1 2 3 4 5 6 7

only people who can afford access to it are the major corporations. For the small, Mom-and-Pop operation to be able to compete equally--one of the dreams of the Net--the data have to be free.

Hoffman and Novak thought they'd gotten their wish when, in early 1995, they got the CommerceNet consortium, of which they are members, to agree to supply up to $100,000 to fund the study. Out of twenty-odd proposals, the committee of five (including Hoffman and Novak) selected A. C. Nielsen, the well-known TV ratings company, which they felt had the best proposal and also had significantly underbid the rest of the field.

Nielsen's methodology was a familiar one: select a nationally projectable sample and conduct telephone interviews using a carefully designed survey questionnaire. In order for the sample to be projected accurately, however, its makeup has to be compared to known census data for the target population (in the Nielsen study's case, the adult population of the United States and Canada). If, hypothetically, 20 percent of your sample were under twenty-five, but official census data shows that in the general population 30 percent are under twenty-five, you need to take this difference into account when you project the results of the survey onto the larger population. While you would probably refine your selection procedures to choose a more representative sample in the case of such an egregious discrepancy, it's typical for samples to vary slightly from the make-up of the larger population. This difficult but well-established process of analyzing the statistics, making these comparisons, and adjusting the results to take these differences into account is called weighting. If this all sounds too complicated and mathematical, think of it like balancing a tire and applying weights to eliminate small imperfections that are unnoticeable at ten miles per hour but make the car vibrate noisily at fifty-five.

When Nielsen weighted the data and released the results in November 1995, they were unexpectedly high: the number of Americans and Canadians sixteen or older with access to the Internet was projected at 37 million, of which 24 million used the Internet, 18 million used the Web, and 2.5 million had actually used the Web to make purchases.[6] Hoffman and Novak, who had proposed the study and were expected to endorse it, immediately challenged these figures. In April 1996, they released a reanalysis of the same data claiming the figures were inflated due to errors in the weights. "The average inflation due to deficient weighting alone is 20.6%, the average inflation due to inconsistency alone is 13%, and the average total inflation in the original CNIDS [CommerceNet/Nielsen Internet Demographic Survey] estimates, when adjusted for the combined effect of these critical flaws, is 38%," they wrote. "As such these estimates lack validity and are of little value to decision makers."[7] Imagine you're running a business, and you're thinking of advertising on Roseanne, and you're basing your marketing plans and advertising expenditures on the assumption that the audience is more than a third larger than it actually is. Hoffman's and Novak's corrected estimates were 28.8 million with access, 16.4 million actually using the Internet, 11.5 million using the Web, and 1.5 million who had used the Web to purchase something. Nielsen, which had priced copies of its full report at $5,000, disagreed with this reanalysis,[8] but Hoffman and Novak's paper was accepted for publication in the peer-reviewed journal Communications of the ACM and Nielsen eventually revised its estimates.

One reason Nielsen's figures seemed so high was that several other surveys, released around the same time, came out with much lower numbers. Specialist Internet publisher O'Reilly and Associates 1995 study came out with 5.8 million adults using the Internet; consulting and research service FIND/SVP estimated 8.4 million adults and 1.1 million children use the Internet; and Times Mirror came up with 25 million adult Americans online.[9]

The wide variation among those numbers illustrates a different problem: there is little agreement on how to define such basic concepts as "Internet access" and "Internet user." All of America Online's 8 million users have Internet access, but that doesn't mean they use it. Leaving net.prejudice aside, AOL members who send email and have chat sessions only with each other aren't using the Internet, they're using AOL. Similarly, a business may connect its network to the Internet but give