These aggregate figures masked wide disparities at the subgroup level. Smokers thought more people smoked than did nonsmokers. Bush supporters thought more Americans approved of the president than did those who disapproved of his performance. Supporters of gay adoptions saw an American public much more supportive of this than did opponents. Churchgoing, however, did not affect perceptions of how many people go to church. On these issues, cross-group differences varied from essentially zero to twenty percentage points. Themes that arguably are more intimately held saw greater degrees of false consensus, as the percentages in Figure 3 show.

To paraphrase the political adage, what you see depends on where you stand. On the questions of Bush's approval rating, gay adoptions, and biblical literalism, believers and nonbelievers saw very different publics. Even on smoking, those who did and those who didn't saw the public in somewhat different ways. Breaking down the respondents in these subgroups by education did little or nothing to close the gap between the separate camps. For instance, college-educated biblical literalists on average estimated that 49 percent of Americans are biblical literalists, versus an average estimate of 29 percent by college-educated nonliteralists-still a twenty-percentage-point gap. Cross-group comparisons on smoking, Bush approval, and gay adoptions produced similar results, with sizable gaps still remaining.

The false consensus effect is not significantly a function of ignorance or knowledge. More highly educated respondents in the Illinois survey clustered somewhat more tightly around their group averages, but those averages still differed significantly across attitudinal groups. Thus, the tendency to exaggerate the degree of consensus around our own positions appears to grow out of perceptions created by our circle of associates. To the extent that young people associate mainly with other young people, who smoke more, they overestimate smoking rates far more than older people, who smoke less (eighteen to twenty-four-year-olds in the survey estimated that 54 percent of Americans smoke, compared to an average estimate of 40 percent offered by respondents over fifty). Who we are determines what we see, and what we see drives what we report.

In sum, these findings challenge the assertion that the error in individuals' opinions tends to be random. Randomness of error is a major part of the argument behind seeing the public as rational, despite individuals' shortcomings regarding knowledge and longitudinal stability. Collectively, these respondents were nowhere near right in their estimations of the public on matters of smoking and gay adoptions. And while collectively they came up with the right answer on Bush's approval rating and biblical literalism, they did so while revealing a lot of bias along the way-bias that was anything but random, and which was largely a function of their systematic misperceptions.

Greg M. Shaw is associate professor of political science, Illinois Wesleyan University. The author would like to thank the students in his spring 2004 public opinion class for their hard work and the people of McLean County, Illinois, for their indulgence.