Video Duration
How long the video is. We log-transform this because views don't scale linearly with length — a 10-minute video isn't twice as good as a 5-minute video by the algorithm's lights.
Duration has a surprisingly strong positive effect for non-Shorts — longer videos get more views on average, even after controlling for topic, title, and production quality. The algorithm rewards watch time, and longer videos generate more watch time per successful impression. There's a floor (very short non-Shorts underperform) but no clear ceiling within the 3-30 minute range typical of political content.
Average Video shows the effect on a typical video. Top 10% Video shows the effect on videos in the top 10% of views — estimated via quantile regression, so you can see whether a factor helps most videos a little, or pushes the top performers much higher.
When present
The video is substantively longer than average.
A 15-minute committee hearing breakdown or deep-dive explainer
When absent
The video is shorter than average (but not a Short).
A 2-minute press-release-style clip