Look Coding with Attentional-Inertia Hazard Analysis

Description

Code every look towards and away from a screen from video, then analyse look lengths as a survival process to show how engagement deepens the longer a look lasts.

Viewers look at and away from a TV about 150 times an hour, and look lengths are lognormal with many short looks. Treating each look as a survival process, the hazard of a look ending peaks at about one second and then falls steadily: the longer someone has been looking, the less likely they are to stop, a phenomenon called attentional inertia found from infancy to adulthood. Distractor probes show longer looks are harder to break, and heart rate or secondary-task reaction time index deepening engagement. Long looks carry viewers past content boundaries, and content seen deeper into a look is better recognised later. Data come from video of viewers in the lab or fixed cameras at home, coded frame by frame. Visual orientation is not comprehension, and home studies predate multi-screen use, so modern applications must code phones and second screens too.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Record viewers with a camera facing them, in the lab or, with consent, a fixed camera at home.
  2. Code the frame at which each look at the screen starts and ends, with a second coder on a sample for reliability.
  3. Compute the look-length distribution and the hazard function: the probability a look ends in each interval given it has lasted so far.
  4. Optionally present timed distractors or a secondary reaction-time probe at different look durations.
  5. Relate look positions to content events such as cuts, segment boundaries or ads.
  6. Test later recognition of content snippets to link look depth to retention.

Sources

  • Handbook of Child Psychology Vol 2 (Lerner et al., 2015)

Also known as

  • Look coding of screen attention with attentional-inertia hazard analysis
  • Attentional inertia analysis
  • Look-length hazard analysis

Categories

  • Analytics & Behaviour
  • Neuro, Biometrics & Affect