Collective Preferences in Democratic Politics: Opinion Surveys and the Will of the People (review)

What national surveys tell us about the nature of public opinion depends on whether we approach the data from a microscopic or a telescopic perspective. Those who look closely at individual responses, whether as interviewers posing questions or as investigators focusing on individual responses, are often struck by the ill-informed basis and unreliability of responses (Converse 1964). Those who examine broad trends produced when answers are aggregated and charted are frequently impressed by change over time and across socioeconomic categories that appears to have considerable regularity and meaning (Page and Shapiro 1992).

The two perspectives, microscopic and telescopic, can be reconciled by assuming that the noise inherent in responses at the individual level tends to cancel out, leaving a kind of collective rationality. However, Scott Althaus disputes the assumption that most individual response error is offsetting and that aggregation simply eliminates random error. He argues instead that survey results are seriously influenced by what he calls "information effects": "bias in the shape of collective opinion caused by the low levels and uneven social distribution of political knowledge in a population." Specifically, the poor, the young, women, and African Americans disproportionately opt out of questions on political issues (i.e., say "Don't Know"), and when they do offer opinions they do not, Althaus believes, express their own predispositions as they would if better informed. Thus survey results fail to represent adequately the attitudes of the total public and its important parts.

Althaus's book is largely an attempt to estimate how aggregate opinion would appear if all relevant social groups were optimally informed about politics. Extending approaches developed by Larry Bartels and by Delli Carpini and Keeter, he employs logistic regression to estimate what "fully informed opinion" would look like "by assigning the distribution of preferences held by the more highly informed members of a given demographic group to all members of that group, simultaneously taking into account the influence of a wide range of demographic variables." The measures of political information that Althaus relies on consist of straightforward factual items, such as asking respondents what office is held by William Rehnquist and which party controls the Senate. Of course, estimates derived on this basis do not guarantee that ill-informed respondents would actually offer the simulated responses if they were themselves well-informed, but the method does provide a plausible set of hypothetical answers.

The core of the book is a comparison of distributions of "fully informed" opinion with the actual distributions produced by the same surveys. Althaus first shows that, of 235 questions drawn from recent National Election Studies, nearly 7% show an average change in marginals when "fully informed" and surveyed opinion are compared. He also finds that this change in marginals is not due much to the effects of item nonresponse (DK answers), but rather to different patterns of answering by respondents with different levels of information. However, small marginal differences are not so telling in themselves, since we already know that changes in wording and in other aspects of surveys can shift marginals substantially, and particular survey items should not be treated as having quasi-official standing as referenda.

More interesting than changes in item marginals are specific analyses. For example, in one case Althaus shows that estimating the effects of greater information makes little difference in the attitudes of men toward abortion rights, but it increases support by women quite substantially and changes conclusions about gender differences. There are a number of other useful analyses reported throughout the book, and some topic areas are shown to be more susceptible to information effects than others. At the same time, the author's careful search for theoretical generalizations about where and why important differences in information levels occur does not bear much fruit. In this sense, identifying information effects becomes another analytic tool we can use in understanding data, not unlike important variables such as education. Indeed, it turns out that differences...