Not So Simple: The Role of Religion in the 2004 Presidential Election

By Leonie Huddy, Stanley Feldman, and Sarah Dutton

Much ink has been spilled over the role played by religion in the 2004 presidential election. In the days immediately following the vote, pundits were quick to conclude that Christian evangelicals had handed President Bush his reelection victory, leading in turn to demands by evangelical leaders for political payback. The conclusion that the religious right was responsible for Bush’s victory in 2004 was based on several different sources of evidence: The popularity of “moral values,” which emerged as the most common reason given by voters for their electoral choice in exit polls (it was selected by 22 percent nationally); the passing of gay marriage bans in all the states in which they were on the ballot; and Bush’s apparent success in geographic areas with relatively large segments of evangelicals, such as rural Ohio, hinting at the effectiveness of campaign “architect” Karl Rove’s strategy to get out the Christian conservative vote in 2004.

But further postelection analysis offers a more complex picture of the role of religion in the 2004 election. A growing number of articles in political science journals and elsewhere provide evidence that the increase in Bush’s support between 2000 and 2004 was no greater in states with a gay marriage amendment on the ballot than those without, that other issues had a greater impact on Bush’s support than views on gay marriage, and that greater turnout among evangelicals was balanced by greater turnout among the electorate overall. Was the talk of a powerful religious voting bloc in the 2004 election overblown?

In a sense, neither side—not those who extol the electoral power of religiosity in 2004, nor those who decry it—is exactly right or entirely wrong in its analysis of the election outcome. Two questions lurk beneath the surface of this debate:

First, to what degree can Bush attribute his victory to evangelicals overall? Pew researchers have concluded that religious voters, especially white evangelicals, formed a powerful base of support for Bush in both 2000 and 2004. Moreover, they found that white evangelicals have identified increasingly with the Republican Party since 2000, a likely consequence of Bush’s well-known evangelical Christian faith.

This leads to a crucial second question: Did highly religious voters hand Bush his enhanced victory in 2004? The answer appears to be no (based on articles by Alan Abramowitz, Barry Burden, Gregory Lewis, and on Pew data) but deserves further scrutiny, given the political importance of such claims.

We delve more deeply into the impact of religion on the 2004 election by employing a statistical technique called logistic regression. We use this multivariate approach because highly religious individuals are not a homogeneous group; they vary on a range of demographic characteristics—such as level of education, income, race, and gender—that also affect vote choice. An analysis of religious voters based solely on crosstabulations of exit poll results does not allow researchers to demonstrate decisively that any observed difference in vote choice by religiosity is due solely to religiosity and is not the result of other demographic factors, such as gender or race, with which it may be confounded. The ability to separate the influence of demographic factors from religiosity is critical to evaluating fully the dynamics of the 2004 election. 

The national exit polls included roughly ten thousand respondents in 2000 and 2004 (reduced to under six thousand in each year who were asked the full battery of relevant demographic questions), but even that sample size is insufficient to assess the effects of religiosity while still controlling for all of the other demographic factors. A table that depicted vote choice by all possible combinations of gender, race, education, age, religion, and religiosity would quickly exceed even the large size of recent national exit polls. A multivariate technique such as logistic regression gets around this problem by controlling simultaneously for many factors that influence vote choice in order to isolate the electoral effect of religiosity independently of the other factors with which it is related.

Data from the 2000 election must also be a part of any attempt to achieve a fuller understanding of the dynamics of the 2004 election, and the use of a multivariate model allows us to examine the “pure” impact of religiosity across elections. It is possible that the characteristics of religious people varied slightly across the two elections—for example, the gender or racial makeup of religious people may have changed slightly between 2000 and 2004. Again, the multivariate approach controls for the effects of other potentially confounding factors, and can do so across both elections to assess whether highly religious voters gave Bush an added advantage in 2004 over 2000.