A number of the audience big data modeling companies including 605, Videoamp and Innovid were the first to bring the idea to ARF. They said they needed to know things they couldn’t know on their own, in order to use set top box and Smart TV data to model viewing behaviors. Things like how many people shared specific devices, and how many shared subscriptions to streaming services and eCommerce companies, so that they could set up their models to take these phenomena into account. Individually they would not be able to afford measuring such things accurately and so they hoped by chipping in through ARF all the companies using return path and/or Smart TV data could use the same parameters.
DASH stands for Device and Account Sharing. The plan is to conduct DASH at least twice a year as the market structure continues to rapidly evolve away from analog to fully digital. A study of the universe based on a probability sample and conducted under ARF supervision by NORC at the University of Chicago, it’s the highest quality level achievable for a survey. ARF plans to seek MRC Accreditation. The sample size of this first wave is 10,452 adults and 410 teens 13-17. The response rate was 12%.
Seven companies are currently subscribing: 605, Videoamp, Innovid, Experian, Verizon, Neustar and NPD. It’s logical to expect that others who seek to compete with Nielsen also take advantage of this cost-effective way of improving their own quality. Nielsen, of course, gathers such information from its 100,000-person national peoplemeter probability sample including minute-by-minute viewing by specific person, device and room type, using measurement rather than modeling. Nielsen response rate is now about 30%.
When I first developed set-top-box data to research grade in the '90s, none of the problems that exist today existed then. There was not mobile device sharing because there were very few mobile devices capable of receiving video. There were no eCom subscriptions or streaming subscriptions. When a set-top box is hooked to a Connected TV (a Smart TV or one that is enabled by a Roku or other such device), and the viewer takes off into the internet to watch something, the set top box acts as if the viewer is still watching the last channel it watched on its set-top box. Nielsen is able to see that and credit properly within the panel, and when it uses big data, it uses AI methods to inform the big data based on the "tells" it finds within the panel (signals in big data associated with the switchover to internet).
Now the analytics companies aggregating big data on audiences are able by using DASH to create models which conform to known reality patterns that significantly affect the accuracy of raw big data. These challenges as I say did not exist at the dawn of STB data, but they exist now for STB data and for Smart TV data.
To give you an idea of the size of these problems, ARF has put out a General Market Report of the first DASH study wave which is pretty much required reading by anyone who wishes to stay up on the vicissitudes of audience measurement. Let’s take a look at some of the findings.
- In the report, ARF also quotes other pertinent recent research, such as a Truthset finding that in a recent big data set they scrubbed, "the accuracy of education level was only 33% while pet ownership was 92%."
The rest of these stats from DASH itself:
- Yesterday viewership of video on any non-TV device was around 50% for 18-34 but only about 20% for 55+.
- The percent who share a non-TV-set device with another household member is over 40% for Non-Hispanic and nearly 50% for Hispanic. This means that any measured ad event involving such a device cannot be assumed to be making an impression on the main user of that device. To whom then should it be credited probabilistically if you don’t have a peoplemeter panel, is the question that DASH answers.
- About 13% share their Disney+ account with friends in other households, and around 32% share Disney+ with relatives in other households. The stats are similar for other streamers, and DASH reports a number of them broken out individually for use in modeling.
- In order to provide parameters for modeling of co-viewing, the DASH General Market Report shows a grid of time of day by household size indicating how much co-viewing is going on in each cell, which the full DASH service granularizes by demographic groups and geographies. This reflects a yesterday recall questionnaire. Of the services studied, Disney+ has highest co-viewing in 4-person households accounting for half of yesterday viewers of that service.
By modeling using DASH, a company can prove its model is working in the dimensions locked in as DASH parameters. This does not guarantee that the program-by-program audiences are all correct, but it increases the probability of proximity to the truth.
It’s interesting that today’s audience measurement replacements for Next Century Media and TRA consider themselves modeling companies, as does the ARF. I always considered my companies measurement companies and used a trivial amount of modeling where Experian demographics were missing in fewer than 8% of the sample. Modeling dilutes measurement and so my preference is to use it as sparingly as possible. However, modeling is a must if one is using only big data and wishes to attribute impressions to human beings rather than to devices. This is probably the bedrock upon which advertisers worldwide decided that panels and big data must always be used together in the World Federation of Advertisers/Association of National Advertisers requirements documents issued in 2020.
Click the social buttons to share this content with your friends and colleagues.
The opinions and points of view expressed in this content are exclusively the views of the author and/or subject(s) and do not necessarily represent the views of MediaVillage.com/MyersBizNet, Inc. management or associated writers.