The issue was brought to a head this year by Procter & Gamble Co. In P&G's words, "With media fragmentation, the diary is just not a viable technique." P&G recommends moving chockablock to people meters for better data and continuous measurement to eliminate sweeps abuse problems.
The goal is right; the cost for accomplishing it is wrong. Pushing it threatens any reasonable plan for improvement. Current revenue from the Nielsen Station Index (local market TV ratings) approaches $200 million, with stations paying more than 90% of it. The likely cost of a people meter panel large enough to report 125 markets approaches $500 million.
People meters are affordable in the largest markets, but a major-market fix doesn't fix national spot. For that, closer to 125 markets are needed, and in the smaller markets the plan unravels.
A sample of at least 300 households is required for each market regardless of size, so the half-billion-dollar-cost falls most heavily on smaller market stations that can least afford it. I've not heard advertisers or agencies offering to cover much of those costs.
RETHINKING THE PROBLEM
There's another way: To get the benefits of meter-based measurement doesn't require people metering the world -- just rethinking the problem. Set meters, which measure what's being tuned in on the set but do not provide viewer demographics, can do it as well and for far less money.
Any good research design puts the dollars where the variability is. In finance, the principle is called "working to the left of the decimal" -- spending dollars where they will have the most effect.
The greatest variability in audience measurement is in the TV channel selected. In a 50-channel universe, whether a household is tuned to a specific channel at a specific time is by far the most important piece of information to have for estimating viewers. This is what the set meter provides.
The variation in viewers-per-set is far smaller. It cannot be less than one and it's seldom more than two. Who is at the set is also restricted by who is in the household. If it's a single-person household, that person is most likely viewing when the set is tuned. Who is in the household is known in a set meter panel.
When there are several household members, location of set, time period and program content can also help identify probable viewers.
Because variation is limited, viewing rates can be modeled from regional people meter data and imputed to specific programs. This includes visitor viewing.
This technique will work even for local news, because the important piece of information -- how many households are tuned -- is measured for each station.
Household sample tuning information used with viewing data from the corresponding region of the national people meter sample can impute specific market viewing with good accuracy. Given time and effort, the power (and cost-effectiveness) of this technique can be proven.
Because of the structure of TV viewing, intelligent management of research dollars calls for large, market-by-market household meter samples and an expanded national people-meter sample. The set meter solution also eliminates sweeps-related problems by providing continuous measurement and, with some modeling, can be used in a single national/local database for optimization. Modeling viewers also makes it possible to use "set-top box" tuning data to report local cable audiences.
MORE MEASURED MARKETS
The trade-off? For fewer people meters we can have more measured markets and the larger samples needed to measure fragmented TV. Both are more important than people meters.
The overall design and timeline for an improved local service might be:
* Stage 1: Set meters expanded to top 75 markets. The national people meter sample increased to 7,500 households.
* Stage II: Set meters expanded to top 100 markets. People meters installed in top 10 markets.
* Stage III: Set meters expanded to the top 125 markets. People meters installed in the top 20 markets.
At each stage, household composition and people-meter viewers-per-set factors from the expanded Nielsen Television Index (national TV ratings) sample and Nielsen Station Index top 20 market sample are used to project viewers. Diaries continue to be used in smaller markets.
Obviously, there are many things to be worked out, including costs, timing and imputation methods, but all are well within our current capabilities. And a set meter/people meter/diary solution is affordable. The signal encoding technology pioneered by Smart TV, which is essential for any future TV measurement, will result in better, lower-cost set meter samples. And diaries need to be improved as we go, because in any realistic scenario they will continue to be used in smaller markets.
There will be resistance from both sides. When a respected agency media researcher was asked about data modeling of the kind I propose, he replied, "Making up numbers is my job." Stations, though not as clever, are just as cautious. They worry about local news numbers and the unique audiences attracted by the high-school basketball team.
We know sampling is a better way to get an accurate count in difficult-to-survey areas. But we also block new and better research methods with our own prejudices. The hard truth is if we don't start using more cost-effective techniques such as modeling, we will continue to argue, pointlessly, about better research no one will pay for.
Mr. Ephron is a partner at Ephron Papazian & Ephron, New York