Limitations section

[martinthomson]

There is a lot of this document that talks about what it can do, but that fails to account for potential misapprehension about what is possible.

This section attempts to enumerate limitations when it comes to using this API for the measurement of advertising effectiveness, particularly when it comes to producing information that is helpful in making decisions about where to invest in marketing.

I've put this up front, so the disclaimer is clear. The section is longer than many of the adjoining sections; I hope that conveys the right sort of message.

Thanks to @rickcentralcontrolcom for raising the underlying issue.

---

Preview | Diff

[rickcentralcontrolcom]

Hello Martin, Thank you for the note, and for taking the concern seriously. I have compared the currently posted 29 May draft ( https://www.w3.org/TR/attribution/) against a copy I saved from 25 May. I may be missing something, but I am unable to identify changes that materially address the issue I raised. The sentence in §1.6 stating that attribution “attempts to measure correlation” appears in both versions. If there is a new or moved section that I should be looking at, please point me to it. My concern is not that the draft lacks any caveat. It is that the current draft still repeatedly characterizes attribution as a way to identify advertising effectiveness, when the API as described produces aggregate information about associations between impressions and conversions. That distinction is fundamental. Before turning to specific passages, I want to restate the underlying issue because I think it risks getting lost in discussions about individual wording choices. The central problem is the longstanding and well-established distinction between correlation and causation. "Correlation does not imply causation" is among the most fundamental principles in statistics, economics, epidemiology, and scientific inference generally. It is often taught as a simple mnemonic because humans are naturally inclined to infer causality from observed associations. Yet that inference is frequently wrong. Roosters crow shortly before sunrise every day, but they do not cause the sun to rise. Ice cream sales increase during summer months, and deaths by drowning also increase during summer months, but neither causes the other. Both are driven by a third factor. The existence of a statistical relationship, even a strong one, does not establish that one event caused another. This confusion is hardly unique to this document. It has been a recurring challenge throughout the history of advertising measurement. Attribution systems observe exposures and subsequent outcomes, then identify statistical associations between them. That does not mean the advertising caused the outcome. Consumers already inclined to purchase are more likely to search, click, visit websites, engage with advertising, and ultimately convert. Distinguishing persuasion from pre-existing purchase intent is precisely the problem that causal inference methodologies seek to solve. Part of the difficulty is that causal assumptions are subtly embedded throughout ordinary language. Words and phrases such as: • effective advertising • ineffective advertising • perform best • works best • leads to • drives • causes • influence • impact • effect • improves advertising performance • enables more effective advertising • what works • what ads perform best • advertising effectiveness • return on advertising • successful advertising all carry an implicit causal meaning. Readers naturally interpret these phrases as statements about what advertising actually changed, not merely what advertising happened to be associated with. For that reason, I believe the Working Group should systematically review the document for language that implicitly converts correlation into causation. The issue is not confined to any single paragraph. It appears throughout the narrative framing of the specification. This matters because Attribution Level 1 is not merely a technical proposal. It is a proposed web standard that could shape industry practice for years. It carries the imprimatur of major standards and industry organizations, including W3C, and will inevitably be cited by vendors, platforms, agencies, consultants, publishers, and advertisers as an authoritative description of what attribution measurement is capable of determining. The Working Group has understandably devoted enormous care to the technical design, privacy safeguards, aggregation mechanisms, and implementation details of the API. I believe the same level of care is required in describing what the resulting measurements actually mean. If the specification repeatedly characterizes attribution as a means of determining advertising effectiveness, identifying what advertising works, or distinguishing effective from ineffective advertising, many readers will reasonably conclude that the standard itself endorses those claims. In my view, that would be a serious mistake. The consequences are not merely academic. Advertising measurement frameworks influence the allocation of hundreds of billions of dollars annually. When correlation is mistaken for causation, budgets tend to migrate toward channels that are especially good at capturing existing demand rather than creating incremental demand. Search, social, and retail media frequently benefit from this dynamic because they operate closest to observable conversion activity and possess extensive behavioral, identity, and optimization capabilities. Meanwhile, media channels whose effects are often delayed, indirect, probabilistic, or difficult to observe through attribution systems—including television, audio, premium video, sponsorships, and broader brand advertising—tend to be systematically undercredited. The result is not simply measurement error. It is the potential misallocation of billions of dollars of advertising investment and further economic pressure on media businesses that are already struggling to survive. A standards document should therefore be especially careful not to elevate an observational attribution framework into an implied measure of causal advertising effectiveness. The following passages in the 29 May draft still seem problematic. In the Abstract: Current language: “This specifies a browser API for the measurement of advertising performance. The goal is to produce aggregate statistics about how advertising leads to conversions...” Concern: “Advertising performance” and “how advertising leads to conversions” both suggest causal measurement. Suggested revision: “This specifies a browser API for privacy-preserving aggregate attribution reporting. The goal is to produce aggregate statistics about associations between advertising-related events and subsequent conversion events...” In §1.2 Background: Current language: “One characteristic that distinguished the Web from other venues for advertising was the ability to obtain information about the effectiveness of advertising campaigns.” Concern: This frames attribution-era web measurement as effectiveness measurement, rather than exposure, conversion, and association reporting. Suggested revision: “One characteristic that distinguished the Web from other venues for advertising was the ability to obtain timely information about ad exposures, interactions, and subsequent conversion events.” In §1.2 Background: Current language: “Having a detailed record of a person’s actions allowed advertisers to infer characteristics about people. Those characteristics made it easier to choose the right audience for advertising, greatly improving its effectiveness.” Concern: This again asserts improved effectiveness without distinguishing targeting efficiency, observed conversion rates, and incremental causal impact. Suggested revision: “Those characteristics made it easier to target audiences believed to be more likely to convert.” In §1.2 Background: Current language: “Advertisers seek to place advertising where it will have the most effect relative to its cost.” Concern: This is a reasonable business objective, but in context it implies that attribution identifies where advertising has the most effect. Suggested revision: “Advertisers seek to place advertising where they expect the greatest return relative to cost.” In §1.3 Goals / §1.4 transition: Current language: “The measurement of advertising performance creates new cross-site flows of information.” Concern: Again, “advertising performance” is broader than what the API establishes. Suggested revision: “Attribution reporting creates new cross-site flows of information.” In §1.4 End-User Benefit: Current language: “Support for attribution enables more effective advertising, largely by informing advertisers about what ads perform best, and in what circumstances.” Concern: This is the clearest overstatement. Attribution can identify which ads are associated with observed outcomes. It does not generally identify which ads caused incremental outcomes or “perform best.” Suggested revision: “Support for attribution provides advertisers with aggregate information about which ads, users, contexts, or circumstances are associated with observed conversion events.” In §1.4 End-User Benefit: Current language: “Connecting that information to outcomes allows an advertiser to learn what circumstances most often lead to the outcomes they most value.” Concern: “Lead to” implies causation. Suggested revision: “Connecting that information to outcomes allows an advertiser to observe which circumstances are most often associated with the outcomes they value.” In §1.4 End-User Benefit: Current language: “That allows advertisers to spend more on effective advertising and less on ineffective advertising.” Concern: This directly claims that attribution identifies effective and ineffective advertising. That is the core issue. Suggested revision: “Advertisers may use this information, together with other measurement methods, to inform campaign diagnostics, optimization, and investment decisions.” In §1.4 End-User Benefit: Current language: “Sites that provide advertising inventory... indirectly benefit from more efficient advertising. Venues for advertising that are better able to show ads that result in the outcomes that advertisers seek can charge more for ad placements.” Concern: This again treats attributed outcomes as evidence of advertising efficiency or causal contribution. Suggested revision: “Sites that provide advertising inventory may indirectly benefit when advertisers have better aggregate reporting about observed outcomes associated with ad placements.” In §1.6 Attribution Using Histograms: Current language: “Different groupings might be used for different purposes. For instance, grouping by creative (the content of an ad) might be used to learn which creative works best.” Concern: “Works best” implies causal creative effectiveness. A histogram can compare attributed outcomes by creative, but not necessarily incremental creative effect. Suggested revision: “Different groupings might be used for different purposes. For instance, grouping by creative might be used to compare observed attributed outcomes across creatives.” In §2 Overview of Operation: Current language: “Not displaying an advertisement (especially for controlled experiments that seek to confirm whether an advertising campaign is effective).” Concern: This is less problematic because it refers to controlled experiments, but it should be clearer that experimental designs are distinct from ordinary attribution reporting. Suggested revision: “Not displaying an advertisement, where the API is used as part of a controlled experimental design intended to estimate incremental advertising effects.” In §3.5 Requesting Attribution for a Conversion: Current language: “A site that observes a conversion might choose to request the measurement of the effect of different stored impressions.” Concern: This is another causal claim. The API allocates conversion value to stored impressions according to attribution logic. It does not measure the causal effect of those impressions. Suggested revision: “A site that observes a conversion might choose to request attribution reporting across different stored impressions.” I would recommend adding a clear statement near the front of the document, perhaps in §1.1 or §1.3, along these lines: “This API produces aggregate attribution reports describing associations between recorded impression events and subsequent conversion events. The API does not, by itself, estimate the causal or incremental effect of advertising. Claims about advertising effectiveness, incrementality, or causal lift require additional methodological assumptions or separate experimental or causal-inference designs.” That clarification would be much stronger than a passing reference to correlation in §1.6, especially while the rest of the draft continues to use terms such as “effective advertising,” “perform best,” “works best,” and “measurement of the effect.” I am not objecting to privacy-preserving attribution reporting as an engineering objective. My concern is that a W3C standard should not unintentionally endorse attribution reporting as a scientifically valid measure of advertising effectiveness unless the specification is much clearer about what the API can and cannot establish. Thanks, Rick Bruner CEO, Central Control, Inc. Newsletter <https://www.centralcontrol.com/newsletter> | ***@***.*** | www.centralcontrol.com Turn Ad Performance Certainty Into Market Advantage Free whitepaper: How to design large-scale geographic experiments for incrementality testing <https://www.centralcontrol.com/how-to-guide-georct?k=2693226e44>

…

On Fri, May 29, 2026 at 3:44 AM Martin Thomson ***@***.***> wrote: There is a lot of this document that talks about what it can do, but that fails to account for potential misapprehension about what is possible. This section attempts to enumerate limitations when it comes to using this API for the measurement of advertising effectiveness, particularly when it comes to producing information that is helpful in making decisions about where to invest in marketing. I've put this up front, so the disclaimer is clear. The section is longer than many of the adjoining sections; I hope that conveys the right sort of message. Thanks to @rickcentralcontrolcom <https://github.com/rickcentralcontrolcom> for raising the underlying issue. ------------------------------ Preview <https://pr-preview.s3.amazonaws.com/w3c/attribution/pull/441.html> | Diff <https://pr-preview.s3.amazonaws.com/w3c/attribution/441/b2461cc...b4863ff.html> ------------------------------ You can view, comment on, or merge this pull request online at: #441 Commit Summary - f447afe <f447afe> Limitations section - b4863ff <b4863ff> Typos, formatting File Changes (1 file <https://github.com/w3c/attribution/pull/441/files>) - *M* api.bs <https://github.com/w3c/attribution/pull/441/files#diff-721a3a36aef527eb7b18f2b40adf9ec0a981167e3fbc8fd4a3f4d8d96f51ad4d> (99) Patch Links: - https://github.com/w3c/attribution/pull/441.patch - https://github.com/w3c/attribution/pull/441.diff — Reply to this email directly, view it on GitHub <#441?email_source=notifications&email_token=BQM5UPTVJVSDUOB2L46EFX345E5WZA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF4ZTONRXGM2TKMZZHGTHEZLBONXW5J3NMVXHI2LPN2SWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BQM5UPRR4772LBNTR2J3WN345E5WZAVCNFSM6AAAAACZSB263GVHI2DSMVQWIX3LMV43ASLTON2WKOZUGU2DMNRXG43TMOA> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BQM5UPUYFUOZ2I5N3KKRDQD45E5WZA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF4ZTONRXGM2TKMZZHGTHEZLBONXW5J3NMVXHI2LPN2SWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/BQM5UPRXD34N4X473LTA7NL45E5WZA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF4ZTONRXGM2TKMZZHGTHEZLBONXW5J3NMVXHI2LPN2SWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you were mentioned.Message ID: ***@***.***>

[bmcase]

@rickcentralcontrolcom you don't see the changes in the spec yet because the PR has not been merged yet. If you look at this preview link https://pr-preview.s3.amazonaws.com/w3c/attribution/pull/441.html#limitations you can see the new limitations section.

[rickcentralcontrolcom]

Thanks, Ben. I found the preview and have now read the new "Limitations and Successful Use" section. I think this is a substantial improvement to the draft. In particular, I appreciate that the text now explicitly acknowledges that attribution can create a false impression of advertising efficacy and that randomized control trials (incrementality experiments) are necessary to measure causal effects. That addresses a significant part of my concern. However, after reading the new section in the context of the rest of the document, I think there remains a consistency issue. The new language correctly distinguishes between attribution (which measures associations) and experimentation (which measures causal effects). Yet other parts of the document still describe attribution in terms that appear to imply causal conclusions, such as learning what advertising "works best," identifying "effective advertising," or helping advertisers determine which actions lead to desired outcomes. If the document now acknowledges that attribution alone cannot establish causality, I think it would be worth reviewing the remaining text for terminology that may unintentionally suggest otherwise. More broadly, my concern is less about any specific implementation detail and more about avoiding a common and consequential misunderstanding: the assumption that observed associations between ad exposures and outcomes are evidence that the advertising caused those outcomes. The new section is a welcome step in that direction, but I believe the same principle should be reflected consistently throughout the specification. Thank you for taking the issue seriously and for adding the new material. Thanks, Rick Bruner CEO, Central Control, Inc. Newsletter <https://www.centralcontrol.com/newsletter> | ***@***.*** | www.centralcontrol.com Turn Ad Performance Certainty Into Market Advantage Free whitepaper: How to design large-scale geographic experiments for incrementality testing <https://www.centralcontrol.com/how-to-guide-georct?k=2693226e44>

…

On Fri, May 29, 2026 at 1:44 PM Benjamin M. Case ***@***.***> wrote: *bmcase* left a comment (w3c/attribution#441) <#441 (comment)> @rickcentralcontrolcom <https://github.com/rickcentralcontrolcom> you don't see the changes in the spec yet because the PR has not been merged yet. If you look at this preview link https://pr-preview.s3.amazonaws.com/w3c/attribution/pull/441.html#limitations you can see the new limitations section. — Reply to this email directly, view it on GitHub <#441?email_source=notifications&email_token=BQM5UPUKSAPWFDDPCCU22ML45HEBRA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINJXHAYTCNBRG4YKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4578114170>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BQM5UPQORK3PIBXNQ3ACFNT45HEBRAVCNFSM6AAAAACZSB263GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DKNZYGEYTIMJXGA> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BQM5UPXGHXEOTEUKWDLSKK345HEBRA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINJXHAYTCNBRG4YKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJKTGN5XXIZLSL5UW64Y> and Android <https://github.com/notifications/mobile/android/BQM5UPXCNYRHFJOHY35TW4345HEBRA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINJXHAYTCNBRG4YKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLTGN5XXIZLSL5QW4ZDSN5UWI>. Download it today! You are receiving this because you were mentioned.Message ID: ***@***.***>

[AramZS]

We will merge this on Monday unless there are any reasonable objections.

[bmayd]

I want to add a voice to the concerns articulated by @rickcentralcontrolcom and suggest that the document be revised to accurately characterize:

• What browser-based attribution is more broadly.
• What intelligence it can provide and what the limits of that intelligence are.
• What can be reasonably inferred from browser-based attribution more generally.
• Where browser-based attribution fits into larger frameworks that can provide evidence suggestive of advertising effectiveness.
• How the outputs from this API, which are an extremely narrow, purposefully limited and generalized subset of the evidence that can be gathered from browsers, compare with more common browser-derived datasets.
• What the outputs of this API can reliably convey.
• Where the outputs from this API meaningly fit into, and what they contribute to, ad-effectiveness measurement frameworks.

As it stands, the document directly asserts in myriad direct and indirect instances, many outlined by Rick above, that the API enables "the measurement of advertising performance"; that's inaccurate.

What the API enables is:

• Recording of very dimensionally limited and nondeterministically incomplete data about observable ad-related events.
• Outputs of a subset of those events, correlated using fixed, client-side rulesets constrained by the limits of available data that are unable to account for confounding factors and don't natively incorporate counterfactuals.
• Reporting of limited, noised, aggregate statistics about the correlated outputs.

In other words, the output of the API provides an extremely limited, approximate and incomplete snapshot of what happened during a campaign, while by design providing insufficient information to make meaningful inferences about why it happened or adjust for user intent or platform optimization bias.

At best it allows marketers to make limited assumptions about the effectiveness of their targeting and inventory supply in correlating with conversions, but it doesn't provide any insight into if or why the targeting or inventory was or was not effective and it shouldn't imply that it does. The API cannot differentiate between ads that induced a conversion (causation) and those that were merely incidental to it (correlation); it is strictly an observational tracking tool, not an incrementality tool.

Per Rick's comment above, the W3C, via this standard, "should not unintentionally endorse attribution reporting as a scientifically valid measure of advertising effectiveness unless the specification is much clearer about what the API can and cannot establish."

I recognize that an ask to adequately frame this proposal's relationship with rigorous frameworks for ad effectiveness is significant and probably beyond the reasonable scope of this primarily technical standard. If that's the case, I suggest as an alternative that the document be revised to clearly indicate the limits of the API regarding campaign effectiveness measurement, that it include references to resources that can clarify those limits and inform the proper use of this API and that any language suggesting it can directly measure ad effectiveness or campaign performance be modified to clearly indicate it can, at best, provide inputs as part of a properly controlled framework for the assessment of campaign effectiveness.

Suggested introductory abstract:

This document defines a client-side API designed to quantify observable, cross-site associations between digital advertising impressions and subsequent conversion outcomes while preserving user privacy. Crucially, the API does not measure advertising effectiveness or establish a direct causal link between an ad exposure and a user's behavior. Instead, it aggregates data on the chronological sequencing of a limited set of observable ad-related events within a single browser instance. The scope and utility of these insights are inherently bounded by the configurations chosen for measurement—specifically what events are tracked, when lookback windows are applied, and how assignment logic correlates those events. To approximate actual advertising efficacy or incrementality, implementations must supplement these associative baselines with rigorous external experimental designs, such as randomized control trials.