• by Tom Hespos , November 4, 2002

Some weeks ago, I weighed in with my opinion on how we should gather the data necessary to bring reach and frequency to the web. My opinion was that actual visitation data should be used, as media properties have ready access to it, and because panel data can be somewhat sketchy.

In looking at the issue a bit closer, I’m inclined to solidify my position that there’s no substitute for actual visitation data. I have two reasons for feeling this way. One reason is that I’ve been hearing mumblings from TV buyers that ratings were somewhat inflated during the upfront this year. Another reason is that I don’t think panel data will give midsize and small websites a fair shake.

Nielsen’s national ratings service projects television ratings by observing 5,000 television households with a device they call the “people meter.” The devices record all television tuning and report the data back to Nielsen. Additionally, the households using the People Meter are asked to have each member of the household press a button when they are viewing, so that Nielsen can marry these data back to demographic data they collected when the People Meter was installed.

I would hesitate to recommend a similar system for measuring Internet media usage. Why? Because we have more media choices open to us on the web. Sure, tons of us use the big portals and search engines fairly regularly. But what about smaller content sites? If we were to use such a system, I think we would have to significantly increase panel sizes to ensure that the smaller sites even register on the radar.

Think about the roster of websites you visit on a semi-regular basis. Odds are, large portals like Yahoo and MSN will be included in that roster. But what about the small sites you visit regularly? What are the odds that a panel of 5,000 would register the visitation on these small sites?

What’s more, I think there’s an issue of statistical stability. What if a situation occurs where the panel registers one user at a specific website during the course of a week, and then nobody during the next week? Can we use these data to project usage for the Internet population? Methinks we can’t.

Actual visitation data would give us a much more accurate picture. An actual tally of users and their demographic data, combined with Internet populations for a particular demographic, would give us the accuracy we need to gauge media consumption within a medium that is more fragmented than any other medium we’ve ever seen.

With panel data, how could we project ratings accurately? What if ratings were projected based on data from an Internet “sweeps” month or week? Small sites would put themselves at risk. When it came time to calculate actuals with a panel system, suppose the smaller site didn’t receive any traffic from the panel that week. What would happen? Would the site need to make good its entire ad schedule for that week?

Sorry, but this media guy’s opinion is still unchanged. Sign me up for a system that includes as much actual usage data as possible. And keep the panel-based stuff to a minimum.