Evaluate brand ads effectiveness on social media causally

Hypothesis: 'Social media (e.g., Facebook) is not effective for brand advertising compared with other channels.' You have historical multi-channel data and can run experiments. Design a causal measurement plan to test this claim. Be specific:

1. Define primary brand outcomes and how to measure them (e.g., aided/unaided awareness surveys, ad recall, branded search lift, direct type-in traffic, share-of-voice). Specify exact definitions and attribution windows.
2. Propose at least two experiment designs (within-platform holdout cell; geo/matched-market lift test) and choose one. State unit of randomization, eligibility rules, cooldown/lag windows for awareness effects, frequency capping, and contamination/spillover handling across friends/markets.
3. Power/MDE: outline inputs (baseline awareness, expected lift, intraclass correlation for geo tests), compute sample size and duration required; note assumptions you’d validate.
4. Analysis: pre-register a model (e.g., difference-in-differences with pre-period, or synthetic control). List covariates (seasonality, prior brand equity, competitor spend, creative quality, device mix). Specify estimand (ATE on exposed), robust SEs/clustering, and how you’ll handle staggered rollout.
5. Budget confounding: advertisers spend less on social brand—explain how you’ll adjust for spend levels and diminishing returns (e.g., log response curves, adstock/lag, saturation modeling) to avoid conflating budget with effectiveness.
6. Guardrails and constraints: CPA/ROAS on DR, site performance, user complaints; set failure stops.
7. Heterogeneity: plan pre-specified subgroup tests (age, market maturity, frequency buckets) and multiple-testing control.
8. Interference diagnostics: proximity/overlap checks, geographic buffers, placebo markets.
9. Decision rule: exact thresholds (e.g., lift ≥ X pp with 95% CI not crossing 0) and how results change next-quarter channel mix.