The wild bounces in the polls have puzzled many observers during this presidential campaign. Some of the exaggerated shifts in the standing of George Bush and Michael Dukakis result from a range of often subtle differences in polling methodology. But the most important source of the variations is the inherent softness of what is being assessed. Many millions of voters simply haven't made up their minds.
Once a voter has thought about his electoral choice and made a commitment, we can measure his decision quite satisfactorily. Should he be polled on his choice, using any reasonably framed set of questions, he will tell us whom he favors - giving the same answer each time he is asked. While he may change his mind as the campaign goes on, he isn't likely to. Considered decisions to support a candidate are quite stable.
Large numbers of people, however, do not make a voting decision until very late in the campaign, and some never make it at all. The latter, nonvoters, often pay so little attention to the race that they never form a preference for one candidate or the other. ``Screening'' questions are used in pre-election polling to separate out likely nonvoters - but these efforts meet with only partial success.
Thus pre-election samples, until very late in the campaign, contain in large numbers (a) people who will eventually vote but haven't yet made a decision, and (b) people who never will vote. When asked in a public opinion survey whether they favor Bush or Dukakis, some in both of these groups will offer a spur-of-the-moment choice. We make a terrible conceptual error when we treat these largely unconsidered declarations as though they are decisions.