Party Identification and Vote Choice in Partisan and Nonpartisan Elections

It is worth noting that neither of these arguments appears to have empirical merit. Glick and Emmert (1987), Bonneau (2001), and Hurwitz and Lanier (2003) all find no differences in the quality of judges between alternative selection systems. Also, Hall (2001) finds that, in the aggregate, nonpartisan elections are driven by political factors, just like partisan elections.

The Political Science course was comprised almost entirely of non-Political Science majors who were simply taking the course to fulfill a general education requirement, meaning that the students were not likely to be unusually politically sophisticated. To make sure that the students in the Political Science class were not uniquely good at applying their partisanship to their vote choice in nonpartisan elections, we re-ran the model interacting enrollment in the political science course with the interaction between partisan format and party id (a three-way interaction) and all interactions among the constituent terms. This effectively allows us to test the hypothesis that the relationships between partisanship and vote choice (conditional on ballot format) are themselves conditional on whether the subject was enrolled in the Political Science class. The results show that the interaction is not significant and that the relationships between partisanship and vote choice (and the extent to which they are conditioned by ballot format) are not conditional on whether the subject was enrolled in the Political Science class. This finding, along with the observational data presented below, mitigates concerns that this portion of the experimental sample was unusually adept at determining party identification from the statements in the candidate biographies.

Consider the following campaign materials or endorsements which reflect the type of information that circulates in a nominally nonpartisan judicial campaign: for Republicans, see the Tea Party endorsement of Nels Swandal (http://​www.​wheatlandteapart​y.​com/​state-elections.​html) and see his ad regarding values and not “legislating from the bench” (http://​www.​youtube.​com/​watch?​v=​8MJI784PwI8), Tim Tingelstad’s campaign site which showcases rhetoric on values and religion (http://​www.​highesthill.​com/​), and Tom Christensen’s website where he commits to “deciding cases based on the constitution, statute and precedent not legislating from the bench.” For Democrats, consider Carol Hunstein’s endorsement from Democratic Congressman John Lewis saying, “Justice Hunstein has shown a true commitment to equality for all Georgians” (see http://​www.​hunsteinforjusti​ce.​com/​), JoAnne Kloppenberg’s statement about fair procedures for the criminally accused (http://​www.​youtube.​com/​watch?​v=​MYO36JAt0R8&​feature=​related), Planned Parenthood of Montana’s endorsement of Beth Baker (http://​www.​ppamt.​org/​elections/​2010-election-results), and Tim Fox’s vote to overturn an Arkansas law banning gay and lesbian couples from adopting (http://​thecabin.​net/​news/​2010-10-10/​open-seat-ark-high-court-draws-veteran-judges#.​Tw4N4xxvZwc).

As a test to ensure successful randomization, we assessed the relative gender and partisan compositions of the two treatment conditions (partisan versus nonpartisan format). If randomization was successful, the gender and partisan composition of the two conditions should differ only by chance. Results show that we fail to reject the null hypothesis that gender and treatment condition assignment are independent (χ² = 0.42, p = 0.517); the test also suggests that party identification and treatment condition assignment are independent (χ² = 1.30, p = 0.521).

Following the convention shown to be most appropriate by Keith et al. (1992), we code independent “leaners” as partisans. Coding partisanship in a single variable with the values we use is equivalent to using two dummy variables that are constrained to have the same effect but with opposite signs. The results are substantively similar if independents are dropped from the analysis and we compare only Republicans and Democrats.

One state, Michigan, also held a contested election, but was dropped because they held their election in a statewide multi-member district.

Identifying the partisanship of nonpartisan judicial candidates is a reasonably straightforward task and has been accomplished for a large number of candidates by previous scholars as part of a measure of ideology (Brace et al. 2001). We coded candidate partisanship by consulting their personal statements of partisanship, service in other elected positions in government, service in government that required appointment by partisan officials, service in state or local party organizations, and, in one case, declared membership in the “Tea Party.”

The weights are applied to account for the sampling design of the CCES.

The crosstab shows an interesting relationship between ballot format and the vote choices of independents. In partisan races, independents tend to vote for Republicans, while they prefer Democrats in nonpartisan races. We suspect this phenomenon is due to the concentration of partisan races in the South, where self-identified independents tend to be more conservative. Thus, when primed with the party identification of the candidate, they prefer Republicans. In nonpartisan elections, which are largely in non-southern states, self-identified independents are more likely to prefer Democratic candidates.

We omit votes for third party candidates.

Given Ohio’s hybrid system where parties may nominate and endorse candidates but party affiliations do not appear on the ballot, one may wonder if it is actually more like a partisan election than a nonpartisan election. To address this concern, we re-ran our model twice, once with Ohio counted as a partisan election state and once with Ohio treated separately with its own dichotomous variable and an interaction between the Ohio dichotomous variable and the voter’s party identification. In both alternative models, we still found no support for the hypothesis that party identification matters more in partisan elections than in the nonpartisan system or the quasi-nonpartisan Ohio system.

The standard percent correctly predicted and PRE statistics are not valid in the presence of sampling weights and are thus not reported here.

We also ran a model that distinguished between elected incumbents and appointed incumbents and found significance only for appointed incumbents in partisan elections; we suspect this finding is an artifact due to having only two appointed incumbents in partisan states (Texas and West Virginia), both of whom won their elections.