Public opinion polls are ubiquitous and influential. In the 1988 presidential election campaign, for example, some 50 national polls were conducted from June to August alone ( Public Opinion, 1988, 11( 3):36-40). Though influence is harder to demonstrate, surveys are widely alleged to affect both electoral turnout and election outcomes. Other far- reaching influences are also suggested, such as the apparent use by Israel of American public opinion data in making military and other decisions ( Gilboa, 1987).
Yet, for all this, polls are often difficult to interpret. Opinions can change with great speed and seemingly little provocation. Gaps may exist between opinions and behavior. More fundamentally, it is often difficult to determine just what public opinion is. Dramatic examples can be found of the sensitivity of opinions to question wording ( Mueller, 1973, chap. 1; Schuman and Scott, 1987). Individuals sometimes respond to questions in almost random fashion ( Converse, 1964). Parts of the public readily respond to questions asking about fictitious events ( Bishop, Oldendick, Tuchfarber, and Bennett 1980).
These and other difficulties make responses to isolated survey questions are an inadequate guide to public opinion ( Schuman, 1986). Instead, poll results must be used in a comparative manner. There are at least four such ways of analyzing public opinion data. The weakest form is to compare results with expectations. For example, if one has always assumed that news broadcasts are of great significance to the general public, it may be shocking to learn that, when telephoned within a few hours after a newscast, few television watchers can remember anything they saw. Such "surprises" are illuminating, but the surprise is mostly related to one's previous misconceptions and not to the precise numbers generated