The Stained Glass Ceiling: Social Contact and Mitt Romney’s “Religion Problem”

Earlier drafts of this paper were presented at the 2009 annual meetings of the Midwest Political Science Association and the American Political Science Association, as well as the 2009 conference, “The Change Election?,” held at the University of Notre Dame. We are grateful for the helpful comments of Jamie Druckman, Geoff Layman, Chris Karpowitz, and three anonymous reviewers. Our participation in the 2008 Cooperative Campaign Analysis Project (CCAP) was supported by the Rooney Center for the Study of American Democracy at the University of Notre Dame, the Ray C. Bliss Institute of Applied Politics at the University of Akron, and by the Center for the Study of Elections and Democracy at Brigham Young University.

More technically, membership in the Church of Jesus Christ of Latter-day Saints. Here we use “Latter-day Saints,” “LDS,” and “Mormon” interchangeably.

Putnam and Campbell (2010), report that the mean feeling thermometer for Mormons within the American population is 48 degrees, lower than the ratings for Jews (59) and Catholics (58) but higher than the feeling thermometer score given to Muslims (44).

Interestingly, these figures are comparable to the percentage who said they would not vote for a Catholic in 1960 (29 %) and to the percentage who said they would not vote for a Mormon in 1968 (also 25 %) (Jones and Jeffrey 2007).

The resulting opinion of the group is likely to be positive simply because of the positive nature of family and friendship ties. However, the same logic would apply for individuals who had negative relationships with family and friends because such individuals are still likely to have made up their minds about the group.

For the technical details of the matching procedure employed by YouGov/Polimetrix, see Jackman and Vavreck (2010). For evidence regarding the representativeness of samples drawn using this method, see Vavreck and Rivers (2008). Note that the survey oversampled battleground states, such that voters in non-battleground and battleground states are represented in equal proportions. Sherkat (2007) presents evidence that fundamentalist Christians are under-represented in the General Social Survey which, if also true for the CCAP, could potentially present a problem for population estimates using these data. However, since one factor for fundamentalists’ non-response to the GSS is the perceived social distance between fundamentalists and the highly educated, female interviewers who conduct the study’s face-to-face interviews, it is not clear that the same would apply to a truly anonymous internet survey. Even more importantly, however, even if the CCAP does under-represent fundamentalist Christians, or any other group in the population, it does not affect the internal validity of our experiment.

Because of the concern that the timing of the experiment might affect how respondents reacted to information about Romney’s religion, we have also interacted the treatment variables with a variable measuring the data of interview. Results are unchanged (details available upon request).

Note that there is no middle category among the response options. Respondents had to indicate that they were more or less likely to vote for the candidate; they could not say that the information had no effect on their vote choice. This was done to ensure that respondents did not reflexively select “no effect” out of a desire to take the path of least resistance. We account for the absence of a middle category by only reporting comparisons between treatment conditions. If the absence of a middle category artificially pushes respondents in one direction or another, we would expect that bias to be the same regardless of the treatment, since the response options always remain the same. Accordingly, we can make comparisons across treatment conditions—the absence of a middle option does not compromise the internal validity of the experiment.

However, we have also run all of our models using a four-category dependent variable, with ordered logit as the estimator. The substantive results are unchanged (and available upon request).

With a cell size of 200 and the observed standard deviation of the dependent variable (roughly .5), a power test reveals that we can detect a difference of 0.15 85 % of the time, with an alpha level of .05 and assuming a two-tailed test. Note, however, that the questions about social contact were only asked of 2/3 of the respondents, thus reducing the cell sizes for those analyses. For those analyses, we can detect a difference of 0.19 85 % of the time. Smaller subsets mean still less power, although the size of the negative reaction to Romney’s Mormonism is large enough that we nonetheless find statistically significant effects.

The randomization check was performed by conducting a Chi-square test on the distribution of each demographic trait across the treatment conditions.

We coded evangelical Protestants using the respondent’s specific religious denomination consistent with Layman and Green (2005), Green (2007), and Steensland et al. (2000).

The eigenvalue for the factor score of frequency of religious attendance and the guidance provided by religion is 1.17.

Republican primary voters were identified with a question on the common content baseline survey (variable BCAP4) that asked registered voters to identify whether they would vote in their state’s Democratic or Republican primary/caucus. Political interest was also gauged using the baseline survey (variable BCAP813)—those who indicated that they are “very much interested” in politics. Knowledge of Romney’s religion was measured with an open-ended item specific to our study. We coded the open-ended responses liberally, counting the many variations of “Mormon,” “Morman” [sic], “LDS,” “Latter-day Saint,” “Church of LDS” and so on as correct. Details for our coding are available upon request. By this measure, roughly half of respondents were aware of Romney’s religion.

Confidence intervals are calculated using CLARIFY (Tomz et al. 2003).

Benson et al. (2011) present similar results to ours for social contact within the context of the presidential primary when political competition is highest between Mormons and evangelicals. However, precise comparisons are difficult because they measure social contact using frequency of contact by splitting the social contact into two groups, high and low. Our question better captures the depth of the social contact and allows us to identify a middle group. We do not have comparable data from a general election period to assess how changing political competition might affect our results.

With an upper bound of −0.18 and a lower bound of −0.38.

For the Separationist frame, upper bound = 0.10, lower bound = −0.29. For the Common Values frame, upper bound = −0.07, lower bound = −0.27.