Measurement science
Advertising measurement glossary
Measurement terms are often used as if they settle the argument. In practice, the same words can describe a clean causal test, a directional diagnostic, or a dashboard convention. This glossary defines the words that decide how much confidence a reader should place in a media claim.
Use it before a vendor call, planning meeting, article edit, or budget readout. The goal is not jargon fluency. The goal is to keep each term attached to the evidence it can actually support.
Jump to the measurement term.
Start with the word that appears in the report, then use the definition to check the comparison, maturity, and claim boundary.
Move from the term to the evidence check.
Start with the word that is doing the most work in the readout, then open the route that tests whether the available evidence can support that wording.
Start with the claim type
| Term | Plain meaning | Ask before trusting it |
|---|---|---|
| Descriptive claim | What was observed in the data. | Is the population, denominator, time window, and tracking limit visible? |
| Comparative claim | How one group, channel, period, or tactic differed from another. | Are the groups comparable before the result was measured? |
| Causal claim | What changed because of the campaign, source, or intervention. | What counterfactual shows what would have happened anyway? |
| Predictive claim | What a model expects will happen next. | Was the forecast tested out of sample, and does it support the decision being made? |
Core measurement terms
Incrementality
The difference between what happened with the campaign and what would have happened without it.
Do not use it as: a synonym for conversions, attributed revenue, or post-launch growth.
Better question: What control, holdout, model, or market comparison estimates the missing baseline?
Counterfactual
The best estimate of the outcome that would have occurred if the measured action had not happened.
Do not use it as: a vague alternative scenario chosen after results are known.
Better question: Was the comparison defined before the readout?
Attribution
A rule or model that assigns observed conversions to touchpoints in a tracked path.
Do not use it as: proof that the credited touchpoint caused the conversion.
Better question: Is this report describing contact history or estimating causal lift?
Attribution window
The lookback or conversion period during which a touchpoint is allowed to receive credit for an observed outcome.
Do not use it as: evidence that the credited outcome was incremental.
Better question: Was the window selected before launch, and how sensitive is the result to shorter or longer windows?
Conversion lag
The time between exposure, click, or visit and the later outcome that appears in reporting.
Do not use it as: an excuse to treat late credited outcomes as caused by the campaign.
Better question: Which cohorts are mature enough to judge, and which outcomes arrived after the first readout?
Marketing mix modeling
An aggregate model that estimates how spend and business drivers relate to outcomes over time.
Do not use it as: proof that every channel estimate is causally identified.
Better question: What controls, priors, calibration tests, and uncertainty ranges were used?
Conversion lift
An estimate of incremental conversions under a defined treatment and comparison design.
Do not use it as: a universal campaign multiplier for future audiences and seasons.
Better question: Were treatment and holdout groups protected from leakage?
Geo lift
A market-level test that compares treated geographies against control or matched markets.
Do not use it as: a clean experiment when pre-period trends diverge.
Better question: Did the markets move similarly before the campaign changed anything?
Brand study
A survey-based readout of changes in recall, awareness, favorability, consideration, or intent.
Do not use it as: automatic proof of sales lift or profitable media impact.
Better question: Are exposed and control respondents comparable, and does the outcome match the decision?
Attention metric
A diagnostic for whether an impression had a stronger opportunity to be noticed.
Do not use it as: a universal currency for business value.
Better question: What buying, creative, or placement decision will the attention signal improve?
Selection bias
A distortion caused when measured groups differ before the campaign, story, or intervention is evaluated.
Do not use it as: a generic criticism for any imperfect sample.
Better question: What pre-existing difference could explain the result without the campaign causing it?
Base rate
The starting frequency or normal level of an outcome before a claim or comparison is interpreted.
Do not use it as: background trivia that can be omitted from a strong-sounding percentage.
Better question: Does the result still matter when the starting level is visible?
Confidence interval
A range that communicates statistical uncertainty around an estimate under stated assumptions.
Do not use it as: proof that the design was unbiased or that the estimate will generalize.
Better question: Is the interval narrow enough for the decision, and are the assumptions credible?
Calibration
The process of checking or adjusting a model against stronger evidence, often experiments or known outcomes.
Do not use it as: a cosmetic model tuning step with no independent check.
Better question: What evidence showed the model's estimates were directionally reliable?
Terms that need boundaries
| When someone says | It may mean | Boundary to add |
|---|---|---|
| "The campaign drove sales" | Sales rose after launch, or a model attributed sales to the campaign. | Name the counterfactual and the comparison group. |
| "The audience performed better" | The selected group bought more, clicked more, or converted more. | Check whether the audience was already more likely to buy. |
| "Conversions came in over time" | The report is using an attribution window, conversion lag curve, or late outcome update. | Separate mature cohorts, credited outcomes, and incremental outcomes. |
| "The model is accurate" | The model predicted observed outcomes well. | Separate predictive accuracy from causal accuracy. |
| "The lift was significant" | The estimate cleared a statistical threshold. | Ask whether the effect size matters for the decision. |
| "The brand improved" | A survey metric moved in the desired direction. | Identify the respondent pool, question wording, and business decision. |
| "Attention was higher" | An impression had stronger viewable or noticed conditions. | Explain what outcome attention is expected to improve. |
How to use this glossary in a readout
- Underline every term that sounds causal: drove, caused, incremental, impact, lift, attributed, optimized.
- Rewrite each term as descriptive, comparative, predictive, or causal.
- Ask whether the evidence design matches the strongest word in the sentence.
- Replace over-strong wording with a narrower sentence when the design is weaker than the claim.
- Link the decision to a method: attribution for path reporting, lift for causal impact, MMM for planning ranges, brand studies for perception movement, and attention for exposure quality.
Takeaway
A measurement term is only useful when it tells the reader what evidence exists, what comparison was made, and what decision the result can support. Clear terms do not make a weak design strong, but they keep weak evidence from sounding stronger than it is.