What’s in a Campaign Logo? Exploring Differences in Candidate Self-Presentation through Campaign Logos

Abstract

Differences in how candidates present themselves to voters is a key concern for scholars of campaigns and elections. Candidate advertisements, speeches, and websites form the basis of our knowledge on the subject, but ignore some of the most pervasive and discernible features of political campaigns: campaign logos featured on signage, stickers, apparel, buttons, websites, and advertisements. This study provides a descriptive examination of gender differences in candidate logos from the 2018 mid-term election. It extends our knowledge of how women and men present themselves by examining how candidates sell themselves in more symbolic terms based on colors, fonts, imagery, and slogans. We argue these logo choices reflect candidate brand positioning strategies, which are connected to attributes of the rival candidate and the preferences of the constituency. We then outline a future agenda for how the study of political campaign logos can inform theories of political science and brand positioning.

Supplemental data for this article is available online at https://doi.org/10.1080/15377857.2022.2040691.

Keywords:

• Elections
• logo
• branding
• gender
• U.S. Congress

Notes

1 The focus of this research is how candidates present themselves to voters. There is an equally large literature on how candidates are presented to voters by the media that works in tandem with direct presentations by the candidate (Milewicz and Milewicz, 2014; Peterson 2018).

2 Although we do not completely rule them out, we do not have predictions for an interactive effect between opponent and constituency factors. We therefore shade that portion of the Venn diagram black suggesting candidates are unlikely to look at the combination of factor attributes within that joint set.

3 In this analysis, we focus on challenger attributes established in the political science literature as important in candidate campaign choices rather than duplicate the candidate factors. In most cases, this decision should have little impact on the results. In a two-party system, for instance, specifying the partisanship of the candidate usually also indicates the partisanship of the opponent. Similarly, specifying they incumbency status of the candidate (e.g., an incumbent) typically indicates the incumbency status of the challenger (e.g., as the non-incumbent) except in the small number of open races.

4 Purposive sampling is a nonprobability sampling method where the researcher selects a sample based on their knowledge of how well selected units fit the purpose or goals of the research (Agresti 1990). The units are selected because the researcher believes they are best suited to address the research question (Frey 2018).

5 In order to assess the reliability of coding, a random subset of logos (50 from the U.S. Senate and 355 from the U.S. House of Representatives) were coded by three coders. Each remaining logo was hand-coded by at least two coders with coding disagreements resolved as a group with the third coder. A variety of coding reliability estimates show a high degree of intercoder reliability with intercoder agreement ranging from 94% (patriotic cues) to 100% (i.e., gendered cues, domestic issue references). Details are in the online supporting material.

6 The PVI measures the partisan leanings of the district or state relative to the nation as a whole. We measure candidate, challenger, and district ideology using data from the 2018 Cooperative Congressional Election Survey (CCES). The CCES asks respondents to place both themselves and each two-party candidate on the 7-point ideology scale. With over 50,000 respondents, we are able to provide Multi-level Poststratification estimates of the mean ideology of respondents in each state or district as well as the mean perception of each candidate’s ideology.

7 None of the models show evidence of overdispersion. The model for gendered language shows a poor fit for the data, but we retain similar results with changes to the model specification or changes to the model (e.g., Negative Binomial) so we report the Poisson estimates here for consistency.

8 The percentage change in the expected count for a standard deviation (sd) change in each predictor xk, holding all other variables constant, can be computed as, $\mathrm{ }100*\mathrm{ }\mathrm{ }\frac{E\left(yx,x_k+\mathrm{ }sd\right)-E\left(yx,x_k\right)}{(E(y|x,x\_k)}=100* [ex-1]$

9 We estimated a model interacting gender with both district competitiveness using PVI (Table 5) and the ideological distance between the candidate and the district (Table SM3) with the expectation that perhaps women were using bipartisan appeals in more swing districts and states or when they diverge ideologically from the district or state. We found no significant interaction effect in either model.

10 In addition,Table SM3 shows support for these results with women candidates more likely to use domestic appeals when there is little to no ideological distance between them and their constituency. In other words, domestic appeals are more likely to appear in a logo for women when they are electorally safe. We suspect being close to one’s district might also allow women candidates to use more gendered language, which is also borne out in Table SM3. In fact, Table SM3 shows that as the ideological distance between the candidate and constituency increases, the use of gendered language in a logo for women candidates decreases.