Is “Popular Rule” Possible? Polls, political psychology, and democracy

Larry M. Bartels
Larry M. Bartels Donald E. Stokes Professor of Public and International Affairs, Woodrow Wilson School, Princeton University

June 1, 2003

The celebrated political philosopher H. L. Mencken once characterized democracy as “the theory that the common people know what they want, and deserve to get it good and hard.” Democratic theorists have mostly focused on the latter issue, without taking seriously the complexities lurking beneath the notion that “the common people know what they want.” The ubiquity of opinion polls probing every conceivable aspect of modern politics and government both reflects and reinforces the notion that the primary problem of modern democracy is to translate definite public preferences into policy. Leaders may ignore the dictates of public opinion, but they are assumed to do so only with good reason—and at their electoral peril.

My aim here is to suggest that this conventional view of democracy is fundamentally unrealistic. Whether it would be desirable to have a democracy based on public opinion is beside the point, because public opinion of the sort necessary to make it possible simply does not exist. The very idea of “popular rule” is starkly inconsistent with the understanding of political psychology provided by the past half-century of research by psychologists and political scientists. That research offers no reason to doubt that citizens have meaningful values and beliefs, but ample reason to doubt that those values and beliefs are sufficiently complete and coherent to serve as a satisfactory starting point for democratic theory. In other words, citizens have attitudes but not preferences—a distinction directly inspired by the work of psychologists Daniel Kahneman and Amos Tversky. My argument extends Kahneman and Tversky’s research, which challenges the behavioral assumptions underlying conventional economic theory, to the realm of politics and emphasizes particularly the challenge it poses to the most fundamental assumptions of democratic theory.

“Framing Effects”

Kahneman and Tversky have called attention to “framing effects”—situations in which different ways of posing, or “framing,” a policy issue produce distinctly different public responses. Framing effects are hard to accommodate within a theory built on the assumption that citizens have definite preferences to be elicited; but they are easy to reconcile with the view that any given question may tap a variety of more or less relevant attitudes. The problem for democratic theory is that the fluidity and contingency of attitudes make it impossible to discern meaningful public preferences on issues of public policy, because seemingly arbitrary variations in choice format or context may produce contradictory expressions of popular will.

Survey researchers have been generating examples of framing effects for several decades in experimental work on question wording and question ordering. But only recently have they begun to think of them as manifestations of more general psychological phenomena—especially of the fundamental context-dependency of attitudes. The normative implications of question-wording and question-ordering effects for our understanding of democracy remain virtually unexplored.

Framing effects can be demonstrated most simply by noting the impact on survey responses of prompting respondents to consider some particular aspect of an issue that might otherwise have been overlooked. In a classic 1950 study by Herbert Hyman and Paul Sheatsley, half of a national sample was asked, “Do you think the United States should let Communist newspaper reporters from other countries come in here and send back to their papers the news as they see it?” To that question, 36 percent said yes. The other half of the sample was asked the same question, but only after being asked whether “a Communist country like Russia should let American newspaper reporters come in and send back to America the news as they see it.” In this second group, 90 percent agreed that American reporters should be allowed in Russia, and 73 percent-twice the share in the first half-sample-said that communist reporters should be allowed to work in the United States. Clearly, asking first about American reporters in Russia prompted respondents in the second group to apply a norm of reciprocity that blunted (though it did not entirely supplant) strong anti-communist attitudes.

What are we to make of this? Let us imagine for a moment that a referendum is proposed to settle the question of whether communist reporters should be allowed to work in the United States. Would the appropriate reflection of public sentiment be the 36 percent support for the proposal considered alone, or the 73 percent support after the norm of reciprocity has been stimulated? I suspect that most observers would consider the latter result more legitimate, on the view that it incorporates at least one additional consideration that citizens should consider relevant—and do consider relevant when it is brought to their attention. Then would a context that stimulated still other considerations be even more appropriate? Which others? Given the practical impossibility of stimulating every consideration that is logically relevant (or might be considered relevant by someone), how might we judge whether one set of considerations is better than another?

What’s in a Name?

Even more perplexing issues arise when the original question is not merely elaborated but altered. Some of the most striking framing effects on record come from question-wording experiments conducted as part of the 1984, 1985, and 1986 General Social Surveys. Respondents were asked whether “we” are spending “too much, too little, or about the right amount” on each of a variety of government programs. Separate random subsamples evaluated essentially similar sets of programs, but with more or less subtle differences in how each was denoted. Some of these subtle differences produced large differences in apparent public opinion. Most spectacularly, while only 20-25 percent of the respondents each year said that too little was being spent on “welfare,” 63-65 percent said that too little was being spent on “assistance to the poor.”

“Welfare” clearly has deeply unpopular connotations for significant segments of the American public and evokes rather different mental images than does “assistance to the poor.” But these different images are attached to the same set of programs and policies; any effort to make subtle distinctions of substance between “welfare” and “assistance to the poor” seems fruitlessly tendentious. Nevertheless, one frame suggests that a substantial majority of the public supports spending more on those programs, while the other—equally legitimate on its face—suggests that the same programs are deeply unpopular. What should a democratic theorist make of this perplexing situation? How might either question—or either outcome—be judged more appropriate than the other?

Sometimes even more arbitrary choices of question wording can produce large differences in opinion. Most people, for example, would presumably acknowledge that “forbidding” an action is substantively equivalent to “not allowing” it. But as Howard Schuman and Stanley Presser have noted, in three separate split-sample experiments in the mid-1970s, between 44 percent and 48 percent of the American public would “not allow” a communist to give a speech, while only about half that share would “forbid” him or her to do so. Substantively identical questions produce markedly different results. Which result reflects the public’s “true” opinion? I can suggest no sensible way to answer that question.

I interpret these framing effects as evidence for the thesis that citizens have attitudes rather than preferences. The contrasting patterns of responses documented in a variety of opinion surveys and experiments reflect real attitudes. The attitudes are neither meaningless nor whimsical nor—at least in any common—language sense-irrational. But neither are they the solid bedrock of comprehensive, logically consistent preferences that most liberal political theorists have taken as a starting point for democratic theory.

No Issue Is Immune

How ubiquitous are these framing effects? Are the examples cited here simply carefully selected anomalies, or are they the tips of icebergs? There is good reason to be wary of overgeneralization, given the wide variety of political issues and choice contexts in any functioning democracy, as well as the striking variation in information, motivation, and cognitive capacity in any mass public. But if these framing effects reflect fundamental aspects of the psychology of attitudes, they should appear with some frequency even in situations that seem ripe for the assumption that citizens have complete and consistent preferences over policy outcomes. This point may be dramatized by drawing examples from two issue areas often seen as prime exceptions to the generally disorganized and fluid character of American public opinion—abortion and affirmative action. The vagaries of survey responses evident even in these realms of unusual salience and concreteness reinforce the notion that public opinion is inherently sensitive to arbitrary aspects of how political issues are framed and political objects denoted.

Some of the complexities running beneath the surface of public opinion even on abortion are suggested by Paul Freedman and Ken Goldstein’s analysis of responses to two questions on the topic in American National Election Study surveys. In response to a general question about abortion in the 1996 survey, about 40 percent of respondents said that “by law, a woman should always be able to obtain an abortion as a matter of personal choice.” However, in response to a different abortion question posed in 1997, 39 percent of these same 1996 pro-choice respondents favored “a proposed law to ban certain types of late-term abortions, sometimes called partial birth abortions.” (Of the remaining pro-choice respondents, 49 percent opposed a ban on partial birth abortions; 12 percent were undecided.) That is, a substantial fraction of those who believed that abortions should “always” be permitted “as a matter of personal choice” also believed that “partial birth abortions” should be banned. While it might be possible to render these two positions logically consistent (for example, by stipulating the availability of some practical alternative to partial birth abortions in situations where they are now being chosen), it seems more straightforward simply to acknowledge that when it comes to public opinion, “always” never means always.

Perhaps these apparent contradictions in public opinion would disappear if political discourse were somehow elevated—but I doubt it. Political elites have had about as much chance of providing a clarifying debate on abortion as they have on any issue before the American public—without producing, at least by the evidence of Freedman and Goldstein, any clear reconciliation of the powerful competing values at stake.

Nor is it obvious how one could clarify public preferences regarding abortion policy through more careful question writing. Some of the ambiguities inherent in any such attempt at clarification are suggested by another analysis of data from the 1997 National Election Study survey, this one by Virginia Sapiro. Half the respondents were asked to rate “opponents of abortion” and “supporters of abortion” on a 100-point “feeling thermometer”; the other half were asked to rate “pro-life people” and “pro-choice people.” It seems fruitless to deny that these are, in essence, alternative ways of tapping exactly the same substantive attitudes. Nevertheless, Sapiro found that they produced significantly different results, with both “pro-life” and “pro-choice” people being rated much more favorably than abortion “opponents” or “supporters,” respectively. The differences appeared consistently among men and women, among more and less politically informed respondents, and among those who were themselves opponents and supporters of abortion. While these differences testify to the success of the rhetorical strategies adopted by abortion partisans in labeling themselves as “pro-choice” and “pro-life,” they do nothing to justify one’s faith in the reality of public attitudes toward them independent of the particular words by which they are denoted.

From Opinion Polls to Referenda

One might still object that questions in opinion surveys are a far cry from real political decisions and that peculiar responses to survey questions even about an issue as salient and fundamental as abortion have little genuine relevance for democratic theory. That objection seems to me misguided in both its aspects. Although most consequential decisions in democracies are made by representatives, and not directly by citizens in policy referenda, theories of representation are almost invariably grounded in analogous choices of policies by representatives, or of representatives by citizens, or both. As long as we continue to evaluate democracy in terms of the correspondence between citizens’ preferences and policy outcomes, all the same theoretical problems will reappear when we attempt to specify what kind of representation is most democratic. In any case, even if the specific formats and substantive content of the questions in some of the examples cited bear little superficial resemblance to the questions put to citizens in policy referenda or to representatives in the legislative arena, there is no reason to suppose that the same conceptual problems do not arise in posing “real” political issues.

The practical reality of these conceptual problems is illustrated by a 1997 referendum on affirmative action programs in Houston. As reported on November 6 in the New York Times by Sam Howe Verhovek, “the future of affirmative action may depend more than anything else on the language in which it is framed.”


“The vote Tuesday came only after a tumultuous debate in the City Council over the wording of the measure. Rather than being asked whether they wanted to ban discrimination and ‘preferential treatment,’ to which voters said a clear “yes” in California last year and to which polls showed Houston voters would also say yes, residents were instead asked whether they wished specifically to ban affirmative action in city contracting and hiring.


The legal effect was the same under either wording, but to this revised question they answered ‘no’ by 55 percent to 45 percent. . . .

Affirmative-action proponents around the nation hailed not just the result of Houston’s vote, but the phrasing of the referendum as a straight up-or-down call on affirmative action, and they said that is the way the question should be put to voters elsewhere.

Its opponents, meanwhile, who are already in court challenging the City Council’s broad rewording as illegal, denounced it as a heavy-handed way of obscuring the principles that were really at stake.”

Who is to decide what principles are “really at stake” in such a policy choice? If we accept, for the sake of argument, that a referendum using the original wording “taken almost directly from the Civil Rights Act of 1964” would have passed, as most observers seem to have believed, would that result have been more or less legitimate than the actual result? These questions are of a piece with those raised by the research of psychologists and public opinion researchers. Political elites who pose referendum questions must frame complex, difficult political issues in specific, concrete language. If citizens had definite, preexisting preferences regarding the underlying issues, any reasonable choice of language might elicit those preferences equally well. But democracy with attitudes requires some more detailed, normatively compelling account of what makes one frame more appropriate than another as a basis for democratic choice. In the absence of such an account, political debate and policy choice become a rhetorical free-for-all—a practical art in which, at best, the ends justify the means.

Democracy Camp?

Some political philosophers have proposed deliberative procedures that might be thought of as embodying normative theories of appropriate contexts for democratic choice. James Fishkin has proposed and implemented a “deliberative opinion poll” bringing random samples of citizens together with experts and moderators to study and discuss important political issues. The views expressed by these citizens after they have deliberated are taken to reflect “what the public would think, if it had a more adequate chance to think about the questions at issue.”

One may be tempted to wonder how adequately a few days of democracy camp would resolve the paradoxes and uncertainties summarized here. Would Fishkin’s moderators reconcile conflicting attitudes toward “welfare” and “assistance to the poor,” much less “forbidding” and “not allowing”? Would they discover some latent public consensus regarding the wording of Houston’s referendum on affirmative action?

More generally, the hopeful assertions of democratic theorists regarding the positive effects of deliberation are largely unsupported by systematic empirical evidence. Indeed, most observers of political deliberation have painted a much less rosy portrait than philosophers of deliberative democracy. New England town meetings apparently involve a good deal of false unanimity, with most important decisions settled in advance through informal networks reflecting preexisting inequalities in social status and political power. The atmosphere of public-spiritedness and mutual respect central to theorists’ accounts of democratic deliberation may be difficult or impossible to achieve in societies burdened by sexism, racism, and fundamental cultural schisms.

Popular Rule—Or Popular Veto?

The realization that attitude expressions are powerfully (and, in my view, intrinsically and unavoidably) context-dependent should spur democratic theorists to specify more clearly how political issues ought to be framed. Theoretical work along these lines may be inspired and informed by relevant empirical research, but first and foremost it will require a more subtle specification of the moral grounds on which one political context or institution might be deemed superior to another. It will not be sufficient to evaluate contexts or institutions by reference to their success or failure in reflecting citizens’ preferences, since that is merely to beg the question.

The most obvious alternative to theoretical progress along these lines is a much-diluted version of democratic theory in which the ideal of “popular rule” is replaced by what William Riker once characterized as “an intermittent, sometimes random, even perverse, popular veto” on the machinations of political elites. If that sort of democracy is the best we can hope for, we had better reconcile ourselves to the fact. On the other hand, if we insist on believing that democracy can provide some attractive and consistent normative basis for evaluating policy outcomes, we had better figure out more clearly what we are talking about.