Friday, October 13, 2017

Renewology: The CW's Premiere Week and Un-Rounded Numbers


Happy Friday. In this post, I will have a quick word on the CW's newbie premieres and a much longer, wonkier word about some things that are going on here with un-rounded numbers this fall.



CW Newbies

NameR%TrueProjTargetAiredSkewMale
Dynasty54%0.340.240.24131%
Valor41%0.310.220.24132%50%

Like several networks during the big four premiere week, the CW had a truly signature story in the explosion of Riverdale. But it was so awesome that it seems highly unlikely the show will find itself in renewal danger. That leaves only a couple shows so far that look interesting from a Renewology standpoint. The CW's fall newbies both come from CBS Studios, so without seeing any ratings it seems unlikely the network will cancel both. Axing both fall newbies in back-to-back years would be rough PR-wise. But after they both got off to weak starts in the ratings, I'm not that optimistic they will renew both either. So it seems like these two may be battling it out for one spot. Right now, Dynasty has the Renewology edge, and that's what I would have thought before the season began.

On the one hand, Valor was a little south of 0.3 in the un-rounded prelims, and Dynasty was probably a high 0.3 after adjusting down. On the other hand, Valor premiered on a somewhat bizarrely high-viewed night; PUT was up by more than a point week-to-week in the first two hours on Monday, which is weird since the Monday Night Football game was down six tenths. This led to everything getting clobbered in True, which was apparent Renewology-wise in how low Lucifer went. Factoring all of that in, I'd still guess Dynasty has the edge, but it might be closer than I would've thought before the season.

Musings On Un-Rounded Numbers

I've been collecting the un-rounded numbers from our incredible commenter RJK and various other sources for a couple years now, but this fall is the first time I'm bringing them out into the open on the SpotVaults and actually using them in seasonal averages, as some of you have noticed.

A big part of the hesitation with this was for neurotic formatting reasons; when un-rounded and rounded numbers are combined in a table, I really don't want the tables to display digits of precision that aren't really there, or cut off digits of precision that are there. In other words, if we have an un-rounded 1.300 for a show, I don't want the table to print "1.3" and cut off the rest. And if we don't have the un-rounded number yet, just a 1.3 for the rounded final, I don't want it to print "1.300" as if we know the next two digits when we don't. So it was more complicated than just printing out the un-rounded numbers willy-nilly. Now, I've finally gone through the logical hoops it takes to get everything displaying correctly.

Why did it take years to actually go through those logical hoops? Mostly because I've historically been a bit of a seller on the real usefulness of un-roundeds. The un-rounded numbers definitely give you more info about things like week-to-week fluctuations. But do those things actually matter? Individual points and individual fluctuations probably have very little bearing on the final decision making. They're more for junkies like us while experiencing the numbers daily.

A good analysis should be aggregating multiple points. And when you aggregate multiple points, the error in a rounded rating is largely cancelled out. I remember looking at this a couple years ago and finding there was only one full-length season on all of broadcast where the average of rounded ratings was even two hundredths (0.02) off of the average of un-rounded ratings (I believe it was Nashville). You have to have crazy adjustment luck to be off by two hundredths in one direction over 20+ episodes. Something like 15%ish of shows were off by a full one-hundredth (0.01). I know we're working with small numbers here, but being off by 0.01 is still not a big deal!

But when navigating the early weeks of a broadcast season, we don't have the luxury of a large sample size of "settled" ratings to filter out the noise. In the first three weeks, Renewology only works off of the single most recent rating, so the difference between being +0.05 from the rounded rating as opposed to -0.05 can be huge, especially with CW and low-rated big four shows. There are also the cases when a show adjusts right around the mark and the rounded numbers perceive a huge difference; this was particularly clear with The Mick on Tuesday, which went from a 48% with its 0.7 prelim to a 71% with its 0.8 final.

This is not necessarily a problem with the model assumptions. When you're at 50% of the network's average, you are almost always dead (about 5% of new shows survive at that level). When you're at 100% of the network's average, you are almost always renewed (about 95% of new shows survive at that level). And in a world of 1.0ish league average, those levels are only five tenths apart. That means about 90% of the R% spectrum is contained within just five tenths. Entire tenths of a point should matter a lot, at least when the sample size is small.

The problem is in the input data. It's not really entire tenths of a point. The perception was that it jumped from a 0.700 to an 0.800, when it was probably more like 0.749 to 0.751. Not only is that a big jump for the raw rating, but it can also distort things like skew, which can further throw off the True formula.

This seems like an easy fix: just use the un-rounded prelims for Renewology based on prelims and un-rounded finals when based on finals. But that plan bumps up against another of my neuroses: I really like having the number that comes out right after finals to be "official," stored in both the Vault and the daily table forever, as an unchanging record of what the verdict was at the time. And un-rounded finals can be a week or more coming in. So the True formula is still explicitly told to work with rounded numbers, to keep the numbers from changing as more data comes in.

Given those constraints, the best solution here is to come up with a formula to guess the un-rounded final, using a mix of the un-rounded prelim, the rounded final, and previous adjustment history. My basic idea isn't that complicated: just take the un-rounded prelim and apply the average of the previous adjustment history. Then, if the rounded final disagrees with the guess, just take it to the edge of adjustment. (The aforementioned The Mick episode would be an example of this; it adjusted down in un-rounded finals both of the first two weeks, so we would predict a downward adjustment, but then it went and managed to adjust up in week three. So we would just go with the smallest possible number that still has it adjusting up: 0.750.) There would be some additional nuances to deal with things like pre-emptions, as well.

Anyway, I haven't actually done this yet, and I may not do it in the super-short term since we have now passed through much of the time when it would be most useful. Since we'll be in week four, Renewology for most big four shows going forward will be incorporating multiple points, and that will help mitigate the effect of rounding errors and adjustments. (Although it would be very helpful for the early weeks of Dynasty and Valor......) But it's certainly something I'd like to get going by midseason, and I may point back to this post at some point as a guide.

If you've somehow read this far, here's a little tease of another SpotVault addition, which should be coming up in the next week or so. Each broadcast page will be adding a "Prelim" and "Adjust" column.

#
DayDateTimeLength
Viewers
A18-49
PrelimAdjust#y2y
Ty2y
1We9/27/20178:000:306.1971.8211.750+0.071
-9%
-9%
2We10/4/20178:000:305.7911.5861.550+0.036
-17%
-17%
3We10/11/20178:000:305.6631.51.444+0.1
-17%
-17%

AVERAGES TO DATE:
Viewers
A18-49
PrelimAdjust#y2y
Ty2y
5.8841.641.581+0.069-14%-14%

As I've said above, day-to-day adjustments/un-roundeds are not that meaningful in the end and I advise not getting too worked up about them. But since I doubt my saying that will actually stop anyone, this will be the definitive way to track how a show tends to adjust, and maybe use that to predict a future adjustment.

A couple notes: these columns do not need un-rounded numbers to function. The Prelim column will list rounded prelims with the same formatting as the A18-49 column if that's all that is available. And if either the final or the prelim is rounded (or both), it will instead list the rounded adjustment. Even when both are un-rounded, you can still get a sense of how many times it had a rounded adjustment because all the ratings that adjusted in rounded numbers are highlighted in bold. Enjoy!

No comments:

Post a Comment

Newer Post Older Post Home
© SpottedRatings.com 2009-2017. All Rights Reserved.