The microfoundations of normative democratic peace theory. Experiments in the US, Russia and China

Democratic peace theory is built on the assumption that liberal- democracy has a pacifying e ﬀ ect on people, a socialization process that is assumed to lack within autocracies. This paper uses an experimental approach to investigate the microfoundations of democratic peace theory among decision-makers of the US, Russia and China. It builds on and extents previous experimental studies by conceptualizing and measuring the presence and in ﬂ uence of liberal norms, by controlling for the perception of threat as induced by the con ﬂ ict, and by testing the in ﬂ uence of hawkishness. The results show that the microfoundations of democratic peace theories do not ﬁ nd support. Neither regime-type, nor liberal norms are of in ﬂ uence on the willingness to attack the opponent, and also the assumed di ﬀ erence in liberal norms between individuals of di ﬀ erent regime types is unsupported. Moreover, hawkish decision-makers are more likely to go to war. The results show that democratic peace theory, which aims to explain why democracies do not ﬁ ght with each other, cannot be used as has been done till today and should be revised. The paper concludes with suggestions for new research avenues. A level of liberal will make it less likely for decision-makers to take military action against a state with a liberal political culture; however, a lower level of liberal norms will make it more likely for decision-makers to take military action against any opposing state, regardless of their political culture.


Introduction
Why do leaders of democracies don't fight with other democracies while they do fight with non-democracies? A substantial body of literature within the field of international relations (IR) argues that liberal norms are the cause of the so-called 'democratic peace', an empirical pattern that indicates an absence of war between democracies. In attempt to create a democratic peace theory, they posit the following mechanism: Democracies supposedly socialize their leaders and citizens with liberal norms. These liberal norms are expected to decrease war-proneness among these individuals, in particular towards societies that share these norms: other democracies (Doyle 1997, 282;Maoz and Russett 1993, 625). Autocracies, on the other hand, supposedly socialize their leaders and citizens with other, more violent and zero-sum, political norms. These norms are expected to increase the tendency of war-proneness among these individuals (Maoz and Russett 1993, 625). The socialization process purportedly nurtures peace between democracies, and thus moreover, to include multiple explanatory factors into our experimental designs, such as hawkish beliefs, and the perception of threat.
In this paper this call is answered. It shows that the current state of the art on democratic peace theory is insufficient to claim that liberal norms cause peace between democracies. My argument builds my previous study (Bakker 2017, 524-527) in which I argue, based on the state of the art, that if we want to know if liberal norms are indeed present in democracies and absent in autocracies, we need to conceptualize and measure them, and study their influence on decision-making processes in both democracies and other regimes types. This paper extends the earlier studies, and in particular my previous study (Bakker 2017). It (1) Offers a stronger conceptualization and measurement of liberal norms, and (2) Uses an experiment on American, Russian and Chinese samples to test the microfoundations of democratic peace theory. These states have different regimetypes but are fairly similar in respect to position on the world stage. The paper then (3) compares the results of the student samples with samples representative of their respective populations, to seek external validity. And lastly (4), following Bakker (2017), the paper also improves the experimental design of previous studies by controlling for the perception of threat, and other explanatory individually based factors that might influence decision-making as posited by other political psychologists (Kertzer and Brutger 2016;Rathbun et al. 2016;Yarhi-Milo, Kertzer, and Renshon 2016).
The results of this paper support my earlier findings (Bakker 2017, 539) about liberal norms and regime-types, and moreover offer new insights into the microfoundations of democratic peace theory. A comparison between representative and student samples of the US, Russia and China show that liberal norms are universal values rather than norms imbued by the super structure of a regime-type. Moreover, nor regime-type, nor liberal norms show to be of influence on the decision to attack the enemy. The evidence of the experiments shows that hawkish beliefs rather than structural factors determine whether leaders choose to go to war. This paper makes the case that structural theories about the causes of war and peace, such as the democratic peace theory, need to be revised. The democratic peace thesis might be correct in its description of an empirical regularity, but the dominant explanation of why democracies do not fight each other is built on empirically shaky foundations.

Liberal norms, the microfoundations of democratic peace theory
The so-called 'democratic peace' is an empirical pattern that indicates an absence of war between democracies (Babst 1964;Rummel 1983). Although there is a general consensus about this pattern, we are still far from a consensus about any explanation (Hayes 2012;Ungerer 2012). Democratic peace theory tries to explain this phenomenon and has become one of the most persistent theories in the field of IR to explain interstate conflict resolution, and is even sometimes called 'an iron law' (Levy 1988, 662). Moreover, democratic peace theory underlies many decisions made in foreign policy (Ish-Shalom 2015), even those that have led democracies to go to war in order to bring peace (Avtalyon-Bakker 2013;Burgos 2008).
The normative explanation of democratic peace theory, the explanation this paper takes issue with, is generally seen as one of the better explanations (Rousseau 2005, 208). It argues that people in liberal-democracies are more peace prone and thus morally superior to people in autocratic regimes. The core argument of this explanation is that liberal democracies have a practice of liberal norms, which leads to trust and compromise within these regimes. Because liberal democracies share these norms with other liberal democracies, the theory also assumes that trust and compromise exist between liberal democracies (Danilovic and Clare 2007;Dixon 1993Dixon , 1994Dixon and Senese 2002, 549;Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Jakobsen, Jakobsen, and Ekevold 2016;Johns and Davies 2012;Kahl 1998;Maoz and Russett 1993, 625;Mintz and Geva 1993;Mousseau 1997;Owen 1994;Rawls 1999;Ray 1995;Risse-Kappen 1995;Rousseau 2005, 27-28;Rummel 1983;Tomz and Weeks 2013;Van Belle 1997;Weart 1998, 75-93). The practices within all other regime-types (by democratic peace scholars referred to as non-democracies) are intrinsically more violent, due to the lack of these liberal norms, so the theory assumes (Maoz and Russett 1993, 625). Thus, liberal democracies are 'forced' to fight with non-democracies because of the lack of liberal norms of the latter. Liberal democracies will, therefore, have to adapt to the more violent norms of the non-democratic states (Kahl 1998, 125-129;Maoz and Russett 1993, 625;Rousseau 2005, 27-28;Russett 1993, 32-33).
What are exactly the microfoundations of democratic peace theory? Most proponents of this explanation refer for theoretical justification to the work by Enlightenment philosopher Immanuel Kant, who is often asserted as the founding father of democratic peace theory (Cederman 2001;Danilovic and Clare 2007;Dorussen and Ward 2010;Doyle 1983aDoyle , 1983bDoyle , 1986Doyle , 1997Doyle , 2005Maoz and Russett 1993;Molloy 2017;Oneal et al. 1996;Ward, Siverson, and Cao 2007). The most cited 1 proponents of democratic peace theory, Maoz and Russett (1993), mention Kant explicitly but rather than refering directly to his work, they cite Doyle (Maoz and Russett 1993, 625), who has translated Kant's philosophy for the field of IR (Doyle 1983a(Doyle , 1983b(Doyle , 1986(Doyle , 1997. Although Doyle offers an insightful translation, he does not specify the concept of liberal norms, nor does he specify a specific mechanism, other than suggesting that liberal democracies will have to assume that non-liberal states are suspect because of their aggressive behaviour to their own citizens (Doyle 1997, 282).
The bottom line is that the expectation that liberal norms are socialized within liberal democracies belongs to the Kantian belief system of liberalism. It is what contemporary Kantian philosopher John Rawls calls 'moral learning': if individuals live long enough under the rules of a self-established just constitutional democracy, they will internalize these rules as intrinsic norms (Rawls 1999, 12-14). A process that is assumed to be intrinsically different within non-democratic settings. It is this assumed difference based on Kantian logic that researchers generally take as a fact to build their theory on. This assumed difference functions as the main building block to explain the empirical regularity of an absence of war between democracies, without empirically testing if these assumptions are empirical facts.
It has been already effectively argued that relying on these assumptions to explain the democratic peace is tautological and should rather be subjected to empirical research (Bakker 2017, 523). That particular study measured, for the first time ever, the actual levels of liberal norms and their subsequent influence on the willingness to attack democratic and autocratic participants. This study found, on average, only a slight difference in liberal norms between Dutch and Chinese participants. Moreover, a multivariate analysis revealed that these norms did not influence the willingness to attack, which was for all participants alike. The empirical tests show that liberal norms cannot simply be assumed absent within non-democratic settings, thereby connecting with the insights from other fields than IR that norms, values and attitudes vary significantly within and between societies (see e.g. Almond and Verba 1963, 22-33;Chilton 1988;Inglehart 1988Inglehart , 2003Inglehart and Welzel 2003;Jackman and Miller 1996;Klicperová-Baker and Košťál 2015;Putnam 1993;Pye 1972;Seligson 2002;Sheafer and Shenhav 2013;Widmaier 2005; Zagrebina 2020).

Bringing Kant back in: conceptualizing liberal norms
The very first measure for liberal norms (Bakker 2017) was a good start, but this paper takes it a step further by offering a clear conceptualization. Since the logic of earlier studies refers repeatedly to Kant (e.g. Cederman 2001;Danilovic and Clare 2007;Doyle 1983aDoyle , 1983bDoyle , 1986Maoz and Russett 1993, 625;Owen 1994), we need to bring Kant's ([1789] 2013) understanding of liberal norms back in. He argues that individuals, enabled by their self-established republic to be free and autonomous, will not only be able to pursue their interests rationally but also learn to listen to the 'the moral law inside': the categorical imperative. Meaning: people would act towards others as they would like them to reciprocate. This can only be achieved when people tolerate others unconditionally and trust them to reciprocate tolerance (see also Wood 1999, 284-285, 295-296). Kant realizes that the freedom to follow one's internal moral law is paradoxical; there is no guarantee that people indeed choose to do so (Kant [1795Wood 1999, 283). Kant, therefore, sees a solution in the formal institutions of the republic that guarantee the rights of individual freedom for everybody in the republic. Subsequently, he expected these institutions to cause a socialization process with the practices of tolerance, trust, and reciprocity. Kant sees it as a process of 'moral learning' that would enable people to become 'reasonable people' (Kant [1795Rawls 1999, 44;Wood 1999, 295-296).
Kant's premises for the perpetual peace he envisaged are built on the rational individual (Wood 1999, 302-303). As argued above, Kant assumes that individuals can only be rational if they are free from oppression by others. That freedom could be realized, in his vision, by the bottom-up creation of republican institutions by all individuals within a society (Kant [1795Wood 1999, 319-320). When that freedom is secured, and individuals could accordingly be rational, a socialization process would initiate and create rationality based on reasonability among the members of the society (Wood 1999, 171-172, 187-190). In other words: the process would create liberal norms. These liberal norms would consist of a feeling of freedom from the state, a feeling of autonomy over life, trust in others, tolerance towards others to act as they like, and the willingness to reciprocate all these norms. Therefore, liberal norms in this study are conceptualized as six dimensions: rationality, freedom, autonomy, tolerance, interpersonal trust, and reciprocity. 2 This paper examines whether Kantian liberal norms are present in democracies and absent in autocracies. It also tests the subsequent influence of these norms on the decision-making during a severe interstate conflict: H1: Decision-makers, born and raised in a consolidated liberal-democracy, have internalized liberal norms in contrast to decision-makers who are not born and raised in a liberaldemocracy.
H2: A higher level of liberal norms will make it less likely for decision-makers to take military action against a state with a liberal political culture; however, a lower level of liberal norms will make it more likely for decision-makers to take military action against any opposing state, regardless of their political culture.

Regime-type
Bakker's (2017) study pinpoints two other elements for which previous studies have not controlled: (1) the conceptualization and operationalization of regime-type, and (2) the conceptualization and operationalization of the conflict and the threat it poses. The two elements are related, because how can we make sure regime-type is separated from the threat perception of the conflict? Democratic peace theory makes a rather 'black and white' distinction between democracy and non-democracy. With democracy they denote a liberal society that enables its citizens to be free and autonomous, in other words: a full-fledged liberal-democracy in which, next to democratic institutions, universal human rights and civic liberties are ensured (Dahl 1971(Dahl , 2000Merkel 2004, 38-42;Møller and Skaaning 2010, 263;Sartori 1987). In contrast, what democratic peace theorists call non-democracy essentially comprises every regime that is not liberal-democratic.
But words such as (non)democracy, autocracy, dictator are multi-interpretable and might have a strong connotation. Political scientists can choose from many different conceptualizations of democracy (Schedler and Sarsfield 2007), how can participants, especially from different cultures, know what we mean with these words? We know, for instance, that Chinese citizens perceive the word democracy completely different than US citizens (Smith 2003). Earlier experimental studies have been quite explicit in the measure for regime-type; they called the enemy a democracy or an autocracy (Tomz and Weeks 2013), or they mentioned an elected president/government versus a (military) dictator (Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005). It is plausible to suspect that negative connotations might have influenced the perception of threat, especially within the non-democratic treatments. The willingness of democratic participants to attack non-democracies, as found in these studies (Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005; Tomz and Weeks 2013) might (partly) have been the result of their description of non-democracy.
Bakker (2017) therefore described the practices of a liberal-democracy and an autocracy, respectively, without mentioning the regime-types explicitly to avoid a possible threat response to the 'bad' or 'good' connotation of an explicit regime-type. This paper follows Bakker (2017, 529) and understands the concept of regime-type as a spectrum with liberal-democracy on the one side and autocracy on the other. The conceptualization of the regime-types, based on Dahl (1971Dahl ( , 2000 lies in the description of the institutions and practices, rather than the actual regime-types. The theoretical expectations for the influence of regime-type on the decision-making process follows the earlier studies: H3: If at least one of the states in an interstate conflict does not rely on democratic practices, its decision-makers will be more likely to take military action against the other state, but if both of the states rely on democratic practices, decision-makers will be less likely to take military action against the other state.

Threat of the conflict
Bakker's findings raise another question. The results show, post-hoc, that the perception of threat of the conflict was an important determinant for participants' willingness to attack the enemy (2017,538). But the question is: what induced that threat? Bakker, and also the other experimental studies, depicted the other state within the conflict as an aggressor who is either building a nuclear weapon or invading the other country (Bakker 2017;Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013). If the core of the conflict is basically about the opponent invading sovereign territory or threatening to use nuclear weapons, it is quite hard to distinguish between the threat of the issue and the threat of the behaviour of the opponent, let alone other factors such as regime-type. When the other state is already perceived as a 'bad guy' it does no longer matter whether that state is democratic or not.
Ideally, to make sure that a test measures the effect of the treatments and not just the treat of the conflict in itself, a distinction has to be made between the issue of a conflict and the actions the enemy takes in relation to that issue. Thus, an interstate conflict consists: (1) the core issue of conflict between the states, and (2) the behaviour of the opposing state during the conflict. The issue of the conflict needs to be as 'morally neutral' as possible, which means: a conflict in with both of the states seem to have a legitimate claim. For instance, a territorial dispute in the classic senseone state comes and takes control over a piece of territorycould create, based on the principles of just war, a 'bad state-good state' atmosphere. That said, the core issue needs to be sufficiently threatening. If the issue in itself is non-threatening, then possible moderating factors will not matter much. Therefore, the starting point must be an issue that is threatening enough to lead to war, but morally neutral enough to distinguish between the actual conflict and the behaviour of the opponent. Accordingly, the selected core issue is a dispute over resources. Indivisible goods are often considered as a cause for war (Fearon 1995, 382) if one of the parties feels that they need full control over the resources to be safe. It follows that an issue over an indivisible good is sufficiently threatening in itself, but is not necessarily connected to the behaviour of the other state.
The behaviour of the opponent to settle the issue is supposed to moderate the severity of the threat of the conflict. The theoretical distinction between the issue of the conflict allows for studying the impact the behaviour of an opposing state can have on an ongoing conflict separate from the issue of that same conflict. Thus, the behaviour of the other state is divided into two behavioural actions: (1) invasiveness (hereafter: invasion) and (2) the use of power. The choice for invasion stems from the earlier experiments which do not control for the invasiveness of the opponent, but equate an invasion (or other aggressive behaviour) as the severe conflict. To consider an invasion as an instrument of the opponent (and thus a separate element of the conflict) rather than an intrinsic part of the conflict will provide more insight into the threat perception of decision-makers, in particular in relation to other explanatory factors. The same logic underscores the choice to conceptualize the use of power as a separate part of the conflict. Invasion is conceptualized as the willingness of states to invade properties (territory or resources) of the other state with the purpose of annexing it. The use of power is conceptualized through a continuum ranging from the use of soft power (e.g. using diplomacy) to the use of hard power (aggressive: e.g. using the military) by the opponent (Nye 1990).
This leads to the following hypotheses: H4: During a severe interstate conflict, decision-makers will be more likely to take military action towards the opposing state that invades their territory over the opposing state that does not invade their territory.
H5: During a severe interstate conflict, decision-makers will be more likely to take military action towards the opposing state that uses hard power over the opposing state that uses soft power.

Controlling for individual factors
Several control factors are included in the design. Following Bakker (2017), hawkishness and gender are included as control factors. That study measured the level of hawkishness and tested whether hawkish beliefs influences individuals within democracies and autocracies in their willingness to attack. Hawkishness was found to be an important explanatory factor for both democratic and autocratic participants. The more hawkish, the more likely they are to attack. Testing for hawkishness might seem close to being tautological, however, theories of IR are structure-specific and assume that agents are affected by structures not their personalities. Following the logic of democratic peace theory, it would not matter whether decision-makers are hawkish or dovish, their behaviour would not be guided by those beliefs. Accordingly, there is good cause to consider hawkishness within this research design: H6: During a severe interstate conflict, hawkish decision-makers are more likely to take military action than dovish decision-makers.
Hawkishness is conceptualized as a continuum that ranges between hawks on one pole and doves on the other. A more hawkish individual is considered likely to use force to resolve conflicts, whereas a more dovish individual is considered to be less likely to use force in a similar case, no matter the context ( Bar-Tal, Raviv, and Freund 1994;Braumoeller 1997;D'Agostino 1995;Kahneman and Renshon 2007;Klugman 1985;Liebes 1992;Maoz 2003;Rathbun et al. 2016;Schultz 2005).

Willingness to attack
Within the democratic peace literature, the dependent variable is war. There is an apparent problem with testing a possible explanation for the democratic peace: war and peace are ex-post assessments and do not inform us in any way about how these outcomes are created. In other words, to empirically test whether regime-type (or any other factor) influences the behaviour of decision-makers in such a way that a war is prevented or not, we have to study that mechanism before a war has even started. Explanations of the democratic peace rely on an ex-ante predicted mechanism: it is argued wars are avoided because individuals within liberal democracies respond differently from those in other regimes to the threats of opposing countries. The focus of research lies thus on the moment before a war: what influences decision-makers to attack, thereby assuming that the willingness to attack has the potential to escalate into a full-blown war. In this study, the willingness to attack is considered as the first step towards war, thereby assuming that decision-makers are well aware of the likelihood that their action can lead to a fullblown war.

Method
To test the microfoundations of democratic peace theory, an experiment was designed in which participants were exposed to a hypothetical severe interstate conflict between two fictitious neighbouring states in which participants had to decide how willing they were to attack the opposing state. The experiment consists of a written scenario in which the factors (1) regime-type of the opponent, (2) the invasiveness of the opponent's behaviour, and (3) the use of power by the opponent are operationalized as randomized experimental treatments. A questionnaire follows the experiment to measure the dependent variable (willingness to attack), the factors liberal norms, hawkishness, gender, and treatment checks. The choice for an experiment follows earlier studies on the micro-level mechanisms of the democratic peace (Bakker 2017;Bell and Quek 2018;Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013). Moreover, an experimental approach is well suited to study the microlevel and also offers the best and transparent control over independent variables in such a way that they can be considered within the analytical design (Druckman and Kam 2011, 44).

Case selection and student samples
The experiments were conducted in 2014 and 2015 on student samples from three different regime-types: United States of America (US) as a liberal-democracy, the Russian Federation (Russia) as a hybrid regime, and the People's Republic of China (China) as an autocracy. The selection of these countries is based on two parameters. First of all, the aim is to study individuals who are socialized within a particular regimetype. The US is one of the oldest full-fledged, bottom-up liberal democracies, and it furthermore occupies a central position in most democratic peace studies. China, with its one-party system and low adherence to human rights, has been a clear example of an autocratic regime for many years. Since Russia blossomed out of the former Soviet Union, it has been travelling from the autocratic regime it once was toward democracy and back and forth, thereby making it an example of a hybrid regime. That indication of the regime-type for these countries is supported by the results of Freedom House 2014 3 and Polity IV 2014. 4 Secondly, since the US, Russia, and China are all powerful players on the world stage, all three countries have massive populations, territories and standing armies with impressive defense budgets. All three are involved in conflict zones, and each has a seat on the UN Security Council. It can be expected that individuals from these countries will not worry about their power position in the world.
This study uses student samples for theoretical and practical reasons. The value of experiments for theoretical clarification lies in teaching us about theory rather than the replication of real life (Druckman and Kam 2011, 44). Druckman and Kam effectively show that the use of students does not 'intrinsically pose a problem for a study's external validity ' (2011, 41) unless the 'size of an experimental treatment effect depends on a characteristic on which the convenience sample has virtually no variance ' (2011, 41). They support their argument with convincing results (Druckman and Kam 2011, 45-52) and contend that when an experiment aims at studying the mechanisms of a theory, and thereby adds to an existing research agenda as most experiments do, the generalizability of the participants should be weighed against the generalizability of many other factors involved, such as setting, timing, context, conceptual operationalizations (Druckman and Kam 2011, 44, 53). A few studies also show that when an experiment is replicated on a non-student sample , or on political leaders (Yarhi-Milo, Kertzer, and Renshon 2016), the results are similar.
Also, like earlier experimental studies have argued, student samples offer the best and most convenient opportunity to come to comparable results (Bakker 2017;Geva and Hanson 1999;Mintz and Geva 1993;Rousseau 2005). It would be impossible to reach large enough samples of decision-makers in all three countries, as decision-makers are too busy making real-world decisions. Students are accessible, in particular within autocratic regimes. And, in particular within Russia and China, students can be considered to be the new elite. Political elites are generally university-educated. The student sample used in Russia, for example, comprises students from the Higher School of Economics, which showed to produce the new political elite of Russia (Mickiewicz 2014). However, to investigate how the student samples relate to representative samples of their respective countries, a separate test is conducted on the latter samples, by using the 6th wave of the World Values Survey (2010-2014) to measure the levels of liberal norms (see further below).
All three experiments used the same research design and identical research material with the only difference that the scenario was offered in their native language. 5

Research instrument 6
The experiment was a 2 × 2 × 2 factorial between-groups design. All participants read the same fictitious story about an ongoing interstate conflict between two hypothetical states: My Country and Other Country, a conflict that is now on the brink of war. To avoid the possible influence of geopolitical and/or economic factors, these are held constant by describing these as similar for both countries (size, population, military and economic power, GPD, no economic interdependence or (regional) trade agreements). All participants are asked to imagine living in My Country and that they are advisors of their government about this conflict today. They all read the same text about the history of the conflict. Figure 1 shows the map of the experimental world that was included in the scenario.

Threat of conflict
The issue of the conflict is described as neutral as possible, although threatening in itself. An ongoing conflict about a seemingly equal right to uranium fields 7 is used, since the possession of uranium and its importance to the world should have a realistic threat level. The scenario describes a contested area of uranium fields that both states seem to have rights to, to indicate that none of the states is a 'bad guy', to avoid contamination with the treatment effects. To control for the behaviour by Other Country, two treatments are used: (1) invasion, and (2) use of power. To measure invasion, half of the participants receives a description that remains close to the initial 'neutral sense' of the conflict (as in both states seem right nor wrong), the other description depicts Other Country as invasive. To measure the use of power, half of the participants receives a description of Other Country using soft power, the other half receives a described hard use of power.

Regime-type
To control for the regime-type of the Other Country, to half of the participants Other Country was described as a liberal-democracy, to the other half it was described as an autocracy. For both regime-types, the practices of the regime were described rather than explicitly being named ad democracy or autocracy, resp. The purpose is, as discussed above, to avoid conceptual misperception and to make sure that the core understandingof what regime-type according to democratic peace theory entails-, is measured as accurately as possible. The core concept of Dahl (1971) is used to describe the practices, and both are described as neutral as possible, without any negative connotation or judgment. To control for the assumed socialization by a specific regime-type, the experiment was executed among student samples within three different regime-types 8 : a liberal-democracy (the US 9 ), a hybrid regime (Russia 10 ), and an autocracy (China 11 ). The participants are asked to imagine to have lived their whole lives in the hypothetical My Country without getting specific information about the regime-type of My Country. The aim is that they will implicitly act from the hinterland of their own country. This variable is called 'socialization'.

Liberal norms, hawkishness & willingness to attack
The experiment is followed by a questionnaire that measures, next to the treatment checks, the levels of liberal norms, hawkishness, gender and the willingness to attack, the latter being the dependent variable. The questionnaire measures liberal norms, hawkishness and the willingness to attack by asking participants to which degree they agree with several statements on a 7-point rating scale that ranges from very strongly disapprove (1) to very strongly approve (7).
The variable 'liberal norms' is measured using five items, which measure freedom, autonomy, tolerance, interpersonal trust, and reciprocity. To recap briefly, 12 the operationalization of these five dimensions roots in statements of existing measures of these concepts (Measures of Political Attitudes 1999; World Value Survey 2010-2014), Table 1 provides an overview of the single items used to measure freedom, autonomy, tolerance, interpersonal trust, and reciprocity. The sixth dimension, bounded rationality, is assumed. Participants indicate on a 7-point rating scale the degree to which they agree with the statements. The average of the sum of these five items constitutes the variable 'liberal norms' that is used in the analyses below. To get some insight how the levels of liberal norms among student samples relate to representative samples of their respective countries, the 6th wave (2010-2014) of the World Values Survey is used to operationalize the dimensions freedom, autonomy, tolerance, and trust for representative samples of the US (N = 2232), Russia (N = 2500), and China (N = 2300). 13  States are generally not trustworthy: they will attempt to expand their territory if they have the chance.
.479 In general, international organizations are ineffective because they lack the power necessary to change the behaviour of powerful states. .447 The use or threat of nuclear weapons is a necessary instrument for states in order to survive as a state.
.366 It is important to teach children to defend themselves physically if necessary.
.450 Everybody thinks of themselves first, so I will have to protect myself and my family before I consider others. .440 The worst way for us to keep peace is by trying to work out agreements at the bargaining table rather than by having a very strong military so other countries won't attack us. (scale reversed) .469 α = .489 The variable 'hawkishness' is measured through the six items related to the beliefs about conflict resolution, as shown in Table 2. The items rely on an existing measure of beliefs about the nature of conflict resolution that range from support for cooperative behaviour to offensive behaviour to solve interstate conflicts (Rousseau and Garcia-Retamero 2007), and on measures that relate to a more personal belief about how conflict should be resolved that range from using physical force or not to protect oneself (World Value Survey 2010-2014). The average of the sum of these six items constitutes the variable 'hawkishness' that is used in the analyses.
The dependent variable, the willingness to attack is measured by asking the participants' level of approval for an attack on the opponent, aslo measured on a 7-point rating scale.

Procedure
The experiments were conducted in regular university classrooms, after lectures. Students were asked to participate on voluntary and anonymous basis. Anyone who did not want to participate was free to leave. Every participant was randomly assigned to one of the 8 experimental groups (combinations of 2 × 2 × 2 treatments). The scenario and following questionnaire were presented as a paper-and-pencil experiment and introduced as a survey about conflict resolution. After completing the survey-experiment, a debriefing followed about the real purpose of the experiment.

Participants
After checking whether participants were born and raised in their respective countries, and all that did not were excluded. In total, 745 participants (US 226 participants, Russia 242 participants, China 277 participants) remained in the analysis, 39% of the total sample was male, 61% was female.

Treatment checks
The treatment checks show that all treatments were perceived accurately by the participants. 14 Participants in the autocratic regime-type treatment find the target country on a 7-point scale significantly (t(770) = −24.49, p < .001) less democratic (M = 1.97, SD = 1.24) than participants in the democratic treatment (M = 4.48, SD = 1.59). 15 Therefore, the measure of the treatment regime-type is included as a binary variable called Regime. Participants in the non-invasion treatment find the actions of target state on a 7-point scale significantly (t(768) = −7.861, p < .005) less violating (M = 4.54, SD = 1.63) than the participants in the invasion treatment (M = 5.43, SD = 1.48). 16 The measure of the treatment invasion is included in the analysis as a binary variable called Invasion. Participants in the hard power treatment find the target country on a 7-point scale significantly (t(771) = −7.73, p < .001) more likely to attack (M = 4.96, SD = 1.55) than the participants in the soft power treatment (M = 4.16, SD = 1.35). 17 Hence, the measure of the treatment Use of Power is considered in the analysis as a binary variable called Use of power.

Results
This study uses an analysis of variance (ANOVA) and an analysis of covariance (ANCOVA) to study the influence of the factors regime-type, invasive behaviour and the use of power of the opponent, the influence of liberal norms in interaction with regime-type and socialization, liberal norms as an individual-based factor, the influence of hawkishness, and the influence of gender.

Presence and absence of liberal norms
Can we find any support for the Kantian assumption that people within democracies have internalized liberal norms in contrast to people who have been raised in other regimetypes? Firstly, Figure 2 shows that liberal norms are present in all three countries. All samples, student and representative, score on average well above the midpoint of their scales. 18 Furthermore, liberal norms are not embraced as one homogeneous mass: liberal norms are varying strongly within each sample; participants range between both poles of the scales. For the student samples, the average levels of China (M = 5.00, SD = .83) and Russia (M = 4.78, SD = .73) differ slightly (but significantly) from the US (M = 5.14, SD = .71). There is a significant difference between the average scores, but only between the US sample and China sample on the one hand and Russian sample on the other hand (F (2, 734) = 13.2, p ≤ .001, r = .18), with a small effect. For the representative samples, the US sample scores on average the highest on liberal norms. With the midpoint of 4.5 in mind, however, the average levels of China (M = 5.27, SD = 1.01) and Russia (M = 5.00, SD = 1.13) do not differ that much from the US (M = 6.16, SD = 1.10) score. The differences in the means are significant (F(2, 6006.15) = 676.55, p < .001, r = .43) with a medium effect.
A comparison between the liberal norms of the representative and the student samples in Figure 2 suggests similar patterns. A pattern that shows that liberal norms are present in all samples, and varying in a similar vein over the populations, student sample and representative samples alike. Although it would lead too far to extrapolate the finding of these similar patterns to indicate that the students might be similar in all aspects of this study, these patterns do give some indication that the results of these student samples have explanatory power that might stretch beyond their own samples.
To conlude, liberal norms indeed exist within liberal-democracy, such as the US. Also, the level of liberal norms in the US is, on average, significantly higher than in other regime-types. However, also within other non-democratic regime-types, such as the hybrid regime in Russia and the autocratic regime in China, liberal norms exist. There are differences, on average, but the averages of all three samples show to have positive scores on liberal norms. It can be concluded that liberal norms also prevail within nondemocratic regimes. We can therefore concluse that the Kantian assumption that liberal norms are absent within non-democracies does not find support, thereby rejecting hypothesis 1. The remaining question is whether the liberal norms are of influencd on the willingness to attack, and whether or not that differs cross-regime.

What influences the decision to attack the opponent?
Democratic peace theory assumes a significant difference between decision-makers of liberal democracies and decision-makers of other regime-types in their willingness to attack liberal democracies. To find support for this assumption, the US decision-makers would have to respond differently towards a democracy over an autocracy, in contrast to the Russian and Chinese decision-makers who would not distinguish between a democratic or autocratic opponent.
Regime-type Figure 3 shows that regime-type is not of influence on the decision to attack. This result is alike cross-regime 19 : nor the US decision-makers, neither the Russian and Chinese decision-makers differentiate between a democratic of an autocratic opponent. The expectation that decision-makers of liberal democracies respond differently to different regime-types than decision-makers of non-democratic regimes does not find empirical support. Although the regime-type of the opponent does not significantly influence the decision-makers of all three countries, there is a different noteworthy result regarding the regime-type of the decision-makers (the factor socialization). Figure 3 also shows that the US decision-makers (M = 3.69, SD = 1.78) are on average significantly more willing to attack any opponent (F(2, 742) = 29.9, p < .001, r = .28) than the Russian (M = 2.67, SD = 1.50) or the Chinese decision-makers (M = 2.70, SD = 1.53). Democratic peace theory would indeed expect such results for the Russian and Chinese decision-makers, however, not for the US decision-makers. The US decision-makers would be expected to be less war-prone, at least towards other democracies, but would not be expected to be more war-prone overall. Table 3 shows the results of a multivariate analysis. Model 1, which only probes the influence of the treatments without the other hypothesized factors, shows this socialization effect. The explanatory power of the factor socialization, however, decreases strongly in model 2, which includes all hypothesized factors. This finding indicates that one of the newly introduced factors in model 2 might be responsible, something that will be explored below. For now, the conclusion is that regime-type does not seem to be of influence on the socialization effect, hypothesis 3 does not find support.

Liberal norms
Democratic peace theory also assumes that liberal norms are of influence on the willingness to attack. Table 3 shows that the interaction between liberal norms, socialization and regime-type has no significant influence on the willingness to go to war (Fmodel2 = .37, η 2 715 † = p < .1, * = p < .05, ** = p ≤ .01, *** = p < .005, **** = p < .001. = .00). Also, interactions of liberal norms with regime-type only (Fmodel2 = .37, η 2 = .00) or socialization only (Fmodel2 = .54, η 2 = .00) are not of significant influence. Hypothesis 2 does not find support. Individual liberal norms do not influence the willingness to take military action. Therefore, the conclusion based on these results is that the core assumptions of the democratic peace theory do not find support.

Behaviour of opponent
An alternative explanation for the willingness to attack another state during a severe interstate conflict is the behaviour of the opponent. This behaviour is operationalized in two other treatments of the experiment: whether or not the opponent invades territory of the opponent, and the opponent's use of power. The perceived difference in invasive behaviour of Other Country does not show to have a significant relationship with their willingness to go to war. Whether or not the opponent invaded the disputed area does not have a significant influence on the willingness to attack (Fmodel1 = 2.8, η 2 = .00; Fmodel2 = 1.58, η 2 = .00). These results are alike for all three samples; cross-regime, the invasion of the opponent did not show a significant relationship with the willingness to attack. Also, when an invasion is considered to interact with other treatments such as regime-type, and use of power, there is no significant influence. Thereby hypothesis 4 does not find support. The treatment 'use of power' shows to have a significant relationship with the willingness to attack. Figure 4 shows a significant difference (t(741,14) = −3.07, p < .01) in the willingness to attack between the different treatments of the use of power: decision-makers that experience the use of hard power (M = 3.20, SD = 1.73) by the opponent show to be significantly more willing to attack than decision-makers that experience the use of soft power (M = 2.82, SD = 1.64). This influence remains within a multivariate analysis: that the use of hard power increases the willingness to go to war significantly (Fmodel1 = 8.38, p < .01, η 2 = .01; Fmodel2 = 6.75, p < .01, η 2 = .01). Hypothesis 5: During a severe interstate conflict, decision-makers will be more likely to take military action towards the opposing state that uses hard power over the opposing state that uses soft power, finds support.

Hawkishness
The relative explanatory power of hawkishness might shed light on the decreasing effect of socialization in model 2. Socialization, when tested in model 1 only along with the other treatments, shows to be of significant influence (Fmodel1 = 29.56, p < .001, η 2 = .08). With the introduction of the factors liberal norms and hawkishness in model 2 the effect and significance of the explanatory power of socialization decreases strongly (Fmodel1 = 29.53, p < .001, η 2 = .08; Fmodel2 = 2.53, p < .1, η 2 = .01). The best explanation for that decrease is the introduction of the individually-based explanatory factors. The factor hawkishness is the only factor that has a significant and substantial influence on the willingness to attack, as evidenced in Figure 5, which finding results in the question if the socialization effect is created by the factor hawkishness. It might be that the hawks of one specific country are more willing to attack than hawks of other countries.
To test for this inductively discovered interaction, model 3 also includes the interaction effect between socialization and hawkishness. Figure 6 shows that this interaction is indeed the explanation for the socialization effect. The results of Table 3 show that the interaction between hawkishness and socialization has a small but substantial effect (Fmodel3 = 5.26, p < .005, η 2 = .02). The factor socialization, which was a strong explanatory factor in model 1, a weak explanatory factor in model 2, looses all significant and substantial explanatory in model 3 (Fmodel1 = 29.56, p < .001, η 2 = .08; Fmodel2 = 2.53, p < .1, η 2 = .01; Fmodel3 = .28, p > .005, η 2 = .00).
Hawkishness, as an independent factor, remains its strong and substantial explanatory power, also in model 3 (Fmodel2 = 104.12, p < .001, η 2 = .13; Fmodel3 = 101.35, p < .001, η 2 = .13). Thus, hawkishness is indeed the main explanatory factor of the willingness of decision-makers to attack. Hawkishness explains for all decision-makers best the willingness to attack, and most strongly for US decision-makers. A quick inspection of the other significant factors in the model shows that there are no further substantial changes. 20

Concluding discussion
The microfoundations, the essential building blocks of democratic peace theory, which are used untested as empirical facts, do not find any support when tested along with alternative hypotheses in a comparative framework. These are important findings for democratic peace theory. Earlier experimental studies of the democratic peace (Bell and Quek 2018;Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Johns and Davies 2012;Maoz and Russett 1993;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013) have instrumentally assumed liberal norms to be present and of influence within liberal democracies. They did not measure whether these were actually present, and they did not test whether they indeed influenced as hypothesized. Although it is assumed throughout the democratic peace literature that liberal norms are of influence on the willingness to attack, in particular among decision-makers of liberal democracies, this research shows that there is no evidence to support that assumption. The assumptions that liberal-democracy does something special with its citizens that makes them 'morally more advanced' (Doyle 1983a(Doyle , 1983b(Doyle , 1986Kant [1795] 2013; Maoz and Russett 1993;Rawls 1999) and that would make them subsequently more peace prone, is empirically simply not supported. Also, regime-type showed to be of no influence on the willingness to attack, or the willingness for other relevant policy options, of decision-makers in all three samples alike. Regime-type did thus not influence decision-makers of liberal democracies significantly, as is expected by democratic peace theory. Note: The scatterplot shows the relation between the level of hawkishness and the willingness to attack. The straight line is the regression line that represents the linear relationship. The curved line is the LOESS regression estimate that gives closer observations more weight and is therefore better able to detect non-linear patterns (if existing). The shades belonging to the lines represent the 95% confidence intervals. The rug, drawn on the horizontal axis, indicates the density.
This non-finding is not in line with earlier micro-level studies. Previous studies showed that regime-type did influence the willingness to attack: individuals of liberal democracies were more willing to attack autocracies over democracies (Bakker 2017;Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013). How can we explain this difference? A comparison of research designs might explain. Firstly, this study disconnected regime-type from the perception of threat by providing the information about the regime-type separate from other factors surrounding the conflict. As the studies of Johns and Davies (2012) and Geva and Hanson (1999) showed, it is hard to pinpoint the exact effect of regime-type when socio-cultural factors are part of the mix and might interact implicitly with regime-type. By separating the behaviour of the opponent from regime-type, it could be suggested that it was not regime-type that triggered participants in earlier studies, Figure 6. Relationship between hawkishness, socialization, and attack. but rather the threat from the conflict itself. Secondly, most of the earlier studies (Geva, DeRouen, and Mintz 1993;Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013) used non-hypothetical countries and real-world conflicts in their scenarios, which might have triggered responses based on real-world perceptions, not only about regime-type but possibly also about other features of these countries. Thirdly, in this study the regime-types were indicated by a neutral description of the practices the regimes, rather than by explicitly naming these. The participants showed to have perceived the regime-types as intended, which means that they got what type of regime was meant. Most studies (Geva, DeRouen, and Mintz 1993;Geva and Hanson 1999;Johns and Davies 2012;Mintz and Geva 1993;Rousseau 2005;Tomz and Weeks 2013) measured the regime-type of the opponent by explicit mentioning the regime-type: democratic and autocratic/dictator, respectively. These words have strong and possibly negative connotations that might have triggered threat responses that are less connected with what a specific regime-type entails. Lastly, the relevance of taking other explanatory factors within the design. This study built strongly on my previous study that found regimetype as a single predictor of the willingness to attack to be of influence, but only for the democratic participants (2017, 538). The multivariate analysis, however, showed that the significant influence of regime-type disappeared and that the perception of threat of the conflict, as well as hawkishness explained the willingness to attack (2017, 539). These results are found in present study as well, which call, if anything, to revisit democratic peace theory.
What would, in the light of these findings, be a good way to revisit democratic peace theory? If we accept that leaders of both democracies and non-democracies are mainly influenced by their hawkish beliefs during interstate conflict resolution, how can we explain that democracies are unlikely to go to war with each other? If we accept that hawkishness is an actor-centric factor of significant and dominant influence on decision-making in general, something that is supported by other studies (see e.g. Kertzer and Brutger 2016;Kertzer and McGraw 2012;Kertzer et al. 2014;Rathbun et al. 2016;Yarhi-Milo, Kertzer, and Renshon 2016), and if we accept that democracies have not ended up at war with other democracies, the question arises: what constraints the hawks in democracies? The findings show that the (mutual) regime-type was not of influence; knowing that the other state was a democracy did not decrease the willingness to attack. What was it then? It might be that there is an interaction between the actor-specific hawkishness and ato be specifiedstructure that causes the democratic peace. In other words, it might be that hawks find themselves more heavily constraint within liberal democracies than in non-democratic regimes.
A new research avenue into the democratic peace project might have a renewed focus on institutional processes. Although this study found that the regime-type did not affect decision-makers in their decision, so separate from the mechanism that is posited by the institutional explanation, and others have shown that the microfoundations of the institutional explanation does not find empirical support (Kertzer and Brutger 2016), an alternative institutional explanation should not yet be ruled out. Such an explanation might lie deeper within the specific and less formal institutional and organizational processes of liberal-democracies. Liberal democracies rely on a large variety of bureaucratic organizations and inherent decision-making processes, which are often less centralized, in comparison to non-democratic institutional settings, and exist of multiple layers and bureaucratic processes. Formal institutions do often not function as intended, due to the dynamics underlying the organizational processes of these formal institutions (Hall and Taylor 1996;Immergut 1998;March and Olsen 1989;Scott 2001). The organizational processes within the formal institutions create new informal institutions, which start to 'live a life of their own', and (partly) take over the formal practices (Powell 1991, 194-200). The norms of an institutional or bureaucratic environment can create a 'logic of appropriateness' (March and Olsen 1989, 23) in which the actors behave according to specific expectations. Expectations inherently connected to the informal institution. The informal institutions can be the result of a (maybe even dialectical) interaction between the structure of the institutions and the individual actors within these institutions. Such an interaction might be the mechanism by which a possible alternative institutional explanation for the democratic peace takes place.
This take on decision-making processes connects with the field of foreign policy analysis. The work by Margaret Hermann c.s. (see e.g. Hermann 1980Hermann , 2001Hermann , 2005Hermann and Hagan 1998;Hermann and Hermann 1989;Hermann and Kegley 1995;Hermann and Preston 1994;Hermann et al. 2001;Kaarbo and Hermann 1998) offers a clear theoretical framework of how hawks could be constraint by institutional arrangements, in relation to their receptivity to context. The relationship between regime-type and leadership style or beliefs has not yet been explored extensively (Cuhadar et al. 2017;Kaarbo 2018;Kaarbo and Hermann 1998;Keller 2005;Schafer and Walker 2006;Walker and Schafer 2006) and a contribution to these studies might be productive.
The question with which this study started, why do leaders of democracies don't fight with other democracies while they do fight with non-democracies, has not been answered, yet. The insights of this study, however, do bring a bit closer to light what possible answers might be available to answer this important question that underlies the enormous research project of the democratic peace. Because results show that democratic peace theory, which aims to explain why democracies do not fight with each other, cannot be used as has been done till today and should be revised. This insight brings scholars that are interested in the democratic peace a bit closer to a deeper understanding. Notes 1. Cited 2100 times dd. 24 March 2020. 2. For a more in-depth description of the conceptualization, see supplementary material. 3. Freedom House (2014); US: free (democracy), Russia: not free (hybrid), China: not free (Autocracy). 4. Polity IV (2014); US: 10 (democracy), Russia: 4 (anocracy), China: −7 (autocracy). 5. The English scenario was translated to Russian and Mandarin Chinese, resp., by native speakers. 6. The research instrument can be found in the supplementary material. 7. Pre-tests among Dutch students showed that a conflict over uranium is perceived as a more likely condition that might lead to war than conflicts over other natural resources (such as sweet water or oil). 8. Which would officially depict it as a quasi-experimental approach, since students are not randomly assigned to the experimental groups, however they are all randomly assigned to the treatments within the experiment. 9. US: 251 undergraduate students from Binghamton University, Binghamton (NY), February 2015.
10. Russia: 250 undergraduate students from Higher School of Economics in St Petersburg, November 2014. 11. China: 280 undergraduate students of Chinese University for Political Science and Law in Beijing, April 2015. 12. For a more in depth explanation of the operationalization of liberal norms, see supplementary material. 13. See the supplementary material for a more elaborate discussion on the operationalization of the WVS study. 14. For a more in-depth description of the treatment checks, please see the supplementary material. 15. Treatment checks of regime-type also differed significantly per country. 16. Treatment checks of invasiveness also differed significantly per country. 17. Treatment checks of invasiveness also differed significantly per country. 18. Please note that the student samples measured the levels of liberal norms on a 7-point scale, the WVS on a 10-point scale. 19. Also when all decision-makers are considered as one sample, there is no significant influence of regime-type (t(774) = .42). 20. A test showed that hawkishness did not have an interaction effect with any of the other treatments.