3. When to use a SAOM
May 20, 2016 Duke Social Networks & Health Workshop 3
• Ques+ons about changes in network structure over +me
– Including mul+ple networks
– Including two-mode networks (selec+ng into foci)
• Ques+ons about how networks affect individual “behaviors,”
such as through peer influence
– Including mul+ple behaviors and possible reciprocal
effects
• Ques+ons about the endogenous associa+on between
networks and behavior
12. j3
ego
j4
j2
j1
Network Decision
€
fego(β,x) = -2
€
xij
j
∑ + 1.8
€
xij x ji
j
∑
outdegree reciprocity
€
fego(β,x) = -2
€
xij
j
∑ + 1.8
€
xij x ji
j
∑fego(β,x) = -2
€
xij
j
∑ + 1.8
€
xij x ji
j
∑
During a micro step, an actor evaluates how changing its outgoing
+e in each dyad would affect the value of the objec+ve func+on
(goal is to maximize the value of the func+on)
ego j1 j2 j3 j4
ego - 1 1 0 0
j1 1 - 0 0 0
j2 0 0 - 0 0
j3 1 0 0 - 0
j4 0 0 0 0 -
May 20, 2016 Duke Social Networks & Health Workshop 12
If… outdegree reciprocity sum
No change -2 * 2 = -4 1.8 * 1 = 1.8 -2.2
Drop j1 -2 * 1 = -2 1.8 * 0 = 0 -2
Drop j2 -2 * 1 = -2 1.8 * 1 = 1.8 -.2
Add j3 -2 * 3 = -6 1.8 * 3 = 3.6 -2.4
Add j4 -2 * 3 = -6 1.8 * 1 = 1.8 -4.2
Given the current state of the network, ego is
most likely to drop the ?e to j2, because that
decision maximizes the objec+ve func+on
13. • Outdegree always present
• Network processes (e.g., reciprocity, transi+vity)
• Adribute based:
– Sociality: effect of adribute on outgoing +es
– Popularity: effect of behavior on incoming +es
– Homophily: ego-alter similarity
– Note: adributes may be stable or +me-changing
(exogenous or endogenously modeled)
• Dyadic adributes (e.g., co-membership)
May 20, 2016 Duke Social Networks & Health Workshop 13
Network Objec?ve Func?on Effects
21. Behavior Decision*
May 20, 2016 Duke Social Networks & Health Workshop 21
If… linear quad age similarity sum
Drop to 0 -.5 * 0 = 0 .25 * 0 = 0 .1 * 10 * 0 = 0 1 * .35 = .35 .35
Stay at 1 -.5 * 1 = -.5 .25 * 1 = .25 .1 * 10 * 1 = 1 1 * .95 = .95 1.7
Up to 2 -.5 * 2 = -1 .25 * 4 = 1 .1 * 10 * 2 = 2 1 * -.55 = -.55 1.45
* Assume covariates uncentered
Second, calculate the contribu+ons for
each of the other effects
22. Behavior Decision*
May 20, 2016 Duke Social Networks & Health Workshop 22
If… linear quad age similarity sum
Drop to 0 -.5 * 0 = 0 .25 * 0 = 0 .1 * 10 * 0 = 0 1 * .35 = .35 .35
Stay at 1 -.5 * 1 = -.5 .25 * 1 = .25 .1 * 10 * 1 = 1 1 * .95 = .95 1.7
Up to 2 -.5 * 2 = -1 .25 * 4 = 1 .1 * 10 * 2 = 2 1 * -.55 = -.55 1.45
* Assume covariates uncentered
These effects pull
ego toward the
extremes
The posi+ve age b
pushes ego’s
behavior upward
Similarity pushes
ego to stay the
same
Altogether, the greatest contribu+on to the behavior func+on comes
from ego choosing to maintain the same behavior level
23. • Necessary for both network and behavior
• Determine the wai+ng +me un+l actor’s chance to make decisions
• Func+on of observed changes
– But not the same as the number of changes observed
– Separate rate parameter for each period between observa+ons
• Wai+ng +me distributed uniformly by default, but differences can
be modeled based on:
• Actor adributes: do some types of actors experience more or
less change
• Degree: do actors with more/fewer +es experience a different
volume of change
May 20, 2016 Duke Social Networks & Health Workshop 23
Rate Func?ons
32. • Helpful to imagine the network func+on as a logis+c regression
– Unit of analysis: dyad
– Outcome: +e presence (keeping or adding) vs. absence
(dissolving or failing to add)
– Each effect represents how a one-unit change in the effect
sta+s+c affects the log-odds of a +e, all else being equal
• Some effects interpretable using odds ra+os, but
– One-unit changes may not be meaningful
– All else is never equal (any change also affects the
outdegree count, at a minimum)
• Behavior func+on specifies how a one-unit change in the effect
sta+s+c affects the odds of increasing behavior one unit
May 20, 2016 Duke Social Networks & Health Workshop 32
Interpre?ng Results
33. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 33
Rate: Each actor is given ~10
micro steps in which to make a
change to its network
• Add a +e, drop a +e, or make
no change
Rate
34. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 34
Outdegree: The nega+ve sign is
typical. It means that +es are
unlikely, unless other effects in
the model make a posi+ve
contribu+on to the network
func+on.
density
35. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 35
Reciprocity: Ties that create a
reciprocated +e are more likely
to be added or maintained. This
effect hovers around 2 in
friendship-type network.
recip
36. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 36
Transi?ve triplets: Ties that
create more transi+ve triads
have a greater likelihood.
• Should also test interac+on
with Reciprocity (usually
nega+ve)
transTrip
37. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 37
Indegree Popularity: Actors with
more incoming +es have a
greater likelihood of receiving
future +es
inPop
38. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
May 20, 2016 Duke Social Networks & Health Workshop 38
Dyadic Covariate: Actors who
share an extracurricular ac+vity
(coded 1) are more likely to have
a friendship +e
X
39. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
Ties driven by similarity on:
Gender (could use “same” effect)
Age
Alcohol use
GPA
Females less adrac+ve as friends
than males.
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
altX
egoX
simX
May 20, 2016 Duke Social Networks & Health Workshop 39
40. Network func?on b SE
Rate 10.26 *** .49
Outdegree -3.91 *** .08
Reciprocity 1.91 *** .09
Transi+ve triplets .52 *** .04
Popularity .29 *** .04
Extracurric. act. overlap .28 *** .06
Smoke similarity .68 *** .12
Smoke alter .14 ** .05
Smoke ego -.04 .05
Female similarity .24 *** .04
Female alter -.11 * .05
Female ego -.04 .05
Age similarity 1.00 *** .13
Age alter -.01 .03
Age ego -.04 .03
Delinquency similarity .15 .08
Delinquency alter -.04 .04
Delinquency ego .02 .04
Alcohol similarity .27 ** .10
Alcohol alter -.03 .03
Alcohol ego -.03 .04
GPA similarity .70 *** .13
GPA alter -.05 .04
GPA ego -.02 .04
From Schaefer, D.R. S.A. Haas, and N. Bishop. 2012. “A Dynamic
Model of US Adolescents’ Smoking and Friendship Networks.”
American Journal of Public Health, 102:e12-e18.
Ties driven by similarity on
smoking behavior.
Smokers more adrac+ve as
friends than non-smokers.
Alter
Nonsmoker Smoker
Ego
Nonsmoker .25 -.19
Smoker -.51 .41
Similarity is an “interac+on” between
ego and alter, thus interpreta+on
requires considering the main effects
Ego-alter selec+on: Contribu+ons to
network objec+ve func+on by dyad type
May 20, 2016 Duke Social Networks & Health Workshop 40
47. Cumula?ve Indegree Distribu?on
Goodness of Fit of IndegreeDistribution
p: 0
Statistic
0 1 2 3 4 5 6 7 8
139
193
282
343
401
437
459
483
491
May 20, 2016 Duke Social Networks & Health Workshop 47
48. Geodesic Distribu?on
Goodness of Fit of GeodesicDistribution
p: 0.001
Statistic
1 2 3 4 5 6 7
1381
2795
5014
7772
10598
12081 11892
May 20, 2016 Duke Social Networks & Health Workshop 48
49. Triad Census Goodness of Fit of TriadCensus
p: 0.114
Statistic(centeredandscaled)
003 012 102 021D 021U 021C 111D 111U 030T 030C 201 120D 120U 120C 210 300
21286492
428358
129429
693
1141
1052
923
625
108
4 171
114
58
39
91
36
May 20, 2016 Duke Social Networks & Health Workshop 49
50. Smoking Distribu?on
Goodness of Fit of BehaviorDistribution
p: 1
Statistic
0 1 2
222
98
182
May 20, 2016 Duke Social Networks & Health Workshop 50
52. Extensions to Basic Model
May 20, 2016 Duke Social Networks & Health Workshop 52
• interac+ons
• event history outcomes
• mul+ple behaviors
• mul+ple network op+ons
• valued +es
• mul+level networks
• two mode networks
• increase vs. decrease in +es and/or behavior
• +me heterogeneity
• simula+ons (test interven+ons)
• ML, Bayes es+ma+on
53. Asymmetric Peer Influence
• Implicit assump+on that effects work the same for:
– Tie forma+on vs. dissolu+on
– Behavior increase vs. decrease
• Unrealis+c for smoking
– Physical/psychological dependence, social learning
• Easy to relax this assump+on
– Separate behavior objec+ve func+on into:
• Crea?on func?on: only considers increases
• Maintenance func?on: only considers decreases
– Could make similar dis+nc+on in the network func+on
May 20, 2016 Duke Social Networks & Health Workshop 53
56. Decomposing Network Homogeneity
Source Selec?on (%) Influence (%) Sample
Schaefer et al. 2012 40 34 U.S.
Mercken et al. 2009 17-47 6-23 Europe (6 countries)
Mercken et al. 2010 31-46 15-22 Finland
Steglich et al. 2010 25-34 20-37 Scotland
• How much network homogeneity on smoking is due to
selec?on vs. influence?
– Systema+cally set selec+on and influence parameters to
zero and simulate network-behavior co-evolu+on (see
Steglich et al. 2010)
May 20, 2016 Duke Social Networks & Health Workshop 56
59. Context Effects
How do these effects depend upon context?
• Randomly manipulate ini+al smoking prevalence
– 25% ini+al smokers up to 75%
• Randomly distribute smokers and nonsmokers across the
network
– Similar results with empirical and model-based
manipula+ons
• Full results in adams, jimi & David R. Schaefer. 2016. “How
Ini+al Prevalence Moderates Network-Based Smoking
Change: Es+ma+ng Contextual Effects with Stochas+c Actor
Based Models.” Journal of Health & Social Behavior 57(1):
22-38.
May 20, 2016 Duke Social Networks & Health Workshop 59
62. • Ties are more or less enduring states
– Plausible for friendship or collabora+ons
– Not useful for “event” data (e.g. phone calls)
• Change occurs in con+nuous +me
• Markov process: future state only a func+on of current state
– No lagged effects, “grudges”
• Actors control outgoing +es and behavior
• One change at a +me
– No coordinated or simultaneous changes
May 20, 2016 Duke Social Networks & Health Workshop 62
Assump?ons
63. • Up to 10% probably ok, more than 20% likely a problem
• Endogenous network & behavior imputa+on
– Missing values at t0 set to 0 (network) or mode (behavior)
– Missing values at t1+ imputed with last valid value if
possible, otherwise 0
• Covariates imputed with the mean
– Other values can be specified
• Imputed values are treated as non-informa+ve, thus not used
in calcula+ng target sta+s+cs
– Convergence and fit are determined based only upon
observed cases
May 20, 2016 Duke Social Networks & Health Workshop 63
Missing Data
64. Good Sources of Informa?on
May 20, 2016 Duke Social Networks & Health Workshop 64
• RSiena manual
• Snijders, van de Bunt & Steglich, 2010
• Steglich, Snijders & Pearson, 2010
• Tom Snijders’ SIENA website
www.stats.ox.ac.uk/siena/
– Workshops
– Scripts
– Applica+ons in the literature
– Latest version of RSiena
– Link to stocnet listserv – important updates announced here
– “Siena_algorithms.pdf”
67. • One mode or two mode network with at least two
observa+ons, each represented as a matrix
– Ties coded 0, 1, 10 (structural 0), 11 (structural 1), or NA
• For each “period” between adjacent waves, stability measured
by the Jaccard coefficient should be at least .25
– Ties persisted / (+es formed + +es dissolved + +es persisted)
• “Complete network data” all actors w/in bounded semng
– Some turnover in set of actors allowed but same actors in
the data for each wave (even if not observed during wave)
– See manual for how to deal with composi+on change
• Recommended N: 30-2000
May 20, 2016 Duke Social Networks & Health Workshop 67
Data Structure: Network
68. • Dependent behaviors
– Time-varying adributes used as dependent variable(s)
– Coded as integer (e.g., 1-10)
– Last +me point is used
• Changing actor covariates
– Time-varying adributes used as independent variables
– Last +me point not used (only applicable for 3+ waves)
• Constant covariates
– Ex: age, sex, race/ethnicity, behavior
• Dyadic covariates
– Ex: semngs that drive contact
NOTE: Covariates are centered by default
May 20, 2016 Duke Social Networks & Health Workshop 68
Addi?onal Data Structures