A website purporting to encourage improved broadband Internet access in the U.S. runs the risk of undercutting its argument by playing games with statistics.

www.speedmatters.org, a project of the Communications Workers of America, has released its second survey of broadband Internet speeds in the U.S., noting that there has been little improvement since last year, comparing the U.S. unfavorably to other countries, and recommending a national high-speed Internet policy. The website also enables visitors to check their own connection speeds and perform comparisons.

My DSL results, 812kbps upload, are even lower than those of Idaho as a whole -- 1224k bps -- as well as lower than those of the U.S. -- 1242kbps. In comparison, the site notes, users in Canada get 7782kbps, users in France get 18022kbps, users in Finland get 22221kbps, users in South Korea get 50688kbps, and users in Japan get 65126kbps. (The good news is that my upload speed, 723kbps, is better than all of those areas except Canada and Japan.)

The media was quick to jump on the report as evidence that the U.S. was losing ground technologically. "The U.S. ranks 15th among industrialized nations in average Internet speed," reported a worried USA Today article.

Now, it is indeed true that the U.S. is behind other developed countries in terms of providing high-speed Internet access, but the technique the website is using is comparing apples and oranges -- and ironically might make U.S. availability look better than it actually is.

Statistics is the sort of thing that makes people's eyes glaze over, but the one factor that's required to make it work is a random sample. If surveyors pick and choose among respondents, or if respondents get to pick and choose whether to answer, then the results are suspect.

So, the speed test is not a 'survey' per se, because it's not random. It's reporting the results from a self-selected group who knew about the speed test, had the time and interest and ability to take it -- and who had Internet access in the first place so that they could. But the people most likely to fall into that group are the Neterati -- the very people who might be 'busting the curve' by having better access than most other Americans.

Moreover, the speed test takes this data, from actual speed tests, and compares it to "typical speeds available in the listed countries." But what is "typical"? What would the "typical" speed be in the U.S., if measured in the same way? Where did this data come from? From the Internet providers in the other countries? What's to keep them from providing the overly rosy results that U.S. communications companies give when asked about broadband penetration?

It also isn't stated how many respondents there were from individual states; a few people with really good or really bad connectivity could slant the results. Looking at the pages in the 65-page report from individual states shows that large parts of a number of states didn't have any respondents at all.

The report also doesn't break down results very much, instead grouping all results into three categories: less than 768kbps, more than 6mpbs, and somewhere in between. Even heavily wired states such as California showed few respondents over 6mbps, and the report doesn't separate the people who have only dialup access from people who have slow DSL or cable Internet.

The report goes on to list "Eight steps to affordable, high-speed Internet access for all," and they are reasonable, which makes it all the more unfortunate that the group is leaving itself open to criticism in this way.