1 of 33

Network Experiments & Interventions

2 of 33

  1. Assessing Peer Effects
    1. Manipulate networks to force connectivity to be exogenous
      1. Roommate studies
      2. Natural Experiments

​

    • Manipulate exposure over existing networks
      • Popularity
      • Voter Turnout

​

  • Interventions: Use networks to affect change
    • Valente’s 4 elements of network intervention
    • Exemplars

​

  • Network interference in “standard” experiments
  • What we’re ignoring: small group stuff
    • network exchange experiments
    • Transmission evolution experiments

​

​

​

​

3 of 33

Network Experiments & Interventions

Intro

Shalizi & Thomas: PI is *generally* confounded

General fear of confounding by unobserved selection features has pushed interest in network experiments. How to make a social relation exogenous?

4 of 33

Network Experiments & Interventions

Problem

5 of 33

Network Experiments & Interventions

Problem

6 of 33

Network Experiments & Interventions

Problem

7 of 33

8 of 33

Roommates assigned randomly within gender, & self-reported smoking, study-style & messiness.

​

No evidence of pair-similarity based on this sorting (i.e. all evidence suggests randomness is real)

Network Experiments & Interventions

Exogenous networks (W)

9 of 33

Assume that one’s one GPA is a function of own ability & peer ability and GPA

Network Experiments & Interventions

Exogenous networks (W)

10 of 33

Assume that one’s one GPA is a function of own ability & peer ability and GPA…results suggest a positive effect.

​

Similar effects on social outcomes (joining a fraternity, etc.)…

Network Experiments & Interventions

Exogenous networks (W)

11 of 33

Network Experiments & Interventions

Exogenous networks (W)

12 of 33

Network Experiments & Interventions

Exogenous networks (W)

13 of 33

Network Experiments & Interventions

Exogenous networks (W)

14 of 33

Note follow-up work (by Salganik & Watts) suggests the effects are weak and not long lasting

Network Experiments & Interventions

Exogenous networks (W)

15 of 33

16 of 33

Network Experiments & Interventions

Exogenous networks (W)

Control

treatment

17 of 33

Network Experiments & Interventions

Exogenous networks (W)

18 of 33

Network Experiments & Interventions

Exogenous networks (W)

SEPTEMBER 2010 VOL 329 SCIENCE

19 of 33

Network Experiments & Interventions

Exogenous networks (W)

20 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

21 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

22 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

23 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

24 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

25 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

26 of 33

Network Experiments & Interventions

Exogenous networks (W): Issues with experimental assignment

Sharique Hassan

27 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

4 types of interventions:

  1. Individuals
    1. Finding opinion leaders or flow blocking nodes that play a key role in the network process. 🡪 usually some centrality score, or an adaptive algorithm. Here highlighted “keyplayer” nodes.

​

  • Segmentation
  • Induction
  • Alteration

28 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

4 types of interventions:

  1. Individuals
  2. Segmentation
    1. Use communities to break the groups into parts, treat some use others as controls.
  3. Induction
  4. Alteration

29 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

4 types of interventions:

  1. Individuals
  2. Segmentation
  3. Induction
    1. Enhance relations & communication
  4. Alteration

30 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

4 types of interventions:

  1. Individuals
  2. Segmentation
  3. Induction
  4. Alteration
    1. Programs that seek to change the shape of the network, add/remove ties or nodes

31 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

32 of 33

Network Experiments & Interventions

Exogenous Behavior (Y)

Exploit the “Friendship Paradox” idea…

33 of 33

Network Experiments & Interventions

conclusion

Many examples of network peer effects driven by experiments.

​

Many examples of how manipulations can change the network itself

​

I think there’s sufficient evidence to start pushing harder against the skeptics: peer influence is clearly real; question is now how do we leverage it most effectively?

​