Manipulation-Proof Machine Learning for Resource Allocation

A large field experiment in Kenya provides the empirical basis for this research. The dataset likely contains behavioral data used to test a new class of strategy-robust estimators designed to remain stable when individuals can manipulate their input features. This work explicitly models the costs of manipulating different behaviors to identify decision rules that are stable in equilibrium.

Use Cases

• Testing strategy-robust estimators based on behavioral manipulation costs.
• Analyzing equilibrium outcomes of transparent decision rules in resource allocation.
• Validating machine learning models for fairness under strategic adaptation.
• Studying the relationship between feature manipulation costs and estimator stability.

Strengths

• Based on a large field experiment, suggesting real-world empirical validation.
• Methodology is explicitly described, modeling costs of manipulating different behaviors.
• Cross-platform presence on PapersWithCode indicates academic relevance and reuse.

Limitations

• Critical metadata such as row count, column names, and file size are completely unknown.
• There is a conflict in license information between sources stating 'Open Access (green)' and 'closed'.
• The dataset's structure and specific variables are not described, limiting reproducibility.

Provenance

Source

Joshua Blumenstock, University of California, Berkeley

Collection Method

Data collected through a large field experiment.

Geography

Kenya