Good points. To follow up on them a bit: 1) There's obviously a lot more separation here, but it's worth noting that there is a lot of data out there available

Good points. To follow up on them a bit: 1) There's obviously a lot more separation here, but it's worth noting that there is a lot of data out there available to broadcasters that doesn't come from Nielsen, such as broadcasters licensing data that comes initially from cable providers. The number of set top boxes here dwarfs the size of the Nielsen data set. 2) You can get more precise if you want. At a basic level, you can look at a lot more options than just changing the channel. You can also train models based on the overlap with something like Nielsen where you have more information about when they are watching. If you can match up comcast data to Nielsen data, you can build a model for active vs. inactive boxes on the comcast data. 3) To backtrack a bit, the fact that Nielsen is a "very small subset" as you said a couple of comments ago is relevant for some things but not others. For the broadcast network prime time shows, it's certainly a big enough sample to see which shows are popular among what demographics without having the level of detailed information that Netflix or cable operators have. All in all, I agree with you that there is more separation and it is harder to use, but there's still a key distinction between Nielsen, which is in something like <~0.1% of homes and the other data sources which are many times larger. They don't all seek it out, but it's possible for broadcasters to acquire the sort of data necessary to answer questions about content popularity like Netflix can consider.