Logistic regression has one other important feature: It allows us to construct predicted probabilities—that is, predictions as to how particular groups of people with identical characteristics on factors such as gender, race, and income would vote. Calculated on the basis of a logistic regression equation, predicted probabilities allow for the isolation of the “pure” effect of religiosity among specific groups of people, such as Jewish women or middle-class blacks, and they let us estimate the electoral effects of one single factor (such as religiosity) within the same demographic group. They also allow for a careful comparison of the effects of religiosity between 2000 and 2004 among very specific subgroups of voters.

Both the exit polls for these elections assessed voters’ gender, age, race/ethnicity, education, family income, religious preference, religiosity, and household union membership. We examined the impact of all of these factors on support for George Bush, as well as whether the effects of religiosity were confined to Protestants or were also exhibited among religious Catholics and Jews. This latter test sought to follow up on preliminary evidence presented by Pew researchers indicating that the political effects of religiosity extended beyond Christian evangelicals in 2000 and 2004 to include other faiths and denominations.

It is worth noting, first, that there was very little change in the electorate between 2000 and 2004, according to weighted exit poll data. Most important, and contrary to speculation concerning an increase in religious voters, individuals who attended religious services at least once a week comprised the same proportion of exit poll respondents in both elections: 42 percent. This finding mirrors evidence reported by a number of researchers that Rove’s efforts to turn out Christian evangelicals were matched by a comparable increase in secular voters.

Logistic regression results indicate that, in addition to making up such a sizeable chunk of the electorate, religious voters who said they attended services at least once a week also voted distinctively, underscoring their importance to Bush’s reelection. In both election years, these voters were more likely than secular voters (who never attended services) to support Bush, although the effects were slightly smaller in 2004 among Protestants, Catholics, and Jews. Not surprisingly, Protestants were more likely to support Bush overall than were Catholics, and Catholics more likely to vote for him than Jews.

But religiosity had much the same effect among all three religious groups, including Catholics. (The only demonstrated interaction between religiosity and religion occurred among Protestants in 2000, where the difference in Bush support between highly observant and secular Protestants was slightly greater than for other religious groups.) Thus, despite Kerry’s well-known Catholicism, highly religious Catholics were less likely than their secular counterparts to support him. These findings provide a definitive answer to our first question on the overall importance of religious voters: They constituted a major source of electoral support for Bush, across religions.