Information design for congested social services Motivation Given - - PowerPoint PPT Presentation

Information design for congested social services

Motivation Given that users of social services have heterogeneous needs, can information design help to target the service to those with high need? In this work:

• stylized queueing model serving users with heterogeneous needs.
• welfare under info. design against simple benchmarks (full-info and

no-info) and the first-best (i.e., centralized admission policies). Criteria: (ex ante) Pareto dominance. Take-away: With sufficient heterogeneity in need, information design can be powerful in improving overall welfare outcomes.

Authors: Jerry Anunrojwong, Krishnamurthy Iyer and Vahideh Manshadi. Paper: https://arxiv.org/abs/2005.07253

Model

Social service provider:

• unobservable FCFS queue
• single server, rate 휇

Heterogeneous need for service:

• high-need (H): must use the service
• low-need (L): have an outside option

No abandonment 푢푖(푘): utility from joining, if 푘 users ahead (zero utility for outside option)

• utside
• ption

휆H 휆L 휇

Low-need users are Bayesians, and maximize expected utility. SSP’s goal: share queue-size information to reduce congestion.

Results

Information design provides Pareto improvement in welfare of all types

• ver the simple mechanisms no-info and full-info
• 1. If 휆H < ¯

휆, then no-info is Pareto dominated.

• 2. With enough demand, full-info is Pareto dominated.

Under sufficient heterogeneity, information design can coordinate users’ actions to achieve the first-best:

• same welfare outcomes as centralized admission policies

Results

0.00 0.05 0.10 0.15 0.20 High-need welfare WH 0.00 0.01 0.02 0.03 0.04 0.05 0.06 Low-need welfare WL ap sm full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full full no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no

Welfare comparison, c = 0.15, λH = 0.8, λL = 0.2, µ = 1

ap sm full no

푢푖(푛) = 1 − 푐(푛 + 1) for each 푖 ∈ {L, H}.