When the survey had been completed, a statistical procedure called chi-squared automatic interaction detection, or CHAID, was used to analyze the results. The measure of respondents' voting preferences indicated that 50 percent of Tennessee's adult population favored Bush, 39 percent favored Kerry, 6 percent were undecided, and the rest preferred some other candidate or refused to answer. In the election held roughly two weeks later, Bush beat Kerry in the state 57 percent to 43 percent.

Meanwhile, responses to the evangelicalism measure indicated that 65 percent of Tennessee adults described themselves as evangelical or "born-again" Christians. Using the same question, the Gallup poll put the figure at 42 percent nationwide in 2003. The evangelicalism measure correlated significantly with views about the Bible, with 59 percent of self-described evangelicals calling the Bible the "actual word of God, to be taken literally, word for word" compared to 22 percent of those who said they were not evangelicals. The relationship was evident despite the controls for demographics and frequency of worship service attendance. With the same controls in place, the evangelicalism measure was also significantly related to "witnessing" behavior. Just under 90 percent of self-described evangelicals said they had "tried to encourage someone to believe in Jesus Christ or to accept Jesus Christ as his or her savior," compared to 45 percent of those who said they were not evangelicals.

On the abortion question, 30 percent of evangelicals said abortion should be illegal under any circumstances, compared to 12 percent of nonevangelicals, although a majority of both groups opted for the "legal under certain circumstances" answer. Differences were sharper on gay marital rights, with 78 percent of evangelicals opposed to gay marriages or gay civil unions, compared to 46 percent of nonevangelicals.

Finally, the measure of ideological constraint indicated that, in keeping with Converse, constraint appeared substantially higher among individuals with higher levels of knowledge about where the candidates stood on the issues examined. Further analysis confirmed that, among both Bush and Kerry voters, the ideology scores of high-knowledge individuals averaged significantly larger than those of individuals in the midrange or low-knowledge groups. The interaction between the two scores, shown in Figure 1, was significant as well, with Bush supporters growing more conservatively constrained as knowledge increased and Kerry supporters growing more liberally constrained as knowledge increased. Score 1 for Converse and for the evangelicals and nonevangelicals.

Meanwhile, the average score of self-described evangelicals on the ideological constraint measure was significantly smaller than the average among nonevangelicals. So was their average issue knowledge. Score -2 for the evangelicals; they both know less and are less consistent.

Figure 2 shows the results from a CHAID analysis using demographic groups, frequency of worship service attendance, and evangelicalism as predictors of voting preference. The analysis identified white evangelicals and white nonevangelicals as two key subgroups whose members differed markedly in their voting preferences, with Bush attracting substantially more support among white evangelicals--especially better-educated ones--than among white nonevangelicals.

To discover whether these voting preferences sprang from different decision-making strategies, white evangelicals and white nonevangelicals were isolated in separate datasets, after which a CHAID analysis was run on each group with measures of ideological constraint, issue knowledge, perceived candidate morality, perceived candidate intelligence, opinions on gay marital rights, opinions on restricting abortion, and demographic group as predictors of voting preference. Figure 3 shows the results for white evangelicals, among whom the variable measure of perceived candidate intelligence was the strongest predictor of voting preference.

However, a majority essentially bypassed candidate intelligence as a consideration by characterizing Bush and Kerry as equally intelligent, and tended instead to choose on the basis of perceived candidate morality. Notably, attitudes on the hot-button issues of abortion and gay marriage did not contribute significantly to the results, compared to Popkin-style assessments of candidate intelligence and morality. Neither did the measures of ideological constraint or issue knowledge.

Meanwhile, Figure 4 shows the same analysis for white nonevangelicals. Here, perceived candidate intelligence was the strongest and only predictor for those who considered Kerry more intelligent than Bush. But among those who perceived Bush as Kerry's intellectual equal or superior, ideological constraint provided at least some guidance. Again, issue knowledge and attitudes toward abortion and gay marriage were not significant. Score .25 for Converse.