08 Inference for Networks – DYAD Model Overview (2017)
1. Modeling networks: regression with additive and
multiplicative effects
Alexander Volfovsky
Department of Statistical Science, Duke
May 25 2017
May 25, 2017
Health Networks
3. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
1
4. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
1
5. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
1
6. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
What models work when the assumptions fail?
1
7. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
What models work when the assumptions fail?
How to develop fail-safes to overcome these problems?
1
8. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
What models work when the assumptions fail?
How to develop fail-safes to overcome these problems?
Where to apply these?
1
9. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
What models work when the assumptions fail?
How to develop fail-safes to overcome these problems?
Where to apply these?
Causal inference
1
10. Why model networks?
Interested in understanding the formation of relationships
Applied fields: sociology, economics, biology, epidemiology
Fundamental theory questions:
What assumptions are made for different network models?
What models work when the assumptions fail?
How to develop fail-safes to overcome these problems?
Where to apply these?
Causal inference
Link prediction
1
13. Some context: Facebook
Facebook wants to change its’ ad algorithm.
Can’t do it on the whole graph
Need “total network effect”
2
Source: Wikimedia
14. How do they solve it?
Interested in estimating
1
N
N
i=1
[Yi (all treated) − Yi (all controls)]
“At a high level, graph cluster randomization is a technique in
which the graph is partitioned into a set of clusters, and then
randomization between treatment and control is performed at
the cluster level.”
Where can we find clusters?
Observable information (e.g. same school)
Unobservable information (“social space”)
3
15. Some context: (im)migration
Want to know how
regime change affects
population.
Politicians during
election years care
about direct effects.
4
Source: http://openscience.alpine-geckos.at/courses/social-network-
analyses/empirical-network-analysis/
17. And one more
Studying taxi rides in Porto
442 taxis
1.7 million rides with (x, y) coordinates at 15 second intervals.
6
Source: Kucukelbir, A., Tran, D., Ranganath, R., Gelman, A., & Blei, D. M. (2017).
Automatic Differentiation Variational Inference. Journal of Machine Learning
Research, 18(14), 1-45.
18. And one more
Studying taxi rides in Porto
Project into a 100 dimensional latent space.
Learn hidden interpretable patterns...
7
Source: Kucukelbir, A., Tran, D., Ranganath, R., Gelman, A., & Blei, D. M. (2017).
Automatic Differentiation Variational Inference. Journal of Machine Learning
Research, 18(14), 1-45.
19. Relational data: common examples and goals
Changes in exports from year to year
−0.30 −0.20 −0.10
−0.4−0.20.00.20.4
first eigenvector of R^
row
secondeigenvectorofR^
row
Australia
Austria
Brazil
Canada
China
China, Hong Kong SAR
Finland
France
Germany
GreeceIndonesia
Ireland
Italy
Japan
Malaysia
Mexico
Netherlands
New Zealand
Norway
Rep. of Korea
Spain
Switzerland
Thailand
Turkey
United Kingdom
USA
−0.25 −0.15 −0.05 0.05
−0.3−0.10.10.3
first eigenvector of R^
col
secondeigenvectorofR^
col
Australia
Austria
Brazil
Canada
China
China, Hong Kong SAR
Finland
France
Germany
Greece
Indonesia
Ireland
Italy
Japan
Malaysia
Mexico
Netherlands
New ZealandNorway
Rep. of Korea
Spain
Switzerland
Thailand
Turkey
United Kingdom
USA
Network regression problems yij = xij β + ij frequently assume
independence of the ij
8
20. Estimating β in network regression
−0.30 −0.20 −0.10
−0.4−0.20.00.20.4
first eigenvector of R^
row
secondeigenvectorofR^
row
Australia
Austria
Brazil
Canada
China
China, Hong Kong SAR
Finland
France
Germany
GreeceIndonesia
Ireland
Italy
Japan
Malaysia
Mexico
Netherlands
New Zealand
Norway
Rep. of Korea
Spain
Switzerland
Thailand
Turkey
United Kingdom
USA
−0.25 −0.15 −0.05 0.05
−0.3−0.10.10.3
first eigenvector of R^
col
secondeigenvectorofR^
col
Australia
Austria
Brazil
Canada
China
China, Hong Kong SAR
Finland
France
Germany
Greece
Indonesia
Ireland
Italy
Japan
Malaysia
Mexico
Netherlands
New ZealandNorway
Rep. of Korea
Spain
Switzerland
Thailand
Turkey
United Kingdom
USA
For Y =< X, β > +E we have
OLS (assume no dependence among ij ):
ˆβ(ols)
= (mat(X)t
mat(X))−1
mat(X)t
vec(Y )
Oracle GLS (assume dependence among ij ):
ˆβ(gls)
= (mat(X)t
(Σ−1
)mat(X))−1
mat(X)t
(Σ−1
)vec(Y )
9
21. Network models
The data
There are n actors/nodes labeled 1, . . . , n
Y is a sociomatrix: yij is a dyadic relationship between node i
and node j.
yii frequently undefined.
Covariates:
node specific: xi
dyad specific: xij
22. Social relations model
Goal: describe the variability in Y .
Sender effects describe sociability.
Receiver effects describe popularity.
Capture this in the Social Relations Model (SRM)
yij = ai + bj + ij
Almost an ANOVA — want to relate ai to bi since the
senders/receivers are from the same set.
23. Social relations model
yij =µ + ai + bj + ij
(ai , bi )
iid
∼N(0, Σab)
( ij , ji )
iid
∼N(0, Σe)
Σab =
σ2
a σab
σab σ2
b
describes sender/receiver variability and
within person similarity.
Σe = σ2 1 ρ
ρ 1
describes within dyad correlation.
12
24. Variability
var(yij ) =σ2
a + 2σab + σ2
b + σ2
cov(yij , yik) =σ2
a
cov(yij , ukj ) =σ2
b
cov(yij , yjk) =σab
cov(yij , yji ) =2σab + ρσ2
How hard is it to fit this model?
fit_SRM <- ame(Y)
13
25. Pictures that pop up
These help capture how well the Markov Chain is mixing and
goodness of fit information.
14
Source: Hoff (2015). arXiv:1506.08237
26. Goodness of fit
Posterior predictive distributions.
sd.rowmean: standard deviation of row means of Y .
sd.colmean: standard deviation of column means of Y .
dyad.dep: correlation between vectorized Y and vectorized Y t
triad.dep:
i jk eij ejkeki
#triangle on n nodes
Var(vec(Y ))3/2
15
Source: Hoff (2015). arXiv:1506.08237
27. Incorporating covariates
Imagine you have some covariates and want to fit
yij = βt
d xd,ij + βt
r xr,i + βt
cxc,j + ai + bj + ij
xd,ij are dyad specific covariates.
xr,i are row (sender) covariates.
xc,i are column (receiver) covariates.
Frequently xr,i = xc,i = xi
When does this not make sense?
(Example: popularity is affected by athletic success, but
sociability is not)
How hard is it to fit this model?
fit_SRRM <- ame(Y, Xd=Xd,Xr=Xr,Xc=Xc)
16
28. Parsing the input
fit_SRRM <- ame(Y,
Xdyad=Xd, #n x n x pd array of covariates
Xrow=Xr, #n x pr matrix of nodal row covariates
Xcol=Xc #n x pc matrix of nodal column covariates
)
Xri,p is the value of the pth row covariate for node i.
Xdi,j,p is the value of the pth dyadic covariate in the direction
of i to j.
29. Back to basics
Can you get rid of the dependencies in the model?
fit_rm<-ame(Y,Xd=Xd,Xr=Xn,Xc=Xn,
rvar=FALSE, #should you fit row random effects?
cvar=FALSE, #should you fit column random effects?
dcor=FALSE #should you fit a dyadic correlation?
)
Note that summary will output:
Variance parameters:
pmean psd
va 0.000 0.000
cab 0.000 0.000
vb 0.000 0.000
rho 0.000 0.000
ve 0.229 0.011
18
30. So what’s missing here?
We have a lot of left over variability.
Common themes in network analysis:
Homophily: similar people connect to each other
Stochastic equivalence: similar people act similarly
19
32. Which is which?
Left: homophily; Right: stochastic equivalence
What are good models for this?
Source: Hoff (2008). NIPS
33. Introducing multiplicative effects
SR(R)M can represent second-order dependencies very well.
Has a hard time capturing “triadic” behavior.
Homophily: create dyadic covariates xd,ij = xi xj
Generally this can be represented by
xt
ri
Bxj,i = k l bkl xr,ikxc,jl
This is linear in the covariates and so can be baked into the
amen framework.
Sometimes there is excess correlation to account.
This suggests a multiplicative effects model:
yij = βt
d xd,ij + βt
r xr,i + βt
cxc,j + ai + bj + ut
i vj + ij
21
34. Fitting these models and beyond
fit_ame2<-ame(Y,Xd,Xn,Xn,
R=2 #dimension of the multiplicative effect
)
22
Source: Hoff (2015). arXiv:1506.08237
35. What happened here?
Why do multiplicative effects help triadic behavior?
Triadic measure is related to transitivity (at least for binary
data).
Turns out homophily can capture transitivity...
yij = βt
d xd,ij + βt
r xr,i + βt
cxc,j + ai + bj + ut
i vj + ij
ui is information about the sender, vj is information about the
receiver
if ui ≈ vj then ut
i vj > 0...
if ui ≈ uj then there is some stochastic equivalence...
36. Lets generalize: ordinal models
Imagine a binary (probit) model:
yij = 1zij >0 zij = µ + ai + bj + ij
Looks like the SRM on the latent scale.
fit_SRM<-ame(Y,
model="bin" #lots of model options here
)
If we go to the iid set up this is just an Erdos-Renyi model:
fit_SRG<-ame(Y,model="bin",
rvar=FALSE,cvar=FALSE,dcor=FALSE)
24
38. Even more general
Consider the following generative model:
zij = ut
i Dvj + ij
yij = g(zij )
ui are latent factors describing i as a sender
39. Even more general
Consider the following generative model:
zij = ut
i Dvj + ij
yij = g(zij )
ui are latent factors describing i as a sender
vj are latent factors describing j as a receiver
40. Even more general
Consider the following generative model:
zij = ut
i Dvj + ij
yij = g(zij )
ui are latent factors describing i as a sender
vj are latent factors describing j as a receiver
D is a matrix of factor weights
41. Even more general
Consider the following generative model:
zij = ut
i Dvj + ij
yij = g(zij )
ui are latent factors describing i as a sender
vj are latent factors describing j as a receiver
D is a matrix of factor weights
g is an increasing function mapping the latent space to the
observed space.
42. Even more general
Consider the following generative model:
zij = ut
i Dvj + ij
yij = g(zij )
ui are latent factors describing i as a sender
vj are latent factors describing j as a receiver
D is a matrix of factor weights
g is an increasing function mapping the latent space to the
observed space.
(Some gs... Normal: g(z) = z, binomial: g(z) = 1z≥0)
25
43. This works for symmetric matrices too!
Imagine that yij = yji then the model looks like:
zij = ui Λuj + ij
yij = g(zij )
26
44. This works for symmetric matrices too!
Imagine that yij = yji then the model looks like:
zij = ui Λuj + ij
yij = g(zij )
ui ≈ uj represents stochastic equivalence
45. This works for symmetric matrices too!
Imagine that yij = yji then the model looks like:
zij = ui Λuj + ij
yij = g(zij )
ui ≈ uj represents stochastic equivalence
Λ is a matrix of eigenvalues:
46. This works for symmetric matrices too!
Imagine that yij = yji then the model looks like:
zij = ui Λuj + ij
yij = g(zij )
ui ≈ uj represents stochastic equivalence
Λ is a matrix of eigenvalues:
positive λi imply homophily, negative ones imply heterophily.
26
47. What is this latent space?
Problem 1: need to select a dimension R.
27
48. What is this latent space?
Problem 1: need to select a dimension R.
This is hard... sometimes there is some intuition.
27
49. What is this latent space?
Problem 1: need to select a dimension R.
This is hard... sometimes there is some intuition.
Problem 2: should the latent positions be interpreted?
27
50. What is this latent space?
Problem 1: need to select a dimension R.
This is hard... sometimes there is some intuition.
Problem 2: should the latent positions be interpreted?
Unclear — maybe think of the distances in this space...
27
51. What is this latent space?
Problem 1: need to select a dimension R.
This is hard... sometimes there is some intuition.
Problem 2: should the latent positions be interpreted?
Unclear — maybe think of the distances in this space...
Problem 3: what about my favorite other models like
stochastic blockmodels?
52. What is this latent space?
Problem 1: need to select a dimension R.
This is hard... sometimes there is some intuition.
Problem 2: should the latent positions be interpreted?
Unclear — maybe think of the distances in this space...
Problem 3: what about my favorite other models like
stochastic blockmodels?
These are just a subclass of models! For example, the
stochastic blockmodel has discrete support for the latent
positions.
53. What is this latent space?
All quotes from Hoff, et al 2002
A subset of individuals in the population with a large number
of social ties between them may be indicative of a group of
individuals who have nearby positions in this space of
characteristics, or social space.
Various concepts of social space have been discussed by
McFarland and Brown (1973) and Faust (1988).
In the context of this article, social space refers to a space of
unobserved latent characteristics that represent potential
transitive tendencies in network relations.
A probability measure over these unobserved characteristics
induces a model in which the presence of a tie between two
individuals is dependent on the presence of other ties.
54. (Tiny portion of the) literature
Nowicki, Krzysztof, and Tom A. B. Snijders. ”Estimation and
prediction for stochastic blockstructures.” Journal of the American
Statistical Association 96, no. 455 (2001): 1077-1087.
Hoff, Peter D., Adrian E. Raftery, and Mark S. Handcock. ”Latent
space approaches to social network analysis.” Journal of the
american Statistical association 97, no. 460 (2002): 1090-1098.
Hoff, Peter. ”Modeling homophily and stochastic equivalence in
symmetric relational data.” In Advances in Neural Information
Processing Systems, pp. 657-664. 2008.
Airoldi, Edoardo M., David M. Blei, Stephen E. Fienberg, and Eric
P. Xing. ”Mixed membership stochastic blockmodels.” Journal of
Machine Learning Research 9, no. Sep (2008): 1981-2014.
Hoff, Peter, Bailey Fosdick, Alex Volfovsky, and Katherine Stovel.
”Likelihoods for fixed rank nomination networks.” Network Science
1, no. 03 (2013): 253-277.
Hoff, Peter D. ”Dyadic data analysis with amen.” arXiv preprint
arXiv:1506.08237 (2015).
55. ame(Y, Xdyad=NULL, Xrow=NULL, Xcol=NULL,
rvar = !(model=="rrl") , cvar = TRUE, dcor = !symmetric,
nvar = TRUE, R = 0, model="nrm",
intercept=!is.element(model,c("rrl","ord")),
symmetric=FALSE,
odmax=rep(max(apply(Y>0,1,sum,na.rm=TRUE)),nrow(Y)), ...)
Y: an n x n square relational matrix of relations.
Xdyad: an n x n x pd array of covariates
Xrow: an n x pr matrix of nodal row covariates
Xcol: an n x pc matrix of nodal column covariates
rvar: logical: fit row random effects (asymmetric case)?
cvar: logical: fit column random effects (asymmetric case)?
dcor: logical: fit a dyadic correlation (asymmetric case)?
nvar: logical: fit nodal random effects (symmetric case)?
R: int: dimension of the multiplicative effects (can be 0)
model: char: one of "nrm","bin","ord","cbin","frn","rrl"
odmax: a scalar integer or vector of length n giving the
maximum number of nominations that each node may make
56. What’s in the ...?
seed = 1, nscan = 10000, burn = 500, odens = 25,
plot=TRUE, print = TRUE, gof=TRUE
seed: random seed
nscan: number of iterations of the Markov chain
(beyond burn-in)
burn: burn in for the Markov chain
odens: output density for the Markov chain
plot: logical: plot results while running?
print: logical: print results while running?
gof: logical: calculate goodness of fit statistics?
64. Data collection examples
PROmoting School Community-University Partnerships to
Enhance Resilience (PROSPER): “Who are your best and
closest friends in your grade?”
National Longitudinal Study of Adolescent to Adult Health
(AddHealth): “Your male friends. List your closest male
friends. List your best male friend first, then your next best
friend, and so on.”
34
65. Notation
Z = {zij : i = j} is a sociomatrix of
ordinal relationships
zij > zik denotes person i preferring
person j to person k
Z =
− z12 · · · z1n
z21 −
... −
zn1 −
35
66. Notation
Z = {zij : i = j} is a sociomatrix of
ordinal relationships
zij > zik denotes person i preferring
person j to person k
Z =
− z12 · · · z1n
z21 −
... −
zn1 −
35
67. Notation
Z = {zij : i = j} is a sociomatrix of
ordinal relationships
zij > zik denotes person i preferring
person j to person k
Z =
− z12 · · · z1n
z21 −
... −
zn1 −
Instead of Z we observe a sociomatrix Y = {yij : i = j}
35
68. Notation
Z = {zij : i = j} is a sociomatrix of
ordinal relationships
zij > zik denotes person i preferring
person j to person k
Z =
− z12 · · · z1n
z21 −
... −
zn1 −
Instead of Z we observe a sociomatrix Y = {yij : i = j}
Different sampling schemes define different maps between Y
and Z (set relations between yij and zij ).
35
69. Notation
Z = {zij : i = j} is a sociomatrix of
ordinal relationships
zij > zik denotes person i preferring
person j to person k
Z =
− z12 · · · z1n
z21 −
... −
zn1 −
Instead of Z we observe a sociomatrix Y = {yij : i = j}
Different sampling schemes define different maps between Y
and Z (set relations between yij and zij ).
Statistical model {p (Z|θ) : θ ∈ Θ} assists in analysis
35
70. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
36
71. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
36
72. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
36
73. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
zi1 zi2 zi3 zi4 0> 0> 0> 0> 0> 0>> > > >
36
74. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
36
75. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
36
76. Fixed rank nominations
yij > yik ⇒ zij > zik
}F (Y )yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
m = maximal number of nominations, di = individual outdegree
Differentiates between different ranks
Captures censoring in the data
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
zi1 zi2 zi3 zi4 zi5 ? ? ? ? ?> > > > >
36
77. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
37
78. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
37
79. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
37
80. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
zi1 zi2 zi3 zi4 ? ? ? ? ? ?> > > >
37
81. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
37
82. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
37
83. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
zi1 zi2 zi3 zi4 zi5 ? ? ? ? ?> > > > >
37
84. Rank
yij > yik ⇒ zij > zik } R (Y )
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
Valid but not fully informative: F (Y ) R (Y )
Cannot estimate row (“sender”) specific effects
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
zi1 zi2 zi3 zi4 zi5 ? ? ? ? ?> > > > >
37
85. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
38
86. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
38
87. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
38
88. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
4 3 2 1 0 0 0 0 0 0
>0 >0 >0 >0 0> 0> 0> 0> 0> 0>
38
89. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
38
90. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
38
91. Binary
yij > yik ⇒ zij > zik
yij = 0 and di < m ⇒ zij ≤ 0
yij > 0 ⇒ zij > 0
} B (Y )
yij = 0 ⇒ zij < 0
F(Y)
R(Y)
B(Y)
Neither fully informative nor valid!
Discards information on the ranks
Ignores the censoring on the outdegrees
In particular: F (Y ) ⊂ B (Y )
zi
yi
1 2 3 4 5 6 7 8 9 10
5 4 3 2 1 0 0 0 0 0
>0 >0 >0 >0 >0 0> 0> 0> 0> 0>
38
92. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
39
93. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
Simulate zij ∼ p(zij |θ, Z−ij , Z ∈ F(Y )):
39
94. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
Simulate zij ∼ p(zij |θ, Z−ij , Z ∈ F(Y )):
1. yij > 0: zij ∼ p(zij |θ, Z−ij )1zij ∈(a,b) where
a = max(zik : yik < yij ) and b = min(zik : yik > yij ).
39
95. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
Simulate zij ∼ p(zij |θ, Z−ij , Z ∈ F(Y )):
1. yij > 0: zij ∼ p(zij |θ, Z−ij )1zij ∈(a,b) where
a = max(zik : yik < yij ) and b = min(zik : yik > yij ).
2. yij = 0 and di < m: zij ∼ p(zij |Z−ij , θ)1zij ≤0.
39
96. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
Simulate zij ∼ p(zij |θ, Z−ij , Z ∈ F(Y )):
1. yij > 0: zij ∼ p(zij |θ, Z−ij )1zij ∈(a,b) where
a = max(zik : yik < yij ) and b = min(zik : yik > yij ).
2. yij = 0 and di < m: zij ∼ p(zij |Z−ij , θ)1zij ≤0.
3. yij = 0 and di = m: zij ∼ p(zij |Z−ij , θ)1zij ≤min(zik :yik >0)
39
97. Bayesian Estimation for Fixed Rank Nominations
Model: Z ∼ p(Z|θ), θ ∈ Θ
Data: Z ∈ F(Y )
Likelihood:
LF (θ : Y ) = Pr (Z ∈ F (Y )|θ) =
F(Y )
dP (Z|θ)
Estimation: Given p(θ), p(θ|Z ∈ F(Y )) can be approximated
by a Gibbs sampler.
Simulate zij ∼ p(zij |θ, Z−ij , Z ∈ F(Y )):
1. yij > 0: zij ∼ p(zij |θ, Z−ij )1zij ∈(a,b) where
a = max(zik : yik < yij ) and b = min(zik : yik > yij ).
2. yij = 0 and di < m: zij ∼ p(zij |Z−ij , θ)1zij ≤0.
3. yij = 0 and di = m: zij ∼ p(zij |Z−ij , θ)1zij ≤min(zik :yik >0)
Allows for imputation of missing yij
39
98. Simulations
We generated Z from the following Social Relations Model
(Warner, Kenny and Stoto (1979)):
zij = βt
xij + ai + bj + ij
ai
bi
iid
∼ normal 0,
1 0.5
0.5 1
ij
ji
iid
∼ normal 0,
1 0.9
0.9 1
Mean model: βtxij = β0 + βr xir + βcxjc + βd1 xij1 + βd2 xij2
xir , xjc: individual level variables
xij1: pair specific variable
xij2: co-membership in a group
40
99. Simulations
We generated Z from the following Social Relations Model
(Warner, Kenny and Stoto (1979)):
zij = βt
xij + ai + bj + ij
ai
bi
iid
∼ normal 0,
1 0.5
0.5 1
ij
ji
iid
∼ normal 0,
1 0.9
0.9 1
Mean model: βtxij = β0 + βr xir + βcxjc + βd1 xij1 + βd2 xij2
xir , xjc: individual level variables
xij1: pair specific variable
xij2: co-membership in a group
βr = βc = βd1 = βd2 = 1 and β0 = −3.26
xir , xic, xij1
iid
∼ N (0, 1) xij2 = si sj /.42 for si
iid
∼ binary (1/2)
40
100. Simulations
We generated Z from the following Social Relations Model
(Warner, Kenny and Stoto (1979)):
zij = βt
xij + ai + bj + ij
ai
bi
iid
∼ normal 0,
1 0.5
0.5 1
ij
ji
iid
∼ normal 0,
1 0.9
0.9 1
Mean model: βtxij = β0 + βr xir + βcxjc + βd1 xij1 + βd2 xij2
xir , xjc: individual level variables
xij1: pair specific variable
xij2: co-membership in a group
βr = βc = βd1 = βd2 = 1 and β0 = −3.26
xir , xic, xij1
iid
∼ N (0, 1) xij2 = si sj /.42 for si
iid
∼ binary (1/2)
40
111. Simulations - information in the ranks
Let C (Y ) be the set of values for which the following is true:
yij > 0 ⇒ zij > 0
yij = 0 and di < m ⇒ zij ≤ 0
min {zij : yij > 0} ≥ max {zij : yij = 0}
We refer to LC (θ : Y ) = Pr (Z ∈ C (Y )|θ) as the censored
binary likelihood.
Recognizes censoring but ignores information in the ranks
112. Simulations - information in the ranks
Let C (Y ) be the set of values for which the following is true:
yij > 0 ⇒ zij > 0
yij = 0 and di < m ⇒ zij ≤ 0
min {zij : yij > 0} ≥ max {zij : yij = 0}
We refer to LC (θ : Y ) = Pr (Z ∈ C (Y )|θ) as the censored
binary likelihood.
Recognizes censoring but ignores information in the ranks
Performs similarly to FRN in the previous study
Less precise than FRN when m is big
113. Simulations - information in the ranks
Same setup as before, but average uncensored outdegree is m
10 20 30 40 50
0.20.40.60.81.01.21.4
m
relativeconcentrationaroundtruevalue
! ! !
! !r
!
!
! ! !c
!
!
!
! !d1
! !
!
! !d2
2: Posterior concentration around true parameter values. The average of E[(β −
(S)]/E[(β − β∗)2|C(S)] across eight simulated datasets for each m ∈ {5, 15, 30, 50}.
censored binomial likelihood. As the censored binomial likelihood recognizes the censoring in
data, we expect it to provide parameter estimates that do not have the biases of the binomial
ood estimators. On the other hand, LC ignores the information in the ranks of the scored
duals, and so we might expect it to provide less precise estimates than the FRN likelihood.
βr : row
βc: column
βd1: continuous dyad
βd2: co-membership
Relative concentration around true value of each parameter:
Measured by E (β − 1)
2
|F (Y ) /E (β − 1)
2
|C (Y ) for each β
45
114. Simulations - information in the ranks
Same setup as before, but average uncensored outdegree is m
10 20 30 40 50
0.20.40.60.81.01.21.4
m
relativeconcentrationaroundtruevalue
! ! !
! !r
!
!
! ! !c
!
!
!
! !d1
! !
!
! !d2
2: Posterior concentration around true parameter values. The average of E[(β −
(S)]/E[(β − β∗)2|C(S)] across eight simulated datasets for each m ∈ {5, 15, 30, 50}.
censored binomial likelihood. As the censored binomial likelihood recognizes the censoring in
data, we expect it to provide parameter estimates that do not have the biases of the binomial
ood estimators. On the other hand, LC ignores the information in the ranks of the scored
duals, and so we might expect it to provide less precise estimates than the FRN likelihood.
βr : row
βc: column
βd1: continuous dyad
βd2: co-membership
Relative concentration around true value of each parameter:
Measured by E (β − 1)
2
|F (Y ) /E (β − 1)
2
|C (Y ) for each β
When m n, most of the information found by considering
ranked/unranked individuals as groups rather than the relative
ordering of the ranked individuals.
115. AddHealth Data - Results
−3.65−3.50−3.35
β
intercept
q
q
−0.050.000.050.10
rsmoke rdrink rgpa
q q q
q
q
q
−0.050.000.050.10
csmoke cdrink cgpa
q q q
q
q q
q q q
−0.050.000.050.10
β
dsmoke ddrink dgpa
q q q q q q q
q q
0.20.40.6
β
dacad darts dsport dcivic
q
qq
q
qq
q
qq
q
qq
0.20.40.60.81.0
β
dgrade drace
q q q
q
q q
646 females were asked to rank up to 5 female friends
Mean model with row, column and dyadic effects for smoking,
drinking and gpa as well as dyadic effects for comembership in
activities and grade, and a similarity-in-race measure.
The CIs are based on binary, FRN and rank likelihoods.
46