Lemmons testified that he compiled the list of sexually explicit sites that should have been blocked by entering the terms “free adult sex, anal sex, oral sex, fisting lesbians, gay sex, interracial sex, big tits, blow job, shaved pussy, and bondage” into the Google search engine and then “surfing” through links from pages generated by the list of sites that the search engine returned.  Using this method, he compiled a list of 197 sites that he determined should be blocked according to the filtering programs’ category definitions.  Lemmons also attempted to compile a list of “sensitive” Web sites that, although they should not have been blocked according to the filtering programs’ category definitions, might have been mistakenly blocked.  In order to do this, he used the same method of entering terms into the Google search engine and surfing through the results.  He used the following terms to compile this list:  “breast feeding, bondages, fetishes, ebony, gay issues, women’s health, lesbian, homosexual, vagina, vaginal dryness, pain, anal cancer, teen issues, safe sex, penis, pregnant, interracial, sex education, penis enlargement, breast enlargement, . . . and shave.” 
        If separate patrons attempted to reach the same Web
site, or one or more patrons attempted to access more than one page on a single Web site, Finnell counted these attempts as a single block.  For example, the total number of blocked requests for Web pages at Tacoma Library during the logged period was 2,812, but Finnell counted this as only 895 blocks of unique Web sites.  Of the 895 unique blocked sites, Finnell was unable to access 59, yielding 836 unique blocked sites for his team to review. 
        The confidence intervals that Finnell calculated
represent the range of percentages within which we can be 95% confident that the actual rate of overblocking in that particular library falls.  We note that these confidence intervals assume that the time period for which the study assessed the library’s internet logs constitutes a random and representative sample. 
        To illustrate the two different methods, consider a
random sample of 1010 web sites taken from a library’s Internet use log, 10 of which fall within the category that a filter is intended to block (e.g., pornography), and suppose that the filter incorrectly failed to block 2 of the 10 sites that it should have blocked and did not block any sites that should not have been blocked.  The standard method of quantifying the rate of underblocking would divide the number of sites in the sample that the filter incorrectly failed to block by the number of sites in the sample that the filter should have blocked, yielding an underblocking rate in this example of 20%.  Finnell’s study, however, calculated the underblocking rate by dividing the number of sites that the filter incorrectly failed to block by the total number of sites in the sample that were not blocked (whether correctly or