CAN WE GET AN ACCURATE ACCOUNTING OF WEBSITE VISITORS?
A Web site is a bit like a rowdy party: it's often difficult to tell just how many people have come and gone.
But while parties are all about levity, the number of visitors to a Web site is a serious matter. Stock prices rise and fall on visitor counts as dot-coms fight for the Internet audience.
Unfortunately, measuring Web site visitors is not a precise science and the two chief methods for keeping track seldom agree.
Research firms such as Media Metrix, Nielsen Netratings and PC Data measure popularity by monitoring Web surfers, while the Web sites themselves look at data recorded by their computers.
Bob Bowman, chief executive of computer retailer Outpost.com, is frustrated by what he says are gross undercounts of the traffic to his Web site by the ratings companies.
"Neither Media Metrix nor Netratings have any clue what they are doing," he said. "We completely ignore them."
Netratings said Outpost.com had 664,000 visitors in December, at the height of the holiday shopping season, when the site had 133,000 sales, according to Bowman.
The likelihood of his Web site making that many sales to just 664,000 visitors was, Bowman said, "the chance of me starting as a shortstop with the Yankees tonight." The true number of visitors, he maintains, was closer to 4 million.
Numbers not comparable
The numbers are not exactly comparable, for many reasons.
The Web-surfing panels that ratings companies use to monitor U.S.-based sites are limited to the United States, since surfers abroad are less important to advertisers and investors. And Web site computers may have a hard time telling visitors apart for technical reasons.
In the case of Outpost.com, Netratings was not counting people who surfed from their workplace. The company has since added a panel of people surfing from work, but this has not lifted the Web site's visitor numbers by much.
TV ratings have been counted for decades by boxes that register which channel a TV set is tuned to in the home. Media Metrix and Nielsen Netratings measure Internet use much the same way -- they pay people to install software that tracks their computer usage.
The main problem with that approach, says researcher Kirthi Kalyanam at Santa Clara University, is that while a TV panel of 5,000 people can give a pretty good picture of the viewership of a few hundred channels, a panel of Internet users is spread much thinner over a practically infinite number of Web sites.
Max Kalehoff at Media Metrix said the company uses an Internet panel of 50,000 people in the United States, and will report traffic for a Web site if a few more than 30 participants visit it in the course of a month.
A study last year by Kalyanam and Yahoo's director of research, Bruce MacEvoy, that compared panel ratings with Web site traffic logs found that panel companies often reported up to 30 times more traffic than log files on the computer servers of small sites.
Netratings' senior vice president of marketing, Tim Meadows, called the study flawed.
He said the Netratings panel of 3,000 people used in the survey had since been discarded as unrepresentative and replaced by 65,000-member panel that he says "can report on the top several thousands of sites with a high degree of accuracy."
But Kalyanam said even the larger panel is too small for reliable readings. He thinks the panels may need to be 100 times larger.
Panel ratings may work fine for large Web properties like America Online that have national coverage and attract tens of millions of visitors, Bowman says, but not for a company like his.
Most of Outpost.com's clientele is on the East and West Coasts -- out of sync with the more diffuse demographic that ratings companies sample.
A University of California-Berkeley study has shown that traffic ratings by Media Metrix are an important factor in the price Wall Street sets for a Web company.
Analyst Timothy Fogarty at Thomas Weisel Partners in San Francisco said he has seen stocks move in response to improvements in ratings -- but personally doesn't treat traffic numbers as hard facts.
"It's important statistical data for us to track, but it's not something we are going to base a recommendation on alone," he said.
Other auditors hired
Untrusting of the accuracy of Web traffic counts, some Web companies are now hiring outfits like ABC Interactive Inc, BPA International and Engage I/Pro to audit the traffic logs recorded by their computers.
The three auditing companies had been counting traffic differently but will launch a Web site this month called Audit Central where they will post reports compiled using the same methods.
Jason Heller at Mass Transit Interactive, a New York company that helps others place Internet ads, welcomes the initiative. While he looks at panel ratings when deciding on ad placement, he says he would be more trusting of a third- party audit.
That does not mean panel ratings will be less important, though. Server logs say nothing about what kind of people visit the site. It matters a great deal to advertisers if visitors are young or old, rich or poor.
Patrick Monahan at auditing company BPA International says there's plenty of room for both census-based audits and panel-based measurement. Magazines have for decades had their circulations audited, he notes, while at the same time contracting with research companies to study who actually reads them.
"You have to look at the two as complementary," he said. "By looking at both sides of the equation, you get a fuller picture of what that Web site has to offer."
Kalyanam said the Audit Central Web site and the auditing companies' joint standard may help bring some order to Internet traffic ratings, but only to a degree.
"The analogy is: we were totally in the Tower of Babel before this, and now some people on one side of the river have decided to speak the same language."