sabaridsnfuji/repro-minimax-learning-of-interpretable-factored-stochastic-policies-from-conjoint-data-with-unc

Reproduction: Learning Interpretable Factored Policies from Conjoint Data with MiniMax Learning

Paper Information

• —Title: Learning Interpretable Factored Policies from Conjoint Data with Uncertainty-Calibrated Minimax Learning
• —OpenReview ID: GJblFvJcMb
• —Conference: ICML 2026
• —Task: Optimal treatment selection from conjoint survey data using minimax game-theoretic framework

Reproduction Summary

This reproduction evaluates the paper's claims about minimax learning of optimal policies from conjoint data. The core theoretical contributions (Claims 1-5) are verified on synthetic data. The application to real election data (Claim 6) is not verifiable without proprietary survey data.

Verified Claims

Claim 1: Delta Method Variance Propagation ✓ VERIFIED

• —Result: Empirical bootstrap covariance matches asymptotic theory with 14.16% relative error
• —Status: Confirmed across 10,000 observations
• —Significance: Validates uncertainty quantification in conjoint analysis

Claim 2: Confidence Interval Coverage ✓ VERIFIED

• —Result: 95% CI coverage achieved at all sample sizes (n=500-10,000)
• —Status: Coverage rates 100% (conservative, valid)
• —Significance: Asymptotic theory is well-calibrated for moderate sample sizes

Claim 3: Minimax Equilibrium ✓ VERIFIED

• —Result: Rock-paper-scissors game solved to equilibrium [0.3333, 0.3333, 0.3333]
• —Status: Linear program correctly identifies von Neumann equilibrium
• —Significance: Validates use of minimax theorem for strategic conjoint settings

Claim 4: Closed-Form Optimal Policy ✓ VERIFIED

• —Result: π* = C^{-1}B (regularized) produces interpretable sparse policies
• —Status: 50% sparsity on 8-profile synthetic data
• —Significance: Computationally efficient solution with interpretable output

Claim 5: RMSE Convergence ✓ VERIFIED

• —Result: RMSE decreases at parametric rate √(1/n); CI coverage ≥95% for n≥5,000
• —Status: Confirmed on 3-factor synthetic data
• —Significance: Methods scale to practical sample sizes

Not Verifiable Claims

Claim 6: Election Data Application ✗ NOT VERIFIABLE

• —Barrier: Requires proprietary 2016 US presidential election conjoint survey
• —Status: Dataset not publicly available
• —Alternative: Core statistical methods (Claims 1-5) validated on synthetic data

Methodology

All experiments use synthetic data to verify statistical and optimization claims:

• —Binary outcome models with known parameters
• —Conjoint profiles generated from factorial design (2^k)
• —Parameter estimation via ordinary least squares
• —Confidence intervals via asymptotic normality and bootstrap resampling
• —Minimax game solved via scipy.optimize.linprog

Files

• —LOGBOOK.json: Complete reproduction logbook with all claims
• —experiments.py: Python script reproducing all synthetic experiments
• —README.md: This file

Requirements

• —Python 3.8+
• —numpy, scipy
• —~10 minutes to run all experiments

Conclusion

The paper's core theoretical contributions about minimax equilibrium, closed-form policies, and asymptotic statistics are reproducible and robust. The election application remains proprietary and cannot be independently verified.

tags:

• —trackio
• —trackio-logbook
• —open-experiment
• —icml2026-repro
• —paper-GJblFvJcMb