Purpose
This article provides an overview of control groups and explains how and why they are used in InnovidXP.
What is a control group?
A control group within the scientific field is a group that is separated from the rest of the experiment, allowing the independent variable being tested to have no influence on the results. A similar approach is used in the InnovidXP platform, where we observe the behaviour of groups of households who have not been exposed to ads to estimate expected baseline visit and visitor rates.
How InnovidXP control groups work
The image below shows a spread of households. All the blue circles here are households that have been exposed to a brand’s TV advertising in the week that a spot airs or in the 30 days prior. All the remaining green circles have not been exposed and so are eligible to be added to the control group.
Each week, we use a household viewership panel (Inscape in the US and Samba in the UK and DE) in this way to determine which households qualify as unexposed to the brand’s TV advertising. The visit behaviour of these households at the daily level forms the baseline for measurement.
How control groups are used
Attribution
The number of visits captured from the control group can tell us how many visits we should expect by chance following a spot of a given size.
The difference between the actual number of visits and the expected number of visits forms the incremental attributed response. Daily control visit rates can vary due to seasonal and weekday patterns. For example, Black Friday would show different volumes compared to an average day.
TV Influence
Whilst attribution is calculated at the spot level, TV Influence is calculated at the campaign level. This metric takes the difference between the visitor rate of the exposed group and the visitor rate of the control group across the whole campaign and expresses it as an uplift percentage. A TV Influence of 20% means that the visitor rate of the exposed group is 20% higher than the visitor rate of the control group.
It is important to note that attribution uses control group visit rates, while TV Influence uses control group visitor rates. Control groups are applied at the spot level for attribution and at the campaign level for TV Influence.
Control group variations
There are variations within the control groups:
Regional control groups
These are filtered by the Market Area so when building control groups for regional campaigns, only households within Market Areas that were served impressions are eligible for the control group. The geographical representation of those impressions is matched.
OTT control groups (synthetic spots)
As there is no such thing as a single spot with multiple impressions for OTT, synthetic spots have to be created by grouping together impressions by day, publisher, and creative.
Hold-out groups
In certain workflows, the client can provide their own hold-out groups. These are more similar to a traditional scientific control group as they can be made up of households that are within the target audience but are not served the adverts.
Related content