Masculine Republicans and Feminine Democrats: Gender and Americans’ Explicit and Implicit Images of the Political Parties

Some scholars have explored citizen’s party images, using the ANES open-ended likes and dislikes questions, but none has focused on gender (Sanders 1988; Baumer and Gold 1995; Trilling 1976); related work on the contents of partisan stereotypes has similarly not focused on gender (e.g. Rahn 1993; Bastedo and Lodge 1980; Hamill et al. 1985). More recently, Danny Hayes has explored the traits that citizens associate with the parties’ presidential nominees, but without an explicit focus on the gendered nature of those trait attributions (2005).

For overviews of the gender gap literature, see Huddy et al. (2008) and Sapiro (2003, pp. 605–610). For an overview of the literature on female candidates, see Dolan (2008).

Interestingly, this process appears to be only partly voluntary; speakers of languages that gender nouns tend to associate a wide range of gendered characteristics with objects depending on the gender their language assigns to the noun (Phillips and Boroditsky 2003).

See, for example, Spence and Buckner (1995), Spence et al. (1978, 1979), Bem (1974, 1981, 1987), and Maccoby (1987). For a review of the vast literature on the conceptualization, measurement, and contents of ideas about masculinity and femininity, see Lippa (2005, Chap. 2). There is considerable cross-cultural consistency in gender stereotypes, amid important cultural variation, though this consistency—and debates about its social or biological bases—is tangential to the purposes of this paper (see, e.g., Ortner 1974, 1996, Chap. 7).

There is an extensive literature in social psychology showing that masculine and feminine traits and other characteristics do not, in fact, form a single bipolar dimension at the individual level (Constantinople 2005); rather, both are multidimensional constructs that vary independently (Bem 1974; Spence et al. 1978). Nevertheless, people generally believe that they form coherent and oppositional packages (Deaux 1987).

Helen Haste argues that the idea of gender difference serves as a sort of master metaphor that gives meaning to myriad dualities at the center of Western culture, including public–private, rational-intuitive, active–passive, hard-soft, thinking-feeling, and many more (1993). On the role of gender ideals in the politics of the American founding and early republic, see Kann (1998), Kerber (1986), Kang (2009), and Bloch (1987).

Huddy and Terkildsen present evidence that the gender associations of issues are not simply the product of the idea that women are more liberal than men; rather, the gender associations flow importantly from stereotyped beliefs about women’s traits and abilities.

I am grateful to an anonymous reviewer for bringing this point to my attention.

The party-candidate master codes are listed in the appendix to the ANES cumulative file dataset. The mentions are in variables VCF0375A–VCF0379A (Democratic Party likes), VCF0381A–VCF0385A (Democratic dislikes), VCF0387A–VCF0391A (Republican likes), and VCF0393A–VCF0397A (Republican dislikes). In 1972 the ANES reported only the first three mentions for each target, although up to five were collected in the interview. The 1972 dataset does report how many mentions each respondent made, up to five; this indicates that about 2% of respondents mentioned more than three things in a each category. Restricting the analysis in other years to include only the first three mentions does not affect the patterns of results in those years, which suggests that the omission of the fourth and fifth mentions in 1972 probably does not substantially influence the patterns observed in that year.

Across presidential years from 1972 through 2004, 28% of respondents failed to mention anything about any party. This ranged from a low of 22% in 2004 to a high of 34% in 1980.

Of respondents with no party likes or dislikes, 25% are pure independents, compared to 6% of other respondents. Ten percent of these respondents refuse to rate the one or both parties on the feeling thermometer (compared to 2% of others), and 29% rate both parties equally at 50 degrees (compared to 5% of others). Finally, these respondents average 0.38 on the zero-to-one ANES interviewer assessment of political knowledge, compared to 0.60. All of this is consistent with Geer’s conclusion that those who fail to answer open-ended questions are, generally speaking, not interested in the question, rather than being unable to articulate a meaningful response (1988).

Of course, some issues and political groups themselves have implicit or explicit gender associations. An important area for future research is the ways that the gendered traits associated with the parties interact with gendered issue and group associations.

Of course, the party-candidate master coding introduces additional distance between the available data and respondents’ own words, and raises both reliability and validity concerns. Here it is somewhat of an advantage that the ANES coding scheme was not developed or deployed with masculinity and femininity in mind; while this probably introduces noise into the coding, hurting reliability, it means that I am not constrained to a particular definition of gender, and the ANES coders are unlikely to have been biased by their own possible gendered associations for the parties. Thus, there is little reason to think that either would bias the ANES coding in favor of my hypotheses. Nevertheless, in the absence of the original verbatim text there is no way to be certain. The ANES is in the process of revising its open-ended coding procedures, and plans to make verbatim text more readily available in future studies, so future researchers may be in a better position to address these concerns directly. See http://​www.​electionstudies.​org/​conferences/​2008Methods/​MethodsConferenc​e.​htm.

The kappa statistic for inter-rater agreement among the three raters was 0.76 for masculine traits and 0.75 for feminine traits; both in the range characterized by Landis and Koch as “substantial” (1977, p. 165; Cohen 1960). Much of the coding disagreement turned out to be over ANES master codes that do not actually appear frequently in the data, so the basic pattern of results presented below hold up when I substitute each individual coder’s initial classifications for the final consensual coding.

The distinction between positive and negative traits was collapsed for the analysis, so stereotypically masculine traits that are culturally sanctioned (e.g., independent, code 315) and those that are not (e.g., cold or aloof, code 438) were both classified simply as masculine, and normatively positive and negative feminine traits (e.g., kind, code 435 vs. indecisive, code 304) were all classified as feminine. In practice, the overwhelming majority of respondents’ party likes were normatively positive traits, and dislikes were overwhelmingly negative, although there were a few exceptions. For example, a small handful of respondents indicated in 2004 that they liked the fact that the Democratic Party lacked a definite philosophy (code 836). This example makes clear that a trait that is often considered a weakness can be a political asset in the right political context, a point to which I return in the conclusion.

Reassuringly, the pattern of results is essentially unchanged when multiple mentions by a single individual are collapsed, which reframes the analysis in terms of the proportion of respondents who mention gendered traits, rather than the proportion of mentions.

The patterns are not any different in the non-presidential years for which party likes and dislikes are available.

Overall, 45.1% of mentions related to issues, 21.2% to groups, and 5.2% to individuals. The proportions in these categories varied somewhat by party: for the Democratic Party, 76.6% of likes and 62.1% of dislikes fell in one of those three categories, as did 70.4% of Republican Party likes and 75.0% of Republican Party dislikes.

Of course, some mentions of issue positions, such as a party being “tough on crime” or “soft on communism” may reflect a respondent’s reaction to a more symbolic masculinity or femininity. As I discuss above, these sorts of issue mentions were excluded from possible coding as traits for two reasons. First, the ANES master codes simply do not provide enough detail about respondent’s actual mentions of issues to code issues in this way, and second, even with verbatim text it would be beyond the scope of this analysis to attempt to ascertain whether a particular reference “really” refers simply to an issue position, or to a possibly-gendered aspect of the approach to the issue, or some combination. As I mention in footnote 12, this is an interesting area for future research.

The slight jumps in masculine Democratic dislikes in 1992 and especially 2000 are driven mostly by references to sex scandals (code 719).

Political knowledge is based on the ANES pre-election interviewer’s assessment of the respondent’s level of political information (VCF0050A). John Zaller reports that this assessment performs very well as a general measure of political knowledge (1992, p. 338); this measure has the added advantage of being reasonably comparable across years, especially in contrast with fact-based measures. The results are somewhat attenuated, but follow essentially the same pattern, when I replace political knowledge with a motivation-based measure of political engagement, based on respondents’ self-reported interest in politics and the campaign, and when I substitute respondent education. This is consistent with Zaller’s comparisons of different strategies for measuring habitual attention to politics (1992, p. 335).

Party affiliation is drawn from the standard ANES party affiliation battery (VCF0301), with independents who lean toward a party classified as independents. The results are substantively unchanged when leaners are reclassified as partisans.

For each like and dislike type, between one-third and one-half of respondents gave no mentions at all. This means that were I to run a model among all respondents, the coefficients would pick up the tendency to mention anything at all—essentially a model of the positivity or negativity of feelings about each party—rather than distinguishing those respondents who mention a gendered trait from those who do not, from among respondents who say something about the party. In any case, the substantive results are essentially the same when each party’s models are run among all respondents who mentioned any likes or any dislikes about that party, although among this broader universe respondent partisanship captures a bit of the tendency of partisans to mention things—including gendered traits—that they like about their own party and things they dislike about the other party. The results are also the same when likes and dislikes are collapsed into a single masculine model and a single feminine model for each party. Results available from the author.

Marginal effects were calculated using the MFX command in Stata. For the dummy variables (party affiliation and gender), the marginal effect is the difference in probability between an otherwise-average respondent who has the characteristic and one who does not. For political knowledge the calculation is the instantaneous marginal impact of knowledge on the probability for an average respondent. Because political knowledge is coded to run from zero to one and because the predicted probability curve is quite linear across the entire range, this marginal effect is almost exactly the difference in predicted probabilities between otherwise-average respondents who are most informed and least informed.

Models that include more extensive sets of independent variables yield entirely consistent results, both for these Democratic Party models and for the Republican Party results I present below. In particular, the probability of gendering the party is essentially equivalent for conservatives, moderates, and independents, for residents of different regions, for white and black respondents, and for older and younger respondents. In addition, there is no evidence of an interaction between engagement and either partisanship or gender, nor between gender and partisanship. As I mention in footnote 21, respondent education acts as a weak proxy for attention to politics, although its effects are washed out when political knowledge is included with education in a single model.

There is considerable debate on the broader role of whites’ racial attitudes in contemporary American public opinion (Kinder and Sanders 1996; Sniderman and Carmines 1997; see Sears et al. 2000 for a recent set of entries in this debate) and on implicit racial priming in particular (Valentino et al. 2002; Huber and Lapinski 2006; Huber and Lapinski 2008; Mendelberg 2008a, b).

There is lively debate on the relatively importance of (unconscious) cognitive accessibility versus (conscious) evaluation of importance in the priming of political attitudes (Valentino et al. 2002; see also Winter 2008, pp. 147–151). The theoretical accounts developed by Zaller (1992) and Mendelberg (2001) both include priming and accessibility as key mechanisms, though neither measures accessibility directly (Miller and Krosnick 2000; Nelson et al. 1997).

The feminine, masculine, and filler words were matched for length and frequency of appearance in the English lexicon (Kucera and Francis 1967). The nonsense strings were created by swapping letters or phonemes in real words, and were matched with the words for length. The LDT portion of the study began with a shorter set of training trails to give participants a chance to get used to the identification task.

The LDT was implemented using PxLab, an open-source software application for psychological experiments, available from http://​www.​uni-mannheim.​de/​fakul/​psycho/​irtel/​pxlab/​index.​html. The web survey was implemented in PHPQuestionnaire (http://​www.​chumpsoft.​com), which was modified by the author to implement streaming video and to interface with PxLab.

As is typical with student samples, the participant pool is not representative of a national sample. The participants are relatively young (age averaged 20 and ranged from 17 to 32). About two-thirds (69%) of participants were women; 54% identified as Democrats, 26% as Republicans and 19% as independent. There were no substantively or statistically significant demographic differences across conditions, and there is no evidence that gender, party identification, or political knowledge moderate any of the findings reported below. The study was approved by the University of Virginia institutional review board, protocol number 2008-0408.

The ads were for the Chevy Malibu and for the Apple iPod. There was also a fourth condition, which included a pair of political advertisements in place of the product commercials. Participants in this fourth condition were omitted from the present analysis.

Because reaction time data are notoriously noisy, following standard practice I exclude trials with extreme outlier response times in calculating the averages, as well as trials in which a respondent misidentified a target word as a non-word.

Thus, I regress individual-level average reaction time to feminine words on individual-level average reaction time to neutral words and a dummy variable for the Democratic condition. Because the estimated coefficients for neutral-word reaction times are very close to one, the approach I take is almost identical to simply subtracting each respondent’s neutral-word average from that respondent’s feminine-word average. Employing this alternate approach generates estimates of the size of the priming effect that are within a few milliseconds of the estimates I present below.

Miller et al. find, for example, that people tend to explain gender differences among voters and professors—both prototypically masculine—in terms of characteristics of women, while explaining gender differences among elementary school teachers—prototypically feminine—in terms of characteristics of men (1991).

This mapping of one binary distinction onto another raises the question of how third parties are understood. Interestingly, Baker notes that during the height of the nineteenth century party era, men who were not committed to either of the major parties were seen as “political impotent” and referred to as the “third sex” of American politics (1984, p. 628). Hoganson cites references from this era to members of third parties as “‘eunuchs,’ ‘man-milliners,’ members of a ‘third sex,’ ‘political hermaphrodites,’ and ‘the neuter gender not popular either in nature or society’” (1998, p. 23). On a related note, Fausto-Sterling (1993) argues that sex is itself not as simple a binary distinction as we often assume.