The ways in which TV has historically been delivered have played a big role in how its measurement has developed. Terrestrial TV, for example, isn’t like digital media, where you can simply count the number of impressions delivered through an ad server — there’s no way of passing data from a TV set back to an operator or broadcaster. Hence the need for panel-based measurement, where special equipment is used to record TV viewing across a representative sample of the total population.
With newer TV delivery techniques, this same limitation doesn’t always exist. Some set-top boxes and smart TVs are able to send back relevant data on TV viewership and ad exposure, known as return-path data. And with measurement bodies increasingly looking to augment panel-based measurement with other large datasets, return-path data is playing a growing role in TV measurement.
The Basics
With terrestrial or over-the-air TV, signals are broadcast across a wide area and picked up by a receiver, which is a one-way relationship. But with cable and internet-delivered TV, there’s a two-way relationship. A set-top box receives data from a TV operator, and can report back which channels were watched and how long for using the same infrastructure (hence the ‘return-path’ label).
There’s a lot of potential scale in this data. The number of people who own a set-top box or smart TV capable of reporting back return-path data in mature TV markets is much larger than the number of people signed up to the major measurement panels.
But the data is fragmented across different operators, and isn’t really fit for ad measurement by itself. Hence, some measurement companies combine return-path data from a number of operators with other datasets, including panel-based data, to give a more accurate view of the entire market.
It’s worth noting that some definitions of return-path data cover any and all viewing data collected by a TV set or set-top box, including automatic content recognition (ACR) data, which is functionally similar in some ways, though has its own unique advantages and disadvantages. Often, these two types of data are defined separately. Nielsen, for example, distinguishes between RPD from cable and satellite set-top boxes and ACR data from smart TVs.
RPD isn’t new to the measurement world. Some companies, including some of the alternative currencies in the US, have used RPD as one of the ingredients in their measurement mix for a long time now. But many of the measurement companies which serve as the primary currencies in their markets, such as Nielsen and Barb, have only begun using or trialling return-path data more recently, seeking to do so in a way which doesn’t compromise their core panel-based measurement.
The Technical Details
Set-top boxes typically track basic information about device usage, such as which channel it was set to, at what time, and how long for. Where there is a return-path back to the operator, this data can be reported back, potentially linked to some sort of identifier about the device it came from.
This data by itself isn’t really useful. A Sky set-top box, for example, might tell you that a particular device tuned into ITV1 at 19:25 on Saturday, and stayed on that channel for 2 hours and 35 minutes. In order to make it useful it must be paired up with listings data and information on which ads shown at that time. So Sky could look up listings for that period, see that the household in question tuned in half way through The Chase Celebrity Special, and stayed for most of Casino Royale. Looking at which ads played during the relevant ad slots, Sky then knows which ads were played on that device during that viewing session. Aggregating this data across all set-top boxes and devices gives granular insights into content and ad viewership across a large audience.
Again, this isn’t really enough information to provide meaningful measurement by itself. This data alone only covers a fraction of the total viewing population, and crucially, unlike panels, this subset of the population isn’t necessarily representative of the wider population. In fact, it’s likely not to be. A costlier operator, for example, would likely have less reach across poorer households, while adoption of newer devices capable of delivering RPD is likely to be higher among younger audiences. Plus, demographic data is very valuable to advertisers in and of itself for helping them understand who exactly is watching their content.
So RPD is generally joined up with yet more data. Some measurement companies map anonymised household or device identifiers included in RPD to identity graphs, in order to glean demographic data, and collect RPD and ACR data from enough providers that they cover the majority of the population. Measurement companies might also make inferences by looking at data about the whole population of households which they collect data on, and those they don’t. A measurement company might know, for example, that lower income households are underrepresented in the pool of households it collects data from, and adjust its measurement accordingly.
Others use RPD alongside traditional panel-based data, which covers a number of gaps left by RPD (including cross-device viewing and co-viewing, as well as demographic information). Insights from the representative panel can be used to judge where an incomplete return-path dataset might be biased or skewed, and correct it accordingly.
Matching up RPD, or indeed any large dataset, with panel data is not a simple process. UK measurement body Barb has been exploring the use of return-path data for a few years now, first issuing a tender last year, but it’s still a work in progress figuring out how to fairly and accurately combine the two datasets. A blog post from Chris Mundy, CEO of RSMB (one of the companies working with Barb on big data integration) highlights some of the challenges and key questions. When not managed carefully, folding RPD data into panel data can actually increase sampling error rates. And, perhaps counterintuitively, using a smaller sample of RPD households who agree to share demographic information can be more accurate than simply using all available RPD data.
The Pros and Cons
The big benefit of RPD generally comes down to its granularity and the size of the datasets available, which can be a strong complement to panel-based measurement (or a better alternative, depending on who you ask).
As TV viewing has become more fragmented, the smaller sample-sizes involved with panel-based measurement have become more problematic. A common issue is ‘zero ratings’, where no panelists tune in to a particular show, meaning it appears that the show in question had no viewers whatsoever (which would be very unlikely to be the actual viewing figure). RPD data, given the much wider population whose viewing is being monitored, can help correct these extremely low ratings (as well as excessively high ratings).
Some proponents of RPD and other big data sets also say they can help correct for biases within panel samples. Comscore for example argues that only a certain type of person will likely be willing to sign up to a panel and put in the necessary work, and these types of people may be more or less likely to watch certain types of content.
Aside from the shortcomings of RPD described above, which may be fixed by combining it with other datasets, there are a few other issues. Set-top boxes don’t necessarily turn off at the same time as a TV screen, meaning RPD may log viewing at times when the TV itself is turned off. There are ways to minimise this issue, for example by recognising and discounting overly long viewing sessions, but it still remains a problem in some cases. RPD datasets are also not always consistent with each other, since they’re captured by a variety of different companies, and there may be differences in the types of data which various operators collect. Not all providers, for example, capture time-shifted viewing.
Finally, RPD relies on audience consent. That’s not an issue at the moment — companies using RPD in the US often measure tens of millions of households, suggesting consent rates are high. But there’s always the possibility for new privacy laws or shifting audience behaviours to limit the ease with which RPD can be collected.
Follow VideoWeek on LinkedIn.
