What problem does it solve? Standard randomized A/B tests are often impossible due to market-level rollouts, network spillovers, or ethical constraints, leaving teams unable to measure the true causal impact of product and policy changes. ## Core Features & Use Cases - Quasi-Experimental Identification: Selects and applies the correct causal method—Difference-in-Differences, Synthetic Control, Regression Discontinuity, Instrumental Variables, or Propensity Score Matching—based on your data structure. - CUPED Variance Reduction: Uses pre-experiment covariates to reduce metric variance and increase statistical power in randomized experiments. - Assumption Verification & Robustness: Validates parallel trends, donor pool weights, and running variable continuity, then runs placebo tests and sensitivity analyses (Oster's delta, Rosenbaum bounds). - Use Case: You launched a pricing policy in three test cities and need to measure its causal effect on conversion. The skill designs a Synthetic Control analysis with a donor pool, placebo permutation tests, and confidence intervals translated into business terms. ## Quick Start Ask the agent to design a difference-in-differences or synthetic control analysis for your market-level rollout, providing the intervention context, outcome metric, and historical baseline data.
