liminfoAttribution Model Reference
Free reference guide: Attribution Model Reference
About Attribution Model Reference
This Attribution Model Reference is a comprehensive guide covering the full spectrum of marketing attribution methodologies. It includes single-touch models (Last Click, First Click, Last Non-Direct Click, Last Platform Click), multi-touch models (Linear, Time Decay with half-life formula, Position-Based U-Shaped, W-Shaped for B2B), and data-driven models (DDA with Shapley Values in GA4, Markov Chain with removal effect, and Shapley Value game theory calculations).
The reference provides in-depth coverage of Marketing Mix Modeling (MMM) including regression model structure, adstock carryover effects with geometric and Weibull decay, saturation curves using Hill functions for diminishing returns, and open-source tools Meta Robyn (R) and Google Meridian (Python). Measurement infrastructure entries cover UTM parameters with naming conventions, conversion windows by platform and product type, MMP (AppsFlyer, Adjust, Branch, Singular), and server-side tracking with GTM Server Container and Meta CAPI.
Performance metrics (ROAS, iROAS with Geo Lift examples, CPA/CAC/LTV ratios), experimentation methods (RCT, Geo Lift Test with synthetic control, Conversion Lift Study), and privacy-centric strategies (SKAdNetwork SKAN 4.0, Google Privacy Sandbox, first-party data, Triangulation Framework combining MTA + MMM + incrementality) complete this marketing analytics reference.
Key Features
- Single-touch models: Last Click, First Click, Last Non-Direct Click (GA UA default), and Last Platform Click with deduplication guidance
- Multi-touch models: Linear (equal credit), Time Decay (half-life formula), Position-Based U-Shaped (40/20/40), W-Shaped (30/30/30/10 for B2B)
- Data-driven attribution: GA4 DDA with Shapley Values, Markov Chain removal effect calculation, full Shapley Value game theory formula
- Marketing Mix Modeling: regression structure, adstock geometric/Weibull decay, Hill function saturation curves, Meta Robyn and Google Meridian
- ROAS metrics: standard ROAS with channel benchmarks, iROAS with Geo Lift examples, break-even ROAS, CPA/CAC/LTV ratio analysis
- Measurement infrastructure: UTM parameter naming conventions, conversion windows by platform, MMP comparison, server-side tracking implementation
- Incrementality testing: RCT design, Geo Lift with synthetic control and CausalImpact, Meta Conversion Lift holdout methodology
- Privacy strategies: SKAdNetwork SKAN 4.0 postbacks, Google Privacy Sandbox APIs, first-party data CDP, Triangulation Framework
Frequently Asked Questions
What single-touch attribution models are covered?
Four single-touch models are documented: Last Click (100% credit to final touchpoint, best for low-consideration retargeting), First Click (100% to first touchpoint, best for brand awareness evaluation), Last Non-Direct Click (Google Analytics UA default, excludes Direct traffic as residual effect), and Last Platform Click (explains how Google/Meta each claim 100% causing double-counting, with MMP deduplication solutions).
How are multi-touch attribution models explained?
Four multi-touch models are covered with formulas: Linear (equal 1/N credit distribution), Time Decay (weight = 2^(-t/half_life) formula with worked example showing 10%/20%/30%/40% distribution over 14 days), Position-Based U-Shaped (40% first touch, 40% last touch, 20% distributed to middle), and W-Shaped (30% each to first touch, lead creation, and opportunity creation, with remaining 10% split among middle touches for B2B SaaS cycles).
What data-driven attribution methods are documented?
Three data-driven methods are covered: GA4 Data-Driven Attribution using Shapley Values to calculate marginal contribution (minimum 300 conversions/30 days required), Markov Chain Model using channel transition probability matrices and removal effect calculation (with Python ChannelAttribution package), and full Shapley Value game theory formula with worked example showing fair credit allocation across all possible channel coalitions with O(2^n) computational complexity.
How is Marketing Mix Modeling (MMM) presented?
MMM is covered across four entries: basic regression model structure (Sales = B0 + B1*TV + B2*Digital + ...), Adstock for carryover effects (geometric decay with lambda parameter, Weibull for asymmetric decay, half-life calculation), Saturation using Hill function S-curves for diminishing returns with optimal spend analysis, and open-source tools comparison between Meta Robyn (R, frequentist) and Google Meridian (Python, Bayesian) with code examples.
What ROAS and performance metrics are included?
Three metric entries are covered: standard ROAS (Revenue/Spend) with channel benchmarks (search 300-500%, social 200-400%, retargeting 500-1000%) and break-even ROAS calculation using margin rates; iROAS (incremental ROAS via Geo Lift showing standard ROAS of 10.0 vs true iROAS of 2.4); and CPA/CAC/LTV relationships with healthy LTV:CAC ratios (3:1+ healthy, 5:1+ growth opportunity).
What incrementality testing methods are documented?
Three experimentation methods are covered: RCT (randomized test/control with PSA ad substitute), Geo Lift Test (region-level ad on/off with synthetic control, 8-week pre-period + 4-week test, Python CausalImpact and Meta GeoLift packages), and Conversion Lift Study (Meta auto-splits 90% test/10% holdout, calculates lift percentage and incremental conversions, Google Ads Intent-to-Treat groups). An experiment design checklist includes power analysis, minimum duration, and p < 0.05 significance.
How are privacy-centric attribution strategies covered?
The privacy entry covers three pillars: Apple SKAdNetwork SKAN 4.0 (coarse conversion values, 3 postback windows, crowd anonymity thresholds), Google Privacy Sandbox (Attribution Reporting API, Topics API for interest targeting, Protected Audiences for remarketing), and first-party data strategies (login/membership data, server-side tracking with Enhanced Conversions, CDP implementation, Data Clean Rooms like Google ADH and Meta AEM).
What is the Triangulation Framework?
The Triangulation Framework cross-validates three attribution approaches: MMM (top-down) for long-term quarterly/annual budget optimization, MTA (bottom-up) for real-time daily/weekly campaign optimization, and Incrementality Testing (experiments) as ground truth for quarterly calibration. When all three methods agree, confidence is highest. In practice, MTA handles daily optimization, MMM guides quarterly budget allocation, and incrementality tests calibrate both models.