Hearing About A Job: Networks, Information, and Labor Market Segregation Katherine Stovel and Christine Fountain University of Washington March
Labor Market Segregation Reskin, Hargens, and Hirsch 2004
Classical explanations for segregation “Demand-side” Employer preferences for- or against- particular groups of workers (blacks, women, immigrants) Includes idea of labor queues “Supply-side” Difference in skill level between groups Difference in preferences of workers for types of jobs
But how do workers and jobs find each other? The seeking/recruiting/matching process… Costly or scarce information Structured channels of information flow Sorting, sifting, and exerting influence
Networks and Job Matching Important features of networks: Structure: Clustered or expansive Granovetter and many others Composition: integrated or segregated Baron and Podolny, Heckathorn, others
A more comprehensive view of where segregation comes from Is a taste for discrimination necessary? Are there other conditions that can generate segregation in labor markets? Skill distributions Network structures
Our approach: Stochastic Dynamic Agent-Based Models Laboratory for exploring theoretical principles Specify simple rules at micro level, examine macro-level patterns Used in studies of collective action, production of order, neighborhood segregation, diffusion, …
Selected recent examples of applying formal models to labor market hiring Calvo-Armengol and Jackson AER 2004 Rubineau and Fernandez unpublished ms 2006 Tassier JMS 2005, unpublished ms 2005 Fountain and Stovel unpublished ms 2005
Big Decision in Agent-Based Models: How to spend complexity chips?
Priorities in our artificial labor market Agents and jobs are heterogeneous Agents linked by (dynamic) network ties Information flows through networks Network structure is variable Preference regimes are variable
Our Framework Workers skills, attributes Workers and Jobs Matched Calculate utilities from regime and LM conditions Save Data Jobs attractiveness Managers attributes 100 iterations
Employer Preference Regimes Baseline All preferences = 0 Discriminatory preferences Employers prefer workers who share their attribute Skill-based preferences Employers prefer workers with higher skill Referral preferences Employers prefer workers referred by a current employee
Structures of Heterogeneity Workers have 2 characteristics: Attribute [0,1] ………………………think race, sex Skill Points [0-100] ………………...think education Managers have 1 characteristic: Attractiveness [0-100] ……………..think working conditions Association between skill and attribute? Association between attribute and network ties?
Association between skill and attribute 2002 CPS Demographic Supplement
Operationalizing association between skill and attribute Simulated Distributions = 50, s.d. = 10 = 50 = 50, s.d. = 10 = 30 = 70
P ij = probability of a tie between i and j Hij = indicator matrix containing 1 if attribute (i) == attribute (j) θ = tunable parameter governing in-group bias θ = 0 out-group bias θ =.5random graph θ = 1 in-group bias p =small underlying probability of a tie p = Simulated Distributions Operationalizing association between attribute and network ties
Levels of network homophily Simulated Networks N = 220, = 5 θ =.5θ =.9θ = 1
Employer Preferences Baseline Discrimination Skill-based Outcomes: Segregation (Index of Dissimilarity) Level of Homophily Population Characteristics Correlation between skill and attribute Experimental Conditions Network Characteristics Level of Homophily Static or Dynamic X X
Primary Outcome: Index of Dissimilarity Low Segregation High Segregation
Simulated Data Full information: Comparing Employer Preference Regimes
Simulated Data
Network restricted information Simulated Data θ Baseline Model Index of Dissimilarity
Network restricted information Simulated Data θ Discrimination Index of Dissimilarity
Network restricted information Simulated Data θ Skill Preference Index of Dissimilarity
Network restricted information Simulated Data θ Skill Preference Index of Dissimilarity
Network restricted information Simulated Data θ Skill Preference Index of Dissimilarity
Network restricted information Simulated Data θ Skill Preference Index of Dissimilarity
What produces firm-level segregation? Summary of results from static network models: Discriminatory preferences Skill preference + group-difference in skill Segregated nets
Where do segregated networks come from? Where:θ tunable parameter governing in-group bias α tunable parameter governing clustering Φ tunable parameter governing co-worker bias and P ij = probability of a tie between i and j 5% of ties re-wired at each iteration P ij = f(θ, α, Φ)
Incorporating a preference for ties to co-workers into a dynamic network model GSS Data
Building Network Segregation Simulated Data
Building Network Segregation Simulated Data
Building Network Segregation Simulated Data
Building Network Segregation Simulated Data
Building Network Segregation Simulated Data
Building Network Segregation Reducing Network Segregation Simulated Data
Building Network Segregation Reducing Network Segregation Simulated Data
Building Network Segregation Reducing Network Segregation Simulated Data
Summary Several mechanisms are sufficient to produce segregation in labor markets When networks are segregated and job- relevant information flows through nets, labor market segregation is substantial Workplace practices can increase network segregation
Implications: Networks and the New Economy
Future Directions Integrate other features of model Institutional variation Influence model (referrals) Examine mixed regimes Refocus data collection efforts at pre-hire phase
Acknowledgements Research Team: Yen-Sheng Chiang Stephanie Lee Anshuman Shukla Jim Moody Peter Hoff Funding National Science Foundation (SES , Stovel, PI) Center for Statistics and the Social Sciences, University of Washington