
#)(2 xS
#)(3 xS
Estim3 <- ergm(BusyNetNet ~ kstar(1:3) + triangles , verbose=TRUE)
mcmc.diagnostics(Estim3)
)()}({ˆ obsxzXzEMLE
?
![Page 105: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/105.jpg)
Simulation, GOF, and Problems
Given a specific model we can simulate potential outcomes under that model. This is used for
• Estimation: (a) to match observed statistics and expected statistics(b) to check that we have “the solution”
• GOF: to check whether the model can replicated features of the data that we not explicitly modelled
• Investigate behaviour of model, e.g.: degeneracy and dependence on scale
![Page 106: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/106.jpg)
Simulation, GOF, and Problems
Standard goodness of fit procedures are not valid for ERGMs – no F-tests or Chi-square tests available
If indeed the fitted model adequately describes the data generating process, then the graphs that the model produces should be similar to observed data
For fitted effects this is true by construction
For effects/structural feature that are not fitted this may not be true
If it is true, the modelled effects are the only effects necessary to produce data – a “proof” of the concept or ERGMs
![Page 107: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/107.jpg)
Simulation, GOF, and Problems
Example: for our fitted model for Lazega
For arbitrary function g
We can look at how well the model reproduces
![Page 108: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/108.jpg)
Simulation, GOF, and Problems
From the goodness-of-fit tab in Pnet we get
![Page 109: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/109.jpg)
Simulation, GOF, and Problems
Effect MLE s.e.
Density -6.501 0.727
Main effect seniority 1.594 0.324
Main effect practice 0.902 0.163
Homophily effect practice 0.879 0.231
Homophily sex 1.129 0.349
Homophily office 1.654 0.254
![Page 110: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/110.jpg)
Simulation, GOF, and Problems
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
0
2
4
6
8
10
12
freq
uenc
y
degree1 2 3 4 5 6 7 8 9 10 11
0
10
20
30
40
50
freq
uenc
y
number of shared partners
![Page 111: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/111.jpg)
Simulation, GOF, and Problems
Effect MLE s.e. MLE s.e.
Density -6.501 0.727 -6.510 0.637
Main effect seniority 1.594 0.324 0.855 0.235
Main effect practice 0.902 0.163 0.410 0.118
Homophily effect practice 0.879 0.231 0.759 0.194
Homophily sex 1.129 0.349 0.704 0.254
Homophily office 1.654 0.254 1.146 0.195
GWEPS 0.897 0.304
Log-lambda 0.778 0.215
![Page 112: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/112.jpg)
Simulation, GOF, and Problems
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
0
2
4
6
8
10
12
freq
uenc
y
degree1 2 3 4 5 6 7 8 9 10 11
0
10
20
30
40
50
freq
uenc
y
number of shared partners
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
0
2
4
6
8
10
freq
uenc
y
degree1 2 3 4 5 6 7 8 9 10 11
0
10
20
30
40
freq
uenc
y
number of shared partners
![Page 114: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/114.jpg)
Independence - Deriving the ERGM
li
j k
m n
heads
tails
li
li
heads
tails
i
k
i
k
![Page 115: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/115.jpg)
Independence - Deriving the ERGM
0.25 0.25
0.25 0.25
AUD
0.5 0.5
SEK
0.5
0.5 li
k
li
k
Knowledge of AUD, e.g. does not help us predict SEK
e.g. whether or
li
j k
m n
![Page 116: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/116.jpg)
Independence - Deriving the ERGM
i
i
k
Knowledge of AUD, e.g. does not help us predict SEK
e.g. whether or
even though dyad {i,l} li
and dyad {i,k}
have vertex i
in common
![Page 117: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/117.jpg)
Independence - Deriving the ERGM
0.4 0.1
0.1 0.4
AUD
0.5 0.5
SEK
0.5
0.5
li
k
li
k
li
j k
m n
May we find model such that knowledge of AUD, e.g. does help us predict SEK
e.g. whether or ?
![Page 118: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/118.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
Consider the tie-variables that have Mary in common
How may we make these “dependent”?
![Page 119: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/119.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
petemary
![Page 120: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/120.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
johnpetemary mary
![Page 121: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/121.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
johnpetemary
paulmary
mary
![Page 122: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/122.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
johnpetemary
paulmary
mary
pete john
![Page 123: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/123.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
johnpetemary
paulmary
mary
paul john
pete john
![Page 124: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/124.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
johnpetemary
paulmary
mary
paul pete paul john
pete john
![Page 125: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/125.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
The “probability structure” of a Markov graph is described by cliques of the dependence graph (Hammersley-Clifford)….
![Page 126: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/126.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
pete
mary
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 127: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/127.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paul
m,pe
![Page 128: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/128.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
johnpete
mary
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 129: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/129.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 130: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/130.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paul
m,pe m,j
![Page 131: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/131.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 132: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/132.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 133: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/133.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paul
m,pe
pe,j
m,j
![Page 134: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/134.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 135: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/135.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 136: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/136.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,j
m,pe
pe,j
![Page 137: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/137.jpg)
Deriving the ERGM: From Markov graph to Dependence graph
mary
johnpete
paulm,pa
pa,pe pa,j
m,pe
pe,j
m,j
![Page 138: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/138.jpg)
From Markov graph to Dependence graph – distinct subgraphs?
too ma
ny sta
tistic
s (par
ameter
s)
![Page 140: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/140.jpg)
A log-linear model (ERGM) for ties
)()()()()Pr(log 2211 xzxzxzxX pp
”Aggregated” to a joint model for entire adjacency matrix
Interaction terms in log-linear model of types
ijX ikij XX jkikij XXX
![Page 141: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/141.jpg)
A log-linear model (ERGM) for ties
By definition of (in-) dependence
)Pr()Pr(),Pr( ikikijijikikijij xXxXxXxX
E.g. and co-occuring i
j
i
j k
i
k
Main effects interaction term
ijX ikX ikij XX
More than is explained by margins
![Page 143: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/143.jpg)
The steps of fitting an ERGM
• fit base-line model• check convergence• rerun if model not converged• include more parameters? GOF• candidate models
![Page 145: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/145.jpg)
Bi-partite networks: cocitation (Small 1973)
1
2 3
41
2
41
2 3
41
2 3
41
2
5
3
41
27
6
cite
texts
![Page 146: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/146.jpg)
Bi-partite networks: cooffending (. Sarnecki, 2001)
offenders
participating
offences
![Page 147: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/147.jpg)
Bi-partite networks
people participating in
belonging to
Social events(Breiger, 1974)
voluntary organisations(Bonachich, 1978)
people
directors sitting oncorporate boards
(eg. Mizruchi, 1982
![Page 148: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/148.jpg)
One-mode projection
Tie: If two directors share a board
Two-mode
one-mode
![Page 149: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/149.jpg)
Bi-partite networks
1
2 3
41
2
41
2 3
41
2 3
41
2
5
3
41
27
6
cite
texts
![Page 150: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/150.jpg)
Bi-partite networks
Researchers
cite
texts
![Page 151: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/151.jpg)
Bi-partite networks
Researchers
cite
texts
![Page 152: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/152.jpg)
One-mode projection
What mode given priority? (Duality of social actors and social groups; e.g. Breiger 1974; Breiger and Pattison, 1986)Loss of information
![Page 153: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/153.jpg)
ERGM for bipartite networks
)()()()()Pr(log 2211 xzxzxzxX pp
The model is the same as for one-mode networks (Wang et al., 2007)
i j
i
k
j
ℓ
i
k
j
ℓ
i
k ℓ k
j
ℓ
Edges ”people” 2-stars
”affiliation” 2-stars
3-paths 4-cycles
![Page 154: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/154.jpg)
ERGM for bipartite networks
i
k
j
ℓ
4-cycles
A form of bipartite clustering
Also known as Bi-cliques
One possible interpretation:
![Page 155: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/155.jpg)
ERGM for bipartite networks
i
k
j
ℓ
4-cycles
A form of bipartite clustering
Also known as Bi-cliques
One possible interpretation:Who to appoint?
How about x?
x
![Page 156: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/156.jpg)
ERGM for bipartite networks
Fitting the model in (B)Pnet straightforward extension
These statistics are all Markov:
i j
i
k
j
ℓ
i
k
j
ℓ
i
k ℓ k
j
ℓ
Edges ”people” 2-stars
”affiliation” 2-stars
3-paths 4-cycles
<BPNet>
![Page 158: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/158.jpg)
Methods for missing network data
Effects of missingness
PerilsSome investigations on the effects on indices of structural properties (Kossinets, 2006; Costenbader & Valente, 2003; Huisman, 2007)Problems with the “boundary specification issue”Few remediesDeterministically “complement” data (Stork & Richards, 1992)Stochastically Impute missing data (Huisman, 2007)Ad-hoc “likelihood” (score) for missing ties (Liben-Nowell and Kleinberg, 2007)
![Page 159: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/159.jpg)
Model assisted treatment of missing network data
missing data
observed data
If you don’t have a model for what you have observed
How are you going to be able to say something about what you have not observed using what you have observed
![Page 160: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/160.jpg)
Model assisted treatment of missing network data
• Importance sampling (Handcock & Gile 2010; Koskinen, Robins & Pattison, 2010)
• Stochastic approximation and the missing data principle (Orchard & Woodbury,1972) (Koskinen & Snijders, forthcoming)
• Bayesian data augmentation (Koskinen, Robins & Pattison, 2010)
![Page 161: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/161.jpg)
Subgraph of ERGM not ERGM
i
j
k
Dependence in ERGM We may also have dependence
i
j
lk
But if
k
?
j
We should include
counts of:
Marginalisation (Snijders, 2010; Koskinen et al, 2010)
![Page 162: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/162.jpg)
Bayesian Data Augmentation
With missing data:
Simulate parameters
In each iteration simulate graphs
missing
Bayesian Data Augmentation
![Page 163: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/163.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
Most likely missing given current
Bayesian Data Augmentation
![Page 164: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/164.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
Most likely given current missing
Bayesian Data Augmentation
![Page 165: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/165.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
Most likely missing given current
Bayesian Data Augmentation
![Page 166: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/166.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
Most likely given current missing
Bayesian Data Augmentation
![Page 167: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/167.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
Most likely missing given current
Bayesian Data Augmentation
![Page 168: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/168.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
and so on…
Bayesian Data Augmentation
![Page 169: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/169.jpg)
Bayesian Data Augmentation
Simulate parameters
With missing data:
In each iteration simulate graphs
missing
… until
Bayesian Data Augmentation
![Page 170: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/170.jpg)
Bayesian Data Augmentation
What does it give us?
Distribution of parameters
Distribution of missing data
Subtle point
Missing data does not depend on the parameters (we don’t have to choose parameters to simulate missing)
missing
Bayesian Data Augmentation
![Page 171: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/171.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
Collaboration network among 36 lawyers in a
New England law firm (Lazega, 2001)
Boston office:
Hartford office:
Providence off.:
least senior:
most senior:
![Page 172: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/172.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers
172
ijx
)( jiij aax )( jiij bbx
)( jiij bbx 1
)( jiij ccx 1
)( jiij ddx 1
323
12
1
)()1(
)()(3
n
nn xtxtxt
Edges:
Seniority:
Practice:
Homophily
Sex:
Office:
GWESP:
with 8 = log()
Practice:
Main effect
t1 :
t2 :
etc.
(bi = 1, if i corporate,0 litigation)
t3 :
![Page 173: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/173.jpg)
Bayesian Data AugmentationLazega’s (2001) Lawyers – ERGM posteriors (Koskinen, 2008)
![Page 174: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/174.jpg)
Bayesian Data AugmentationCross validation (Koskinen, Robins & Pattison, 2010)
Remove 200 of the 630 dyads at random
Fit inhomogeneous Bernoulli model obtain the posterior predictive tie-probabilities for the missing tie-variables
Fit ERGM and obtain the posterior predictive tie-probabilities for the missing tie-variables (Koskinen et al., in press)
Fit Hoff’s (2008) latent variable probit model with linear predictor Tz(xij) + wiwj
T
Repeat many times
![Page 175: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/175.jpg)
Bayesian Data AugmentationROC curve for predictive probabilities combined over 20 replications (Koskinen et al. 2010)
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
False Positive Rate
Tru
e P
ositi
ve R
ate
![Page 176: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/176.jpg)
Bayesian Data AugmentationROC curve for predictive probabilities combined over 20 replications (Koskinen et al. 2010)
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
False Positive Rate
Tru
e P
ositi
ve R
ate
![Page 177: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/177.jpg)
Bayesian Data AugmentationROC curve for predictive probabilities combined over 20 replications (Koskinen et al. 2010)
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
False Positive Rate
Tru
e P
ositi
ve R
ate
![Page 178: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/178.jpg)
Bayesian Data AugmentationSnowball sampling
• Snowball sampling design ignorable for ERGM (Thompson and Frank, 2000, Handcock & Gile 2010; Koskinen, Robins & Pattison, 2010)
• ... but snowball sampling rarely used when population size is known...
• Using the Sageman (2004) “clandestine” network as test-bed for unknown N
![Page 180: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/180.jpg)
Spatial embedding – Daraganova et al. 2011 SOCNET
306 actors in Victoria, Australia
![Page 181: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/181.jpg)
Spatial embedding – Daraganova et al. 2011 SOCNET
306 actors in Victoria, Australia
... all living within 14 kilometres of each other
spatially embedded
![Page 182: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/182.jpg)
Spatial embedding – Daraganova et al. 2011 SOCNET
306 actors in Victoria, Australia
... all living within 14 kilometres of each other
Bernoulli conditional on distance
Empirical probability
![Page 183: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/183.jpg)
Spatial embedding – Daraganova et al. 2011 SOCNET
Edges -4.87* (0.13) 1.56* (0.65) -4.79* (0.66) -0.20 (0.87)
Alt. star -0.86* (0.18) -0.86* (0.2)
Alt. triangel 2.74* (0.15) 2.69* (0.14)
Log distance -0.78* (0.08) -0.56* (0.07)
Age
heterophily-0.07* (0.01) -0.07* (0.01) 0.001 (0.07) 0.002 (0.06)
Gender
homophily-1.13* (0.61) -1.13 (0.69) 0.09 (0.83) 0.07 (0.47)
ERGM: distance and endogenous dependence explain different things
![Page 185: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/185.jpg)
Other extensions
There are ERGMs (ERGM-like modesl) for- directed data- valued data- bipartite data- multiplex data- longitudinal data- modelling actor autoregressive attributes
![Page 186: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/186.jpg)
Further issues – relaxing homogeneity assumption
ERGMs typically assume homogeneity
=
=
=
=
(A)Block modelling and ERGM (Koskinen, 2009)(B) Latent class ERGM (Schweingberger & Handcock)
![Page 187: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/187.jpg)
Further issues – model selection
Assessing Goodness of Fit:- Posterior predictive distributions (Koskinen,
2008; Koskinen, Robins & Pattison, 2010; Caimo & Friel, 2011)
- Non-Bayesian heuristic GOF (Robins et al., 2007; Hunter et al., 2008; Robins et al., 2009; Wang et al., 2009)
Model selection- Path sampling for AIC (Hunter & Handcock,
2006); Conceptual caveat: model complexity when variables dependent?
- Bayes factors (Wang & Handcock…?)
![Page 188: An introduction to exponential random graph models (ERGM) Johan Koskinen The Social Statistics Discipline Area, School of Social Sciences Mitchell Centre](https://reader038.vdocuments.site/reader038/viewer/2022102805/551ae0d15503466b6a8b5bec/html5/thumbnails/188.jpg)
Wrap-up
ERGMs- Increasingly being used- Increasingly being understood- Increasingly being able to handle imperfect data (also missing link prediction)
Methods- Plenty of open issues- Bayes is the way of the futureLegitimacy and dissemination- e.g. Lusher, Koskinen, Robins ERGMs for SN, CUP, 2011