Democracies in Conflict

The Origins of the Democratic Peace Debate

The first test of the proposition that democratic states are less likely to engage in war with one another appears in Babst (1964) who concluded that the absence of democratic states opposing one another in both World War I and World War II was unlikely to be a result of chance. Small and Singer (1976) built upon this result with their observation that the entire Correlates of War data set did not include wars between any pair of democratic states between 1816 and 1965. The authors were somewhat cautious about the robustness of this result, however, due to the relatively small number of wars and democratic states during the period of study.

The democratic peace proposition began to gain some measure of notoriety in the literature with Rummel’s (1983) claim in JCR that democratic states are generally more peaceful than nondemocracies. This study represented the first appearance of the democratic peace literature in the JCR and was among the very early contributions on the topic. Rummel’s findings attracted responses in JCR from Weede (1984), who contended that this result was not statistically robust, and Chan (1997), who found little support for the claim that democracies were generally more peaceful. Importantly, however, Chan found that democracies were unlikely to engage in war with one another, as suggested by Babst (1964) and Small and Singer (1976). Doyle (1986) brought even greater attention to this result with his development of a Kantian theory of a “separate peace” among democracies.

By the end of the 1980s, Levy (1988, 88) made his well-known assessment that the lack of military conflict between democratic states is “the closest thing we have to an empirical law in the study of international relations.” This claim seems rather strong in retrospect given the strength of the evidence available at the time, but Levy’s confidence was upheld by subsequent studies. A flurry of findings in the late 1980s and early 1990s moved the literature quickly toward a consensus that (1) democracies are involved in military conflict just as often as other types of states and (2) democracies are rarely involved in military conflict with one another (Maoz and Abdolali 1989; Morgan and Campbell 1991; Morgan and Schwebach 1992; Bremer 1992). Maoz and Russett (1993) contributed to the democratic peace becoming one of the central research agendas in the field (see also Russett 1994).

The collapse of the Warsaw Pact in 1989 and of the Soviet Union in 1991 fundamentally altered the structure of the international system and launched a wave of democratization that led some scholars toward rather enthusiastic forecasts of the “end of history” as civilization marched inevitably toward democracy and peace (Fukuyama 1989). This confluence of events created the opportunity for new approaches to take hold in the study of international relations. Rather than debating contrasting theoretical paradigms, the field began to focus on explanations for empirical puzzles (Zinnes 1980). The democratic peace literature and the JCR, in particular, became two of the primary proving grounds for this different approach. During the thirty-four cold war years from 1957 through 1990, JCR published a total of four articles on the impact of democracy on military conflict. In the twenty-six years since the end of the cold war in 1991, that number skyrocketed to more than seventy articles.

Realist scholars immediately pushed back vigorously against the empirical claim that domestic politics affected the use of force. International Security devoted an entire special issue to the democratic peace and gave especially prominent space to its critics. Layne (1994), for example, found little evidence of reticence to attack democratic states in his historical review of several military crises between Great Power democracies. Spiro (1994) argued that temporal dependence problems rendered the statistical significance of the democratic peace finding suspect. And Farber and Gowa (1995) argued that common interests rather than common polities account for the lack of military conflict between democracies.

Numerous rebuttals to these empirical critiques demonstrated the robustness of the democratic peace result (Russett et al. 1995; Rousseau et al. 1996; Maoz 1997; Thompson and Tucker 1997), but a skeptic could still argue that this literature consisted largely of an empirical finding in search of an explanation and, as a result, had become a gaggle of theories all seeking to explain the same result. Generally, these theories could be categorized into one of two camps: normative and structural. The first set of arguments emphasized shared norms and values among democratic leaders as the causal mechanism explaining both the lack of military conflict among democracies and the willingness of democratic states to attack nondemocracies (see, e.g., Doyle 1986; Owen 1994; Dixon 1993, 1994). ¹ The second set of arguments emphasized the role of democratic structures and institutions in constraining leaders from using military force (see, e.g., Small and Singer 1976; Rummel 1983; Bueno de Mesquita and Lalman 1992). Fundamentally, this latter group of theories was founded on the assumption that citizens would punish leaders who forced them to endure the costs of war. “International action in a democratic political system requires mobilization of both general public opinion and a variety of institutions that make up the system of government, such as the legislature, the political bureaucracies, and key interest groups” (Maoz and Russett 1993, 626).

All of these theories were empirically plausible, since they had been constructed in order to explain the democratic peace phenomenon. Yet they remained fundamentally unpersuasive for the same reason. Maoz and Russett (1993) attempted to resolve this dilemma by deriving separate measures of democratic norms and structures, but their empirical operationalizations of both norms and structures were so intertwined as to make that effort unsatisfactory as well.

Audience Costs, Democratic Efficacy, and the Turn to Alternative Implications

The scholarly prominence and empirical robustness of the democratic peace finding soon attracted interest from formal theorists seeking to provide a more coherent conceptual foundation. Fearon (1994) offered an explanation of the democratic peace that launched an extensive literature on the impact of audience costs on military conflict (see also Smith 1998; Schultz 1998). Consistent with “structural” theories of the democratic peace, the central assertion of these audience costs models is that constituents will punish (i.e., remove) leaders that they view as incompetent. Both Fearon (1994) and Smith (1998) focus intensely on the importance of backing down from military threats as the central indication of leadership incompetence. The models explain the democratic peace by arguing that citizens in democratic states have significantly greater opportunities to punish their leaders than do citizens in nondemocracies. Most importantly, these models highlighted additional implications of the argument that should also be supported if this explanation of the democratic peace is valid. Specifically, the audience costs model directed democratic peace scholars to focus on crisis bargaining and crisis outcomes.

Early analyses of the audience costs model seemed to yield strong empirical support. For example, C. F. Gelpi and Griesdorf (2001) and Schultz (2001a) both found evidence that democratic states were more likely to emerge victorious in a crisis when they make escalatory threats. More recent analyses, however, have cast doubt on the audience costs hypothesis. Downes and Sechser (2012) criticized the data sets utilized by both C. F. Gelpi and Griesdorf (2001) and Schultz (2001a). Meanwhile, Snyder and Borghard (2011) relied on substantially different data and research methods from any of these scholars and found little evidence of audience costs as construed by the Fearon (1994) and Smith (1998) models. ²

These critiques exposed an important weakness of the audience costs argument: its narrow focus on empty threats as the critical indicator of policy maker incompetence. Failing to follow through with a threat could be a sign of incompetence; that same behavior could also indicate the opposite if a policy maker has obtained new relevant information. ³ More generally, it is not obvious why constituents should place greater weight on policy consistency than on policy outcomes. For example, while Snyder and Borghard (2011) find little evidence of domestic constituencies punishing leaders for inconsistency in foreign policy, they find ample evidence of domestic punishment for policy failure. Moreover, it is worth noting that the earlier studies supporting the audience costs hypothesis (C. F. Gelpi and Griesdorf 2001; Schultz 1999, 2001a) cannot distinguish between the hypothesized effects of policy inconsistency and policy failure because of their reliance on observational crisis bargaining data. Moreover, recent literature reframing audience costs as “competence costs” suggests the utility of this broader framing of the concept (C. Gelpi and Grieco 2015).

Fearon’s (1994) perspective on the empowering effects of constituency constraint also led scholars to investigate democratic efficacy in other areas. Lake (1992) first noticed the tendency of democracies to prevail in military conflict, followed by Bueno de Mesquita and Siverson (1995) and D. S. Bennett and Stam (1996). Reiter and Stam (1998a, 1998b, 2002) brought together various competing explanations of democratic efficacy in warfare in an influential project that brought this new literature to greater prominence.

Reiter and Stam’s strongest result is that democracies are substantially more likely to win wars that they initiate but only slightly more likely to win wars when they are attacked. They interpreted this result as evidence in support of the structural theories of the democratic peace. Specifically, they conclude that “The finding that democratic initiators are more likely to win is strong circumstantial evidence of their concern about how foreign policy outcomes will affect domestic political fortunes” (Reiter and Stam 1998a, 387).

As noted above, the earliest structural arguments about the democratic peace assumed that citizens would we unwilling to suffer the costs of military conflict and would punish leaders who engaged in war. However, this argument appeared inconsistent with the facts that democracies initiated numerous wars (Maoz and Abdolali 1989), that democratic leaders often became more popular after the initiation of military force (J. E. Mueller 1970, 1973; Oneal and Bryan 1995; Baker and Oneal 2001; Baum 2002; Chapman and Reiter 2004), and that hawkish candidates were often electorally successful. Moreover, theories based on a popular aversion to the costs of war could not explain that democracies were successful in battle (Reiter and Stam 1998a, 2002) and in crisis bargaining (C. F. Gelpi and Griesdorf 2001; Schultz 1999, 2001a).

Audience costs models could account for some of these findings but did not speak clearly to the questions of efficacy in war and rally ‘round the flag effects. Moreover, audience costs—construed as punishment for an inconsistent crisis bargaining policy—were difficult to measure with observational data (Schultz 2001b). After more than a decade of research, the democratic peace agenda had developed several robust empirical findings as well as a variety of theories to explain these phenomena. But the ability of the theories to explain the data remained unsatisfying.

As the empirical work on democratic efficacy gained momentum, a second game theoretic approach developed in an effort to unify this literature within a single theoretical approach. Like the audience costs models that preceded it, this new framework also focused on policy making success to explain the phenomenon but anchored the theory in different assumptions about what constituted policy “incompetence” and about how autocracies differed from democracies. Selectorate theory (Bueno de Mesquita et al. 1999, 2003) began with the assumption that all leaders seek to retain office by satisfying constituents and that constituents are by definition capable of removing their selected leaders. The critical difference between democracies and autocracies, according to this view, is the size of their constituencies—or “winning coalitions.” Democratic leaders tend to require much larger winning coalitions than their autocratic counterparts. While autocrats can satisfy their constituents with bribes and other “side payments,” democratic leaders must provide successful public policies in order to remain in office.

The selectorate theory of the democratic peace (Bueno de Mesquita et al. 1999, 2003) focused attention more directly on the importance of policy success rather than previous structural theories that had emphasized the costs of conflict and the consistency of military policy. The primacy of policy success provided a single theoretical foundation for the democratic peace, democratic efficacy, and rally ‘round the flag literatures. Like audience costs theory before it, selectorate theory made sense of existing empirical knowledge about democracy and military conflict and also led to important additional implications about the behavior of democracies during war and international crises. In this instance, however, the new theoretical framework pushed scholars toward observing the domestic political processes that were theorized to influence democratic foreign policy.

Scholars interested in the democratic peace began to investigate the operation of public opinion in democratic states engaged in military conflict. As noted above, early structural theories of the democratic peace were founded on the assumption that democratic publics were unwilling to suffer the costs of war (Maoz and Russett 1993). This assumption is consistent with key findings about the public’s response to casualties in military conflicts (J. E. Mueller 1970, 1973). But selectorate theory suggested that policy success should drive public responses to military conflict as strongly as the costs themselves. This theoretical insight brought the literature on the democratic peace into conversation with the long-standing literature on public opinion and foreign policy.

Selectorate Theory, Public Opinion, and the Foreign Policy of Democracies

Selectorate theory requires that members of the selectorate in democracies (i.e., voters) reward successful public policy efforts and punish failures. In order to accomplish these tasks with regard to military conflict, voters would need to satisfy several criteria. First, citizens would have to form coherent attitudes about foreign policy issues. Second, citizens would need to form and hold those attitudes in ways that are at least partially independent of the elite actors that they would allegedly constrain. Finally, citizens would need to act on these attitudes by casting ballots or placing pressure on their leaders in other ways (Aldrich, Sullivan, and Borgida 1989; Aldrich et al. 2006).

By the time that selectorate theory drew the attention of conflict scholars to these questions, the literature on public opinion and foreign policy had been addressing them for several decades. JCR, for example, published eighteen articles on public opinion and foreign policy before Rummel’s first study on the democratic peace in 1983. The growth of this literature—while not as striking as that of the democratic peace—has remained healthy with more than forty articles published in JCR since Rummel’s famous democratic peace finding. Much of this work focused on public opinion in the United States, largely because of the widespread availability of survey data. The earliest work on American public opinion and foreign policy was highly skeptical of the public’s ability to form coherent attitudes about foreign policy (Almond 1960; Lippmann 1922; see also, Campbell et al. 1960; Converse 1964). However, the Vietnam War brought about a dramatic shift in our understanding of public opinion and foreign policy. Verba et al. (1967) demonstrated that attitudes toward the Vietnam War were ideologically coherent. Caspary (1970) and Achen (1975) found that public attitudes toward foreign policy engagements were much more stable than had previously been suspected. And most importantly, J. E. Mueller (1970, 1973) demonstrated that the American public had responded in consistent and systematic ways to the experience of military casualties (measured as deaths) in both the Korean and the Vietnam Wars.

These startling insights led to a nearly complete reversal in our understanding of the public and foreign policy (Holsti 1992). A number of studies indicated that American attitudes toward foreign policy had a relatively sophisticated internal structure (Wittkopf 1990; Hurwitz and Peffley 1987; Peffley and Hurwitz 1993). In fact, over the next two decades, a model of “the rational public” came to dominate this literature.

Page and Shapiro (1982) analyzed data from thousands of survey questions collected in the United States over nearly a half-century and found that aggregate attitudes toward foreign policy issues were often quite stable over time, were just as stable as attitudes toward domestic issues, and generally shifted in reasonable ways in response to salient world events. Specifically, the authors, “maintain that Americans, as a collective body have done well with whatever information has been provided, and that they have formed and changed their policy preferences in a reasonable manner” (Shapiro and Page 1988, 211).

This optimism about the rationality of the American electorate grew into a confidence in the “prudence” of the public (Jentleson 1992) that appeared to have a sense of American national interests that comport with a relatively restrained realist perspective on world politics. Even more persuasively, Herrmann, Tetlock, and Visser (1999) found in an experimental context that individuals combined dispositional preferences with specific contextual facts about military operations in order to form reasoned judgments about the use of force. Moreover, while scholars revisited and built upon J. E. Mueller’s (1970, 1973) insights about public responses to casualties and war (Gartner, Segura, and Wilkening 1997; Gartner and Segura 1998), they continued to find that the public responded to real-world information about battlefield casualties in reasonable and systematic ways.

Just as this literature seemed to be reaching a consensus on public aversion to casualties and its role as a restraint on American military policy, the intersection of the democratic peace debate with the war in Iraq pushed this literature in new directions. The 2003 Iraq War represented America’s first substantial experience with military casualties since Vietnam. J. Mueller (2005) argued that America’s experience in Iraq mirrored that of Vietnam and Korea, except that casualty aversion had grown even more severe in the intervening years. This result seemed to confirm some of the early arguments about the democratic peace based on the public’s aversion to the costs of war. But other scholars argued that the public’s response to the costs of war was contingent on other contextual variables. Kull and Destler (1999) argued that a new internationalism among the public meant that international law and the consent of international organizations played a key role in the public’s response to war. Gartner and his coauthors argued that legislative structure and the geographic arrangement of casualties and constituencies determined the level of public constraint on war (Gartner, Segura, and Wilkening 1997; Koch and Gartner 2005; Gartner and Segura 2008). And a number of scholars began to focus on policy success as a determinant of casualty tolerance.

C. Gelpi, Feaver, and Reifler (2005, 2009) conducted a series of surveys of American public attitudes toward the war in Iraq and their analysis of opinion at both the individual and aggregate level suggested that “beliefs about the likelihood of success matter most in deter-mining the public’s willingness to tolerate U.S. military deaths in combat” (C. Gelpi, Feaver, and Reifler 2005, 8). Similarly, Eichenberg’s (2005) review of aggregate American support for military operations from 1981 to 2005 suggested that victory rather than casualties was the most influential determinant of public support. Importantly, Croco (2011, 2015) demonstrates that these perceptions of success combine with judgments about culpability for the military conflict in order to determine whether and how both the public and elites punish leaders who initiate unsuccessful military operations. These results are strongly consistent with selectorate theory’s (Bueno de Mesquita et al. 1999, 2003) structural explanation for the democratic peace which expects constituents in large winning coalition polities will be strongly focused on public policy success.

Elite Cues, the Elasticity of Reality, and the Partisan Limits of Constraint

Much of the literature on public opinion and American foreign policy evolved with little attention to the impact of partisanship. This focus was partly due to a focus on aggregate trends in public opinion—where partisanship would seem to have less explanatory power—rather than individual attitude formation. Additionally, however, as noted above, numerous studies had indicated that foreign policy views did not generally align strongly with party identification.

But stark divisions over the Iraq War brought foreign policy views into greater alignment with partisan politics (Jacobson 2008) and brought American Politics scholars into the literature on American foreign policy. Many of these scholars challenged the notion of a “rational” public, noting that public attitudes toward the war could be manipulated by framing effects and elite cues (Gershkoff and Kushner 2005; Boettcher and Cobb 2006). Drawing on themes raised by Zaller (1992) and Larson (1996), Berinsky (2007) challenged the “battlefield success” argument forwarded by C. Gelpi, Feaver, and Reifler (2005, 2009), Gartner and his coauthors (Gartner, Segura, and Wilkening 1997; Gartner and Segura 1998; Gartner 2008), and Eichenberg (2005). Specifically, Berinsky questioned the extent to which the public had knowledge of battlefield events such as casualties and also emphasized the biased processing of information based on the impact of cues from partisan elites. Berinsky and Druckman (2007) also challenged the findings of C. Gelpi, Feaver, and Reifler (2005) on methodological grounds. C. Gelpi and Reifler (2008) rebutted the specific methodological critiques, but the broader concern about partisan bias in opinion formation remained salient. For example, Gelpi et al. (2009) demonstrate that perceptions of success have a dramatic impact on support for military operations even after controlling for party identification but how do individuals construct those perceptions of success? Does partisanship shape individual perceptions of success? If so, does the impact of these partisan elite cues threaten the public’s ability to constrain or punish their leaders?

These more difficult questions of causal inference pushed scholars increasingly toward the use of experiments to test arguments about the formation of attitudes rather than the previous heavy reliance on aggregate data. Gartner (2008, see also Gartner and Gelpi 2016) conducted a series of experiments on public responses to casualties and found consistent evidence in support of what a “rational expectations” model of opinion formation whereby individuals updated their attitudes toward war based on reasoned inferences from the exposure to new battlefield information. Similarly, in an experiment comparing the impact of battlefield information as reported by news media to elite rhetorical cues, C. Gelpi (2010) found that individuals updated their attitudes in response to surprising information that challenged their prior beliefs rather than rhetoric from the president.

Skeptics of the rational expectations approach were not always persuaded by these results. For example, Paolino (2015) contended that Gelpi’s data contained evidence of the impact of elite cues. C. Gelpi (2016), however, rebuts this claim. And Nyhan and Reifler (2010) made perhaps the sharpest critique of the rational expectations approach when they concluded that exposure to new and surprising information actually “backfires.” That is, individuals tend to become more entrenched in their prior beliefs when faced with new information that conflicts with partisan predispositions. Wood and Porter (2016), on the other hand, conducted a series of experiments which consistently indicated that “[b]y and large, citizens heed factual information, even when such information challenges their partisan and ideological commitments” (Wood and Porter 2016, 1). In fact, the authors demonstrate that the backfire effect that Nyhan and Reifler observe appears to be an artifact of the particular question wording used in their survey item. ⁴

While scholars continue to explore and debate issues surrounding the formation of mass opinion, several results seem to emerge clearly from these exchanges. First, it is clear that partisan elite cues influence opinion formation at the individual level. Even scholars within the rational expectations camp find clear evidence of these effects (C. Gelpi, Feaver, and Reifler 2009; Grieco et al. 2011; Golby, Feaver, and Dropp 2017). Second, when presented with new information, individuals tend to update their beliefs in ways consistent with rational expectations theory, although the effects are more muted for strong partisans (Gartner 2008; C. Gelpi 2010, 2016, Wood and Porter 2016). And third, while the formation of individual level attitudes can be influenced by partisan biases, American public opinion in the aggregate tends to respond to international events much as the “rational public” scholars would expect (Shapiro and Page 1988; Erikson, MacKuen, and Stimson 2002). For example, Berinsky (2007) finds substantial variance in public knowledge of the number of US military deaths that had occurred in Iraq as well as some modest evidence of partisan bias in the errors that individuals make in their casualty estimates. However, at the same time, he also finds that the modal public response is quite accurate, and the median estimate from his sample of respondents is extremely accurate. Specifically, the median response was 900 US soldiers killed, and the correct answer increased from 901 to 915 during the fielding of the study.

How can partisan biases shape attitudes at the individual level while the public is generally reasonable in the aggregate? ⁵ There appear to be important mechanisms working at both the micro and the macrolevels. At the microlevel, Lupia and McCubbins (1998) persuasively demonstrate that individuals can use elite cues to form attitudes that reflect their interests while gathering and retaining much less information than would be necessary in the absence of cues. Moreover, Baum and Groeling (2010) argue that the impact of elite cues depends heavily on the informational advantage that elites have over the public with regard to political issues. In the absence of any independent basis for judgment, individuals tend to rely on partisan cues as trusted experts. As an issue remains salient in the political environment over time, however, individuals become more informed and have a stronger internal basis for judgment. Consequently, “reality asserts itself” over time, and real-world events begin to shape opinion. ⁶ This insight is especially important because—like Bueno de Mesquita et al. (1999, 2003)—it connects the extensive literature on “rally ‘round the flag” effect with the “rational expectations” literature on casualties, success, and democratic constraint. Democratically elected elites have great leeway to use military force at their discretion and the public will initially be permissive in these efforts. Over time, however, if they mission is viewed as unsuccessful or excessively costly, the public will turn sour on the mission and be more inclined to punish their leaders as a result.

At the macrolevel, the key mechanism appears to be that individual biases as well as individual errors or lack of factual knowledge tend to cancel one another out (Erikson, MacKuen, and Stimson 2002). The opinions of uninformed individuals will become “white noise” as one aggregates to the macrolevel, while the biased attitudes of strong partisans will cancel one another out when competing parties mobilize their bases on opposing sides of an issue. As a result, even in an environment with a substantial number of uninformed or highly partisan voters, changes in the median attitude are driven disproportionately by individuals who are attentive to political events and who do not have strong party loyalties. This mechanism of aggregation allows the public to behave “as if” all of its members were rational because a sufficiently large number of them are, and the remaining individuals roughly balance one another out. “The net result is that the more informed, thoughtful, and attentive citizens contribute disproportionately to aggregate movement” (Erikson, MacKuen, and Stimson 2002, 5).

A reasonable public in the aggregate, however, gets us only halfway toward the expectation that public opinion may constrain foreign policy. Specifically, voter dissatisfaction will not put pressure on leaders unless opposing parties are willing to tap into this dissatisfaction and attract popular support by advocating for alternative policies. Page and Brody (1972), for example, find that American voters had clear and coherent attitudes toward the Vietnam War and were willing to take them into account in evaluating candidates. However, the fact that Nixon and Humphrey were both vague and took similar positions on this issue meant that voters could not use these attitudes to make their choices. Similarly, Aldrich, Sullivan, and Borgida (1989) found that the impact of foreign policy issues varied substantially depending on the salience of foreign policy issues in the presidential campaign and the perceived policy differences between the candidates. Thus, when partisan elites provide the public with foreign policy alternatives, voters are able to use their evaluations of foreign policy performance to select (or remove) their leaders (Aldrich et al. 2006).

Data from recent elections confirm that attitudes toward the Iraq War had a substantial impact on voting behavior in 2004 and 2008 (Hillygus and Shields 2005; Norpoth and Sidman 2007; C. Gelpi, Reifler, and Feaver 2007; Hill, Herron, and Lewis 2010), and that casualties from the war—which was widely perceived as unsuccessful—cost the Republican party a significant number of votes (Karol and Miguel 2007; Gill and DeFronzo 2014). However, the existence of an opposition candidate that is willing to highlight foreign policy failure of the incumbent as part of its platform is an important scope condition for the popular constraint on foreign policy. For example, as President Obama faced reelection in 2012 the war in Afghanistan had become increasingly costly, unpopular, and viewed as unsuccessful. Nonetheless, Mitt Romney’s failure to articulate any substantial critique of Obama’s Afghanistan policy meant that the president paid no discernible electoral price for the more than 1,700 American soldiers killed between 2009 and 2012 in two unsuccessful wars.

Media Access, Media Bias, and the Availability of Foreign Policy Information

The final critical link in the chain of democratic structural constraint is the public’s willingness and ability to gather relevant information on foreign policy issues. Numerous studies have documented the fact that members of the public do not retain large amounts of factual knowledge about policy issues. But the failure to retain factual information does not imply that attitudes are not based on facts. Online processing models of decision-making suggest that individuals update their attitudes in response to new information but only retain the new judgment without the content of the information that generated it (Bizer et al. 2006).

Nonetheless, even if members of the public do not retain factual information about foreign affairs, they do need to be exposed to this information if it is to have an impact. Since only a tiny proportion of the population has any direct experience with military combat, the media plays a critical role in the public’s ability to construct informed judgments. Early research on the impact of the mass media often found very modest effects (Lazarsfeld and Merton 1948; Hovland, Lumsdaine, and Sheffield 1949). Some prominent studies questioned the “minimal” nature of media effects (Entman 1989; Allen et al. 1994; W. L. Bennett and Paletz 1994). But by the mid-1990s the so-called agenda setting hypothesis had become the dominant view of the impact of news coverage on public opinion (McCombs and Shaw 1972; Iyengar, Peters, and Kinder 1982; Iyengar and Kinder 1987; Krosnick and Kinder 1990; Iyengar and Simon 1993; Miller and Krosnick 2000).

The agenda setting research was largely based on evidence collected in the “Golden Era” of news media, where news consumption was dominated by a small number of television (TV) outlets with a very similar editorial slant and tone. Moreover, these fairly similar outlets tended to mirror elite partisan rhetoric with their coverage. This practice of “indexing” elite rhetoric limited the room for the independent influence of media cues (W. L. Bennett 1990, 1994). However, the 1990s also brought a major change to the structure of news media in the United States. The advent of cable TV news, and especially the rapid rise of Fox News as a major TV outlet with a distinctly conservative editorial slant, fundamentally altered the informational environment for Americans (Baum and Kernell 1999). This transformation has continued as news sources have proliferated across the Internet. Greater choices with regard to media consumption have led to greater variance in exposure to political information (Prior 2005).

A decade after the launching of Fox News, Groeling (2008) discovered “a surprisingly pervasive pattern of bias across media outlets” (Groeling 2008, 650). In a related study, Baum and Groeling (2008) found even stronger evidence of bias across “new media” sources such as online blogs. Notably, both studies found substantial evidence of partisan bias from both left- and right-wing sources. The increasingly partisan media environment raised important questions about the potential for larger media effects through the mechanisms of persuasion and self-selection. The former argument implies that news coverage directly causes individuals to change their minds, while the latter argument implies that self-selection of news exposure insulates individuals from the surprising information that is central to the “rational expectations” model.

With regard to persuasion effects, DellaVigna and Kaplan (2006) leveraged a research design built around the gradual expansion of Fox News across TV markets to make an even stronger claim that access to Fox News caused an increase in Republican voting of 0.4 percent to 0.7 percent between the 1996 and 2000 presidential election. This result implies that between 3 percent and 28 percent of Fox viewers were persuaded to vote Republican in 2000 by exposure to Fox. Coverage of the Iraq War also became an important proving ground for studying persuasion effects in the new media environment. For example, Aday (2010) found that while 90 percent of the network newscast stories on Iraq were neutral in tone, nearly 40 percent of Fox News stories had a positive valence. Meanwhile, Kull, Ramsay, and Lewis (2003) found in the wake of the invasion of Iraq and the failure to locate weapons of mass destruction (WMD), that Fox News viewers were significantly more likely than other Americans to continue to believe that Saddam Hussein had nonetheless possessed WMD and had substantial links to Al-Qaeda.

With regard to self-selection of news and the ability to insulate oneself from the “real world” of news events, Morris (2005) found that the partisanship of Fox News viewers did not differ significantly from Cable News Network (CNN) viewers in 1998. But by 2004—in the wake of the Iraq War and its whirlwind of coverage—Fox viewers were significantly more likely to identify as Republicans. By the presidential election of 2008, a Pew Study indicated that Fox viewers were about twice as likely to be Republicans as compared to network news viewers, and Republicans were about twice as likely to approve of the coverage of Fox News as Democrats. Baum and Groeling (2008) also demonstrate significant partisan differences in news selection across a variety of new media platforms.

The polarization of the news media in the wake of the Iraq War created a second layer of partisan cues on top of the long-standing literature on partisan politicians. Now citizens needed to interpret the information that they receive in terms of both the partisan identity of the speaker (i.e., the politician) in a news story and the partisan identity of the reporter (i.e., the news outlet). This additional level of partisan filtering undoubtedly creates opportunities for avoiding relevant information. However, as was the case with partisan elites (Lupia and McCubbins 1998), partisan polarization also creates opportunities for the public to distinguish between credible claims and “cheap talk” in the media (Baum and Groeling 2009).

Public Opinion, Selectorate Theory, and the Scope of Democratic Constraint

Selectorate theory provided an elegant theoretical explanation for most of the major empirical patterns regarding democracy and the use of military force (Bueno de Mesquita 1999, 2003). Skeptics of the democratic constraint argument, however, contended that publics are not capable of providing such a constraint on foreign policy decision-making because of biases created by partisan cues from political elites and increasingly polarized media outlets. But while these cues can have a significant impact on attitude formation at the individual level, they do not eliminate or overwhelm a reasoned process of attitude formation and updating with regard to foreign policy issues for a substantial segment of the public. As a result, when one examines public opinion in the aggregate, it broadly conforms to the expectations of the rational public framework. Moreover, the public seems to care deeply about the success of foreign policy efforts, as expected by selectorate theory.

The literature on the role of foreign policy issues in American electoral politics, however, points to a very important caveat for the democratic constraint argument (Aldrich et al. 2006). Political parties must provide voters with alternative platforms if voters are to provide any constraint on policy making. Additionally, the literature on media effects reminds us that individuals must have access to information on foreign policy via the news media if they are to be able to form attitudes that are independent of elite rhetoric. These two results appear to provide the scope conditions for the operation of democratic constraint as anticipated by the literature on media, partisanship, public opinion, and foreign policy.

Baum and Potter (2015) explicitly build on these two scope conditions in their extensive cross-national study of democratic foreign policy constraint. Specifically, Baum and Potter focus on the interaction of the effective number of political parties and the accessibility of mass media as the core causal mechanism at the root of the democratic peace. Their study finds that these two conditions—rooted deeply in the analysis of public opinion data—provide a persuasive account of democratic constraint in terms of the decisions of democratic polities to initiate military conflict, to reciprocate when they are threatened or attacked, and to collaborate with other democracies to prevail once they are engaged in military conflict.

Like Russett’s (1994) Grasping the Democratic Peace, and Reiter and Stam’s (2002) Democracy and Victory, Baum and Potter’s (2015) War and Democratic Constraint seems to represent an important milestone in our understanding of democracy, partisanship, public opinion, and foreign policy. Nonetheless, important questions remain in our evolving understanding of democracy and conflict. First, while Baum and Groeling (2009) give us hope that even a polarized media environment can yield informative cues for citizens, recent attention to “fake news” and its potential effects on voting behavior in the 2016 election raises concerns about the limits of the rational public. False news stories are certainly not a new phenomenon. But widespread access to the Internet and social networking platforms raises the prospect that nearly any individual can distribute false information very widely in a short period of time. Moreover, while scholars have uncovered substantial evidence that biased cues can be informative is persuasive, it is not obvious that false cues can be informative as well.

Second, while our understanding of structurally based sources of democratic constraint has grown dramatically over the past three decades, these results do not imply that democratic norms do not matter for foreign policy behavior. Unfortunately, progress on the potential impact of democratic norms has been slower. Much of this difficulty is rooted in the challenges involved in measuring democratic norms in a systematic way. Early studies often relied on institutional data to measure the presence of norms by proxy (Maoz and Russett 1993; Dixon 1994), but this strategy has obvious limitations given the prominence of alternative arguments about democratic structures. Recent advances in our ability to code text and generate systematic data open the possibility for further progress in our measurement of democratic norms at both the mass and the elite levels. Moreover, recent experimental work on the democratic peace suggests a normative or ideational component to public aversion to attacking democracies (Tomz and Weeks 2013; Cuhadar and Druckman 2014). ⁷ Both of these avenues may be fruitful for future research.

Finally, while we have extensive observational data relating democratic political structures to foreign policy behavior, and we have extensive public opinion data—including experimental manipulations—that document the causal mechanisms linking public opinion to democratic foreign policy, we know much less about the direct causal impact of democracy on foreign policy. Estimating an average treatment effect for democracy supposes that one could—in principle—randomly assign states to be democratic or autocratic (Pearl 2009). Such experimental control is, of course, impossible, and the presence of democracy is seriously confounded with a number of other factors that are plausibly related to military conflict. Thus regression models that estimate independent effects for democracy are dependent on strong assumptions about functional form (King and Zeng 2007). Even more flexible estimators such as statistical matching models may struggle to find “nearest neighbor” comparisons that represent plausible control cases for the “treated” democracies. Identifying a treatment effect for democratic institutions remains a difficult research design problem.

The difficulty in estimating a treatment effect for democracy is more than a merely academic concern. Presidents Clinton, George W. Bush, and Obama all explicitly relied on the assertion that “democracies don’t fight each other” as a core principle of American grand strategy. This belief has led policy makers to try to spread democracy around the world without a strong understanding of how one should “reassign” states to be democratic, nor what the treatment effect of such a reassignment is likely to be, given that the democracies that we have observed emerged through a very different causal process. Recent American experiences with imposing elections in Iraq, Palestine, Afghanistan, and elsewhere suggest that assigning states to have democratic political structures may not have the same result as democratic institutions that emerge from an endogenously driven process. Thus scholars must be careful to be circumspect about the policy implications from this research and have a responsibility to articulate the limits of our knowledge regarding causal inference.

Lessons from the Democratic Peace Literature

I have argued that the democratic peace literature has generally been characterized by a productive exchange between inductive observation and deductive explanation. In particular, scholars in this literature have gathered information about the behavior of democratic states and democratic citizens, constructed theories that comport with these observed patterns, and then tested those theories through an examination of alternative implications of the argument. The empirical pattern of the dyadic democratic peace led to the construction of domestic audience costs models. These models led scholars to investigate the efficacy of democracies in conflict. The results of these additional tests laid the foundation for the construction of selectorate theory, which, in turn, made predictions about the conditions under which the public would punish involvement in military conflict. The resulting attention on public opinion and foreign policy revitalized this literature and brought it more productively into conversation with long-standing literatures in American politics and communications. These exchanges, in turn, helped to formulate a set of scope conditions for the operation of democratic constraint.

Thus, the democratic peace literature represents an excellent example of how a progressive research program—in the Lakatosian (1970) sense—can contribute to our understanding of international politics. The great progress that this literature has enjoyed over the past three decades stands in some contrast to previous debates over broad paradigms for understanding international politics. Fully articulating the reasons for the progressive nature of the democratic peace research program are beyond the scope of this article. However, two features of this literature appear to stand out as potentially important sources of scientific progress. First, the literature stayed closely focused on identifiable empirical puzzles (Zinnes 1980). The initial puzzle, of course, was the democratic peace itself, but this focus evolved to include democratic efficacy, and finally the formation of public attitudes toward war. And second, each of these puzzles was investigated with common data sets that provided scholars with an agreed empirical “target” that they needed to hit. Whatever their inevitable flaws, the Correlates of War data sets on Militarized Disputes and Interstate Wars as well as the Interstate Crisis Behavior Project became a useful proving ground for theories. Similarly, arguments regarding American public opinion were interrogated with common data sources on presidential approval and war approval. Even when scholars conducted their own surveys or experiments, the data and survey instruments were made publicly available for replication and extension.

This combination of empirical puzzles and common data sets disciplined scholars to frame their questions and their analyses in compatible ways that enabled the formation of cumulative knowledge. This practice is reflected in the many direct exchanges between scholars over the interpretation of common data sources (Russett et al. 1995; Maoz 1997; Thompson and Tucker 1997; Downes 2009; Berinsky and Druckman 2007; C. Gelpi and Reifler 2008; Paolino 2015; C. Gelpi 2016. Paradigmatic conflicts often lacked this kind of focus and so missed the chance to generate cumulative knowledge (Vasquez 1997). As the contributors to the JCR forge ahead into the next sixty years of research on conflict and violence, scholars could do much worse than to rely on the model of the democratic peace research program as a model for deepening and broadening our understanding of the sources of peace and war.