October 7, 2009

The Dawn of Hybrid Audience Measurement

Magid M. Abraham, Ph.D.
Co-Founder

It is often said that the Internet is the most measurable medium. This is certainly true of measuring ad impressions and click through rates, which are critical elements of advertising measurement. However, measuring audiences has been far more challenging, particularly in light of the ongoing differences between server logs, which represent a census of usage, and panel based estimates provided by third party companies including comScore. The purpose of hybrid measurement is to derive audience projections that reconcile to census usage counts and deliver the key metrics of unduplicated reach and audience demographics required for media research.

An obvious question is why not use server data alone since it represents a census of usage? Server data accurately measures usage tonnage: page views (PV) and sessions, but its measurement of usage time is inaccurate. Furthermore, unique user counts are primary based on unique cookies. Cookies are well known as a poor surrogate for persons. Multiple users on the same machine are represented by one cookie, whereas a single user using multiple browsers on the same machine or using multiple machines can be represented by multiple cookies. When users delete cookies they get counted multiple times, which can be as often as they visit a site, if they set their browsers to delete their cookies after every session. Finally, there is a problem with machines that reject cookies. In addition, a key requirement for audience measurement is to measure the audience of not just one site, but the unduplicated reach of multiple sites, to enable the evaluation of media plans t spanning multiple sites. Server logs typically record usage of sites owned or operated by a single publisher, and therefore do not allow for cross-site overlap with third party or competitive sites that could be used as part of campaign.

Panel measurement tracks an individual person with known demographics across sites and time without reliance on cookies. Audience reports include demographic composition, unduplicated reach and frequency, and rich segmentation based on behavior such as content consumption or purchasing activity. However, a panel is based on a sample of people and therefore is subject to bias and sampling errors. In particular, it is difficult to represent work usage at large enterprises where IT security policies prohibit the download of panel metering software. Finally, niches sites may be difficult to represent via a panel designed to mirror a mainstream audience.

Hybrid measurement is a solution that combines the best aspects of server side census measurement, and of person based panel measurement. The basic idea is that census measurement provides an actual count of PVs which can be used to anchor audience measures. On the other hand, panels provide rich information on person-level usage and demographics. The combination allows to match the PV counts and to estimate number of unique persons visiting a site or a group of sites, in a manner that reconciles with the census level page views. It offers a ‘one number’ solution that eliminates the inherent conflict in estimates between web analytics and audience measurement reports and allows publishers to focus on building their business rather than argue about metrics.

There is general industry agreement that this the right conceptual approach. However, the devil is always in the details and a number of requirements must be met for this to fulfill industry needs:

  1. Audience counts must reflect people, not machines, browsers or cookies.
  2. Rigorous standards must be used to ensure that a site’s tags fairly represent its audience. Tagging is an error prone process, and care must be exercised to ensure that all pages are tagged, but also that only legitimate pages, those that are actually requested by user, are tagged once and only once in order to prevent rival web sites from gaming the system to boost their audience. Audit safeguards must be systematically deployed to ensure accuracy and fairness across sites.
  3. The hybrid estimation methodology must fundamentally account for the drivers of differences between panel and census pages views include not only differences in unique visitors, but also differences in usage frequency and usage intensity. Just patching census usage metrics and panel unique visitors on the same report is naïve and misleading.

Hybrid measurement can be used to measure all types of digital media where a census count is available. This includes the web, online video, the mobile web, online games, and digital advertising campaigns. It will be also increasingly applicable to TV measurement as census counts of TV usage become readily available from digital set-top boxes. It is the dawn of a true revolution in media measurement.

More about

TV