Do Partisan Types Stop at the Water s Edge?

Do Partisan Types Stop at the Water s Edge? January 8, 2017 Joshua D. Kertzer, 1 Deborah Jordan Brooks, 2 and Stephen G. Brooks. 3 Abstract: Do distinctive partisan types exist in the public s eyes in foreign policy? A growing number of analyses presumes that Democrats and Republicans in Washington are, in fact, seen as systematically di ering across an array of foreign policy issues, with major implications for questions ranging from the future of liberal internationalism, to the ability of leaders to send credible signals by going against their party s type. Yet there is a surprising absence of work that has investigated the microfoundations of partisan types, and how we know them when we see them. Building on the stereotype literature in social psychology, we explore the scope conditions of partisan types using a national survey experiment, which finds that partisan types are greatly attenuated at the water s edge. Our findings have important implications for a number of literatures, most notably those that examine against type models and the role of (bi)partisanship in foreign policy. 9881 words 1 Assistant Professor of Government, Harvard University. Email: jkertzer@gov.harvard.edu. Web: http:/people.fas.harvard.edu/ jkertzer/ Associate Professor of Government, Dartmouth College. Email: deborah.j.brooks@dartmouth.edu. Web: http://dartmouth.edu/faculty-directory/deborah-jordan-brooks 3 Professor of Government, Dartmouth College. Email: stephen.g.brooks@dartmouth.edu. Web: http://dartmouth.edu/faculty-directory/stephen-g-brooks

1 Introduction Do the Democratic and Republican parties have distinct types in foreign policy in the eyes of domestic audiences? A growing amount of work on the domestic determinants of foreign policy and International Relations (IR) presumes they do, arguing that Republicans are from Mars, and Democrats are from Venus: Republicans are hawks, while Democrats are doves (e.g. Gries, 2014); Democrats favor working multilaterally, while Republicans are more willing to go it alone (e.g. Rathbun, 2011); Republicans are more likely to favor free trade, while Democrats are more likely to be protectionist (e.g. Milner and Judkins, 2004), and so on. If foreign policy was once characterized by a bipartisan liberal internationalist consensus which Arthur M. Schlessinger, Jr. called the vital center (Schlesinger Jr., 1949) it is now commonly thought that the center no longer holds (e.g. Kupchan and Trubowitz, 2007; Hurst, 2014). The potential existence of distinct partisan types in foreign a airs has especially important stakes for two important debates in IR. The first concerns the durability of the liberal international order the United States helped construct following the Second World War. Given liberal internationalism s liberal origins, assessments of its fate typically hinge on its underlying degree of support among the American public, resulting in a large, rapidly growing literature evaluating the extent to which a bipartisan liberal internationalist consensus still holds in public at large (e.g. Kupchan and Trubowitz, 2007; Busby and Monten, 2008; Chaudoin, Milner and Tingley, 2010; Bafumi and Parent, 2012). If the public perceives the two parties as espousing systematically di erent foreign policy views, both top-down and bottom-up models of public opinion in foreign policy imply this liberal internationalist consensus is likely to break apart (Baum and Groeling, 2009; Kertzer, 2017). The second concerns the domestic politics of costly signaling. As a swiftly proliferating literature influenced by formal models of legislative bargaining tells us, if political parties have distinct types in foreign issues in the eyes of domestic audiences, and voters are uncertain about the merits of a policy proposal, parties can attempt to send more credible signals and induce greater public support by going against type. Parties with reputations for hawkishness or interventionism, for example, can more credibly sell rapprochement or retrenchment, while parties with reputations for dovishness or isolationism can more persuasively argue for the use of military force (e.g. Cukierman and Tommasi, 1998; Schultz, 2005; Fehrs, 2014; Saunders, 2015a). Yet parties need to have distinct types in in order to be able to profitably go against them a claim that has yet to be systematically explored. 1

Despite the frequency with which partisan types are invoked in the IR literature, there is an absence of work that has systematically investigated the microfoundations of partisan types in foreign policy, explored their scope conditions, or rigorously conceptualized what a partisan type is in the first place, or how we know one when we see it. We suspect the absence of systematic empirical work on this issue should be surprising to many readers. There is a large body of scholarship on the role of partisanship in public opinion about foreign policy, but it tends to focus on first-order beliefs (what do Republicans and Democrats think about foreign policy?) rather than the second-order beliefs inherent in partisan types (what do people think Republicans and Democrats think about foreign policy?) (e.g. Holsti and Rosenau, 1990; Rathbun et al., 2016). Similarly, when IR scholars have studied stereotypes, they have tended to focus on the mental images we have of international actors rather than of domestic ones (Herrmann and Fischerkeller, 1995). There is a massive amount of research on party brands and issue ownership in American politics, but it tends to think primarily about party reputations in terms of a reputation for competence on a particular issue, rather than issue positions per se. 1 In this article, we o er what we believe to be the first systematic exploration of partisan types in foreign a airs, seeking to make both a theoretical and empirical contribution. First, we conceptualize partisan types, borrowing from a diverse body of literature on the structure and content of stereotypes in social psychology to suggest an empirical strategy political scientists can use to study partisan types along four di erent dimensions: content, prevalence, intensity, and stereotypicality. Although we hope this typology and measurement strategy will be of use for the study of partisan types more generally, we focus here specifically on foreign a airs, describing the experimental design of an original national survey fielded on 1007 adult Americans in August 2014 examining the range of issues in which the mass public perceives the Republican and Democratic parties as having distinctive types. We then present our findings, which suggest that partisan types in foreign policy are relatively weak, less prevalent, less intense, and less distinct than in domestic politics. We further suggest that our results are not due to the ignorance of the mass public, but rather to partisan types simply being less distinct in foreign a airs than many political scientists assume. Our findings are thus consistent with the argument emphasizing the continued stability of the bipartisan tradition in 1 Indeed, issue ownership is frequently measured with questions like which party do you think would do a better job in handling this issue?, which while not completely divorced from issue positions (Walgrave, Lefevere and Tresch, 2012), is nonetheless conceptually distinct from the kind of partisan types we are interested in exploring here (Petrocik, 1996). Moreover, although the party brands literature has explored issue ownership in the context of foreign policy, it has tended to reduce foreign policy to national security an issue on which Republicans are generally more trusted (Gadarian, 2010) ratherthanexploringaricherarrayofforeignpolicyquestions. 2

foreign policy (Chaudoin, Milner and Tingley, 2010), while also raising important scope conditions for signaling models relying on the assumption of distinct partisan types. 2 Partisan types and their significance Political scientists who ask whether partisanship matters in foreign policy are typically referring to one of two theoretical mechanisms. The first is a direct pathway, in which political parties either because they represent di erent coalitions of interests, or are composed of di erent ideological factions have di erent foreign policy preferences, potentially leading to divergent foreign policy behaviors and outcomes (e.g. Fordham, 1998; Gowa, 1998; Rathbun, 2004; Milner and Tingley, 2015). The second is an indirect pathway, in which audiences either at home, or abroad have beliefs about the foreign policy preferences of political parties, which in turn a ects parties strategic incentives (e.g. Foster, 2008; Trager and Vavreck, 2011; Saunders, 2015b). It is this latter mechanism, which we refer to as partisan types, that we focus on here. In order to clarify how they a ect parties incentive structures and the contours of domestic politics, however, it is first important to define what we mean by partisan types. We define types more generally as socially shared beliefs about the characteristics, attributes, and behaviors of particular actors. (Hilton and von Hippel, 1996, 240). Three general points are thus worth emphasizing for our purposes. For one thing, types are socially shared. Although types may be built upon past actions (previous reluctance to work with the United Nations, for example, may give Republicans reputations for unilateralism), for them to have any traction, they must be socially shared by the audience; in this sense, types are social facts (Searle, 1995); they can also be thought of as reputations, in that they are beliefs about an actor that exist in the minds of others (Dafoe, Renshon and Huth, 2014; Brutger and Kertzer, 2017). Above all else, they are stereotypes, in the sense that they are beliefs about the characteristics of other groups (Hilton and von Hippel, 1996, 240), specifically the groups policy preferences. 2 In addition, types reside at multiple levels of analysis: we can understand types as operating at the individual-leader level based on leaders policy stances, and at the level of political parties 2 As with the stereotype literature more generally which argues that stereotypes need not be accurate in order to be widely held (Allport, 1954; Judd, Park and Kintsch, 1993) itispossiblefortypestobecompletelyunmoored from actual previous policy positions, though we find relatively little evidence of this in the results we report below. 3

more broadly. Although both variants are significant, our focus here is on partisan types, beliefs about the policy preferences of Republicans and Democrats. We choose to focus on partisan types for two reasons. First, partisanship is a powerful force in American politics (e.g. Zaller, 1992). Whether because of selection e ects ex ante or legislative constraints ex post, the scope and strength of partisan types determines how much latitude individual leaders have to establish types of their own. Second, and relatedly, partisan types are typically understood as more enduring: individual leaders come and go, but parties persist. Especially in foreign a airs, it often takes time for leaders to build up independent types, as most political candidates do not have the chance to develop clear and distinctive types on foreign policy issues before entering o ce, compounded by electoral incentives for candidate ambiguity (Tomz and Van Houweling, 2009), and the tendency of the media to devote little attention to reporting candidates positions (Conover and Feldman, 1989, 912). It is for these reasons that party brands are typically seen as powerful heuristics in American politics: voters lack the time and capacity to familiarize themselves with each individual candidate s position on every issue, and thus turn to parties instead (Rahn, 1993; Lupia and McCubbins, 2000). 3 Thus, in against type models in IR like Trager and Vavreck (2011) and Saunders (2015a), for example, the relevant type is really at the party-level, rather than the leader-level. Indeed, partisan types are also thought to di use internationally as well: Foster (2008), for example, argues that the United States is more likely to be targeted by foreign challengers when a Democratic White House is facing challenges from a Republican congress, out of the assumption that foreign leaders carry around the same partisan types in their heads as domestic audiences. Finally, the strength of partisan types has important implications for a variety of di erent questions about the domestic politics of IR. The first concerns democratic theory and the nature of public opinion. When partisan types are stronger, elite cues are easier for the public to follow (Levendusky, 2010). Moreover, because strong partisan types cause specific policy stances to be seen as a badge of membership within identity-defining a nity groups, citizens presented with them are more likely to engage in partisan motivated reasoning (Kahan, 2016, 2), causing them to express more certainty about their opinions, and engage in the various biases that follow from it (Bolsen, Druckman and Cook, 2014). The second concerns the fate of liberal internationalism. Whether because of American hege- 3 It is perhaps for a similar reason that the voluminous literature on stereotypes in social psychology inevitably thinks of stereotypes as something that refers to groups rather than a discrete individual, since the e cacy of stereotypes in person perception hinges on the perceiver drawing inferences about an individual through social categorization. 4

mony, embedded liberalism, or some interaction between the two (Ruggie, 1982; Deudney and Ikenberry, 1999), debates about the stability of the rules-based liberal international order often turn to American domestic politics, hence the large, rapidly growing literature evaluating the extent to which a bipartisan liberal internationalist consensus still holds among the American public (e.g. Kupchan and Trubowitz, 2007; Busby and Monten, 2008; Chaudoin, Milner and Tingley, 2010; Bafumi and Parent, 2012). The distinctiveness of partisan types bears directly on this question. If public opinion in foreign policy is shaped by elite cues particularly the presence of elite consensus or polarization (Zaller, 1992; Baum and Groeling, 2009; Saunders, 2015b, though see Kertzer and Zeitzo, 2017) and the public sees partisan elites as espousing fairly similar foreign policy views, this likely creates a natural limit on how far public attitudes can veer away from the center. In contrast, if there are in fact strong partisan types in foreign policy, this creates the potential for a vicious cycle, as a progressively larger cleavage emerges between the foreign policy views of the supporters of the two parties. For one thing, if the public perceives party elites as di ering greatly on foreign policy issues, then public attitudes are likely to follow and become more polarized. In turn, if party elites see their base supporters as shifting away from the center on foreign policy issues, they have incentives to follow suit, which would likely prompt the partisan supporters in the public to shift further from the center, thereby furthering the cycle. Lastly, the potential presence of partisan types also has important consequences for a burgeoning body of literature exploring the informative value of actors going against type in order to send credible signals (Schultz, 2005; Trager and Vavreck, 2011; Fehrs, 2014; Saunders, 2015a; Kreps, Saunders and Schultz, 2016; Kane and Norpoth, 2017). At their most general level, the logic of these models is relatively straightforward: an actor (the receiver ) is uncertain about the merits or outcome of a potential policy being recommended by another actor (the sender ), and thus relies on knowledge it has about the sender in order to evaluate the credibility of its claims. 4 Whether because we are particularly attentive to incongruent or surprising information (Maheswaran and Chaiken, 1991), or because of the inherent value of costly signals over cheap talk (Schelling, 1960), signals are stronger if they come from unlikely or biased sources, who thus may give the most credible advice (Calvert, 1985; Kydd, 2003). If even the Pentagon says defense spending is too high, defense spending 4 In this sense, these models simultaneously assume both the presence and absence of uncertainty: the legislator is both uncertain about the outcome of a policy, and certain about the bias of her advisors (e.g. Calvert, 1985); the public is uncertain about the merits of a policy, but knows (or at least has a rough estimate of) the ideal point of the cuegiver (e.g. Chapman, 2011). 5

should likely be cut (Krehbiel, 1991); if even the United Nations approves of a military intervention, the intervener s intentions are likely good (Thompson, 2009; Chapman, 2011); if even Fox News praises a Democratic policy, it is probably meritorious (Baum and Groeling, 2009). Leaders whose support of a policy goes against type are thus more persuasive (Cukierman and Tommasi, 1998; Schultz, 2005). While some variants of these models focus on types at the leader-level the adage that only Nixon can go to China many others rely on types at the party-level. Yet leaders can only gain from going against their party s type in foreign policy on issues where distinct partisan types exist in the first place something that has yet to be systematically established. That such partisan types exist in the foreign policy realm is widely assumed in much of the existing literature. Yet despite the ubiquity of the assumptions that Republicans are seen as being from Mars and Democrats as from Venus, there are reasons to question the distinctiveness of partisan types in foreign a airs. Foreign policy was for a long time assumed to be a domain in which there was relative bipartisan agreement, both among political elites, and the public at large (Schlesinger Jr., 1949; Gowa, 1998; Chaudoin, Milner and Tingley, 2010). Public opinion scholars like Holsti and Rosenau (1990) and Wittkopf (1990) turned to foreign policy orientations like militant internationalism and cooperative internationalism to explain foreign policy attitudes precisely because conventional political variables like partisanship explained relatively little of the variance in either elites or the mass public s foreign policy views. The era of Scoop Jackson and Nelson Rockefeller has long passed, but the mainstream foreign policy establishment in Washington remains su ciently congealed that one Obama administration sta er disparagingly referred to it as the Blob. 5 Of the eleven Preferential Trade Agreements entered into between 2001-2016, for example, eight passed along bi- or cross-partisan lines (Kucik and Moraguez, 2016). Similarly, as we show in Appendix 3, whether measured by militarized interstate disputes, international crisis behavior, or militarized compellent threats, Democratic administrations have not been systematically less hawkish or interventionist than their Republican counterparts. Many of the fiercest debates in foreign policy occur within parties rather than between them, as foreign policy is often characterized by cross-partisan baptist-bootlegger coalitions, in both security (e.g. liberal internationalists and neoconservatives joining forces to support military interventions), and economics (e.g. both the critical left and the nationalist right opposing free trade 5 http://www.nytimes.com/2016/05/08/magazine/the-aspiring-novelist-who-became-obamas-foreign-policy-guru. html?_r=0 6

and globalization). As a result, even though the two parties often adopt di erent stances on specific issues (e.g. Democrats were more favorable towards the 2015 nuclear deal with Iran; Republicans are now more favorable towards the 2017 immigration travel ban), contemporary public opinion data often shows a fair amount of bipartisan consensus about more general foreign policy goals. The 2016 Chicago Council on Global A airs survey report on American public opinion towards foreign policy, for example, announces that: American support for US engagement in the world remains remarkably stable and crosspartisan. The US public also remains united around combatting a similar set of top threats, including terrorism and nuclear proliferation. Americans of both parties share a similar view of how to deal with Russian and Chinese power, and both support the US military presence in key allied countries such as Japan and South Korea. And crosspartisan majorities favor a continued shared leadership role for the United States internationally. 6 In other words, then, if partisan types accurately reflect either the historical record or the degree of political polarization on various issues, it is not immediately clear how distinctive partisan types should be across the board in foreign a airs. Moreover, regardless of how objectively similar the two parties are in foreign policy issues, partisan types ultimately have an important intersubjective component, which raises a variety of questions given how far removed foreign a airs is for many members of the mass public. Formal models of against-type dynamics were originally developed in legislative signaling games, where it was reasonable to treat types as common knowledge, since the types in question are relatively well-defined, both because of the nature of the senders (e.g. legislative committees, specialized by design), and the sophistication of the receivers (e.g. legislators) (Krehbiel, 1991). It is unclear how well these models translate to the context of public opinion about foreign policy, where sender preferences may be less distinct, and the receiver much less knowledgeable (Delli Carpini and Keeter, 1996). To sum up, despite the popularity and significance of the concept, we actually know relatively little about which (if any) foreign policy issues have distinct partisan types, or even what approach political scientists should take if they wanted to explore the topography of partisan types. It is this question that we explore in the next section. 6 Chicago Council on Global A airs 2016, 8 7

3 Operationalizing partisan types We argued above that we should think of partisan types as beliefs about the policy preferences of Republicans and Democrats. In this sense, types are stereotypes, so for a measurement strategy we turn for inspiration to the rich literature on stereotypes in social psychology. There are a great many di erent ways to characterize a partisan type, but in this section we introduce four di erent properties or dimensions of partisan types that we believe will be of particular use to political scientists, and discuss how to measure them. First, partisan types have content. The literature on stereotype content in psychology is vast (for a summary, see Stagnor and Lange, 1994; Hilton and von Hippel, 1996; Fiske et al., 2002), but for our purposes we might think simply of the content of a partisan type as the policies associated with a particular party. Based on the discussion above, one might associate hawkishness, unilateralism, and free trade with the Republican party, for example, and dovishness, multilateralism, and protectionism with the Democratic party. The question of the content of a type is independent of the other characteristics we discuss below, which focus not on what a type is, but rather, on how widely held and intense it is. Second, for partisan types to be meaningful, they must have prevalence; partisan types should be widely held. There are debates amongst psychologists about how crucial consensus is for stereotypes (e.g. Jussim, 2012; Judd, Park and Kintsch, 1993), but in a political science context, just as social facts are predicated upon intersubjectivity, partisan types are the most powerful when they are widely held. If only a small segment of the population associates an issue with a particular party, it may still be politically relevant, but is of less interest to us than if a majority of the population makes the association. Third, partisan types vary in their intensity, based on whether a policy proposal is highly or only weakly associated with a particular party. We care about the intensity of partisan types because it provides another way of speaking to their power, and thus, how much traction a political leader can derive from going against type. For example, a leader is unlikely to procure much political advantage from going against a policy position that is only weakly associated with the party. The intensity and prevalence of partisan types are individually significant, but can also be usefully combined into a single metric. Indeed, it is reasonable to consider intensity and prevalence as interrelated: the stronger the partisan type is on a given issue, the more likely the partisan type 8

is also widely held. The partisan types with the highest levels of overall strength are the ones for those issues where there is both a high willingness to assign a type, and where the type assigned is highly associated with a particular party. Below, we use the term stereotypicality as a shorthand for this aggregate measure. We can make the above discussion more concrete using the example in Figure 1. Suppose a group of study participants (in our case, a national sample, but the logic remains the same regardless of the sample composition) is given five di erent policy proposals. For each policy proposal, the respondents are asked to imagine that leaders from a political party were taking the issue position being presented; in other words, which party would they guess was the one taking the position? Because partisan types are stereotypes, in order to study them it is not enough to simply focus on first-order di erences in the policies endorsed by Democrats and Republicans, but to also study second-order di erences as well: the beliefs that audiences have about what Democrats and Republicans believe. Figure 1: Exploring di erent measures of partisan type Count 50 40 30 20 10 0 50 40 30 20 10 0 50 40 30 20 10 0 50 40 30 20 10 0 50 40 30 20 10 0 Definitely Democratic Probably Democratic Responses Probably Republican Definitely Republican Both Neither Proposal A Proposal B Proposal C Proposal D Proposal E Stereotype Measure Content Prevalence Intensity Stereotypicality Content Prevalence Intensity Stereotypicality Content Prevalence Intensity Stereotypicality Content Prevalence Intensity Stereotypicality Content Prevalence Intensity Stereotypicality 0.00 0.25 0.50 0.75 Measure Proposal A Proposal B Proposal C Proposal D Proposal E The left-hand panel displays hypothetical distributions of responses for five policy proposals; the right-hand panel calculates each of our four di erent quantities of interest for each proposal, described in detail in the text. For example, the stereotype content of proposal A is relatively Democratic, and proposal B relatively Republican; proposal C has the highest level of stereotype prevalence, and proposal D the lowest; proposal E has the highest stereotype intensity, proposal C has the highest overall level of stereotypicality, and so on. For the results in numeric form, see Appendix 1. 9

The left-hand panel in Figure 1 displays hypothetical distributions of responses for each of the four policy positions (ranging from Definitely Democratic leaders to Definitely Republican leaders, as well as Both and Neither categories to allow for the possible absence of distinctive types), while the right-hand panel of Figure 1 presents the four di erent quantities of interest previewed above. First, we can simply look at the content of each partisan type: for those participants who did associate a policy with a particular party, on average, how Republican or Democratic a type is it? Thus, as measured by stereotype content, proposal A is a relatively Democratic stereotype, and proposal B a relatively Republican one. Second, we can look at the prevalence of each partisan type: what proportion of the sample associated the policy proposal with a particular political party? This measure is independent from a stereotype s content: Proposals A and B have diametrically opposed stereotype content, for example (0.27 vs 0.73), but the same level of stereotype prevalence, as 62% of the respondents were willing to assign a type to each. Similarly, proposal C has the highest level of stereotype prevalence, since 90% of the sample associated it with a particular political party, and proposal D had the lowest level of stereotype prevalence, with only 40% of the sample associating it with a particular political party. Third, we can look at the intensity of each partisan type: of those who assigned a distinct type for a particular policy proposal, how extreme a stereotype was it? Of the five proposals in Figure 1, proposal E has the highest level of stereotype intensity, while proposal D has the lowest, though the intensity levels for the first four proposals are all relatively similar to one another. Finally, we can look at the overall stereotypicality of the proposal, which is a function of both the stereotype s intensity and its prevalence. Since proposals A and B have identical prevalence and intensity scores, they have identical (moderate) stereotypicality scores as well. Proposal D has the lowest stereotypicality score, and proposal C the highest. Each of these four measures thus captures something subtly di erent: content tells us what parties a policy is associated with, prevalence tells us how frequently the policy is associated with parties in general, intensity tells us the strength of the associations made, and stereotypicality provides an aggregate measure of how crystallized the stereotype is. The measures are deliberately simple, and there are, of course, countless other ways to characterize partisan types. 7 Our claim, 7 The prevalence score presented above, for example, tells us the proportion of participants willing to assign a distinct type on a particular policy, rather than how much consensus there was about the types assigned. Similarly, the intensity measure above is akin to measures of attitude extremity (Miller and Peterson, 2004) inthatitfocuses 10

then, is not that these are the only ways to operationalize partisan types, but rather, that they constitute four simple measures that are likely intuitive to many political scientists, and which we believe will be useful in many empirical applications. In the next section, we use these measures to map the topography of partisan types in foreign policy. 4 Method To o er what we believe to be the first systematic analysis of partisan types in foreign policy, we employ an original survey experiment, fielded on a national sample of 1007 American adults in August 2014 by Survey Sampling International (SSI). 8 The main survey instrument consisted of two questionnaires. At the beginning of the first questionnaire, participants were instructed: For the first set of questions, we re going to present you with a series of policy proposals. Please indicate the degree of support you would feel towards the proposed policies if politicians in the US took each position. Participants were then presented with a list of 12 policy proposals covering a mix of domestic and foreign political issues (discussed in greater detail below, and presented in full in Appendix 2). For each proposal, participants indicated their degree of support on a Likert response scale ranging from 1 (extremely unsupportive) to 7 (extremely supportive). After participants completed the questionnaire indicating their support for each proposal, they then were presented with a second questionnaire, in which they were instructed: Now, we would like for you to think about these issues in a di erent way. If you heard that leaders from a political party were taking the issue positions described below, which party would you guess was probably the one taking that position? on the strengths of the associations made rather than the direction of the association. 8 See Appendix 2.1 for further discussion of the sample. 11

Participants were then presented the same 12 policy proposals as before, but this time, asked to indicate which party was more likely to be the one taking the position, using the response options Definitely Democratic leaders, Probably Democratic leaders, Probably Republican leaders, Definitely Republican leaders, Both Democrats and Republicans, and Neither Democrats nor Republicans. Thus, whereas the first questionnaire measures participants own feelings towards these proposals, the second uses these proposals to tap into the partisan stereotypes participants hold about each of the two major political parties. Finally, participants completed a short demographic questionnaire. Although the layout of the survey was relatively straightforward from the perspective of the participants, it contained a relatively complex randomization protocol. First, to avoid potential order e ects, we randomized the order in which each of the policy proposals were presented within each questionnaire. Second, for eight of the policy issues (listed in full in Table 3 in Appendix 2), we randomly varied the content of each proposal: for trade policy, for example, half of the participants were presented with a protectionist policy proposal, and the other half with a free trade policy proposal. This technique not only avoids conflating partisan types with issue ownership, but allows us to study a wide variety of policy proposals without inducing concerns about respondent fatigue, or the demand e ects that would likely arise if each respondent were evaluating multiple policy proposals on the same issue. This randomization carried over across both questionnaires, so participants who were given a protectionist proposal in the first questionnaire, for example, were also given a protectionist proposal in the second. Third, because it is possible that partisan types manifest themselves not with the goal of a policy, but with the tactics, for four of the foreign policy proposals (listed in full in Table 4 in Appendix 2), we held the purpose of each policy fixed, but varied the approach: namely, whether the policy was conducted multilaterally or unilaterally. Here, we employed a nested randomization structure. One third of participants were assigned to a pure control condition, in which, for each of these four policies, we only presented the central purpose of the policy but did not mention how it would be conducted. For the remaining two-thirds of participants, respondents were randomly presented with either a unilateral or a multilateral version of each policy; each approach was randomized at the item-level, such that some participants were in the multilateral condition for some policies, and the unilateral condition for the others. 9 9 In this manner, we avoid potential contamination e ects that would result if the pure control was also assigned 12

Because random assignment allowed for di erent participants to evaluate di erent versions of each policy, we obtain results for 28 di erent policy statements overall. Although the list is not exhaustive, it nonetheless reflects a wide range of domestic and foreign policy issues, from economic issues to social ones, from general foreign policy predilections, to specific foreign policy interventions. This breadth not only bolsters the generalizability of our results, but also enables us to test whether participants espouse systematically di erent partisan types in domestic issues than in foreign policy ones. 5 Results: what do partisan types look like in foreign policy? We present our initial results in three stages. First, we look at stereotype prevalence, showing that relatively low proportions of Americans assign distinct partisan types to foreign policy issues compared to domestic ones. Second, we look at stereotype content, showing that of the participants who do assign types to foreign policy issues, the types tend to be less stark, although some foreign issues display starker types than others. Third, we show that for the 28 political issues we examine here, the intensity and prevalence of the partisan types are highly intercorrelated, reconfirming the relative weakness of partisan types in IR. 5.1 Stereotype prevalence Figure 2 depicts the stereotype prevalence measure of each of our 28 policy proposals, with 95% bootstrapped confidence intervals derived from B = 1500 bootstraps. The results show that domestic issues generally display relatively high levels of stereotype prevalence: regardless of whether the policy statements are in favor or opposed, over three quarters of our participants assign a partisan type to abortion, taxes and gun control, for example. In contrast, apart from military spending, the stereotype prevalence levels for many of the foreign policy issues are relatively low: half of our participants don t assign a partisan type to a proposal for arms control, and barely more than that for interventionist or isolationist policies and trade policies. The multilateral/unilateral treatments also appear to have relatively little e ect when measured by stereotype prevalence. at the item level. To increase the dosage of the multilateralism treatment, it includes both quantitative ( cooperating with other countries ) and qualitative ( seeking backing from the United Nations ) measures of multilateralism. See Ruggie (1992). 13

Figure 2: Stereotype prevalence: the proportion of participants who assigned a partisan type on each issue Domestic Pro-abortion Anti-abortion Pro-tax hike Anti-tax hike Pro-environment Anti-environment Pro-gun control Anti-gun control 0.00 0.25 0.50 0.75 1.00 Foreign Pro-arms control Anti-arms control Interventionist Isolationist Free trade Protectionist Increase military spending Decrease military spending 0.00 0.25 0.50 0.75 1.00 Troops humanitarian Troops humanitarian (unilateral) Troops humanitarian (multilateral) Troops oil Troops oil (unilateral) Troops oil (multilateral) Enviro sanction Enviro sanction (unilateral) Enviro sanction (multilateral) Stop piracy Stop piracy (unilateral) Stop piracy (multilateral) Foreign (Approach) 0.00 0.25 0.50 0.75 1.00 Stereotype prevalence The proportion of respondents who assigned a type to each of the 28 di erent issues, with 95% bootstrapped confidence intervals. 14

Figure 3: Stereotype content: average partisan types for each issue Domestic Pro-abortion Anti-abortion Pro-tax hike Anti-tax hike Pro-environment Anti-environment Pro-gun control Anti-gun control 0.00 0.25 0.50 0.75 1.00 Foreign Pro-arms control Anti-arms control Interventionist Isolationist Free trade Protectionist Increase military spending Decrease military spending 0.00 0.25 0.50 0.75 1.00 Troops humanitarian Troops humanitarian (unilateral) Troops humanitarian (multilateral) Troops oil Troops oil (unilateral) Troops oil (multilateral) Enviro sanction Enviro sanction (unilateral) Enviro sanction (multilateral) Stop piracy Stop piracy (unilateral) Stop piracy (multilateral) Foreign (Approach) 0.00 0.25 0.50 0.75 1.00 Definitely Stereotype content Definitely Democratic Republican The average stereotypes participants suggested for each of the 28 di erent issues, with 95% bootstrapped confidence intervals. The light grey lines conect pairs (or triads) of treatments for a given issue: the longer the line, the more distinct the stereotypes, and the greater the e ect of switching from one policy stance (e.g. interventionism) to another (e.g. isolationism). 15

5.2 Stereotype content To more deeply probe the topography of partisan types, Figure 3 plots the stereotype content measures for each of the 28 policy proposals, revealing what the average partisan types were for each issue amongst those participants who did assign a party to the issue. As before, the lines around the point estimates depict 95% bootstrapped confidence intervals; the figure also includes light grey lines connecting each pair of treatments (or, triad of treatments, for the foreign policy approach conditions) on a given issue. Thus, the longer the grey line, the more distinct the stereotypes, and the greater the e ect of switching from one policy treatment (e.g. interventionism) to another (e.g. isolationism). The results show that the magnitude of the di erences in stereotype content between policy pairs varies dramatically across issues. Domestic issues display a large and intuitive symmetry between opposing policy proposals: for example, a pro-choice policy has a relatively Democratic type, while an anti-abortion policy has a relatively Republican one. Foreign policy approaches whether an intervention is conducted multilaterally or unilaterally display extremely small treatment e ects; there is some indication that unilateral missions are more strongly associated with the Republican party, but not consistently so, and the e ect sizes are modest; the principal policy objective (Jentleson, 1992) matters more for stereotype content than the approach itself. Of the foreign policy issues in the middle panel, partisan types on military spending are relatively distinct, an e ect similar in magnitude to many of the domestic issues in the top panel. Otherwise, the magnitude of the di erences tends to dissipate: although partisan types significantly di er for arms control, the size of the treatment e ect is roughly 2.5 times smaller than for the domestic issues, and partisan types for interventionism and free trade are barely discernible. On the whole, then, although partisan types don t entirely stop at the water s edge, the content of partisan types is notably less distinct for many foreign issues. 5.3 Stereotype intensity We argued above that stereotype prevalence and stereotype intensity are interrelated: the more willing people are to assign a policy to a political party, the stronger the association people should see between the policy and the party. We test this hypothesis by estimating a simple bivariate linear regression model, in which the stereotype intensity of each of the 28 issues is regressed on the stereotype prevalence scores for each issue. Importantly, not only is the stereotype prevalence 16

Figure 4: Stereotype prevalence and intensity are highly interrelated 1.00 Stereotype intensity 0.75 0.50 0.25 Anti-tax hike Anti-abortion Pro-tax hike Pro-abortion Anti-gun control Pro-gun control Decrease military spending Anti-environment Increase military spending Isolationist Pro-environment Pro-arms control Anti-arms control Interventionist Protectionist Free trade Issue type Domestic Foreign 0.00 0.00 0.25 0.50 0.75 1.00 Stereotype prevalence The proportion of respondents who assigned a type to each of the policy proposals, versus how intense the stereotypes were; to improve legibility, the foreign policy approach issue labels are omitted from the plot. A simple bivariate linear regression model indicates that 82% of the variance in stereotype intensity can be explained by variance in stereotype prevalence. Domestic issues are in shown in red, foreign issues in turquoise. Stereotype intensity is scaled such that a value of 0 indicates the weakest possible stereotype (in which Republican and Democratic attributions cancel one another out), and a value of 1 indicates the strongest possible stereotype (in which all respondents would have indicated that a policy was definitely Republican (Democratic)). 17

measure highly significant (p <0.000), but the model fit is impressively high, with an R 2 statistic of 0.820, suggesting that 82% of the variation in stereotype intensity can be explained by variation in stereotype prevalence. We demonstrate the same point visually by plotting stereotype prevalence and intensity scores in Figure 4, this time estimating separate regression models for domestic issues (in red) and foreign policy issues (in turquoise). The plot illustrates two points: first, the strong covariation between stereotype prevalence and stereotype intensity. Second, the cluster of foreign policy issues is generally are generally further down than the domestic political issues; in general, then, the foreign policy issues we include here display weaker partisan types than their domestic counterparts. 5.4 Modeling overall stereotype strength (stereotypicality) The above analysis suggested that although partisan types are detectable in foreign a airs, the e ects tended to be quite modest, apart from that of defense spending. In general, foreign policy issues displayed relatively low levels of stereotype prevalence, and the types that the public perceived tended to be relatively weak and indistinct. The findings should thus reassure those scholars concerned about the potential collapse of bipartisanship in foreign policy, while also raising some important questions about the viability of against type models in foreign policy in which the receiver is the mass public. If partisan types are social facts, the question of their objective accuracy is at most a secondary one, since socially shared beliefs can have real consequences regardless of their veracity (Searle, 1995). For party-based against type e ects to be substantively strong in foreign a airs, for example, it matters less how objectively distinct the two parties are from one another, and more how distinct the parties are perceived to be by the domestic audience. Nonetheless, the question of accuracy suggests two diametrically opposed interpretations of the findings presented above. The first is that the generally weak findings for partisan types in foreign policy simply show how ignorant or inattentive the public is about world a airs: although political scientists may know that Republicans are from Mars and Democrats are from Venus, the public itself may be too disconnected to recognize these clear partisan gaps (e.g. Guisinger, 2009; Kertzer, 2013). A very di erent interpretation is that the weakness of partisan types in foreign policy is not an indictment of the public, but rather, a relatively accurate approximation of reality. There are two types of benchmarks one can use to measure stereotype accuracy. The first is attitudinal: assessing the congruence between participants second-order beliefs and their first-order 18

preferences, thereby testing whether the issues where respondents perceive the greatest partisan gaps are the issues where the gaps themselves are the greatest. The second is behavioral: analyzing the historical record, and assessing how di erently the two parties have behaved when in o ce. Given space constraints and the sheer number of issues examined in the experimental design, it is well beyond the scope of this article to o er a systematic behavioral test though we o er a first cut at one in Appendix 3, where we turn to a set of frequently used conflict datasets in IR to suggest that the two parties are perhaps less distinct in regards to their conflict behavior than some of our models in IR assume. In the space below, then, we o er an attitudinal test, investigating the individual- and issue-level correlates of stereotypicality in our data using a set of mixed e ect models, assessing the accuracy of respondents second-order beliefs by analyzing the relationship between issue-level polarization and stereotypicality, and examining whether the stereotypicality gap between domestic and foreign issues shrinks among more politically sophisticated respondents. These tests have important implications both positively and normatively. If issue-level polarization is strongly correlated with stereotypicality, it suggests our respondents stereotypes about Republicans and Democrats in foreign a airs largely track with what Republicans and Democrats in the mass public actually think. If more politically sophisticated respondents who tend to be more ideological, and more likely to receive cues from elites (Zaller, 1992) are no less likely to perceive a stereotypicality gap between foreign and domestic policy issues than their less sophisticated counterparts, it suggests that this stereotypicality gap isn t simply due to mass ignorance. Given the strong intercorrelation between the extremity of the partisan type held, and willingness to assign a distinct partisan type in the first place, we integrate the two to produce a composite measure of stereotypicality at the respondent-level. 10 This approach enables us to calculate a stereotypicality score for each of the 28 issue areas for each of the 1007 participants, looking across all of the responses simultaneously and examining the extent to which respondent-level characteristics (e.g. political sophistication, and a participant s own attitudes regarding the issue proposal) and issue-level characteristics (e.g. whether the issue is a foreign issue or not, and the the degree of partisan polarization in the issue) shape assessments of stereotypicality. Given the clustered structure of the data, we estimate a series of mixed linear models with random e ects on each participant and 10 Unlike the issue-level measure of stereotypicality shown in Figure 1, where we calculate the average level of stereotypicality for a given issue, we calculate our respondent-level measure of stereotypicality by abs(x i 3); as before, we recode the both and neither responses to form a neutral scale midpoint. 19

each issue area, presented in Table 1. Table 1: Mixed linear models: respondent-level and issue-level correlates of stereotypicality (1) (2) (3) (4) (5) Age 0.001 0.002 0.001 0.001 (0.001) (0.0005) (0.001) (0.001) Male 0.049 0.041 0.049 0.049 (0.015) (0.015) (0.015) (0.015) Income 0.0001 0.0002 0.0001 0.0001 (0.003) (0.003) (0.003) (0.003) White 0.025 0.022 0.025 0.025 (0.018) (0.017) (0.018) (0.018) Education 0.020 0.020 0.020 0.020 (0.005) (0.005) (0.005) (0.005) Partisanship 0.004 0.004 0.004 0.004 (0.003) (0.003) (0.003) (0.003) Political interest 0.095 0.082 0.095 0.095 (0.009) (0.008) (0.009) (0.009) Strength of preferences Foreign policy issue Polarization 0.148 (0.009) 0.212 (0.025) 0.807 (0.150) Constant 0.473 0.136 0.106 0.287 0.035 (0.023) (0.042) (0.040) (0.042) (0.044) N 12,074 12,038 12,037 12,038 12,038 AIC 8,419.018 8,285.000 8,037.908 8,257.042 8,268.846 BIC 8,448.613 8,366.354 8,126.657 8,345.791 8,357.595 p <.1; p <.05; p <.01. All models include random e ects for both respondents and issues. The first model in Table 1 estimates a simple one-way ANOVA, simply partitioning the variance in the responses to determine how much of the variation in stereotypicality can be attributed to characteristics of respondents, rather than characteristics of the policy proposals themselves. Consistent with other work emphasizing the considerable heterogeneity of the public (e.g. Kertzer, 2013), an analysis of the intraclass correlations finds that there is 4.16 times more variation in the data between respondents than between issues, thereby reinforcing the importance of incorporating respondent-level predictors to explain this variation theoretically. Thus, the second model in Table 1 adds a series of individual-level covariates: respondents age, gender, income, race, education, partisanship and interest in politics (all of which are described in greater detail in Table 2 in Appendix 2). The results show that, on average, more educated 20