Description: This article is an overview of IPv4 and IPv6 with some FAQs, including their differences and our methodology. This AVG web page provides helpful information on IPv4 and IPv6.
What are IPv4 and IPv6?
The primary household identifier in our digital measurement is the household IP address (which we hash when using within our platform).
IPv4 and IPv6 are different versions of IP addresses, and they differ in structure, functionality, and, most importantly, the rate of rotation when measuring reach, frequency, and outcomes.
For any network, IP addresses change over time. The longer an IP address remains the same, the more accurate it is as a digital representation of a household. IPv6 addresses rotate more quickly than IPv4 IPs, although it can vary.
Our research shows that IPv4 addresses rotate on average every 2-4 weeks, and IPv6 addresses rotate on average every 1-2 days.
The following diagram outlines the rotational differences between IPv4 and IPv6:
The rotation rate impacts the calculation of reach and frequency at the household level. In the examples above, the IPv4 scenario indicates one household would see three adverts, whereas the IPv6 scenario indicates three households would each see one advert.
How do the differences impact InnovidXP methodology?
The risk with short rotation cycles is overcounting household reach and undercounting ad impression frequency, therefore, our InnovidXP methodology relies on IPv4 IPs. Innovid is investigating how to potentially treat or stabilize IPv6-related data to include these in our measurement methodology.
Why use IPv4?
While we know that using IPv4 may reduce the number of devices we capture measurement data from, it results in a more stable device universe. This is because the IPv4 data is rotated less.
IPv6 must be sufficiently stable to identify recurring impressions from the same household, as it degrades rather than enhances the quality of deterministic matches during the household resolution process.
How does Innovid manage any devices that are filtered out?
By filtering out IPv6-only devices from our reach and frequency calculations, we can use the more stable IPv4 addresses. We then apply extrapolation modeling approaches to provide a representative census-level view of overall household reach and frequency metrics.
Is Innovid confident with reach accuracy using the IPv4 methodology?
The reach extrapolation methodology was tested and validated using simulated datasets representing real-world data with known “unmatched” households.
When applying the extrapolation approach to these datasets, the upweighted reach values were consistently within +/- 1% of the expected simulated totals.
Multiple other reach extrapolation methodologies were tested against the simulated data. However, they resulted in more significant divergences from the expected values.