This paper shows that the time trend of HIV infections among gay men could be forecast from persuasive information in the press. The principal methodologies are detailed in Fan (1993). The mainstream press was likely to be a good gauge of the persuasive information acting on the public for AIDS related topics because the news media could be used to forecast time trends of public opinion and behavior for over a dozen topics (e.g. Fan 1988, Fan and McAvoy 1989, Fan 1993). The analysis began with the electronic retrieval from the NEXIS commercial database of the texts of 2462 out of 8728 AIDS stories in the New York Times and the Washington Post from January 1, 1981 to February 12, 1992. The chosen stories contained the word ‘AIDS’ within five words of one of the word roots ‘disease’, ‘illness’ or ‘immune’. The 8.0 million characters of retrieved text were scored by computer using the Info Trend computer programs previously described (Fan 1988). The algorithm depended on the occurrence of key words, their orders in paragraphs and the distances between them. For the scoring, the analyst enters a topic specific dictionary and a set of word relationship rules, usually about 50 in number. The resulting scores are in the form of numbers of paragraphs containing user specified ideas at specific times.

The computer first discarded all paragraphs except those containing the word AIDS' or the word pair ‘acquired immune’. Then, the AIDS paragraphs were scored by machine for their views on HIV transmission through: sex, blood, nonsterile needles, the mother, casual contact, not through casual contact, and miscellaneous including mosquitos.