Tyler M. Ransom and Dr. Richard Butler, Economics

Research has shown that voters show partisanship—propensity to vote along party lines—in Presidential elections. My project explores whether a person’s religiosity, as measured by affiliation variables, has an impact on partisanship and vote choice. My coauthors and I develop a religious partisanship index (RPI) that gauges the relative importance of religion between Democrats and Republicans in Presidential elections from 1980 to 2000. We find that, in every election of the 1980s, religion increased Republican partisanship, but since President Clinton (the elections of 1992, 1996 and 2000), religion increased partisanship among Democratic voters.

While there have been many studies on the topic of partisanship, there has not been a clear, widely used definition of it. We rigorously define partisanship as the conditional probability of voting along party lines given party affiliation (e.g. P(R|R) is the probability that a Republican votes for a Republican candidate). We then look at the difference between P(R|R) and P(D|D) using a Gini-type index of inequality and paying special attention to how religion affects the disparity measure.

We use poll data collected by the Center for Public Interest Polling (CPIP), a stratified, random sample of New Jersey residents. This survey-based data is derived from questions to New Jersey residents about their demographics, political and party alignments, and which candidates they voted for in major elections. We use only those surveys that contain data relevant to Presidential elections and study how an individual’s party and religious affiliation influences his or her tendency to vote for one Presidential candidate over another, paying special attention to the party membership of each candidate. The religious affiliation variable is divided into Catholic, Protestant, Jewish, other religion, and no religion.

We measure the disparity between the two probabilities via regression analysis. In terms of multivariate regression notation, the sample partisanship voting equation regression for a presidential election can be written as

1)

where XiB is the vector of non-religious variables with associated estimated coefficients (B) and ZiY are the religious variables with associated coefficients (Y). The difference in the likelihood of partisanship voting between Republicans and Democrats (P(R|R) and P(D|D), for the typical ith Republican and typical jth Democrat, can be written as

2)

To separate religious differences out we add and subtract from (2) the following two terms, dropping the “^”-notation from the predicted β and γ coefficients for simplicity:

3)

Equation 3 gives the “as if” measures for the other party. In other words, one can observe how a Democrat would respond if he or she had the religious characteristics of a Republican, and vice versa. The foundation of the RPI lies in these counterfactual, “as if” terms.

By adding and subtracting the counterfactual terms in (3), the “typical” partisanship difference in outcomes on the right hand side of Equation 2 can be rewritten as

4)

where 0 ≤ α ≤ 1, a weighting factor possibly determined by the relative number in each group. The RPI is measured by focusing on a pair of the terms in (4): the (relative) Republican RPI would be the percentile change in voting likelihood when going from XjDβD + ZjDγD to XjDβD + ZjDγR, calculated from a Democrat baseline, or going from XiRβR + ZiRγR to XiRβR + ZiRγD, calculated from a Republican baseline. Once we calculate the percentile change, we then look at the distribution of voting likelihoods and find out how many percentage points correspond to that percentile change. This puts the partisanship difference into more intuitive, measurable terms.

Our expectations of more religious partisanship among Republicans than Democrats during the Reagan Administration holds, but then diminishes and reverses during the Clinton Administration. In the ’80 and ’84 elections, religiosity amounted to a 4% partisanship difference in favor of the Republicans. By 1992, that difference was 4% in favor of the Democrats, agreeing with existing research that claimed President Clinton had received a religious boost in his victory. The election of 1996 showed a diminished Democrat advantage at just fewer than 2%, and 2000 showed virtually no partisanship difference attributable to religion, perhaps because voters followed party lines irrespective of religious affiliation.

A manuscript analyzing in detail the above study has been submitted for publication in the American Journal of Sociology. The associated computer code used to calculate the disparity index is available from the author upon request.

References

• Butler, Richard J. and James B. McDonald. 1987. “Inter-distributional Income Inequality.” Journal of Business and Economic Statistics 5:13-18.
• Kellstedt, Lyman A., John C. Green, James L. Guth and Corwin E. Smidt. 1994. “Religious Voting Blocs in the 1992 Election: The Year of the Evangelical?” Sociology of Religion 55:307-26.
• Oaxaca, Ronald L. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14:693-709.