0 Network Effects in Coordination Games Satellite symposium “Dynamics of Networks and Behavior” Vincent Buskens Jeroen Weesie ICS / Utrecht University.

Slides:



Advertisements
Similar presentations
Assumptions underlying regression analysis
Advertisements

Computational Statistics. Basic ideas  Predict values that are hard to measure irl, by using co-variables (other properties from the same measurement.
How to Schedule a Cascade in an Arbitrary Graph F. Chierchetti, J. Kleinberg, A. Panconesi February 2012 Presented by Emrah Cem 7301 – Advances in Social.
Network Matrix and Graph. Network Size Network size – a number of actors (nodes) in a network, usually denoted as k or n Size is critical for the structure.
Gibbs sampler - simple properties It’s not hard to show that this MC chain is aperiodic. Often is reversible distribution. If in addition the chain is.
Tacit Coordination Games, Strategic Uncertainty, and Coordination Failure John B. Van Huyck, Raymond C. Battalio, Richard O. Beil The American Economic.
CSE 522 – Algorithmic and Economic Aspects of the Internet Instructors: Nicole Immorlica Mohammad Mahdian.
Calibrated Learning and Correlated Equilibrium By: Dean Foster and Rakesh Vohra Presented by: Jason Sorensen.
Negotiation A Lesson in Multiagent System Based on Jose Vidal’s book Fundamentals of Multiagent Systems Henry Hexmoor SIUC.
Learning in games Vincent Conitzer
Socio-economic Factors influencing the use of coping strategies among Conflict Actors (Farmers and Herders) in Giron Masa Village, Kebbi State, Nigeria.
Describing Relationships Using Correlation and Regression
Correlation & Regression Chapter 15. Correlation statistical technique that is used to measure and describe a relationship between two variables (X and.
1 Duke PhD Summer Camp August 2007 Outline  Motivation  Mutual Consistency: CH Model  Noisy Best-Response: QRE Model  Instant Convergence: EWA Learning.
Variance and covariance M contains the mean Sums of squares General additive models.
Resampling techniques Why resampling? Jacknife Cross-validation Bootstrap Examples of application of bootstrap.
BA 555 Practical Business Analysis
Risk Attitude Reversals in Drivers ’ Route Choice When Range of Travel Time Information is Provided Jin-Yong Sung Hamid Hussain.
6/2/2001 Cooperative Agent Systems: Artificial Agents Play the Ultimatum Game Steven O. Kimbrough Presented at FMEC 2001, Oslo Joint work with Fang Zhong.
Outline  In-Class Experiment on a Coordination Game  Test of Equilibrium Selection I :Van Huyck, Battalio, and Beil (1990)  Test of Equilibrium Selection.
Joint social selection and social influence models for networks: The interplay of ties and attributes. Garry Robins Michael Johnston University of Melbourne,
Dynamic Network Security Deployment under Partial Information George Theodorakopoulos (EPFL) John S. Baras (UMD) Jean-Yves Le Boudec (EPFL) September 24,
© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license.
Chapter 19 Data Analysis Overview
Discrete Probability Distributions
Cornerstones of Managerial Accounting, 5e
Analysis of Individual Variables Descriptive – –Measures of Central Tendency Mean – Average score of distribution (1 st moment) Median – Middle score (50.
1 BA 555 Practical Business Analysis Review of Statistics Confidence Interval Estimation Hypothesis Testing Linear Regression Analysis Introduction Case.
CS401 presentation1 Effective Replica Allocation in Ad Hoc Networks for Improving Data Accessibility Takahiro Hara Presented by Mingsheng Peng (Proc. IEEE.
CORRELATIO NAL RESEARCH METHOD. The researcher wanted to determine if there is a significant relationship between the nursing personnel characteristics.
Relationships Among Variables
Multiple Linear Regression A method for analyzing the effects of several predictor variables concurrently. - Simultaneously - Stepwise Minimizing the squared.
CPS Learning in games Vincent Conitzer
Learning dynamics,genetic algorithms,and corporate takeovers Thomas H. Noe,Lynn Pi.
Models of Influence in Online Social Networks
Introduction to Inferential Statistics. Introduction  Researchers most often have a population that is too large to test, so have to draw a sample from.
Presented by Qian Zou.  The purpose of conducting the experiments.  The methodology for the experiments.  The Experimental Design : Cohesion Experiments.
Chapter 10. Sampling Strategy for Building Decision Trees from Very Large Databases Comprising Many Continuous Attributes Jean-Hugues Chauchat and Ricco.
Model Selection1. 1. Regress Y on each k potential X variables. 2. Determine the best single variable model. 3. Regress Y on the best variable and each.
Module 1: Statistical Issues in Micro simulation Paul Sousa.
International Environmental Agreements with Uncertain Environmental Damage and Learning Michèle Breton, HEC Montréal Lucia Sbragia, Durham University Game.
A Graph-based Friend Recommendation System Using Genetic Algorithm
Various topics Petter Mostad Overview Epidemiology Study types / data types Econometrics Time series data More about sampling –Estimation.
Biostatistics Class 6 Hypothesis Testing: One-Sample Inference 2/29/2000.
The Group Lasso for Logistic Regression Lukas Meier, Sara van de Geer and Peter Bühlmann Presenter: Lu Ren ECE Dept., Duke University Sept. 19, 2008.
© 2012 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part, except for use as permitted in a license.
© Copyright McGraw-Hill Correlation and Regression CHAPTER 10.
Yongqin Gao, Greg Madey Computer Science & Engineering Department University of Notre Dame © Copyright 2002~2003 by Serendip Gao, all rights reserved.
Chapter 11 Statistical Techniques. Data Warehouse and Data Mining Chapter 11 2 Chapter Objectives  Understand when linear regression is an appropriate.
© 2006 by The McGraw-Hill Companies, Inc. All rights reserved. 1 Chapter 12 Testing for Relationships Tests of linear relationships –Correlation 2 continuous.
Chap 18-1 Copyright ©2012 Pearson Education, Inc. publishing as Prentice Hall Chap 18-1 Chapter 18 A Roadmap for Analyzing Data Basic Business Statistics.
Designing Games for Distributed Optimization Na Li and Jason R. Marden IEEE Journal of Selected Topics in Signal Processing, Vol. 7, No. 2, pp ,
Introduction to Statistical Models for longitudinal network data Stochastic actor-based models Kayo Fujimoto, Ph.D.
Networked Games: Coloring, Consensus and Voting Prof. Michael Kearns Networked Life NETS 112 Fall 2013.
Classification Ensemble Methods 1
Repeated Game Modeling of Multicast Overlays Mike Afergan (MIT CSAIL/Akamai) Rahul Sami (University of Michigan) April 25, 2006.
IJCAI’07 Emergence of Norms through Social Learning Partha Mukherjee, Sandip Sen and Stéphane Airiau Mathematical and Computer Sciences Department University.
MAIN RESULT: We assume utility exhibits strategic complementarities. We show: Membership in larger k-core implies higher actions in equilibrium Higher.
Vincent Conitzer CPS Learning in games Vincent Conitzer
Statistical Methods. 2 Concepts and Notations Sample unit – the basic landscape unit at which we wish to establish the presence/absence of the species.
Games, Strategies, and Decision Making By Joseph Harrington, Jr. First Edition Chapter 4: Stable Play: Nash Equilibria in Discrete Games with Two or Three.
Introduction to ERGM/p* model Kayo Fujimoto, Ph.D. Based on presentation slides by Nosh Contractor and Mengxiao Zhu.
Overfitting, Bias/Variance tradeoff. 2 Content of the presentation Bias and variance definitions Parameters that influence bias and variance Bias and.
OVERCOMING COORDINATION FAILURE THROUGH NEIGHBORHOOD CHOICE ~AN EXPERIMENTAL STUDY~ Maastricht University Arno Riedl Ingrid M.T. Rohde Martin Strobel.
Break and Noise Variance
Adjustment of Temperature Trends In Landstations After Homogenization ATTILAH Uriah Heat Unavoidably Remaining Inaccuracies After Homogenization Heedfully.
CASE − Cognitive Agents for Social Environments
I. Statistical Tests: Why do we use them? What do they involve?
An Introduction to Correlational Research
Jump-Shot Or Drive? (Using Mixed Strategy Nash Equilibria to Predict Player Behavior) By Patrick Long.
Presentation transcript:

0 Network Effects in Coordination Games Satellite symposium “Dynamics of Networks and Behavior” Vincent Buskens Jeroen Weesie ICS / Utrecht University

1 Introduction Actors have interactions while they are organized in networks How can we analyze the co-evolution of networks and behavior? –First, fixed networks –Second, dynamic networks –An example using coordination games

2 Introduction Examples of coordination problems –Driving on left or right side of the road –Meeting a friend in a train station with two meeting points –Smoking behavior among friends –More generally, emergence of conventions and norms

3 The Coordination Game Player 2 Player 1 b < c < a < d RISK = (a – b)/(a + d – b – c) XY Xa,ac,b Yb,cd,d

4 The Equilibria (X, X) and (Y, Y) are both Nash equilibria There is also a mixed equilibrium (Y, Y) is the payoff-dominant equilibrium (X, X) is the risk-dominant equilibrium if RISK > 0.5; (Y, Y) is the risk-dominant equilibrium if RISK < 0.5. The mixed equilibrium is risk dominant if RISK =.5.

5 The Problem Payoff-dominant equilibrium is better for both players, however, under some conditions the other equilibrium may emerge, especially when this is the risk- dominant equilibrium What is the role of the structure of the network in this process?

6 Theory on Local Interaction Depending on noise and type of learning –either “the risk-dominant equilibrium will emerge” (Ellison 1993, Young 1998: Ch.6) –or “payoff-dominant” or “mixed” absorbing states remain possible (Berninghaus and Schwalbe 1996, Anderlini and Ianni 1996). Closed neighborhood better than circle Neighborhood size: no effect (?) Neighborhood overlap promotes the payoff-dominant equilibrium

7 The Model Actors located on graphs (undirected ties) Actors play repeatedly coordination games with all neighbors At each point in time, actors play the same move against all their neighbors. Actors receive information about the proportion of neighbors that played X and Y

8 The Model Actors start with propensity 0.5 to play Y After each round, this propensity increases or decreases with 0.1 depending on the best-reply against the neighbors in the last round. In this simulation: 100 replications until convergence for each starting propensity.

9 The Networks and Risk The set of non-isomorphic connected networks with 2 to 8 actors (N = 12,112) Selection of networks with 9 to 25 actors (N = 100,502) Payoffs: integer values such that 0 = b < c < a < d = < a / (20 + a – c) = RISK <.905

10 Analytic Results RISK has a negative effect on reaching the payoff- dominant equilibrium (Y,Y); the effect is not linear but a step-function If RISK = 0.5, i.e., a – b = d – c, there are no network effects towards the payoff-dominant equilibrium Comparing RISK and 1 – RISK, all network effects are reversed; effects that work for RISK > 0.5 towards (Y,Y) work in the other direction for RISK < 0.5 We restrict ourselves to RISK > 0.5, i.e., where the risk- and payoff-dominant equilibrium do not coincide.

11 Analyses Predicting the expected proportion of actors in a given network that play Y after convergence for 14 categories of RISK >.5. Independent variables –Network size –Density (proportion of ties present) –Centralization (degree variance) –Segmentation (P 3 /P 2, where P i is de proportion of distances in the network larger than or equal to i) –Proportion of actors with an odd number of neighbors –Maximal degree in the network –Proportion of times not converged to ALL X or ALL Y

12 Regression for RISK-values

13 Network dynamics: Why Actors will avoid ties in which coordination fails and seek ties in which coordination succeeds Networks may segmentize, with different behaviors in segments. Potentially different network effects

14 Network dynamics: What limits number of ties? Few models adequately deal with explaining number of ties Theoretically, we should argue from goal attainment through ties, not through ties directly We know of no satisfactory simple solution

15 Networks dynamics: Assumptions At each time, with some probability, actors have the opportunity to relocate a tie one-sidedly. –No switch costs –Sequential changes, in random order Myopic decisions: relocate tie if this increases payoff. –Relocate tie to actor with whom coordination fails to one with whom coordination succeeds –No change in ties if payoff-irrelevant; otherwise network would never converge Obviously: Size and density do not change Unknown consequences for – Degrees and degree-variance change – Connectedness and segmentation

16 Simulation Initial networks : all non-isomorphic networks of size<=8, including disconnected networks One RISK value: maximal static network effects For each of these networks –Initial behavior and adaptation of propensities: as before –Iterate until convergence No actors wants to change behavior No actors wants to change ties –Convergence attained in all simulations; exceptions are possible (for instance 2-cycles)

17 Questions for analysis How does the proportion of Y-choices depend on the initial network and the tie- change rate? How does the probability that equilibrium consists of two norms (both X and Y choices) depend on the initial network and the tie-change rate? How does the final network depend on the initial network and the tie-change rate?

18 Regression of proportion of Y-choices in equilibrium Variable | Initial Final InitialFinal Size | Density | Initial DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Final DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Dynamics change rate | r2 |

19 Logistic regression of Multiple norms in Equilibrium Variable | Initial Final InitialFinal Size | Density | Initial DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Final DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Dynamics Change rate | + - -

20 Properties final networks Size and density are constant by construction Degree variance slowly increases with tie change rate Segmentation stays more or less the same for small tie-change rates but decreases rapidy for larger tie-change rates MaxDegree does not change for any tie-change rate The percentage of nodes with an odd number of neighbors does not really change

21 Associations of Initial and Final Network Properties higher tie-change rate correlation NoChange > DegreeVar MaxDegree PropOddDegree Segmentation Tau-b > Connected %final nets

22 Analyses to Be Done Repeated simulations: separate random variation from lack of fit/misspecification Larger networks, other values of risks Effects of other network characteristics (e.g., betweenness,..) Non-linearities in the effects Interaction effects between network characteristics Sensitivity of the analyses related to the sample of networks and the specification of the statistical model

23 “Methodological” conclusion We can derive testable hypotheses of network effects in interactions by –A large “systematic” sample of networks –Simulating an interaction process on this network –Calculate relevant network characteristics –“Predict” characteristics of (the equilibrium state of) the interaction process from initial network characteristics (network fixed) Similar approach with dynamic networks Selection appropriate statistical models is often non-trivial

24 The distribution of degrees of the final network Variable | DegreeVar MaxDegree PropOddDegr Density | Size | Initial DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Dynamics DYN2 | DYN3 | DYN4 | DYN5 | DYN6 | _cons | r2 |

25 Regression of Properties Final Network (continued) Variable | Connected Segmentation Density | Size | Initial DegreeVar | Segmentation | MaxDegree | PropOddDegree | Connected | Dynamics DYN2 | DYN3 | DYN4 | DYN5 | DYN6 | _cons | r2 |