TiVo posted a graph of audience size and viewership second-by-second during the Superbowl this last weekend. According to TiVo's website, this is based on data collected from 10,000 viewers who were using their TiVos.

Here's the graph of the percentage of audience in "play" mode each second, see if you can see why I'm very suspicious of their data (note that the large size of this image may cause some browsers to lay out this page oddly. My apologies. I didn't want to shrink the image down because that would obscure the details which are important to understanding the point of this article):


As one would expect, the graph is noisy, as people pause and play their TiVos.

But the odd thing is that the noise is always in increments which are exactly the same size: each jump up or down is exactly the same size as every other jump up or down (or a multiple thereof), with a smoother curve superimposed on top.

I measured the size of the jump, and it is about 3 percentage points, or 300 people in their sample.

In other words, the data is suggesting that there was a group of about 300 people out of the 10,000 who were pressing "pause" and "play" on their TiVos in perfect unison throughout the entire game. To call this improbable would be more than a mild understatement.

There are three possible explanations, from most likely to least likely:

1) TiVo screwed up badly in creating this graph. Odd that nobody inside TiVo scratched their heads and said, "hmmmm, this does look a little odd, maybe we should double-check our data before we embarrass ourselves in public." This is exactly the kind of graph which makes physics professors send their freshman students back to the lab to redo the experiment...

2) The graph is correct, but there's something really weird about the way the data was collected. For example, maybe data was gathered from 10,000 TiVos, but only a subset of about 30 were actually polled each second (this would explain the results, since in a group of 30 a single person pressing play or pause would cause a jump of the size observed).

3) The graph is correct and the data is correct and some mysterious mechanism makes precisely the same number of people (3% of total viewers) press play or pause at precisely the same instant. This is the least probable possibility, and will inevitably lead to discussions of orbiting mind-control lasers and tinfoil hats.

If TiVo has an explanation for this weird data, I haven't seen it on the website. Anyone know the answer?

[Aside, before someone suggests this: One obvious, but insufficient, explanation is that viewers' pausing and playing is naturally synchronized by the action in the game and the commercials. This is true, but if that was the cause of these mysterious jumps, then one would expect each jump up or down to be a different size. For example, 300 people might fast forward through one play, but 400 would fast forward through a different one. The only way this explanation could work is if every moment in the game and every commercial has exactly the same appeal to that subgroup of TiVo watchers who actively pause/rewind/fast-forward through the game. Otherwise, there should be significant variation in the size of individual spikes.]