A transferability-based estimator combines censored purchase records with roughly 26,000 experimental observations to isolate incremental revenue across four distribution channels.
DML Quantifies Representation Error in Causal Estimates
Cross-fitted DML turns AI-learned controls into an inferential target, separating valid representation-dependent estimates from conditions for causal coverage.
An error-as-signal audit combines PatentSBERTa disagreements with two-model climate-function judgments, reducing the Y02 estimate from 592,387 to 441,468 patents.
Inventory A/B Tests Show Predictable Interference Bias
A pairwise randomization over items and time separates carryover from capacity crowding, reducing bias mechanisms that affect switchback and item-level designs.
Outlier Choices Rarely Move Means but Flip Marginal Decisions
A pre-registered reanalysis of 358 behavioral-science meta-analyses shows median mean shifts of at most 0.047 Cohen’s d but decision reversals up to 15.9%.
