📐 Causal inference and econometrics: Smoothing event-study estimates, Low-rank generalized least squares, Monte Carlo tests for estimators, etc.
🧪 ML predictions as measurements: Classifier error in exposure comparisons, Label audits for machine-scored experiments, etc.
📋 Data collection: Efficient labeling for share regressions, Optimal survey sample allocation, etc.
🎯 Calibration and cutoffs: Probability calibration with fewer ties, Streaming survey weights, Optimal classification thresholds, etc.
⚓ Stability: Auditing prediction instability, Bootstrap consistency training, Reducing forgetting during retraining, etc.
🧭 Less greedy methods: Bagged forward stepwise regression, Decision trees from recurring splits, etc.
🔎 Joins: High-precision record linkage, Record linkage that preserves groups, Randomization inference with uncertain linkage
〰️ Smoothing, binning, and dimension reduction: Local trends in noisy time series, Kernel bandwidth selection, Dimension reduction by projection pursuit, etc.
💬 Measuring political knowledge and learning: Learning estimates adjusted for guessing, Hidden political knowledge?, Survey design and partisan knowledge gaps, etc.
🗳️ Measuring opinion and deliberation: Who has unstable political attitudes?, Auditing deliberative polls, Measuring support for militant groups, etc.
⚖️ Benchmarks: Designing and auditing ML benchmarks, AI benchmark for econometrics
📚 Other: R statistical tools for AI assistants, Learning from data, Discriminant adaptive nearest neighbors, etc.