The New York Times has an interesting flash application that breaks down the text of yesterday’s Democratic debate (there was a debate? UPDATE: And it was in my own city??) by speaker and shows visually the distribution of who spoken when through the debate. I mention it here because it’s one of these data transformations very much in the same spirit of what I keep pushing here. They took the transcript, made it visual and interactive, and the end result is a vastly different view onto the debate than anyone had before. It uses the same transcript as anyone else, but adds something very new and informative.
One can’t help but notice that the different candidates are not getting the same amount of speaking time. Clinton spoke more than 3.5 times more words, and the same for speaking time, than Biden. For that matter, basically so did the moderator, who held the floor for more time than anyone but Clinton. It’s no wonder that Clinton is considered “the Democrat to beat” considering she’s in our face more.
If the numbers weren’t so vastly different between the candidates, we’d chalk it up to some random variation that happens from debate to debate. But, from the numbers, the speaking times are clearly planned. It’s so clear that I feel like maybe I missed something. Is it common knowledge that the debates are proportioning time out to the candidates based on their poll numbers (or something equivalent)? It’s not just that the front-runners are getting more time. The statistical correlation is ridiculously high (speaking time versus FOX News/Opinion Dynamics Poll. Oct. 23-24: r=.96). That is, the debate organizers are basically using this formula to determine how much time each candidate should get:
Speaking Time = 8:26 minutes + 25 seconds * Latest Poll Number (%)
Of course, debate organizers can’t control exactly how long each candidate talks for, but the candidates only deviated from the formula by at most two minutes and twenty seconds (Biden, who spoke less, and
Dodd, who spoke more).
So now I’m getting off topic a bit, but in any case: transformations on data can be very revealing!