What about noisy conversions? [question]

[maudnals]

Question from developer:

Does the fact that there’ll be conversions that are not reported and that fake reports will be generated mean that the total number of conversions will be different from the actual measured number? Or will it be marginal?

[maudnals]

Answer:

Because of the randomized response in the new proposal, the total number of conversions that will be reported via the API will be different from the true number of conversions that did take place. It will be possible to recover an estimate of the true conversion count; the proposal editors will provide a script to help you do so. You can find a first version of this script here.

The noise correction script needs to be based on the noise value that is applied by the browser. In the new version of the API, this value is now easily visible to API users: it will be embedded in the reports.

Also note: In the next origin trial, consider using the temporary debugging /performance analysis mechanism. With this, you can measure differences between cookie-based and attribution-reporting-based measurement. But note that the randomized response noise will only account for some of the differences — rate limits, network details, or user-initiated data clearing will also create some differences.