Investigate visit–report correlation causality

You notice a positive correlation between visits to an ad's page and the likelihood that the ad is reported as bad. Determine whether higher page visits actually cause more reports, or whether exposure or selection confounds the relationship.

Give: (1) at least three concrete hypotheses (for example, a pure exposure effect, targeting bias toward sensitive cohorts, low-quality ads driving both attention and reporting, or instrumentation effects), and the falsifiable prediction each one makes; (2) a diagnostic plan with exact metrics and cuts: per-visit report rate versus visit-volume buckets; within-ad fixed-effects trends; cohorting by traffic source; funnel-based conditional probabilities; (3) a statistical model to estimate effect size while controlling for exposure: for example, an ad-day logistic or Poisson regression with an offset of $\log(\text{visits})$, ad fixed effects, hour-of-day and geo controls; report the coefficient interpretation and how you would check overdispersion and multicollinearity; (4) at least one quasi-experimental approach (for example, holdout geos, randomized delivery caps, or an IV using exogenous traffic shocks such as site outages) and how you would run a parallel A/A to validate it; (5) action criteria: define thresholds for an operational alert (for example, at least a 20% lift in per-visit report rate with $p<0.01$ after multiple-testing control) and what product or policy levers you would recommend under each outcome.

Overview: This question assesses a candidate's skills in causal inference, experimental design, diagnostic analytics, and statistical modeling with ad-level observational data.