Media Metrix Inc., a New York Internet measurement firm, has raced into the lead in this vital arena, becoming a hot Web stock in its own right. But lately, the company is getting puzzled looks from some of the Web companies whose traffic it measures.

"It's just wrong," says Bruce MacEvoy, director of research at navigation site Yahoo! Inc., of Media Metrix's report that the number of "unique visitors" to Yahoo dropped 5 per cent between August and September. Yahoo's own records, MacEvoy insists, showed "very substantial increases." Media Metrix executives stand by their data, saying Web sites such as Yahoo lack their sophisticated analytical methods. Thestakes are high, since Yahoo's success as an advertising medium depends on delivering hefty audiences.

Old-fashioned methods
The Internet was supposed to be the most measureable medium ever, able to instantly track and record millions of clicks. Yet most independent measurement companies use surprisingly old-fashioned methods.

Media Metrix, for example, recruits "panels" composed of thousands of users who agree to install special software that tracks every online move they make. They function much like the legendary squads of television viewers monitored by Nielsen Media Research, a unit of Dutch publisher VNU NV.

Increasingly, Web publishers and online merchants are questioning those findings and gathering vast quantities of information on their own by examining the digital trail visitors leave on their sites. They say that these "log files" that record visitors' movements are exact representations of their audiences and not projections.

Media Metrix says this approach can't distinguishbetween "unique visitors" and those who may log on twice in one day from different computers. What's more, the Web sites' own counts may include hits by "crawlers," the software programs that scour the Internet on behalf of search engines. Those can inflate recorded traffic even though no one is really looking.

"This is the only way to measure people" who visit Web sites, says Steve Coffey, vice-chairman and executive vice-president of Media Metrix. He adds that his firm's independence is another guarantor of the data's accuracy.The discrepancy between traffic figures logged by the two methods is sometimes significant, leading to charges of undercounting.

For example, Family Education Network, a Boston educational Web company, was startled to see its September "page views" recorded by Media Metrix - a measurement of how many Web pages are requested by users - plummet 40 per cent from the previous month. Its own files showed growth of 5 per cent in the same period.

The figure was all the more surprisinggiven that the back-to-school season in the US is typically strongest for Family Education Network. The company says it advertised heavily during that period, which should have built traffic. "You don't have any month that's better than September" for an education site, says Jonathan Carson, the chief executive officer of Family Education Network. "If you're dropping from August to September, you have life-or-death problems."

Media Metrix's executives say they don't have an immediate explanation for the drop in its traffic measurements from the Family Education Network site. One factor may be the makeup of Media Metrix's panel. It doesn't track audiences in school computer labs, from which Family Education Network says it gets a portion of its traffic, particularly once US student-users return to school in the autumn.

Until recently, Media Metrix also didn't track international traffic, which penalised companies that receive most of their visitors from overseas. That has been an issue for CMGI Inc.'s AltaVista, the popular Web search engine. More than half of AltaVista's visitors are from other countries, which the company says partly explains the discrepancy between its own data and that of various panel measurement companies. "Those page views aren't going to be captured," says Ralph DiMuccio, industry relations manager at AltaVista.

A study by FAST, an Internet advertising association, compared traffic counts by individual Web sites with those by panel-measurement companies on more than two dozen top Web sites. In some cases, the measuring agencies counted as much as five times more traffic than the sites did themselves. On other occasions, the tracking services counted only about 85 per cent as many visitors as the sites did. Media Metrix's Coffey says the study relied on year-old measurement data from Media Metrix that was gathered when its panels were smaller and less reliable than they are now.

Slinging mud
Media Metrix remains the most widely used measuring company. But competitors arenipping at its heels. Most rivals take the same approach, using random sampling techniques to recruit computer users who agree to install their tracking software. The measurement companies also conduct regular telephone interviews to find out how many users were on the Internet during a particular time period. They arrive at a number for the total online population by extrapolating from US census figures.

Even some of the rival measurement companies are slinging mud at each other over who has the most reliable data. NetRatings Inc., which is partially owned by Nielsen Media Research, says its method of recruiting panel volunteers and its use of telephone calls produce a more random sample than direct mail solicitations, a technique Media Metrix uses for its panel recruitment.

"I believe that Media Metrix has gotten away with methodological murder in this marketplace," says Tim Meadows, vice-president of marketing at NetRatings, which filed for an initial public offering recently. Media Metrix, for itspart, says it uses both telephone and mail solicitations.

Executives at the firm insist that mail solicitation is critical to reaching households that tie up their phone lines when they're online.