because that would be introducing a lot more statistical uncertainty. you'd have all the lurking variables of 'CNN viewers' vs 'FOX viewers' and no way to know whether any observed effect was a product of being a regular viewer of the particular news outlet or a product of the particular commentator.
you'd also pretty much just be measuring democrat vs. republican stats because FOX is so overwhelmingly an obvious mouthpiece for the GOP -- but you'd be doing a poor job of examining that because CNN is not as partisan. you'd be better off comparing FOX vs. MSN, because MSN more closely mirrors the degree of partisan bias FOX has. still if you're going to all that trouble what you'd really want to do is just survey a random sample of people and ask them if they're members of the DNC or the GOP -- take out all the inherent variability of people getting news from some particular format and cut directly to actually comparing party division.
by identifying a time period through which one network had two roughly equally popular commentators that took divergent approaches to the topic, the researchers eliminated alot of error and were able to focus on the key question of the influence of mainstream media political commentary. this is good statistical design, screening out variability.