In the Illinois survey, respondents' ability to estimate the prevalence of the behaviors and attitudes mentioned in the false consensus questions varied dramatically across issues. Figure 1 shows the average estimates of Americans' opinions and habits on these issues, along with the correct figures drawn from national surveys, and respondents' reporting of their own opinions and habits on these points.

Respondents correctly estimated President Bush's approval rating and the percentage of Americans who call themselves biblical literalists. They greatly overestimated the percentage who smoke, somewhat overestimated church attendance, and significantly underestimated support for gay adoptions. In fact, these respondents were more like Americans on the whole than they realized, in that their reporting of their own habits and opinions produced results that more closely approximated national averages than did their estimations of others' behaviors.

Why did some estimates deviate from reality? It's difficult to say objectively how much public visibility each of these topics is given. Presidential approval rates likely receive the most media coverage; hence, the near-perfect estimation here might be expected, especially in an election year. It is less likely, though, that respondents have as many opportunities to learn the percentage of Americans who are biblical literalists. Despite this, respondents got it right. And smoking, a particularly visible habit, received the least accurate estimates among items. An issue's visibility, therefore, does not seem to provide a satisfactory explanation of how closely these respondents could estimate what others do and think.

Education offers another possible explanation for the accuracy of estimates. Respondents with college degrees might be expected to perform better at this task than those without. More educated respondents should be able to learn about their environment more efficiently than less-educated ones. Figure 2 shows estimates on these five items by education group.

As it turns out, in this simple bivariate correlation, college-educated respondents out-performed their less-educated counterparts on only two of the five items: church attendance and smoking. Neither group estimated the percentage of Americans who are biblical literalists with any particular precision. No significant cross-group difference appeared on the other two questions. Hence, the percentages in Figure 2 do not support a claim that education is a strong determinant of correct estimations of Americans' habits and opinions on these points.

Multivariate analysis of the impact of education on accuracy of estimations, simultaneously controlling for the possible effects of age, gender, household income, and the frequency with which respondents said they discussed politics, revealed that education was significantly associated with greater accuracy on only two issues (smoking and Bush approval). On Bush approval, the difference between those with college degrees and those without amounted to a mere one percentage point. On smoking it came to just over eight percentage points. Thus, on four of the five items, education made little to no difference between offering a close estimate and one wide of the mark. To estimate correctly the habits and opinions of others, respondents needed to overcome their own bias on those issues. However, even highly educated respondents failed to do so.