Polarization and the Top-Two Primary: Moderating Candidate Rhetoric in One-Party Contests

Abstract

Top-two primaries reshape the electoral process by changing the mix of opponents that candidates face, thus altering the electorate to which they must respond. Specifically, when top-two primaries produce two same-party opponents for the general election, candidates cannot simply rely on party-based voting to win. Advocates of the top-two primary system contend that if safe districts offer voters the choice between ideologically extreme and moderate candidates of the same party in the general election, voters should choose the more moderate option. Research to date suggests that this is not the case. If the top-two primary does invoke moderation, I argue that it should be because it is self-imposed by candidates in response to new electoral incentives. The change in primary rules should cause candidates to self-moderate in hopes of broadening the range of potential voters that they may capture. I test this proposition by examining the rhetoric found on state legislative candidate websites during the 2016 election. Results show that those facing same-party opponents use more moderate, bipartisan, and vague messaging when compared to those facing opponents of the opposite party.

Keywords:

• electoral rules
• state legislative elections
• state politics
• polarization
• political communication

Acknowledgments

The author would like to thank Tom Carsey, Pam Conover, Mike MacKuen, Chris Clark, Virginia Gray, Claire Dunn, Jelle Koedam, Eric Hansen, Dan Gustafson, Kelsey Shoub, three anonymous reviewers, and the State Politics Working Group at the University of North Carolina at Chapel Hill for their helpful comments on previous drafts of this article. Thanks also to Vanessa Schoning for her contributions as a research assistant for this project.

Supplemental material

Supplemental data for this article can be accessed on the publisher’s website at 10.1080/10584609.2019.1579772.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1. The top-two primary is currently used in California, Louisiana, and Washington. In Louisiana, if one candidate receives more than 50% of the total vote in the primary, that candidate is declared the winner and no runoff election is held.

2. A contest is considered monetarily competitive if the candidate with fewer campaign funds raised at least half as much as their opponent.

3. Louisiana is not included because it did not hold state legislative elections in 2016.

4. Krippendorff’s alpha reliability coefficients are calculated to compare the results of hand coding by the author and by the research assistant. In computing Krippendorff’s alpha, the comparison made is between the total number of mentions coded by the author and the total number coded by the research assistant, cumulative across all included pages of each candidate’s website. Computations are made using assumptions for ratio-level data. Krippendorff’s alpha for each of the seven dependent variables are as follows: party affiliation mentions, 0.80; bipartisan statements, 0.70; ideological statements, 0.76; endorsements, 0.71; party-owned issues, 0.54; vague issue positions, 0.50; party-congruent issue positions, 0.67.

5. The economy is notably absent from the list of party-owned issues in online supplemental Appendix A, yet it is an issue that is likely mentioned by candidates of both parties. It is not included here because the state of the economy is a performance issue rather than a party-owned issue. As Banda (2013) notes, performance issues and party-owned issues are distinct in that parties enjoy only unstable, short-term advantages on performance issues, but benefit from stable, long-term advantages on party-owned issues.

6. Several tests were used to evaluate model fit, including Akaike’s information criterion (AIC), Vuong statistics for non-nested models, and likelihood ratio tests. Negative binomial was selected as the best-fitting count model for all three dependent variables.

7. Candidate issue positions may fall into one of the following three categories: party-congruent, vague/nondirectional, or party-incongruent. The proportions for dependent variables relating to issue positions are calculated as a proportion of all issue positions.

8. http://www.americanideologyproject.com.

9. I choose Tausonovitch and Warshaw’s ideology scores as the best available measure for district-level voter preferences. Party registration data offer one alternative; however, this measure cannot be used because voters in Washington do not declare a party affiliation. Another possibility is to use district-level presidential vote share, but as others have shown, this is problematic because presidential vote share is subject to strong influence from short-term factors (Levendusky, Pope, & Jackman, 2008; Tausonovitch & Warshaw 2013).

10. Previous margin = 50 – absolute value (winning candidate’s vote share – 50).

11. I performed F-tests to evaluate the addition of the interaction term for each of the seven dependent variables. The interaction term improves model fit with the first four dependent variables (party ID, bipartisan statements, ideological statements, and endorsements) but does not improve model fit for the final three dependent variables (the types of issues discussed, vague issue positions, and party-congruent issue positions). The presence of a one-party contest conditions candidates’ mentions of party ID, bipartisan statements, ideological statements, and endorsements with a strongest effect in districts with larger previous win margins, while the presence of a one-party contest uniformly conditions how candidates talk about issues across all contests.

12. Indicator variables for challengers are dropped from all models to avoid perfect multicollinearity.

13. One possibility is that front-runners or incumbents may make candidates less inclined to broaden their appeal than challengers or those running for open seats. While adequate polling data for state legislative seats are not available, I test this possibility with an interaction term between incumbent status and the presence of a one-party contest. F-tests indicate that the interaction term does not improve model fit for any of the dependent variables. I provide comparisons of predicted outcomes for incumbents and non-incumbents in one-party contests, presented in Figures 11 through 17 in online supplemental Appendix D. For six of the dependent variables, the expected outcomes for incumbents and non-incumbents are statistically indistinguishable. For cross-ideological endorsements, there is a significant difference but it is in the opposite direction: incumbents are expected to have a greater proportion of cross-ideological endorsements than non-incumbents. This is unsurprising, however, given the high reelection rates of incumbents and the benefits that interest groups receive during the legislative process by supporting incumbent officeholders, even when they are from the opposition party (Hall & Wayman, 1990).

14. Ho, D., Imai, K., King, K., & Stuart, E. (2011). MatchIt: Nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 42, 1–28. Retrieved from http://www.jstatsoft.org/v42/i08/.

15. Observations were matched according to the following variables: district voter ideology, previous win margin for the seat, candidate party, state, and candidate type (incumbent, challenger, open seat).

16. In Table 4, coefficients for the presence of a one-party contest are positive for Models 1 and 3; however, the interaction term complicates direct interpretation. Expected counts and matching models in Table 5 allow a more direct interpretation of differences in candidate behavior in one-party and two-party contests.

Additional information

Notes on contributors

Steven Sparks

Steven Sparks is a PhD candidate, Department of Political Science, University of North Carolina at Chapel Hill.