Here's the table that will appear on each Vault page.
Info | Overall | Viewing/Competition | Lead-in | Seasonal | |||||||||||
# | Date | A18-49 | True | Sitch | PUT | Comp | Defl | C+D | Sitch | LI | eLI | Sitch | Mult | Sitch | |
1 | 9/16/2014 | 1.7 | 1.71 | -0.01 | -1% | 29.6 | 4.5 | 1.9 | 6.5 | -0.05 | 1.0 | 0.5 | -0.25 | 0.860 | +0.28 |
2 | 9/23/2014 | 1.3 | 1.51 | -0.21 | -14% | 30.9 | 9.1 | -1.2 | 7.9 | -0.19 | 0.8 | 0.4 | -0.26 | 0.866 | +0.23 |
3 | 9/30/2014 | 1.3 | 1.53 | -0.23 | -15% | 31.3 | 8.9 | -1.3 | 7.6 | -0.16 | 0.8 | 0.6 | -0.21 | 0.916 | +0.14 |
4 | 10/7/2014 | 1.4 | 1.47 | -0.07 | -5% | 32.4 | 7.8 | -1.2 | 6.6 | -0.06 | 0.9 | 0.7 | -0.19 | 0.892 | +0.18 |
5 | 10/14/2014 | 1.2 | 1.43 | -0.23 | -16% | 30.8 | 7.2 | 0.0 | 7.2 | -0.12 | 0.6 | 0.4 | -0.26 | 0.902 | +0.16 |
6 | 11/4/2014 | 1.6 | 1.50 | +0.10 | +7% | 33.2 | 4.3 | 0.4 | 4.7 | +0.13 | 1.9 | 1.1 | -0.12 | 0.945 | +0.09 |
7 | 11/11/2014 | 1.3 | 1.37 | -0.07 | -5% | 31.0 | 7.5 | -0.3 | 7.2 | -0.12 | 1.9 | 1.7 | -0.01 | 0.960 | +0.06 |
AVERAGES: | 1.40 | 1.50 | -0.10 | -7% | |||||||||||
STDDEV: | 0.18 | 0.11 | |||||||||||||
STDDEV%: | 13% | 7% |
Overall
These have always been included in the tables: the True score as well as the "Sitch," which is the difference between True and A18-49 (or how much the timeslot is affecting the rating according to this formula). In this table, it's expressed both in numerical and percent terms, as the numerical one is clearer for breaking down how much each individual section matters. Some other numbers are included and will be explained below for the diehards, but all you really need to know is that the Sitch in each of the three sections adds up to get the total effect of the formula.
Viewing/Competition
This part of the formula was the most drastic overhaul for 2014-15. In previous years, there were three components: overall viewing (PUT), competition, and what I called a "holiday" number, which was designed to offset the very low competition figures in low-viewed situations. The basic idea behind "holiday" was that there are two types of competition: competition from other things on TV, and competition from things that pull people away from TV. "Holiday" would account for the second kind.
I decided that the PUT adjustment and the "holiday" one were kind of redundant and endeavored to make it more simple: expressing "competition" as one number that combines "TV competition" and... "non-TV competition" or "holiday" or what it's currently called, "deflation." Taking the overall viewing component out of the formula has a couple significant benefits: it puts less emphasis on the rough PUT estimates, making the formula a little less volatile on a weekly basis. And it also flattens out the weeknights, reducing the viewing-based bias against Sunday shows and toward Thursday shows that has been here from the beginning. (Though it's possible that it flattens out the effect too much for Friday shows; we'll see how it goes this year.)
So here's how viewing/competition are treated: the key number here is the sum of competition + deflation (C+D). The "normal" C+D is 6.0, with "normal" competition expected to be roughly 6.0 and "normal" deflation expected to be roughly 0.0. the True formula gets a bonus for each point above 6.0. This bonus is 0.1 per point for shows with a rating of 2.0 or less, and 5% of the A18-49 rating per point for shows above 2.0.
Here's a vague explanation of what competition and deflation actually are:
Comp (competition): This number should be close to the sum of the A18-49 ratings on the other broadcast networks, but there are some major adjustments in play as well. The biggest is that the 10:00 hour has a significant addition to account for the lack of available Fox/CW numbers. Other sizable exceptions: there's a small addition for all Sunday shows due to the glut of cable competition; broadcast sports events are only counted as 75% of their ratings; and some cable sports events (Monday Night Football, Thursday Night Football and some postseason events) are added in at 50% of their ratings.
Defl (deflation): This number combines two formulas that test the relationship between overall viewing and broadcast viewing. Basically, it asks what should the PUT be given the broadcast viewing, and what should the broadcast viewing be given the PUT? When these numbers are lower than they "should" be, Defl returns a positive number that estimates how much overall viewing is deflated. It tends to be sharply negative in high-competition situations (especially football Sundays and Mondays), tempering the effect of competition. I'd like to be able to put that negative portion into the competition and have Defl be close to zero on all weeknights, but I don't have a good way of doing that yet. Maybe next year.
I will note that I held off on releasing all of this for awhile primarily because I was concerned about how low the overall viewing estimates were early in the season. Very early in the fall, I adjusted the normal C+D from 5.0 up to 6.0. And it could probably be another point higher. The good news is that this is affects everything equally. So I might be concerned that everything should be a tenth or so lower in True, but it doesn't really matter from a power rankings standpoint.
Lead-in
This part of the formula is almost unchanged vs. last year. The big change last year was the incorporation of a compatibility adjustment, which treated a show's lead-in as lower if it had a significantly different 18-49 skew.
"LI" is the literal rating for the half-hour before a program aired.
"eLI" (or "effective lead-in") is the number including this compatibility adjustment.
A half-hour show loses 0.1 in True for every 0.5 that its eLI is above the projected league average (1.7), while an hour show loses 0.1 in True for every 0.6.
Seasonal
This part got pretty heavily overhauled as well. In previous editions, I've broken the season apart into two or three pieces and applied one multiplier to each chunk based on "fall hype"; in other words, it's always seemed like early-season episodes are inflated even after applying situational adjustments. I'm trying something new this year, looking at this general idea not as "hype" but as another way of expressing what I call the "league average decline."
Broadcast entertainment original ratings are down about 10% every 52 weeks; when comparing things year-to-year, that 10% gets baked into the cake as a one-time event, but it's surely a more gradual process when breaking it down within a 52-week season. This multiplier ("Mult") applies roughly a 10% "league average decline" evenly across 52 weeks. I wanted it to center around the middle of the regular season, so the break-even point is week 18 rather than week 26. This means that an average week 1 show gets about a 3.5% deduction and an average week 35 show about a 3.5% bonus. (And a show at the end of the summer (week 52) gets about a 7% bonus.) For most shows, this is a fair amount less drastic than the old "fall hype" adjustment; far fewer shows will be exploding in True in the second half of the season. It may ultimately prove to be not drastic enough.
There's an additional skew-related component to this, recognizing the tendency of younger-skewing shows to collapse late in the season because viewing drops more among the younger set. Young-skewing shows like New Girl and The Vampire Diaries get much more severe seasonal adjustments, while ancient-skewing shows like Blue Bloods have a very tiny multiplier across the whole season. In the early season numbers currently available, that means the seasonal adjustment is hurting those young-skewing shows' True numbers considerably, but it will be repaid to them in the spring when their ratings will be expected to drop more. (So if you're wondering why Mult doesn't go up uniformly for every episode, it's due to fluctuations in skew.)
Standard Deviation
Over the years, I have always used the A18-49 and True standard deviations as a test of how the formula is working. The idea is to create a number that's more consistent over the course of the season than raw ratings, so True should have a lower standard deviation than A18-49 in most cases. I've decided at my own peril to put that test into these tables as well, so you will be able to see how much value True is actually adding. Many shows already see significantly lower standard deviations in True (including the one cherry-picked for the above example!), but there are a fair number of shows for which True is doing slightly worse at the moment. For series that have had the same viewing/competition/lead-in pretty much all season, the formula does not add much value right now, so the standard deviation is worse just due to PUT-related noise. A great deal of True's value is in accounting for declines later in the season; so ideally, by the end, the overwhelming majority of shows will have a "better" (lower) standard deviation in True.
I also added the standard deviation as a percentage of the average. (Real statisticians call this the "coefficient of variation.") The percentage works best for comparing across multiple shows, though the actual standard deviation may be better for comparing A18-49 vs. True for one show (since the % might be lower simply because the True numbers are higher).
This is about all I have to say. For the most part, I don't want to get much more specific than this, but if you search the archive of True-related posts you can probably get some sense about some of the older components not detailed here. As of this post going up, the second tables are now live for 2014-15 originals on Tuesday, Wednesday and Thursday. The rest should be up by the end of the day. For now, the True/Sitch numbers are also included in the first table on each Vault page, but that's going to change pretty soon.