First, define the causal estimand as the effect of the surge recommendation on completed trips, accounting for both noncompliance and interference. Then, propose a two-stage approach: use instrumental variables (IV) to handle noncompliance and a design-based or model-based method to address interference, such as cluster randomization or spatial regression discontinuity. Finally, outline sensitivity analyses to test assumptions.
Pro tip: Emphasize that in ride-hailing, interference is often the norm, so proactively suggest designs like switchback or geo-based randomization, and clarify that IV estimates a local average treatment effect (LATE) for compliers.
Clarify that the goal is the effect of the surge recommendation on completed trips, and state assumptions like exclusion restriction for IV and no interference between clusters.
Use random assignment as an instrument for actual surge acceptance, estimating the LATE via two-stage least squares or Wald estimator.
If interference is present, consider cluster randomization (e.g., by zone) or spatial models that account for spillovers, such as a spatial autoregressive model or difference-in-differences with spillover terms.
Combine methods: e.g., use IV within a cluster-randomized design, or apply a spatial IV approach. Estimate the effect on completed trips, possibly using a two-stage least squares with spatial lags.
Test sensitivity to interference assumptions, check for weak instruments, and validate with placebo tests or alternative specifications.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.