That’s where the second regression model comes in. It exploits a pattern the authors identified in television viewership: It tends to be highest 2.5 hours after the sun sets, regardless of what’s on the air. This makes sense: People like to be outside or doing other stuff during daylight hours, settle in at home to watch TV for a bit after the sun sets, and then tend to go to bed within a couple of hours.
Around the country, Carlson’s show is broadcast in the hour before Hannity’s. This sets up a random experiment: In counties where the sun sets earlier, Carlson viewership will be higher (and vice-versa when the sun sets later). This isn’t because people prefer Carlson to Hannity for any particular reason, but simply because they want to watch something on Fox and Carlson’s show happens to be on.
Studying this random pattern allows them to remove the possibility that it’s something about the kind of people who watch the shows, rather than the programing itself, that’s driving the results.