MARCH 2014

UNDERSTANDING DIGITAL
AUDIENCE MEASUREMENT

EXECUTIVE SUMMARY

THIS REPORT WILL DETAIL

Audience measurement is the basis for half a trillion

dollars of advertising spend. It addresses two basic
questions about audiences of media properties: How
large are they and who are they? These seemingly
simple questions have become increasingly more
difficult to answer as advertisers look to reach specific
targeted audiences, requiring media sellers to measure
ever more granular audiences. This short paper reviews
current digital audience measurement methods and
how they meet the demands of the modern
media environment.

201 Third Street, San Francisco, CA E [email protected]

201
Third
Street, San
CA E [email protected]
2014
Quantcast.
All Francisco,
Rights Reserved.
2014 Quantcast. All Rights Reserved.

T 415.738.4755
T 415.738.4755

1 DIFFERENCES BETWEEN PANELBASED AND DIRECT MEASUREMENT

2 CHALLENGES WITH
PANEL-BASED MEASUREMENT

3 QUANTCASTS UNIQUE

W quantcast.com
W quantcast.com

METHODOLOGY

THE EVOLUTION OF AUDIENCE MEASUREMENT

Audience measurement was first used in the 1920s to measure
radio listenership. Participants in a measurement companys
panel kept diaries of their radio listening, and the collected
data was used to estimate the total listenership for a program.
That estimation was primarily done via extrapolation. While the
data collection techniques changed, this same basic method,
extrapolating from a sample, was applied to television in the 50s,
then online in the 00s.
Only in the last decade, with lower data costs and more
computing power, has direct measurement become feasible.
As the name implies, direct measurement measures the media
property directly via a digital beacon. The publisher incorporates
the beacon into their content so it is activated with every
content consumption event. The beacon can provide data about
every directly measurable attribute, such as visits, geography
and platform. While digital traffic measurement has largely
transitioned to direct measurement, many measurement services
still rely on panel-based extrapolation for audience attributes,
such as demographics.

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com

CHALLENGES WITH PANEL-BASED MEASUREMENT

Today, most digital publishers sell advertising with some form of targeting. Advertisers typically ask that their campaigns appear
in specific sections or in front of specific target audiences. Panel-based measurement can struggle to accurately represent these
more granular audiences, since a panel might comprise only 1% of the population being measured. Here we discuss some of the
challenges with panel-based measurement.

1. SAMPLE BIAS
Panel-based methodologies rely on the sample being representative of the population. For instance, if 5% of the population are
tennis fans, the panel should also be comprised of 5% tennis fans. When the panel is not representative of the population being
measured, it is known as sample bias. One of the significant challenges of panel-based measurement is recruiting a panel that does
not result in sample bias. For example, if the panel is recruited using free NASCAR tickets, it would overrepresent the populations
enthusiasm for racing and likely underrepresent the populations interest in tennis.

2. MEASURING SMALLER AUDIENCES: SECTIONS AND AUDIENCE SEGMENTS

As advertisers and agencies demand more
targeted media buys, publishers wish to
measure audiences for specific sections, such
as the sports pages, or audience segments,
such as High-Income Moms or In-App
Purchasers. By their nature, panel-based
measurement solutions lose accuracy when
measuring smaller audiences, because it is
hard to ensure correct representation of the
smaller audience in the panel.
Example (Fig 1.): Of an Internet population of
100 million, yoursite.com typically receives
200k users, and 20k users visit the sports
section. With a panel representing 1% of the
population, only 200 users from the panel are
likely to visit the yoursite.com sports section.
Even if the gender of those 200 users were
known, it would be challenging to accurately
infer gender for the whole section, and
inferring age or income would be even
less accurate.

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com

3. MEASURING AUDIENCES WITH MULTIPLE ATTRIBUTES

When advertisers want to reach college-aged females
from Texas, they arent interested in reaching females,
Texans, and college-aged people separatelythey
want to reach an audience with all three of those
attributes. Panel-based measurement quickly loses
accuracy as you layer on more attributes, because the
representation of that desired audience on the panel
is smallthe panel will include females, but it will
include fewer who are college-aged, and fewer still
who are from Texas. Extrapolating against this small
sample is likely to produce a poor result.
Example (Fig 2.): LargeApp typically has 1MM users,
but only 1,500 are Texan, college-aged females. With
a panel representing 1% of the population, only 15
panel members meet these criteria, not allowing any
reasonably accurate extrapolation.

4. AUDIENCE DE-DUPLICATION
The goal of audience measurement is to measure
people, not cookies or other identifiers. The challenge
for audience measurement services is that people
typically use multiple devices and browsers, which
generate multiple cookies and identifiers. The
measurement service must de-duplicate those
cookies and identifiers to arrive back to an accurate
count of people. Panel-based audience-measurement
services manage this de-duplication by examining
the usage characteristics of panel members who visit
a property. For instance, if the audience in the panel
tends to clear their cookies more often than the general population or owns more mobile devices than average, that would inflate the
de-duplication factor and result in undercounting the number of people visiting the property.

QUANTCASTS UNIQUE METHODOLOGY

Quantcast employs direct measurement to capture traffic and other directly measurable attributes, then determines other audience
characteristics, such as demographics, through a technique called statistical modeling. For example, starting with a set of users with
a known gender, Quantcast infers the gender of new users based on their similarity to the known users, scored against hundreds of
data points. Quantcast can employ this technique because it sees each U.S. online user on average 600 times a month, and those
additional data points, while not providing direct demographic data, provide a strong signal about user similarity.

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com

HOW DOES QUANTCAST STATISTICAL MODELING WORK?

1. Start with reference data from registrations and surveys
2. Examine characteristics of reference and measured users (which can include sites visited,
apps used, app usage frequency, content categories, device type and many more)

3. Infer attributes for each user based on similarity to reference users

4. Validate inferences against reference data and external census data

Note: example for illustration only; actual models include hundreds of factors

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com

OVERCOMING CHALLENGES FACED BY PANEL-BASED MEASUREMENT

By taking a different approach, Quantcasts methodology overcomes many of the challenges encountered by panelbased measurement.
Panel-based measurement is sensitive to sample bias because it assumes the panel represents the audience being measured.
Quantcasts methodology is based on an assumption that users who share an attribute, such as gender, behave similarly in a
detectable fashion.
Limited by their size, panels cant accurately
represent the smaller, more targeted audiences
that advertisers demand. Compared to a panel,
Quantcasts statistical modeling can accurately
infer attributes for a relatively larger group of
people, enabling accurate measurement of even
the smallest audiences.
Finally, panel-based measurement services
de-duplicate users based on the attributes of their
panel, which is subject to sample bias. Quantcast
de-duplicates users by examining multiple data
points across all sites and apps, such as time
period, observed visit frequency, visit source and
property type, so is less subject to the potential
bias of a sample.

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com

THE MIGRATION TO DIRECT MEASUREMENT

Until recently, direct measurement has been cost-prohibitive to
do at scale. Panel-based measurement has been a cost-effective
means to measure a large population, using a small set of
sample data.
Today, lower computing and data costs have changed that
equation, where direct measurement is not only possible but
necessary, given how publishers are increasingly adopting
segment-based audience sales and new media platforms.
Quantcast has been providing direct measurement coupled
with statistical modeling since 2007 and in that time has refined
its collection and modeling techniques to provide consistent,
accurate traffic and audience-profile data across any digital media
platform, for free.

LEARN MORE ABOUT AUDIENCE MEASUREMENTCONTACT US AT

[email protected].

201 Third Street, San Francisco, CA E [email protected]

2014 Quantcast. All Rights Reserved.

T 415.738.4755

W quantcast.com