How 'Democratic' is the Democratic Peace? A Survey Experiment of Foreign Policy Preferences in Brazil and China

Recent research has shown that British and American respondents are less willing to advocate the use of force against fellow democracies than against non-democracies (TOMZ and WEEKS, 2013). These findings may contribute to understandings of the 'democratic bias'—unwillingness to attack democracies. A critical next step is assessing whether publics beyond the US and the UK have similar attitudes. To address the scope of popular preferences for peace with democracies, we conduct survey experiments using online panels in two emerging powers, one a democracy (Brazil) and one a non-democracy (China). Our survey randomly varies the hypothetical target's regime type and authorization by the United Nations for military action. We find that Brazilian respondents are significantly less likely to support the use of force against a democracy than a non-democracy. However, after controlling for UN approval, Chinese respondents do not appear to distinguish between democracies and non-democracies when considering whether force is justified. In addition, for both countries, UN approval has a larger effect than democracy on public support for the use of force.

(2020) 14 (1) e0002 -2/38 tudents of politics have devoted enormous attention to exploring the role that domestic publics play in formulating foreign policy. Evidence of the impact of public opinion has been difficult to unearth, however. Given difficulties with inference and identification, researchers have begun to apply experimental techniques to assist in better understanding relationships.
In a recent example, Tomz and Weeks (2013) demonstrate experimentally that subjects in the United States and the United Kingdom are less willing to advocate attacks against democracies than against non-democratic countries.
Popular preferences might thus account for the democratic peace -the observation that democracies seldom fight each other.
Yet while experiments can establish strong causal linkages, ambiguity persists in the interpretation of results and in connecting individual-level findings to macro-level processes. It is not clear, for example, that asking citizens in a democracy whether they are willing to go to war with another democratic country necessarily implies that respondents in a non-democracy will behave differently, or that the populations in all democracies are equally reticent to make war on other democracies. Much remains before researchers can confidently tie findings about a subject's stated preferences to patterns of conflict involving liberal republics. A critical step, as Tomz and Weeks (2013) point out, is to determine the generalizability of the linkage between public opinion, regime type, and war.
How 'democratic' are popular preferences for peace with liberal states? To answer this question, we conducted survey experiments involving subjects in two emerging powers, China and Brazil. The United States and the United Kingdom are 'exceptional' nations, possessing extraordinary wealth, power, social status and with closely linked histories and cultures. The foreign policy attitudes of citizens in these two English-speaking democracies might prove equally exceptional.
The two countries in our survey were chosen with considerable care. Each nation is a rising regional power, capable of acting aggressively if it so chooses.
Using military force is thus more than a mere abstraction for publics in either country. Each nation's interests are also at odds with the global status quo; questions about the use of force are unlikely to be confused in the public mind with hegemonic leadership or acting 'as the world's policeman'.

S
(2020) 14 (1) e0002 -3/38 Brazil is a young democracy, exhibiting key differences from established leading nations. As in many younger democracies, support for democracy is not strong as in the United States or Western Europe; in some surveys, fewer than half of respondents identify democracy as the best form of government. Popular enthusiasm for cooperating with other democracies may be affected by the degree to which democracy is consolidated domestically.
China is a non-democracy. The democratic peace is defined by critical differences in the foreign policy behaviors of democratic and non-democratic states. It is thus essential to examine the determinants of foreign policy in both types of regimes. If public opinion has an impact on the use of force by regimes, then we must examine opinion in both types of systems. Of course, it may well be that public preferences have less impact on foreign policy in authoritarian than in non-authoritarian countries. But this would imply an interaction between institutions and public opinion as the causal mechanism that explains the democratic peace.
Either way, a critical first step is empirical -examining public attitudes toward democracies in both democratic and non-democratic regimes. These cases thus offer important variation in regime type, development, status and culture needed to evaluate the generality of the link between public opinion and democratic peace.
A second concern has to do with meaning. Publics may imbue the word 'democracy' with content that researchers ignore at their peril. It is possible that subjects interpret democracy, not as a set of political institutions and norms as understood by academic researchers, but as coded language for a 'good', 'friendly' or 'responsible' country. To find out, we included a second experimental variable in our survey. Subjects were randomly informed that the United Nations had, or had not, authorized using force against a target nation. While only an initial step in determining how subjects perceive democracy, the treatment addresses concerns that democracy may be interpreted by subjects as an authoritative cue indicating quality or virtue.
As it turns out, the public preference for peace with democracies is widespread but context dependent. Respondents from both Brazil and China were generally less likely to endorse military violence against a state when it was randomly identified as a democracy in our experiment. At the same time, however, United Nations authorization -or a lack thereof -proved much more important in predicting public preferences for using force.
After reviewing relevant literatures, we detail the benefits of a broader assessment of the connection between public opinion and the democratic peace, across cultures, regime type and economic development. We then discuss details of our experimental design and present the results from the Brazilian and Chinese samples. We conclude by reviewing implications of our findings for democratic peace theory.

Literature: democracy and peace
Democracies are much more peaceful with each other than are other pairings of states, though democracies are about as war prone as other regimes in general (RUSSETT and ONEAL, 2001) 1 . This implies that democratic dyads are the most cooperative, followed by non-democratic dyads, while mixed dyads (democracy and non-democracy) are the most conflictual. Numerous studies observe or document a significant reduction in conflict in democratic dyads (e.g. BABST, 1964;DOYLE, 1997;HUTH and ALLEE, 2003;LEVY, 1988;MAOZ and RUSSETT, 1993;RUSSETT, 1993;SMALL and SINGER, 1976) Theorizing the democratic peace has proven to be a greater challenge.
Constructivists claim that force in the international system is becoming socially ______________________________________________________________________________________________ 1 Some debate over whether democracies are generally less warlike, though even advocates admit that this is a weaker relationship (e.g. ROUSSEAU et al., 1996).
Others assert that mature democracies fail to fight states they perceive as democratic (WEART, 1998) 4 . Many authors have focused on the informational aspect of democracies, viewing them as more transparent (e.g. SMALL, 1996;VAN BELLE, 1997) or possibly more credible due to the 'audience costs' or opposition groups that enable democracies to signal resolve (e.g. FEARON, 1994;SCHULTZ, 1998SCHULTZ, , 1999SMITH, 1998) However, constraint theories have been criticized as ad hoc and deductively flawed (MESQUITA et al., 1999;ROSATO, 2003). Moreover, scholars have noted that efforts to avoid circularity between theory and evidence would benefit most from new empirical content (HUTH and ALLEE, 2003). Work by Mousseau (2000) and Hegre (2000) shows that the pacifying effect of a treatment for whether the target country supports human rights subsumes the effect of the treatment for democracy.
Evidence that citizens care more about whether a hypothetical target is humanistic than democratic suggests a role for social affinity. Common preferences or values may be a key contributor to democratic peace (GARTZKE, 1998(GARTZKE, , 2000HUTH and ALLE, 2003). The term 'democracy' may also have important socially constructed connotations for respondents, reflecting subjective normative 'goods' in addition to a nation's actual political attributes.
The notion that democratic peace can be explained by elite or popular affinities is also attractive because it is uncomplicated (FARBER and GOWA, 1997 merely be a 'permissive condition'. Again, a critical first step is empirical: measuring public attitudes in both democracies and non-democracies.
Our analysis also explores the role of international institutions in opinion formation. Specifically, our survey experiment varies whether or not the proposed use of force has been endorsed by the United Nations. The inclusion of this second treatment helps separate procedural democracy from 'good' countries.
Although academics have learned to use terms like 'democracy' with considerable analytical precision, it does not follow that our subjects have in mind the Freedom House definition. Democracy may simply serve as a convenient proxy for things respondents deem to be good or similar to themselves or their nation. Subjects may treat the researcher's use of the term as an authoritative cue that force is unwarranted, and that recommending military action will be frowned upon.
Subjects may also view the approval of international institutions as an important mechanism for assessing the legitimacy of proposed uses of force, as suggested by sources in the literature.
Just as 'princely virtues' were once presented as the standard by which the behavior of political leaders was to be evaluated-even though very few princes actually exhibited these virtues-so too 'democracy' has now come to represent a broad and amorphous set of desirable national qualities. Almost every country claims to be a democracy, even those that clearly do not qualify by any reasonable definition. At the same time, enemies are capable of misrepresenting regime type.
Saddam Hussein and Fidel Castro each claimed that their regimes were democratic, and each offered a skeptical view of democracy in the United States.
Symbolic or socially constructed interpretations of democracy are bound to appear among subjects from powerful western nations, where a country's virtues will tend to be associated with a willingness to accept the status quo. Rather than capturing the effects of the political institutions of democracy on the willingness of subjects to advocate war, experimental research may really be measuring whether the hypothetical opponent is perceived to be in good standing with the international order or is even hostile or friendly.
A partial solution may stem from evidence that international institutions help to shape public preferences involving the use of force 7 . Experimental work on the effects of IOs (international organizations) on public opinion (TINGLEY and TOMZ, 2012) has shown that the United Nations Security Council's (UNSC) approval boosts public support for war via a process of 'legalizing' the proposed action by granting it moral authority, serving as a useful heuristic for the public.
While the regime type of the target country may be perceived as 'cheap talk' on the part of strategic actors, approval from IOs may serve as a more reliable and costly cue from an authoritative entity for determining individuals' stances on the use of force. Existing literature has documented the 'second opinion role' of IOs (GRIECO et al., 2011); it shows that the American public is affected by IO approval of the use of force because it provides a credible cue that such use of force is good policy (CHAPMAN, 2011;GRIECO et al., 2011).
We use international institutional approval of the use of force to control for the virtues of a potential target. Specifically, we vary whether or not the use of force has been approved by the United Nations. We thus measure the impact of regime type on public attitudes in an environment where the international community has judged action necessary, and in one where it has not.
Previous attempts to tie public opinion to the democratic peace have not considered that this relationship may be mediated through international institutions, or that 'democracy' itself may be interpreted by subjects as an authoritative cue to the effect that they should oppose the exercise of military violence. Combined with the exceptional nature of the samples used in previous studies, the danger is that the impact of popular preferences is either too ubiquitous or too unusual to conform to the dyadic macro observation that democracies do not fight each other, while other combinations of regimes continue to interact through force. We explore these possibilities by means of the survey experiment that we detail below.
______________________________________________________________________________________________ 7 An international resolution also serves as a commitment mechanism, encouraging domestic publics to 'rally round the flag' and may even lead foreign publics to advocate caution from their own governments (THOMPSON, 2006). International institutional approval further implies greater support and lower costs for states or coalitions authorized to use force, making contests less objectionable to domestic publics.

Experimental design
The ideal experiment on public opinion and the democratic peace would measure the support of subjects for their government's use of force against every country. An opponent's regime type would be randomly assigned, and each country would face identical scenarios and geopolitical contexts. Our resources are too limited to survey every country, and each country faces unique security environments that no doubt affect public perspectives (compare, for example, Iceland and Israel). However, we believe that careful case selection and an experimental design that decontextualizes security threats helps to advance the literature.
Our study involved survey experiments in China and Brazil. As with previous work (LACINA and LEE, 2013;TOMZ and WEEKS, 2013), we use internetbased polling. Subjects in each country were asked to read short scenarios ('vignettes') about crises involving two hypothetical countries ('A' and 'B') and to express their support for using force. Two treatments were randomly assigned: the regime type of Country B and UN authorization for Country A's use of force.
Our survey used the following format. The script was translated into the local language 8 . Prior to reading the script, subjects were advised that the scenario is hypothetical and should not be read as if it referred to any particular country 9 : a country in the same part of the world as 'Country A' is developing nuclear weapons and will have its first nuclear bomb within six months. This country (Country B) could then threaten other countries in the region with possible nuclear attack.
'Country A' has attempted to resolve the situation peacefully, but 'Country B' refuses to stop or even discuss the issue. The second treatment varies UN authorization for Country A's use of force against 'Country B'. Extensive theoretical research identifies the approval of international institutions as a key factor in determining popular support for war.
One strain of thought emphasizes the legitimizing effect of authorization by an international institution (FINNEMORE, 2003;HURD, 2007). A second perspective argues that international approval plays an informational role, reducing uncertainty about the likely reaction of the international community to a state's use of force (BOEHMER et al. 2004;CHAPMAN, 2011;FANG, 2008;GRIECO et al., 2011;VOETEN, 2005). The combination of regime type and international institutional support for using force defines 2x2=4 treatments. or non-democracies, not about particular countries in specific contexts. We create a much more general framework for assessing the willingness to use force -one that more nearly reflects the axiomatic nature of democratic peace theory -by using generic country names in our vignettes. Our approach is also useful in simplifying the process of conducting surveys in locations where government officials might reject more specific or pointed survey questions addressing national policy.
The choice to use generic country names also has a more practical motive. were provided with an online link to the survey experiment, which was programmed in the local language, and routed back to the survey firm's website where subjects were compensated for participating in the survey. We adopted many of Peifer and Garrett's (2014) recommended best-practices for online panels, and data were screened for duplicate responses. Following Tomz and Weeks (2013), we also collected subjects' demographic information -age, gender, education, income, religiosity, and interest in international news -and foreign policy attitudes, such as militarism, internationalism, and nationalism.
While each sample of respondents is not perfectly representative of the population of the two countries, they give us a good picture of the opinions of middle class, well-educated citizens, a sample population that is particularly well disposed to reflect the values sought in democratic peace research.  Consequently, excluding those with no internet access or with the lowest educational levels will probably inflate the impact of democracy, and given the similar relationships between internet access and democracy observed in the World Values Survey (INGLEHART et al., 2014), this effect will likely be similar in

Results
The sections below review the major findings for our study. The results reveal surprising differences and remarkable similarities across two populations with very different cultures and political structures. A final section attempts to make sense of these findings.    These results suggest several initial conclusions. First, there is an effect of democracy in Brazil, and evidence of a suggestive but not significant effect in China. Second, in both cases, there is an even larger impact of UN approval on the willingness to use force. Finally, our Chinese subjects are generally more supportive of using force than are our Brazilian subjects. These differences between Brazil and China are striking but may reflect demographic differences or other features of sample variability. For example, China's sample is younger and more male than the Brazilian sample, variables associated with a willingness to advocate force. We next conduct multivariate analysis to address a variety of demographic and attitudinal variables.

Robustness checks
We complement our basic analysis with robustness checks in the presence of control variables. We adopt two strategies. First, we use logistic regression to   (HURWITZ and PEFFLEY, 1987;HERRMANN et al., 1999;TOMZ and WEEKS, 2013), and 'Nationalism' (JOHNS and DAVIES, 2012)

______________________________________________________________________________________________
13 See Appendix for more information about each sample's summary statistics (Table S06, Table  S07, Table S08). In both countries, self-reported levels of militarism, internationalism, and nationalism had a sizable effect on the likelihood of a respondent advocating the use of force. Not surprisingly, a respondent with strong militaristic attitudes is more likely to support military action than are less militaristic individuals. Nationalistic respondents in both countries tend to be more favorable toward to the use of force than less nationalistic respondents. Higher levels of these two variables in China may explain the greater overall willingness of Chinese respondents to use force. Those who thought of specific cases in response to the vignettes were also more likely to favor aggressive foreign policy action, which may explain the relatively smaller effect sizes in studies with hypothetical situations than in studies that rely on real cases 14 .
Perhaps the most striking distinction between the two samples occurs in the context of the internationalism measure, which produced large, significant but contrasting effects for Brazil and China. Internationalism substantially increases support for using force among Brazilian subjects of the survey experiment, whereas it is negatively associated with a willingness to war among Chinese subjects. We speculate that this difference reflects contrasts in how subjects in the two countries interpret the role of the United Nations. China's permanent seat on the UN Security Council may alter the meaning of UN authorization for some Chinese, as UN authorization requires, at a minimum, China's acquiescence. ______________________________________________________________________________________________ 14 For our large two samples, we checked for heterogenous treatment effects due to compliance by testing our models on those who followed our instructions to think of a generic case and those who admitted to thinking of specific cases. The results were largely similar, and the direction of the treatments' effects were consistent across the four subsamples. Including a model with interaction terms for the two treatments and a variable for specific cases also showed that there were no interaction effects between the two treatments and specific cases at the 0.05 level. The appendix includes Table S09 listing the ten countries most often listed by subjects who admitted they were thinking of a specific country, rather than a generic one as instructed. 15 Note that scholars have criticized the use of regression and logistic regression to analyze experimental data and proposed adjustments and alternatives for examining predicted values. We separately calculated Freedman's 2008 plug-in estimator and obtained very similar results. The predicted probabilities generated reveal how these citizens react to the target country's regime type and IO endorsement. Again, the rallying effect of UN approval is clear whereas the pacific effect of democracy is weaker. Given UN approval, the predicted probability of the Brazilian respondent advocating the use of force against a non-democratic nuclear proliferator is 0.43 and against a democratic proliferator is 0.37. Without UN approval, however, the probability of supporting war decreases to 0.28 when the target is a non-democratic regime and further to 0.21 when the target is a democracy. Similarly, the predicted probability that a Chinese subject in our first sample backs military action sanctioned by the UN is 0.57 against a non-democratic target and 0.53 against a democratic target. Without authorization by the United Nations, the likelihood that Chinese respondents support an attack is 0.47 if the target regime is non-democratic and 0.43 if the target is said to be a democracy. In our follow-up study in China, the predicted probability that a Chinese subject supports an UN approved attack is about 0.58 regardless of the target's regime type. Without UN approval, the predicted probability is 0.45 against a non-democratic target and 0.39 against a democracy. Of course, the average respondent does not represent the average citizen. For this reason, we also generated predicted values using population medians and percentages for age group, education and gender (not shown) (Minnesota Population Center, 2017 16 ). The comparative patterns were identical as were the hypothesis tests. The only difference was that in every cell the predicted probability of support for using force was about 0.05 higher for the population mean than for the sample mean. This was true for both Brazil and China, reflecting greater support for the use of force among poorer and less educated subjects.
As Our study does not provide the power needed to explore all these differences, but several are intriguing and worth mentioning. The lowest education cohort actually had a positive treatment effect -they were more likely to support attacking a democratic Country B. The impact of democracy was very large for subjects with low nationalism, but again, the sample size in this category is very small.
We also estimated the treatment effects using randomization inference (RIGDON and HUDGENS, 2015)   Note: Overall N is the sum of complete and partially complete entries. Our survey included a feature to prevent respondents from taking the survey multiple times by placing a cookie on their browser. Some respondents had the same IP addresses, presumably sharing the device on which they took the survey, as in the case of members of the same household participating in the survey. We wanted to allow this possibility and drop only those respondents who got the same treatments and produced same responses repeatedly from one IP address.
For Sample 02, YouGov created a sampling frame representative of Internet Users in China based on gender, age, educational attainment, and income using the annual report by the China Internet Network Information Center (2014). YouGov then recruited a total of 2,723 respondents and matched those respondents to a sampling target of 2,500 based on gender, age, and income.    Total N of Compliers 3,131 4,136 1,851 Source: Prepared by the authors with their own data. Note: The table lists the ten countries mostly frequently listed by non-complying respondents who admit-ted they were thinking of a specific country, rather than a generic one, defying our instruction to think of the latter in the beginning. These non-compliers were asked to list the specific countries in at the end of the survey. We did not explicitly state a limit to the number of countries they can list and respondents were free to list as many countries as they can fit into the blank. Here we show only the first country they listed in their open-ended response, analyzing only one observation per respondent.
(2020) 14 (1) e0002 -37/38 Source: Prepared by the authors with their own data. Note: The table displays the percentage of respondents who supported military action against a non-democratic target and the effect of democracy, controlling for attitudinal and demographic variables. The difference in the percentages is estimated as the effect of democracy. Asterisks (*** p<0.01, ** p<0.05, *p<0.1) show the statistical significance of the effect.