Apply instrumental variables under interference

Assume a clean A/B test cannot be run for a new ride-sharing feature because of interference. Suggest an instrumental-variables design to estimate the feature's causal impact on trip volume. Specify and support all IV assumptions rigorously: relevance, exclusion, independence (as-if random), and monotonicity if the estimand is LATE. Provide at least two concrete, plausibly exogenous instruments—for example, staggered eligibility for driver app versions, or exogenous weather shocks that move demand without directly touching the feature—and write down the first-stage and second-stage (2SLS) equations. Explain how you would check for weak instruments using the first-stage F-statistic, conduct over-identification tests such as Sargan or Hansen, deal with clustering and heteroskedasticity, and evaluate exclusion-rule violations when marketplace spillovers are present. Would a setting with effectively unlimited supply strengthen or weaken the credibility of the exclusion restriction, and why? If some assumptions fail only partially, sketch sensitivity analyses or bounds, such as Conley-type approaches.

Overview: The question tests causal inference with instrumental variables when interference is present, checking whether candidates can define units and aggregation levels for market-level spillovers, state IV assumptions (relevance, exclusion restriction, independence, monotonicity), and set up estimation frameworks like two-stage least squares. It is common in Statistics and Math interviews for data scientist roles because networked marketplaces make simple A/B tests invalid; the item belongs to econometrics/causal inference and mainly measures practical use of IV methods, along with conceptual grasp of identification, robustness diagnostics, and sensitivity analysis.