Dave, I understand the point about stability--or consistency. Obvously if you become overly reliant on the ratings that's what you want, not  abrupt changes in methodology--like the PPMs for radio or the people meters in TV  --in both cases replacing diaries.

What's striking, to me, at least, is  that while previous dramatic alterations in radio and TV "audienve currencies" didn't make everyone happy, they were accepted and the buyers and sellers made the required adjustments to cope with the changes in the data. But now, we seem to be going overboard on our reliance---and expectations--from Nielsen. I guess that's largely a function of the  wave of "digital" thinking that's sweeping the industry. We are demanding levels of micro-precision in our TV ratings that can not be attained. Sad.

Good points Ed. However, when adveriteers spend the better part of $80 billion a year on linear and streaming ads on TV, a tenth of a ratings points is many tens of millions of dollars a month, so it matters. Wehter it is perfect or not is less of an issue than it is the best that an be done with the available data and technology and that it is stable and comparable to other TV programming, other media alternatives and the past and, critically, trusted (as Jack taught me a long time ago).

Dave and Jack, of course, nobody can say   that  Nielsen is perfect --after all it, too, is only a survey--or now, a collection of surveys all mashed together. I think that our industry has become far too reliant on the ratings--carried down to a decimal point--as if they were reporting reality. Truth is that we will never know from any survey what reality is. Even when a seller like Amazon can pinpoint exactly how many sets  got its content, we don't know who watched.

The Gauge was a nice effort by Nielsen to promote itself and at the same time to inform everybody, in a general way, what's happening. Unfortunately it became a promotional toy for the pro -streaming, anti- "pay TV" folks.

Ed,, there is no question that Flks have complained -- for good reason -- about Nielsen numbers. My point here is that a big part fo the complaint was around sample bias and making sure panel and the big data were "fit" to population realities. DASH helps there. Yes. It is small. But the panel is much bigger than it used to be and its big data is enormous. Plus, the ARF created DASH and the MRC accredited it for the purposes that Nielsen uses it. We will never have perfrect TV ad measurement. But no question that it just got a good bit better.

Spot on Jack.

Certainly we know the basic trend will continue Ed, but the point here is the old way exaggerated non linear viewing and the temporary discontinuity will immediately fix that. Of course, smart advertisers and sources will know the change is only in the metrics, not the behavior. Still there will be bragging rights gains for some and lost for others.  

Again, I apologize for my "typos". My new computer has a hair trigger and deletes parts of words without any warning. Sigh!

 Dave, while I agree with most of your comments about how advertisers should look at TV and I have no beef agaist the ARF's DASH sstudy, all it is, using afairly small sample of 10,000 per year, is a survey. Various sets of respondents are interviewed and asked questions about how they get their TV and the results are melded together to estimate the way the whole country gets its TV--linear only, over-the-air or cable, streaming-only, etc. Calling it a "measurement"  implies a level of precision in obtaining the answers that may be unwarranted. 

As for what is happening, Nielsen has tradionally been forced to estimate the size of the TV-owning universe as reliable and current tat on this subject is rarely. I recall long ago--1960, to be exact--when I was very, very young---tht Nielsen pegged U.S. TV ownership at 90% but when the census came out the "correct" figure was 89%. So Nielsen had been putting out slightly "wrong" audience numbers for a while--but there was no big fuss about it. Nielsen merely revised its projection base downward slightly to conform to the "truth". 

Nielsen is doing more or less the same thing again,  only now the various segments --linear-only, streaming-only, both, etc.--- are being reweighted according to the DASH estimates. 
 
As a result, there will be  changes in individual show and seller ratings and in highly generalized general reports like The Gauge, but, as the folks at Nielsen will point out, the basic trends will  no doubt continue--slow but, for a time, steady growth in Streaming's share of total viewing.

Artie, very good points. Thanks very much for raising them.
As I called out in the headline of the piece, it is the DASH enhancement of Nielsen's core viewing data from its panel and from its Big Data set of second-by-second viewing from set-top boxes, smart TV's and apps that makes the difference. DASH's value is to help remove macro sample biases in those data sets, which capture viewing from many tens of thousands in the first case and more than a hundred million in the second. The almost 10% undercount  number is what is relfected in the new Nielsen Gauge numbers from a viewing reach number. And, as all who use Nielsen data know, the recalibratoins have had signifcant impacts in increasing total TV rated impressions. Thus, I felt comfortable describing it that way.
For sure, all samples and porojections carry error rates. On the antenna issue, the fact that it is not down says a lot. Of course, growth in antennas is also validated by consumer sales data of digital antennas on Amazon and smart TV sales with installed digital antennas.

Hi DAve, I'm not sure the conclusions here follow from the research. DASH is credible, but it estimates household access to TV services and devices; it does not measure what people actually watch. A household with an antenna or cable connection may still do most of its viewing via streaming. Likewise, the increase in antenna access from 16% to 17.1% is small and could easily reflect sampling variation without context (sample size, statistical significance, etc.) More broadly, Nielsen is introducing several methodological changes at once and explicitly says they are not guaranteed to increase ratings. Recalibrating the estimated universe is not, by itself, proof that broadcast and cable viewing had been undercounted by 10%.

I agree that it behooves advertisers to embrace linear and streaming in their strategies but to frame these numbers as a comeback story for linear seems premature at best.

And Josh, another 9-10% got over-the-air reception back then so these folks, too, could find any game they wanted if it was aired by a broadcast TV network or station.

The problem is that pre-streaming, 85% of us had cable, and all the games were on broadcast or cable. So finding the game wasn't a problem for 85% of us.

Now, in addirtion to broadcast and cable, major sporting events are spread across Netflix, Peacock, ESPN, Prime, Paramount+, Apple TV, and Max.  Plus local platforms (I subscribed to Gotham before the season so I could watch the Knicks; best $240 I ever spent.) 10 years ago, with one subscription you were set; now you need 7. According to Deloitte, 90% of US HHs. subscribe to at least one streaming service; among this 90% the average is 4 subscriptions. 

So of course people can't find games. 

Very interesting Joe, and thank you for Ed and Josh (and also Pat).

Accurate data is virtually imposible ... by the time you finish that start will have changed.

As a coincidence, last Monday the Australian Bureau of Statitics started our 2026 Census.   It is door-knocked to collect the data that the people in the home provided.   In our country is large but only ~28m people living in ~11.5 homes.   It takes over a year to get as accurate as possible for all people, homes, ages etc.

Ironically, if you have 'overnight data' based on a well organised statisically (and monitored daily) panel you can get acceeptable data ... often within 3% to 5% variations from day-to-day etc.

The key between daily data statistical methods is whether that the panellists are consistent.

Secondary, is the method that other data souces are often used.   An example (which I noted as a research method check) produced 'TV Ratings' that felt difficulty high.

Why?   If you were watching content on a TV (e.g. from a 3rd-party content source) and the viewer turned the TV off but left the 3rd-party source still providing content that wasn't seen on the TV or prior-active source.

Thanks, Josh. That's very interesting. 

I've written abiout this before. At VideoAmp we of course had no Nielsen data, and because they constrain clients from supporting comparisons of Nielsen data with other data, no aid was forthcoming. So the analyses we could do were limited. However, because many outlets report on Nielsen program ratings for persons 18+ and 18-49, we were able to amass a bunch of that from the Internet, and compare prime time program ratings for 18+, 18-49, and (derivable) 50+, for a bunch of programs.

We found that in general, we were ballpark close on 18+, but that VideoAmp credited a materially higher % of 18+ viewing to 18-49 year-olds, while Nielsen (this pre-dated BD+P) credited a materially higher % to persons 50+ 

We did a bunch of investigation, looking for context and external validation, and quickly ended up convincing ourselves that we were "right." One thing I put in a bunch of decks was that, when Nexstar acquired the CW, the trades reported that the median age for the CW was 58 (obviously based on Nielsen). This meant that for every 25 year-old watching The Flash, there was also a 91 year-old. Not surprisingly, we showed younger median ages across the board for networks, including news networks, which still skewed old for us, but less so.

Josh were you ever able to make a comparison with what the people meter was finding on  a total viewing basis as well as show by show or by genres? For example, let's say that according to Nielsen, the average person aged 55+ did 7 hours of viewing per day while the average 18-34 did only 3. Were your numbers, when aggregated, similar? Or for prime time mystery dramas did you find that 18-34s were 34% of the average minute audience while Nielsen's figure was close-say, 32%? 

The way we did personificaation at both VideoAmp and Comscore was essentially to start with a known roster of HH members (say from Experian), and take what wa on the set, and clculate viewing probabilities within the HH based on the emogrphy of the persons in the HH and what was on the screen. This assignment was trained by a panel-- at COmscore, Arbitron PPM; at VideoAmp, TVision.l While this might get individual HHs wrong, in aggregate (which was how both companies report) it works pretty well. 

There are age skew biases in footprints but this is where DASH becomes useful (and ideally you have a sufficient diversity of big data sources to mitigate the boias of any single one.) With DASH you can see the demographic distribution of, e.g., Set Top Box HHs, so the skew may be known. (You can even see demographics by provider; Comcast versus Spectrum e.g. Vizio vs. LG.)

Nielsen wasn't using the PPM in the TV panel. It's in radio metros and not overlapping HHs. They use it in TV for OOH and this is why they no longer treat guest viewing as in-home viewing-- because if they did, using PPM for OOH and meters fopr guest viewing double counts guest viewing (the PPM knows if you're in your own home but not in someone else's.)

Josh, if one of the big data panels Nielsen is using is based on set-top-box data doesn't this bias it in favor of linear, hence givimg  an older slant to the total household's set usage pattern. Also, the big data panels do not provide viewer info--only set usage--which, itself, can be mislading demographcally. 

Re the new wrist  watch people meter methodology, I am assuming that the old carry around or wear pager sized  devices will be abandoned in the 26,000 home people meter panel. If that's true and Nielsen attains a high level of cooperation re panel members always wearing their "watches", then the issue of people leaving the room while "watching" is no longer a concern. The CIMM study seemed to indicate that younger panel members were more likely ignore their button pressing obligations under the old system--thus uderstating the audience somewht plus slanting them slightly older. So their level of cooperation with the "watches" is faily important.

Regarding bullet 2, ARF announced this week that DASH is going to 2 releases a year, each with the most recent 12 months of data. While I know mwasurement companies don't like to change universe estimates during a broadcast year, this would enable Nielsen, if so inclined, to refresh universe estimates 2X a year. Users would likely balk at a mid-year universe chaange,but it could be desirable if a weighting variable is changing rapidly (e.g., streaming penetration.)

Regarding bullet 3, on HDAM: I'm somewhat surprised at the assertion that big data skews old. I'm very much under the impression that it is panel data that skews old. Perhaps Nielsen personification of big data, driven by the panel, was driving the older skew in the big data, so an AI model could be seen as remedial to reliance on an older-skewing panel. 

Bullet 4, improved weighting, the reader can't really parse without more detail. Better weighting is obviously... well, better... and because of the scale of big data, such data can support more granular weighting than a panel, without a problematic loss of reliability.*

Regarding Hispanics: I believe the National Hispanic Enumeration Survey is Nielsen's own survey, which their press doesn't seem to make clear. With Roberto Ruiz (long time Univision research exec) now atop Nielsen's research team, Hispanic representation will be well-stewarded. 

I'm guessing the ACR monitored tuning adjustment looks at sets they have metered, which are also in their ACR footprint, and using the meter as a truth set to correct for misidentification or (probbly more common) missed viewing sessions in the ACR data (e.g. if some content isn't fingerprinted.) If I understand this correctly, this seems like best practice.

Trying to parse "provider B householding": I don't think Nielsen wants to make HHs out of individual providers (I'm 99% sure the ACR providers can household their own devices with near-absolute certainty, since they get the ACR data back from the set via the HH's internet, so they know sets with common IP addresses.) Rather, I think Nielsen wants to household devices ACROSS providers. Provider B probably upgraded the signal/metadata provided to Nielsen, improving the quality of the match of their devices to the Nielsen HH graph that combines devices into HHs across providers.

_____________ 
*For example, hypothetically: 2 genders X 8 age cells X 4 race/ethnicity cells X 5 HH income breaks X 3 TV HH types would yield 960 weighting cells. In a panel of say 60,000 people that's an average of 62.5 persons per cell. For say 20 million HHs with 3 persons each on average, for the same weighting schema that would be 62,500 persons per cell. Way more lattitude in weighting granularity. That's more people per weighting cell than the entire population of Glendive.

People meter non-compliance cuts both ways. A viewer might log in when beginning a viewing session, but not logging off and back on when they briefly leave the room. Better tracking of egress and return could conceivably reduce audience to the commercial minutes.

There is data that shows that push-button meter panels skew older than more pasive collection. Dr. Pat Pellegrini did one such study in Canada for Numeris, comparing a people meter panel to Arbitron PPM; Yan Liu of TVision told me that there was a good study on this in Japan, but alas it's in Japanese. And CIMM did a study, I believe with TVision.

Dave, the main reason why linear TV is not dying as fast as the forecasters are predicting--and may not die out at all--is that it is designed to work for advertising as well as consumers, using many of the lessons learned from the networks' experieences with radio. So commercials almost always appear in well planned in-show breaks, a single standard rating source supplies independent estimates of the audience, the metrics are the same for  everybody, independent auditing is availble, etc.

In contrast, streaming was primarily oriented to allow each platform to do whatever it chose re ad positoning, audience measurements, verification that the spots ran and where, etc. on the ass-backwards assumption that it was the advertisers' responsibility to deal with most of the issues that arise, not the sellers'. Couple this with dubious ad view and "audience" metrics, total lack of standarization, and often bogus targeting mechanisms--plus the very high costs of programmatic buying and problems about fraud and you can see the contrast.

On the viewer side the differences are also huge. With linear TV viewers get to know exactly when their favorite shows are on--and where; with streaing, viewers are expected to decide exactly what content--often down to the episode by episode level, they want to watch and know where to find it every time they decide to watch some TV. Sorry, for many that's just too much of an every time hassle.

It will be interesting to see how much the numbers change when Nielsen (finally) updates their methodology for the Gauge in September. Also, their primary "Gauge" figure should be 18+ & Ad Supported, as that is what the Ad/Media industry cares more about than Persons 2+ and Ad Free & Ad Supported combined.

Spot on Gavin. Linear TV is getting $50 billion of ad spend for a simple reason, it works!