Renewology: What's New In Season Three

Hello! After an improved performance for my renew/cancel model last season, I didn't have a ton of time/motivation to change things up this summer. And yet... there is a long post ahead about the things that have changed!

Fox is Harsh

Like every year, I did some tweaking to the network-specific adjustments. After really underestimating NBC's generosity for a second straight year, they are now perceived to be a few hundredths more generous than before. I also made a tweak of a similar size on ABC, making them a little harsher overall than before. (They will be especially more harsh on comedies, since their comedy/drama imbalance was rather crazy last year.) CBS and CW, which have generally worked pretty well, are basically unchanged.

But the most consequential question was (and still is) Fox, as that network transitions into what will reportedly be a much more "live event"-focused strategy. As I said after their upfront, even though they maintain a "full" schedule of entertainment programming in 2018-19, it was kinda plain to see that they are teeing up a lot of cutbacks in the future.

So Renewology's expectation is that Fox's renew/cancel targets will be 27.5% higher than what their ratings say they are. This is a huge departure from the network-specific adjustments of the past, at least on the big four; we've never even had one as high as 10% in either direction. (It's not that far from the one used for the CW... though that one goes in the other direction.) This is another 20% on top of the Fox adjustment from last year, essentially a declaration that the network will wipe out at least one full night of entertainment programming. We pretty much know that will happen, thanks to Fox's deal with the WWE. So you can make a good argument that even this is too conservative, but I just don't feel comfortable going any farther than this. I might make further adjustments later if more news breaks about what Fox's "live events" will actually be.

The new Fox Renewology actually wouldn't have been a huge miss even using last year's shows. Fox was a bit harsher than we expected, and with these targets they would be more generous than expected, but not by much more than we missed in the other direction last year. The light-green shows (Brooklyn Nine-Nine, The Orville and The Resident) would drop down to low-end bubble shows (40-50%), while the true bubble shows of last year (The Gifted, Gotham, Lucifer, LA to Vegas, The Mick) would become more like 20% now. Will the network still find room for shows that perform like Orville and Resident did a year ago? My guess is that some might sneak through, but I wouldn't feel safe at those levels. Again, it's possible this will turn out to be way too conservative, but we're all kind of guessing when it comes to what this network will look like.

Lead-ins

The main thing I wanted to change in the True formula was an overhaul of how it perceives lead-ins. I think it's fine for the most part, but there are clearly some cases where it doesn't get the job done, especially involving comedies with large lead-ins. There were some cases where I made a manual correction to the lead-in weighting before publishing the True numbers last year (the early weeks of Roseanne come to mind) and some others where I wish I had; ones that really stand out in my memory include the NFL playoff episodes of The Last Man on Earth (they inflated its R% for much of the rest of the season) as well as the premiere of Me, Myself and I, which was listed as a renewal favorite on premiere night and tanked the next week.

I studied a lot of live-only numbers, both from the most recent season as well as from the trove on Sitcoms Online circa 2010. (Search "final ratings.") The actual new lead-in component is still a work in progress, and might be something I incorporate later this season or next season; right now, it solves some problems but creates enough other problems that I don't want to roll with it.

But I did add a couple of the lessons of the new version into the old version, for now. The main one is that reality shows actually have less live viewing than I thought. When just looking at Live+3 and Live+7 gains, you could easily get the impression (as I did) that unscripted series have very few people DVRing them at all. But looking at live-only viewing, it becomes clear they still have a lot of same-day DVRers; in fact, the ratio of live viewers is not that different between scripted and unscripted. So while we used to penalize a show with an unscripted lead-in by about 20% vs. a show with a scripted lead-in of the same size, now that gap is more like 5-10%.

Estimated Final Ratings

This is something I first spoke about wayyy back at the beginning of last season when I began putting un-rounded ratings in the Vault. The plan was to stop using rounded finals for the True formula and start using something that I call a "guessed final," which is an educated guess at what the un-rounded final will be that doesn't require waiting an extra week to get the actual un-rounded final. I never actually got around to incorporating this last year, but finally got a simple version of it done this season.

There are just two rules here:
1. Start with the un-rounded prelim, then add or subtract 0.1% for every 1% that a show grows or declines relative to its lead-in. This is the "guessed final" that will be used for Renewology if it gets posted before finals come out.
2. After the rounded finals come out, if a show adjusts in a way that disagrees with the number in Step 1, it is perceived to "barely" adjust. In other words, if a show has a 1.13 un-rounded prelim, and Step 1 guesses that it will adjust up to 1.14, but it hits a 1.2 in the rounded finals, then the "guessed final" will be a 1.15: the lowest possible number that still squares with a 1.2 final.

I've played around with much more complicated versions of this, but haven't been able to add any real value on top of the simple lead-in-based adjustment. It's possible it will be improved in the future, but this is probably what we'll have at least for this season.

Anyway, I'm sure this seems like extreme minutiae, but it actually improves the performance of the True formula in all three years where un-rounded prelims are available. For 2017-18, here's the average error between the actual un-rounded final and the other possible methods:

Rounded finals: 0.25
Un-rounded prelims: 0.21
"Guessed final," Step 1 only: 0.19
"Guessed final," Steps 1 & 2: 0.13

The moral of the story is that using rounded finals like I did in the last two years was actually the worst possible method, in terms of having numbers that are close to the actual finals. It also led to some iconic moments like that time The Mick had a 0.749 prelim, adjusted up to 0.8, and went from like 40% to 68% or something in Renewology. There should not be nearly as much movement from prelims to finals anymore, and especially with super-low rated shows there probably shouldn't be as much movement from week to week either. (The main one I'm thinking of here is iZombie, which had a double digit movement each of the first seven weeks in R%, but has less volatility in the new version.)

Unfortunately, any improvement we might have gained from this will probably be cancelled out by...

Estimated Viewing Levels

This is one I was not anticipating having to write about, but here we are. Since it looks like there may be some difficulty in getting viewing levels this year, I'm at least starting off estimating them based on the ratings and shares available for broadcast programs. I had something like this in fall 2017, when I was unsure if viewing levels would be available back then, but never had to use it. This method essentially comes up with two separate estimates:

1. An estimate using the 18-49 rating/share numbers in each half hour.
2. An estimate applying last year's average year-to-year PUT decline (-8.5%) to the PUT numbers from the year-ago evening.

And it averages these two things together.

In comparing this with actual known PUT levels, this method seems to perform a bit better than using either one individually. However, it's definitely going to be a lot worse than having the genuine article. The estimate for Premiere Monday in Step 1 had PUT levels down 15% from Premiere Monday last year, which would be a really steep decline. Combining that with Step 2 comes up with a projection of like -12%. But what if it was actually -8.5% or -15% instead? It can make a somewhat sizable difference with a bubbly premiere rating like Magnum P.I.; instead of its currently listed 52%, using a -8.5% PUT decline results in 46%, while using the 15% decline that the shares suggest results in 57% for Magnum.

So, that's it for now! The numbers for Premiere Monday, Tuesday and Wednesday have been updated, and Thursday will be along later today. I will have some thoughts on the actual numbers themselves probably at the beginning of the next week.