Heinz College Events

Statistical & Machine Learning Approaches to Network Experimentation Workshop

Statistical & Machine Learning Approaches to Network Experimentation Workshop

Co-sponsored by the Living Analytics Research Centre, Microsoft Research and Facebook

Workshop Details
The problem of how to design valid networked experiments has challenged statisticians and researchers across the social and computer sciences.  Networks by their very nature involve dependence among the nodes. So the units in a network are dependent. Treatments allocated to nodes propagate via network links and dynamics over time. Thus we expect substantial violations of the stable unit treatment value assumption (SUTVA) and treatment interference. Thus we ask: what does randomization buy us in experiments on networks? Or alternatively, how should one design a network-based experiment with randomization so that we can statistically analyze the results?
With partial support from Facebook and Microsoft Research through a grant to the School of Computer Science, Heinz College and CMU’s Living Analytics Research Centre are sponsoring a two-day workshop on Network Experimentation, to be held on April 22-23, 2013, beginning with a reception on the previous evening, April 21. We will be bringing together top researchers from economics, statistics and machine learning with special interests in (a) the design of network experiments; b) the evaluation of results of experiments carried out in a networked environment in order to infer causation; (c) inferring latent network structure from transaction data distributed over time and space. While inferring latent network structure is interesting in and of itself, can we connect this to network experiments? For example, should we infer latent structure as a first step in designing network experiments so that we are aware of observed and latent networks?
We invite you to present your related work as part of the workshop, and we will cover your travel expenses and accommodations in Pittsburgh.
We look forward to your response and to what we hope will be your participation in this enterprise.

Manuscript link >>

~ Sunday, April 21, 2013 ~

6:00-7:30 pm:  Dinner at Avenue B restaurant (reservations now closed)

                         5501 Centre Avenue, Pittsburgh, PA (412.683.3663)


~ Monday, April 22nd ~

9:00am – 9:15:  Welcome and Introductions – Steve Fienberg and Ramayya Krishnan

Morning Session/Day 1:

9:15 – 11:55am: Context Setting Presentations

            1. Network Models and Dependence (David Krackhardt) 9:15 – 10:00 am

            2. Randomization and Experimentation (Donald B. Rubin) 10:10 – 10:55am

            3. Economic Perspectives (Charles Manski) 11:10 – 11:55am

Noon to 1:15pm:  Lunch

Afternoon Session/Day 1:

1:30 – 3:30: Network Experimentation Approaches I

            1. Edo Airoldi: (Title Forthcoming)

            2. Dean Eckles: “Estimating peer effects with mechanism experiments, observational data, and    

                 encouragement designs”

3. Johan Ugander: “Clustered Graph Randomization: Network Exposure to Multiple Universes”

            4. Andrew Thomas/Michael Finegold:” Protocols for Randomized Experiments to Identify

                Network Contagion”

3:30 – 4:00pm:  Coffee

4:00 – 5:00pm:  Network Experimentation Approaches II and Discussion

            1. Deepak Agarwal: “Content Recommendation on Web Portals”

            2. Sharique Hasan: “The Longitudinal Effect of Contact Opportunity on the Formation of Inter-caste

                Social Relations”

5:45 – 7:45: Workshop dinner in Connan Room (1st floor)/ University Center (reservations now closed)


~ Tuesday, April 22 ~

Morning Session 1/Day 2:

9:00 – 10:00am:  Network Experimentation Approaches III        

1. Akhmed Umyaro: “Do Your Online Friends Make You Pay? A Randomized Field Experiment in an
Online Music Social Network”

2. Ee-Peng Lim/Michael Finegold: “Reviving Dormant Ties in an Online Social Network Experiment”

10:00 – 10:15:  Coffee

Morning Session 2/Day 2:

10:15 – 11:15am: Network Experimentation Approaches IV

            1. Pedro Ferreira: “The Impact of Likes on the Sales of Movies in Video-on-Demand: a Randomized


            2. Adrian Dobra: “The interplay between networks and Bayesian models in the analysis of cell phone data”

11:15 – noon:  Open Discussion

Noon – 1:15pm:  Lunch

Afternoon Session/Day 2:

1:30 – 3:00pm:  Towards a Research Agenda for Network Experimentation

            Designated Discussants:  Guido Imbens and Paul Holland

3:00:  Open comments

4:00:  Workshop ends

Event InformationLocation Information
All occurrences of this event have expired. Please use the contact information to contact the event organizer for information regarding upcoming occurrences of this event.