Diffuse Musings James Moody Duke University, Sociology Duke Network Analysis Center http://dnac.ssri.duke.edu/ SAMSI Complex Networks Workshop Aug/Sep 2010
Mar 27, 2015
Diffuse Musings
James MoodyDuke University, Sociology
Duke Network Analysis Centerhttp://dnac.ssri.duke.edu/
SAMSI Complex Networks WorkshopAug/Sep 2010
Consider two degree distributions: A long-tail distribution compared to one with no high-degree nodes.
The scale-free network’s signature is the long-tail
So what effect does changes in the shape have on connectivity
1. Shape MattersConsequences of Degree Distribution Shape
1. Shape MattersConsequences of Degree Distribution Shape
Volume
D
ispe
rsio
n x
Ske
wne
ss
1. Shape MattersConsequences of Degree Distribution Shape
Search Procedure: 1) Identify all valid degree distributions with
the given mean degree and a maximum of 6 w. brute force search.
2) Map them to this space3) Simulate networks each degree distribution4) Measure size of components &
Bicomponents
1. Shape MattersConsequences of Degree Distribution Shape
1. Shape MattersConsequences of Degree Distribution Shape
Consider targeting high-degree nodes by cracking down on commercial sex workers:
Interventions can have very different effects depending on where you sit within this field
Does this matter?
1. Shape MattersConsequences of Degree Distribution Shape
Network Sub-Structure: Triads
003
(0)
012
(1)
102
021D
021U
021C
(2)
111D
111U
030T
030C
(3)
201
120D
120U
120C
(4)
210
(5)
300
(6)
Intransitive
Transitive
Mixed
1. Linking ShapesFrom “motifs” to “structure”
PRC{300,102, 003, 120D, 120U, 030T, 021D, 021U} Ranked Cluster:
M MN*
M MN*
M
A*A*
A*A*
A*A*
A*A*
11
11
1
1111
011
11 0
0
00 0 0 0
0 00 0
And many more...
1. Linking ShapesFrom “motifs” to “structure”
Structural Indices based on the distribution of triads
The observed distribution of triads can be fit to the hypothesized structures using weighting vectors for each type of triad.
llμlTl
T
T
)()(l
Where:l = 16 element weighting vector for the triad typesT = the observed triad census T= the expected value of TT = the variance-covariance matrix for T
1. Linking ShapesFrom “motifs” to “structure”
-But such an image conflates many temporally distinct events. A more accurate image is something like this:
In general, the graphs over which diffusion happens often:
• Have timed edges • Nodes enter and leave • Edges can re-occur multiple times• Edges can be concurrent
These features break transmission paths, generally lowering diffusion potential – and opening a host of interesting questions about the intersection of structure and time in networks.
The Cocktail Party Problem
3. Time MattersDynamics of affect dynamics on
Source: Bender-deMoll & McFarland “The Art and Science of Dynamic Network Visualization” JoSS 2006
3. Time MattersDynamics of affect dynamics on
Relationship timing constrains diffusion paths because goods can only move forward in time.
a b c dStandard graph: - Connected component - Everyone could diffuse to everyone else
3. Time MattersDynamics of affect dynamics on
Relationship timing constrains diffusion paths because goods can only move forward in time.
a b
Dynamic graph: - Edges start and end - Can’t pass along an edge that has ended
Time
b c
c d
3. Time MattersDynamics of affect dynamics on
Relationship timing constrains diffusion paths because goods can only move forward in time.
a b
Dynamic graph: - Edges start and end - Can’t pass along an edge that has ended
Diffusion is asymmetric: a can reach c (through b) and b and reach d (through c), but not the other way around.
Time
b c
c d
3. Time MattersDynamics of affect dynamics on
Relationship timing constrains diffusion paths because goods can only move forward in time.
Time
a b c
c d
Concurrency, when edges share a node at the same time, allows diffusion to move symmetrically through the network.
This can have a dramatic effect on increasing the down-stream potential for any give tie.
3. Time MattersDynamics of affect dynamics on
Implied Contact Network of 8 people in a ringAll relations Concurrent
Edge timing constraints on diffusion
Reachability = 1.0
3. Time MattersDynamics of affect dynamics on
Implied Contact Network of 8 people in a ringSerial Monogamy (3)
1
2
1
1
2
1
2
2
Reachability = 0.43
Edge timing constraints on diffusion
3. Time MattersDynamics of affect dynamics on
1
2
1
1
2
1
2
2
Timing alone can change mean reachability from 1.0 when all ties are concurrent to 0.42.
In general, ignoring time order is equivalent to assuming all relations occur simultaneously – assumes perfect concurrency across all relations.
Edge timing constraints on diffusion
3. Time MattersDynamics of affect dynamics on
Timing constrains potential diffusion paths in networks, since bits can flow through edges that have ended.
This means that:
•Structural paths are not equivalent to the diffusion-relevant path set.•Network distances don’t build on each other. •Weakly connected components overlap without diffusion reaching across sets.•Small changes in edge timing can have dramatic effects on overall diffusion•Diffusion potential is maximized when edges are concurrent and minimized when they are “inter-woven” to limit reachability.
Combined, this means that many of our standard path-based network measures will be incorrect on dynamic graphs.
3. Time MattersDynamics of affect dynamics on
Solution? Turn time into a network!
Time-Space graph representations“Stack” a dynamic network in time, compiling all “node-time” and “edge-time” events (similar to an event-history compilation of individual level data).
Consider an example:
a) Repeat contemporary ties at each time observation, linked by relational edges as they happen.
b) Between time slices, link nodes to later selves “identity” edges
3. Time MattersDynamics of affect dynamics on
So now we:
1) Convert every edge to a node2) Draw a directed arc between
edges that (a) share a node and (b) precede each other in time.
Solution? Turn time into a network!
3. Time MattersDynamics of affect dynamics on
So now we:
1) Convert every edge to a node2) Draw a directed arc between edges that (a) share a node and (b) precede each other in time.
3) After the transformation, concurrent relations are easily seen as reciprocal edges in the line-graph. Becomes this:
Solution? Turn time into a network!
3. Time MattersDynamics of affect dynamics on
4. Universal or Particular?
When do we need to bring in domain-specific information?
“diffusion” as transmission between nodes seems universal; but the content of the graph likely interacts with the structure.
H W
CC
C
Provides food for
Romantic Love
Bickers with
How does information move here?
Generality depends on:a) Transmission directionality: does
“passing” the bit affect the sender?b) Relational Permeability: Does
transmission move differently across different relations?
c) Structural Reflexivity: does transmission affect the structure?
http://dnac.ssri.duke.edu/