Thursday, August 16, 2012

True Strength 2.0: The New True Brew So Far

Last summer, I did a post about the development of my timeslot metric True Strength almost every week. This summer, not so much; my last one was an introductory post about two months ago!

Does this mean I'm slacking? Maybe? But I'd say I have been slaving over a hot spreadsheet a pretty hefty amount. I just haven't been blogging about it. As I've said on a few occasions, the creation of True 1.0 was not very "photogenic"; in other words, very few people even by my very low traffic standards seemed to want to read about it. The True2 project feels even worse in that regard. It's not about starting from scratch and laying out lots of theory-type stuff. I initially thought that was how it was gonna be, but I decided pretty early on that repeating the whole process and blogging every minute of it was not a very efficient way to go about this. This is about making changes and digging deeper into things that tripped me up last summer. Tweaking ain't that pretty.

So anyway, I'm gonna begin my rollout of True2 information today, talking a bit about what has taken up most of my True2 time this summer to date.

The Big bc/PUT Theory

The vast majority of my True2 time this summer has gone into thinking about the three-way relationship between these things: ratings for individual shows, combined ratings for everything on broadcast, and overall TV viewership. It's a bitch to unravel, and I sort of got into those complexities last summer before retreating to a series of shortcuts. Worst case scenario, I can still retreat to a bunch of those shortcuts for the 2012-13 edition, as I think they largely worked out OK. But I'm trying my best to come up with something a little more unified. I do think I have something that would be an improvement over last year if the summer ended today, but it's not quite where I'd like it just yet.

If you remember some of my early posts from last summer, I tried devising this thing called "tendency PUT," which was essentially an effort to strip the influence of broadcast ratings out of overall viewing levels. I posted about it a couple times and then basically dropped it when I realized that ratings seemed to correlate better with actual PUT than with that.

For version 2.0, I'm bringing it back into the mix in a slightly revised form. Essentially what I want to be able to do is boil those the combined broadcast ratings and overall viewing levels into two new, "purer" components:

1) Tendency PUT. How many people would "tend" to watch TV on the evening, taking other TV events ("competition") out of it?
2) Tendency Competition. How much of the TV-viewing audience is tied up in other programming, taking the influence of non-TV events ("viewing levels") out of it?

Step One: Tendency PUT

Step one was relatively easy. I defined Tendency PUT a bit differently from last year, when I tried to assign some arbitrary broadcast level in each case. This time, I simply said: How many people would be watching TV if nobody were watching the big-five broadcasters? To get that number, I started with the same idea I came up with last year; for every 2.0 rating points that get injected into (or taken from) the combined broadcast ratings (bc), that tends to translate to a 1.0 increase (or decrease) in overall viewing. The only difference was that I found the ratio was more like 2.5/1 this time. (May try to illustrate this when I go into more depth later.) So, here we go: Tendency PUT = PUT - (0.4 * bc).

Why is Tendency PUT better than plain ol' PUT? The place where it really makes a huge difference is in super-competitive situations. If you look at week two of each of these three shows, you can see that the situations with a super-high bc also tend to have a super-high PUT, because those huge TV events like Oscar and NFL championship games bring a ton of new people to the TV. So those shows took substantial hits due to high PUT levels. And it really shouldn't be that way, since those extra people are all gonna be watching that big event. They're not really "in play" so to speak. There was no way a reasonable competition calculation could offset that PUT inflation, so those shows all had TRUE scores in week two that seemed way too low. Now, those shows all have relatively typical PUT levels, so the bonus for competition looks a lot more helpful.

Step Two: Tendency Competition

Step two is where I find myself mired at present. It turns out that explaining depressions in overall viewing is just plain hard. Explaining huge competition is much easier because, you know, you can see what the actual rating is. But the nature of holiday viewing and weekend viewing is a little more nebulous. Yes, you can look at how much the overall viewing goes down, but that number doesn't usually tell the tale of what happens to individual shows. While last year I essentially concluded that the drops were pretty close, the real problem in a timeslot metric that adjusts for competition is the additional hit because of how low the competition is on weekends/holidays. And these situations often get even worse using my "Tendency PUT" because there's a further penalty because of the low competition volume. (The Tendency PUT equation above subtracts a percentage of bc, meaning when bc is really low, that'll drive the Tendency PUT up.)

So I've been trying to find a way to "normalize" the competition level in really low-viewing situations the same way I normalized the viewing level in really high-competition situations in Step One. Preferably I'd like something a little less hand-wavy than what I did last year (making two separate rather arbitrary baselines for weekends and weeknights). I've tried a lot of different permutations of what I call a "holiday addition" to the competition number. The one I'm using right now seems to be on the right track, but it's just not big enough, and it's a negative number in other situations that are OK right now but would get wrecked if I double or triple the number, for example.

Other Tidbits

I've applied some other changes that I'll discuss later, but most of them were at least roughly mentioned in the post a couple months ago. Generally speaking, some of the overly swingy "rich get richer, poor get poorer" adjustments are out, and so I'm guessing the Sitch numbers typically won't be as big in True2. Holiday problems aside, I'm happy that my current version of the number exhibits a lot less volatility across the course of the season. The vast majority of the time, True2 has a lower standard deviation across the season than the first version and the raw 18-49 ratings. As I said, if the summer ended today I could live with this number and I think it'd work better than last year's.

My plan is to really drill down into this in the coming days and try to finish this thing up. I would really like to have a True2 ready to go and pretty much completely laid out by the beginning of my epic 22-day fall preview on August 26. That seems realistic but it's not exactly a guarantee. Still, it really feels like the week of open real estate between the end of the War of 18-49 season (tomorrow! *sniff*) and the beginning of the fall preview would be a great time to vomit out a bunch of boring True2 posts. So that's... not so much the plan, but it's the hope. Whether I have True2 or not, I believe this fall preview is gonna be pretty fun, and I absolutely can't wait to really start focusing on something other than bc-levels and PUT-levels. Till next time!

No comments:

Post a Comment

© 2009-2022. All Rights Reserved.