Monday, September 25, 2017

Renewology: What's Changed for 2017-18


While there are some tweaks for 2017-18, the basic Renewology structure is the same. So I recommend looking back on the intro and especially the summary of the process from this time last year. This just works as an amendment to that.

Just want to say off the top that Renewology will be back, and is ready to run as we speak. But it's possible it could be delayed a bit to start the season, depending on whether we will have a source for viewing levels this season. If not, the estimates for viewing levels (which I get using shares) can be kinda sketchy, as I found out when I put out some R% using the estimates during premiere week last year. I have a rough idea for something that will hopefully make these estimates a little more reliable, but I didn't want to waste much time on it till I knew if I would have to.

Vectorized True

The big project this summer on the True front was to create a more dynamic formula that could stretch across all the seasons of the A18-49+ era. There are a lot of interesting possibilities with 16+ years of True. It could be used to create an even more "fair" historical measure, True Plus, which can account not just for the league average but also the conditions of each individual timeslot. It could even be used to run something like Renewology deeper into the past, and get much more robust data about network decision-making dating back into another generation.

I was never quite convinced this would be worth all the trouble, but I finally went for it this year, in part on the grounds that this might make the formula much easier to update. Every year so far, I've had to go through a bunch of different functions, generate new constants from the most recent season, and plug those into the formula. But now the formula automatically generates those numbers within each season (including a projection of where the numbers will land this season), and it will be very easy to tell it to just move on to the next season.

Anyway, this whole project will not be all that obvious right now, but could bear some fun fruit down the road. And having to come up with these things across such different eras of TV was an interesting learning experience that I hope may have improved True somewhat even for this year.

Changes to True

I did very little to the structure of True itself in the off-season, but the one big thing I did may have some interesting effects. It came from creating the aforementioned "vectorized True." One thing that such a project has to account for over time is just the general importance of timeslots. It becomes pretty clear when comparing across 16 years that all the timeslot factors - viewing, competition and lead-in/lead-out - were more important in the pre-DVR era than they are now. And the weight True was putting on the importance of viewing/competition in 2017 had gotten to be too much; it was really more of a circa-2010 weight, which I guess makes sense because it isn't something I've changed too much since first creating True around that time. Dialing back the viewing/competition did a couple things:
  • A lot of shows in the 2016-17 True had their lowest scores around the highest-viewed parts of the year (late fall and early winter). And I also got the sense that many shows were getting too much of a bounce in the late spring; it's how some stuff like Powerless and Making History scored much closer to the bubble in Renewology than they really were. Weighing deflated viewing and competition less helped to smooth that out a bit.
  • Last year's True was extremely kind to shows in the 10:00 hour. The general principle goes down as a win; it is hard to get shows like Agents of SHIELD, Code Black and The Blacklist into the renewal favorite column unless you account for the increased difficulty in that hour. But again, there were some clear flops (in this case I would highlight The Catch, Pure Genius, and Conviction) that made the effect seem like a bit much. So this has been dialed back a bit as well. I actually didn't even set out to do this part, but it was just a side effect of valuing timeslots less in general.
Changes to Renewology

The formula has gotten a bit harsher simply because the networks seemed a bit harsher in 2016-17, and those 2016-17 numbers are now in the model. As I said in the last post, two networks (ABC/Fox) were much harsher than expected, one (NBC) was more generous, and the other two (CBS/CW) were close to correct. This meant that the "bubble" (AKA the point of 50% renewal), which was right at 75% of the network's adjusted average last year, is now more like 77%. Not a huge difference, but every point counts with the shows in truly uncertain territory. There are some other subtle changes like this that have emerged from simply incorporating the 2016-17 numbers.

Here are a few other things tweaked in the process:
  • Renewology randomly generates thousands of scenarios for where a show's ratings might end up, which creates a lot of uncertainty. In this version, I've also added that same mechanism for where the network's ratings might end up, based on the error that Renewology's target projections had in previous seasons. This will add some even more extreme cases, like if a show holds up really well while the network average tanks or vice versa, and should make things a tiny bit more uncertain in the early weeks. This is something I knew about last year and would've been in the formula from the get-go if there had been a little more time before premiere week.
  • Not sure if you remember, but around the time CBS' Thursday comedies came back I had to do an emergency fix with Renewology to make the comedy R% look more reasonable. This is because the formula expected the network's comedy average to be super-high, due to it being super-high the previous season. While it is still very difficult to hone in on where these category averages are going end up until at least halfway through the season, I have noticed that at least the year-to-year trends are usually a good indicator early on. So while we still start with what the average was the previous season, there is also an adjustment based on how the category ratios are looking compared with last year. This should hopefully get it closer to reality in the first couple months of the season.
  • I did a little more digging into how much shows drop post-premiere. I'm pretty sure I said somewhere last year that other than the huge difference between new/returning shows, there aren't a lot of ways to subset these declines in a meaningful way. Even comedies and dramas have about the same post-premiere trajectory, on average. However, I did split them up this year into "CBS" vs. "everyone else." CBS newbies hold up noticeably better post-premiere than the big five average pretty much every year. So CBS newbies are compared with the CBS average decline only, while the other four networks are compared with the average decline among the other four networks.
  • As I mentioned in the last post, I wanted to have a rule that would account for major seismic "intervening events" that are outside of the renew/cancel decisions on this year's scripted shows. The rule is basically that every additional or removed 40 hours are worth 0.1 points to the target. The pickup of Idol seemingly made ABC's "true" target a tenth higher last year, and in fact the cancellation of 38 Idol hours in 2015-16 had almost the exact opposite effect on Fox's target that season. So that's where I'm getting the number. This is also something I could apply to situations like Fox last year, where I could justify cobbling together all the stuff that wasn't on the 2016-17 sked (the extra hours of Lethal Weapon, Lucifer and The X-Files in 2017-18) to add 0.05 or so to the target. Another example might be a change in the NFL packages, like when CBS added Thursday Night Football a few years back.

No comments:

Post a Comment

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