Attribution bias
Attribution bias is the systematic, predictable way a given model mis-assigns credit relative to true causal contribution. Last-click over-credits closing and demand-harvesting channels; first-click over-credits discovery; view-through can over-credit cheap impressions. Recognizing each model's bias is essential because no observational model recovers causation on its own.
What this means
Every observational attribution model imposes a credit rule, and every rule has a characteristic bias. Last-click systematically over-credits whatever tends to be last — retargeting, brand search, direct — because those channels intercept users already converging on a purchase. First-click over-credits discovery channels and ignores the closing work. Position-based hard-codes a belief about which touches matter.
These are not random errors; they are structural tilts that recur on every path of a given shape, which is why they qualify as bias rather than noise.
Why no model escapes it
Observational models only see correlations between recorded touches and conversions; they cannot observe the counterfactual of what would have happened otherwise. So they cannot, by themselves, distinguish a channel that caused a conversion from one that merely co-occurred with it. Data-driven models reduce some arbitrariness but still infer from observed paths, not experiments.
The corrective is to combine attribution with incrementality testing and media-mix modeling, which approximate the counterfactual. Treat attribution as a credit-allocation lens with a known tilt, validate its conclusions experimentally, and never read its credit as a causal claim. This page is educational, not statistical advice.
- Last-click over-credits closing/harvesting channels
- First-click over-credits discovery touches
- Observational models cannot see the counterfactual
How it appears in analytics and logs
Credit that consistently favors closing or harvesting channels under last-click is bias, not proof those channels caused the conversions.
Diagnostic use case
Anticipate a model's known biases before acting on its credit — for instance discounting last-click's favoritism toward retargeting and brand search.
What WebmasterID can help detect
WebmasterID's transparent first-party paths help you see which touches a model is over- or under-crediting and reason about its bias.
Common mistakes
- Reading model credit as proof of causation.
- Scaling spend toward channels that bias favors, like retargeting.
- Believing data-driven attribution removes all bias.
Privacy and accuracy notes
Attribution bias is a property of credit-assignment logic over recorded paths, not a privacy mechanism. This page is educational, not legal advice.
Related pages
- Attribution vs incrementality vs MMM
Attribution, incrementality testing, and marketing-mix modeling (MMM) are three distinct measurement approaches often confused. Attribution distributes credit across observed touches; incrementality experiments measure causal lift versus a control; MMM uses aggregate, often top-down regression on spend and outcomes. They answer different questions and should be used together, not treated as interchangeable.
- Incrementality testing: what attribution cannot tell you
Incrementality testing measures the lift a channel actually causes by withholding it from a control group and comparing outcomes. It answers the question every attribution model dodges: would this conversion have happened anyway? It is causal where attribution is merely correlational, but it requires deliberate experiment design.
- Last-click attribution: simple, and what it hides
Last-click attribution assigns 100% of a conversion's credit to the last touchpoint before it. It is simple, deterministic, and the historical default — which is exactly why it misleads: it ignores every earlier touch that created demand, systematically overrating bottom-funnel channels and underrating discovery.
- Attribution analytics
Inspect paths to see where a model tilts credit.
Sources and verification notes
- Google Analytics Help — Attribution and attribution modelingDescribes how different models assign credit, illustrating their biases.
Last reviewed 2026-06-24. Facts are checked against primary/official sources where available; uncertain specifics are marked “Data not yet verified” rather than guessed.