Spillover and Congestion Effects of Large-Scale Information Interventions

Abstract

We study the equilibrium impact of scaling personalized application guidance in Chile’s centralized school assignment system (SAE). A combined individual-level and market-level randomized design isolates the direct effect of information on treated families, congestion stemming from many neighbors reacting simultaneously, and spillovers onto untreated households. Personalized risk reports and tailored recommendations lead families—especially those initially at highest risk of remaining unassigned—to add and re-rank schools in ways that systematically lower their non-assignment probability. Treating entire local markets further reduces aggregate non-assignment rates without creating adverse congestion or negative spillovers, showing that large-scale information provision is a safe, effective complement to strategy-proof assignment mechanisms.

@article{ spillover-congestion-information_2025,
  title = "Spillover and Congestion Effects of Large-Scale Information Interventions",
  author = "Allende, Claudia and Neilson, Christopher A. and Ochoa, Fernando",
  journal = "Work in progress",
  year = "2025",
  note = "wip"
,
  abstract = "We study the equilibrium impact of scaling personalized application guidance in Chile’s centralized school assignment system (SAE). A combined individual-level and market-level randomized design isolates the direct effect of information on treated families, congestion stemming from many neighbors reacting simultaneously, and spillovers onto untreated households. Personalized risk reports and tailored recommendations lead families—especially those initially at highest risk of remaining unassigned—to add and re-rank schools in ways that systematically lower their non-assignment probability. Treating entire local markets further reduces aggregate non-assignment rates without creating adverse congestion or negative spillovers, showing that large-scale information provision is a safe, effective complement to strategy-proof assignment mechanisms."
,
  url_article = ""
}

Research design and key findings

This project evaluates the spillover and congestion effects that emerge when personalized application information is provided at scale within Chile’s centralized school assignment system (SAE). We combine an individual-level randomized control trial with a market-level cluster RCT covering 379,460 applicants. Local “markets” are defined via a DBSCAN clustering algorithm that groups applicants by home location and nearby school supply; clusters are randomly assigned to receive the personalized treatment or to remain as controls. A separate small-scale individual RCT keeps treated families at least 0.5 km apart to benchmark partial-equilibrium effects.

All families receive a generic report about the SAE process, while treated units also obtain a personalized email and WhatsApp delivery with nearby (≤2 km) schools and model-based admission probabilities. The dual experimental design lets us separately identify five forces: (i) individual treatment effects on application portfolios, (ii) average policy effects when entire markets receive guidance, (iii) spillovers on untreated neighbors, (iv) equilibrium effects on assignment risk and match quality, and (v) congestion from many households reacting simultaneously.

We find that treated families systematically expand and re-rank their rank-order lists, substantially reducing non-assignment risk, especially for applicants who started with the highest baseline risk. Market-level treatment reduces the share of unassigned students relative to control clusters, and we detect no adverse congestion or negative spillovers; if anything, small-scale RCTs modestly overstate risk reductions relative to market-level impacts. Overall, large-scale personalized information and recommendation policies can be implemented safely as a complement to centralized, strategy-proof assignment mechanisms.

Figures

• Coauthors: Claudia Allende, Fernando Ochoa
• Published: Work in progress
• Date: 2025-11-16
• PDF