• by Charlene Weisler , Op-Ed Contributor, February 24, 2010

While there are many companies that are examining set-top-box data, Bruce Tyroler, VP Research for Scripps, has been especially focused on analyzing the data from several sources. His work has expanded our knowledge of set-top-box data for practical use. In my interview with him, Bruce talks specifically about his work in this field and shares some of his ground-breaking findings.

Below is a short excerpt from the interview. Direct links to the full interview videos can be found at http://weislermedia.blogspot.com/search?q=tyroler

CW: Can we go into detail on your work on set-top-box data?

BT: Sure. A lot of the work I have done on set-top-box data has been focused on validating its utility to us. And I would say that the data is useful to us if it reflects something in the real world. That is important. That is the key thing.

So in part that involves looking at different set-top-box datasets. I looked at several of them so far -- four different ones -- and how they compare with Nielsen.

But I am not saying that Nielsen is therefore the gold standard. A dataset is not valuable to the extent that it correlates with Nielsen. If that were true there would be no point in using set-top-box data because then the gold standard becomes Nielsen and Nielsen is always better than the set-top-box data.

So I have tried to develop ways of looking at how the data does relate to Nielsen but also whether it might be measuring certain things better than Nielsen does. And in some cases I think that the answer might be yes.

I believe that because there are two things going on. There are certain kinds of viewing behaviors that tend to be pretty universal for people that are viewers of a show or of a network. And I think that the proof of that is that there is a lot of commonality in the up and down movement that you find of viewers to individual programs. I am talking about up and down movements on a minute by minute level - a lot of commonality across different datasets -- really different datasets from completely different footprints.

And what that shows is that, conditional on them viewing a certain program, they tend to view it in similar ways in different populations. I think that the benefit of set-top-box data is because the scale of it is so much greater, there is less noise in that measurement.

The proof that I found that set-top-box data can provide incremental value over Nielsen is -- and I found this with two kinds of data and I found this to be true a couple of years in a row -- that for a number of different networks that you can predict the change, the minute by minute change, in the Nielsen ratings better with the set-top-box data and how it is moving, than from the Nielsen data itself.

CW: Can you go into more detail on that?

BT: So in other words, let's say that you want to predict what is going to happen to the rating on MTV at 1:04 a.m. For Nielsen you have what happened at 1:03 and 1:02 and 1:01 -- it went up, it went down, it went up this much, it went down that much. And given that, you want to predict what is going to happen in the next minute.

Of course you should be able to come up with a reasonably good prediction because those things are pretty correlated. For example, if you are heading into a break, then the following minute you have a pretty good idea of how much that is going to fall and the minute after that how much the rating is going to begin to rise again. You have pretty good predictions.

But the question is, could you have as good or an even better prediction if you took your data for 1:03, 1:02 and 1:01 from a different dataset -- one that is drawing not from nationally represented homes, but from a much larger panel of homes.

And what I have been able to show over and over again is that the answer is yes. That means that two things are true. It means that both the set-top data and the Nielsen data are pointing at something in common, and that something is common is some kind of universal viewing patterns. There is no other reason why these two things would be related in any way.

So that is true, but it is also true that the set-top-box data is doing so with less noise. That is why its predictions are better. There is no other reason why that could be. This is not a criticism of Nielsen. It is just talking about the limits of a sample-based methodology where you are talking about 10,000, 20,000 homes as compared to a panel that might be a million homes or more....