Collecting naturalistic attention data at scale
The attention data in this booklet has been collected in two slightly different ways.
The TV attention data has been collected by TVision in the US. TVision has recruited a panel of 5,000 US households who are asked to install a camera on top of their main TV and set up an accompanying recording box.
Once installed, their system records when people enter the room and, crucially, whether or not their heads are turned towards the screen. They even record how long people look at the screen: if their heads are in the same position for too long, their system assumes that the panellist has fallen asleep. The box attached to the screen records what was on the screen – what channel, what show, what ad, and so on. They can then understand how much visual attention was directed towards whatever was on the screen at the time.
The digital advertising data comes from Lumen. Lumen has recruited a thousand-strong panel in the UK.
The panellists install software that turns the webcams on their computers and phones into high-quality eye tracking cameras. The software also records what ads were shown on the screen, in what location, and for how long. It then uses this information to calculate which were viewable, which ads were viewed (a big difference), and for how long they were viewed.
The systems used by the two companies are slightly different, but comparable. TVision’s data is, technically speaking, head tracking rather than eye tracking; it measures when people’s heads are turned towards the TV. But given that TV ads take up 100% of the space of the screen, if you’re looking at the screen when an ad is on, then you’re looking at the ad. Things are more complicated online, when both advertising and editorial content can be on the screen simultaneously. This is why we need to use eye tracking for these environments.
Both TVision and Lumen take data quality and data privacy very seriously. Panels are recruited to be nationally representative, with the appropriate mix of genders, ages, and ethnicities. Panellists are fully informed and properly incentivised for their time and involvement. No personally identifiable information is collected or retained and no respondent-level data is ever released. By design, neither company has the ability to store or upload any video from any of their panellists. In both cases, all the data processing is done locally on the panellists’ machines, with only summarised lines of data uploaded. TVision hardware is made of components that have been certified by the US Federal Communications Commission.
