Share

Copy link LinkedInXEmail

A parameter is a value a plan has to fix before a reading, a rule, an instrument or a move can be computed: a window, a threshold, a share or a cap, named once so every system that applies it applies the same one. The parameter is the slot and is governed here; the number in it belongs to the business and is a record, set over a window on a stated basis. Each parameter says what it is a parameter of, a metric, an attribution model, an evidence design, an instrument or a processing purpose, or which tactic it constrains.

What you can do with a parameter

• tableEvidence Plan · Settings

Its parameters— 11

• Attribution lookback windowThe number of days before a conversion within which an exposure may receive credit under an attribution model. It is disclosed with every attributed reading, because the same conversions attributed over one day and over thirty are two different numbers.
• Viewable pixel shareThe share of an ad's pixels that must be in the viewable area of the screen, for the required time, before an impression counts as viewable. The standard fixes it at half the pixels, and at less for the largest units.
• Viewable timeThe continuous time an ad's pixels must stay in view before an impression counts as viewable: one second for display and two for video under the standard.
• Minimum detectable effectThe smallest lift an experiment is designed to detect, stated before it runs. The sample and the holdout are sized to it, and an effect below it is not evidence of absence.
• Statistical powerThe probability an experiment detects an effect of the minimum detectable size when it is there. The convention is eighty percent, and a test run below it reads noise as no effect.
• Significance levelThe probability an experiment reports an effect that is not there. The convention is five percent, and it is set with the power rather than after the result.
• Holdout shareThe share of eligible people withheld from the activity in a holdout experiment, sized with the minimum detectable effect and the power rather than chosen for comfort.
• Incrementality factorThe multiplier a holdout establishes between an observed reading and its incremental value, applied to an observed rate or an attributed figure so a plan reports what the activity caused rather than what it touched.
• Frequency capThe most deliveries of a tactic one person may receive within a window, so reach is not bought by repetition and a stated preference is not worn down.
• Session timeoutThe interval of inactivity after which a visit ends and the next act starts a new one. Thirty minutes is the common default, and a reading of visits changes with it.
• Retention periodHow long personal data processed for a purpose is kept before it is erased or anonymised. The law requires a period and leaves its length to the purpose.

Why it matters

What goes wrong without it, as the commitments that license this kind say it, each standing on its sources.

• A value a plan fixes is a governed parameterA plan that runs on numbers nobody declared runs on numbers nobody can trace: a lookback window set in one platform's defaults, a frequency cap in another's, a visit timeout in a third's and a lapse rule in a spreadsheet, each changing a reading without changing its name. The standards bodies fix some of these values outright, the experimentation literature says the design parameters are set before a test and never after, and the regulation requires a retention period without naming one. Declaring the slot as vocabulary and the value as a record with its basis is what lets a business supply its own context without touching the model, and what lets an export carry the value a computation actually needs. The private platform this model learned from wrote the same rule, that a value store never defines meaning, and it rotted where the slots were named for the engine rather than the marketing and the citations were never made.
  Viewable Ad Impression Measurement Guidelines, Version 2.0 · Outcomes and Data Quality Standards · Digital Audience-Based Measurement Standards · Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing · The Unfavorable Economics of Measuring the Returns to Advertising · A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook · Regulation (EU) 2016/679, Article 5: Principles relating to processing of personal data · ISO 80000-1:2022 Quantities and units, Part 1: General