Abstract
This analysis focuses on the effects of domestic public pacifist opinion and international security threats on foreign policy outputs. Much work has suggested that governments’ foreign policy outputs are responsive to public opinion in advanced democratic countries. Using the cases of several Western democracies, this article offers a theory of the effect of public pacifism on foreign policy. It employs a cross-sectional time-series analysis over a period of a quarter century to test the theory and the generalizability of the hypothesis of an opinion–foreign policy nexus using new measures and broader data. Results here contradict literature on expected public opinion and policy outputs in the Cold War period, yet are supported after. The findings indicate that the predicted effect of public opinion on foreign policy outputs to be conditional on the presence of security threats. Convergence between leaders and public opinion in post-Cold War Western democracies is likely to make hawkish foreign policy less tenable in the West.
Introduction
In democracies, the link between government policy and public opinion seems intuitive, yet the claim that democratic governments acquiesce to the public’s wishes is not universally accepted. Traditional views have long held that public opinion, being inadequately informed and capricious, does not or should not influence foreign policy (Almond, 1950; Converse, 1964; Lippmann, 1925). On the other hand, elected officials need the support of their constituencies and several scholars have made the case that the public opinion constrains (Sobel, 2001; Sobel and Shiraev, 2003) or otherwise affects policy (Holsti, 1996). Empirical research in the United States (US) has generally supported this ‘democratic responsiveness’ or opinion–policy nexus hypothesis (Page and Shapiro, 1983; Shapiro and Page, 1988). 1 However, tests of the generalizability of the hypothesis to Western democracies more broadly have not been able to conclusively support results found in the US. 2 This inconsistency poses a particularly vexing query: Does public opinion influence a government’s foreign policy?
In answering this question, the article assesses several scenarios which might shape the opinion–policy relationship and offer a more general assessment of this nexus within the context of changing security environments. It is argued that an external threat conditions the influence of public opinion on policy-maker decision-making. From this perspective, a lack of opinion–policy congruency, due to the effect of the presence of a higher security threat as was experienced during the Cold War, would pose some challenges to the opinion–policy nexus hypothesis. These suppositions are tested in eight US allied countries under various conditions of threat using new measures and broader data. The analysis includes Belgium, Canada, Denmark, France, Germany, Italy, the United Kingdom (UK) and The Netherlands. 3 These countries were all North Atlantic Treaty Organization (NATO) allies with strong liberal democratic institutions similarly affected by the systemic threat environment of the Cold War 4 and differ with regard to the intensity or degree to which the Cold War threat affected them, however. This cross-validation will help uncover common patterns, where differing opinion–policy relationships may be due to the external variable: threat.
This article is organized as follows. It first examines the relevant literature on the relationship between public opinion and foreign policy in order to assess the evidence in support of a generalizable theory that security threat conditions the influence of pacifist public opinion on foreign policy outputs. Second, it outlines a theory of public opinion using measures of public pacifism and government policy. Third, a research design to test hypotheses from the theory is presented suggesting that threats affect a potential relationship between public opinion and foreign policy outputs. It is expected that during the higher threat conditions, such as the Cold War, security threats confound the establishment of an opinion–policy nexus in all of the countries to varying degrees. With the end of the Cold War, the reduced threat environment allowed for a reassessment of the traditional foreign policy, which had placed security concerns above strict compliance to public opinion in favour of one which more closely mirrored public opinion. In the conclusion, the implications of this research for the relationship between the US and its former Cold War alliance partners are discussed.
Hypotheses on public opinion and foreign policy
Most of the research that has supported a democratic responsiveness model has been conducted in the US over the past quarter century. Page and Shapiro (1983), for example, have demonstrated that a positive empirical relationship between public sentiment and government policy may exist. 5 They later showed that this basic relationship exists with regard to foreign policy as well (Shapiro and Page, 1988). Continuing work has modified this somewhat, but the general hypothesis remains intact. 6 Investigations of its generalizability have mainly focused on claims of the stability and rationality of public opinion and have been replicated cross-nationally to a certain degree. 7 The thesis presented in this article focuses on the relationship between public opinion and foreign policy outputs, of which there have been fewer analyses, and, of those, the findings have been contradictory.
Eichenberg and Stoll (2003), for example, test the generalizability of opinion–policy relationships that had been previously supported solely with US data (i.e. Shapiro and Page, 1988). They compare defence spending and public support for defence spending in five countries: the United States, Great Britain, France, Germany and Sweden. Their findings are supportive of the claims of democratic responsiveness and their model is an excellent assessment of a particular policy issue, defence spending, in order to test hypotheses of democratic responsiveness. Tests of Page and Shapiro’s rationality of public opinion hypothesis, for example, have been supported by Isernia et al. (2002) with data from France, Germany and Italy and by Everts (2000). This suggests that the hypothesized rationality and the opinion–policy nexus hypotheses are generalizable. However, Eichenberg and Stoll do not consider the effect of public opinion on foreign policy more generally. Isernia et al. and Everts do not directly attempt to address Page and Shapiro’s opinion–policy nexus hypothesis. Furthermore, these analyses do not assess specifically the role of threat on the effect of public attitudes on policy outputs.
In contrast to Isernia et al., Everts and Eichenberg and Stoll’s work supporting an opinion–policy nexus, Brooks (1990) finds little support. Brooks analyses the relationship between opinion and foreign policy in Britain (1985), France (1987) and Germany (1990) in an attempt to assess the democratic responsiveness of the relationship. He provides evidence to suggest that there is no apparent correlation between mass public opinion and the adherence to a consequent policy. Brooks’s findings for the period he studied, the last 10 years of the Cold War, are problematic for assertions of an opinion–policy nexus. However, Brooks’s time period may be noteworthy because it is limited temporally to Cold War data only, and his overall findings may be a function of this temporal limitation. 8 This is because, according to the theory outlined in the following section, Cold War security necessities may have required a closer association with US policy preferences regardless of the preferences of the public, whereas after the Cold War, there may have been greater latitude among US allied governments to be responsive to their respective publics’ preferences. It is therefore necessary to include the post-Cold War period of relatively low levels of security threats.
Previous models consider neither foreign policy more broadly nor the direct effect of exogenous factors, broadly defined as security threats. 9 In this analysis, therefore, the article tests claims that public opinion informs public policy during and after the Cold War using a broad set of democratic allies of the US. What is more, by examining these relatively comparable countries, it is possible to assess variation on the level of threat allowing for the test of the effect of threat intensity on a posited opinion–policy relationship. This may also expose any country-specific idiosyncrasies.
Germany, for example, is distinctive in many ways due to the peculiarities of its history. 10 Germany’s role in a succession of wars with France and much of Europe going back to 1870 sets it apart. Germany was also a divided nation during the Cold War and experienced a transition from an aggressor to a pacified partner. Moreover, and in particular its supportive role in the struggle against Soviet hegemony over Europe in the aftermath of World War II and during the Cold War is unique in that unlike the United Kingdom (UK) or France, for example, it shared a common border with Warsaw Pact states.
Dorff (1997) makes a compelling case that German foreign policy displays ‘elements of exceptionalism’, sometimes overly assertive in the pre-World War II period, acquiescent to the US in the Cold War period and evolving toward ‘normalcy’ in the post-Cold War period. 11 Normalcy, in Dorff’s view, refers to states that pursue their self-interest. Additionally, Gordon argues that ‘the normalization of German foreign policy has already begun’ (1997: 241). More than a decade later the claim appears prescient.
While it is necessary to recognize country differences, it can be maintained that domestic characteristics do not entirely negate the essential relationships suggested by the theory that external threat is causally related to a breach in a congruent relationship between a country’s public opinion and its foreign policy outputs. Germany’s geographic placement, adjacent to the Warsaw Pact border, may have had the effect of producing higher threat levels for Germany as compared to the UK and France, for example, but the causal process, that security threats influence the effect of public opinion on foreign policy outputs, is the same for each of these countries.
A theory of the effect of public opinion on foreign policy
In this analysis, both international as well as domestic influences typically associated with either realist or liberal ideas into an integrated theory are incorporated. 12 This integration is unusual because, for realists, systemic anarchy makes state security particularly important. Anarchy, which leads to the assumption that state is the primary unit of analysis, produces a state of self-help where a state’s purpose and its primary interest is survival. 13 It is this state-centrism which leads to the omitting of domestic variables that is often criticized by opponents. 14
Liberal explanations more typically emphasize the ‘institutions and practices’ which aggregate domestic interest group preferences into state policy. 15 Since the actors and processes that determine foreign policy are found mostly at the domestic level, liberal analyses favour domestic over systemic variables. 16 It is argued in this article, however, that it is essential that we know both what the state’s primary systemic function is, i.e. survival, and how well a state performs its function, i.e. its foreign policy.
A theory on the public opinion–foreign policy relationship thus must integrate both international-level and domestic-level variables. For example, to explain the foreign policy-making process, it is suggested that leaders want to retain office, and to retain office they are concerned with both external threats, i.e. security, and internal threats, i.e. public opinion. For purposes of this study, external security threats are considered to be both military in nature as well as non-military political conflicts (Baldwin, 1997; Pfetsch and Rohloff, 2000a), which captures the changing nature of these threats during and after the Cold War. Domestic threat, for a leader and/or their party, is electoral in nature due to oppositional public sentiment. 17
Security threats matter
During the Cold War the threat environment was heightened due to the presence of Soviet and NATO forces at the ready on either side of the demarcation between East and West Germany and the constant threat of nuclear annihilation this presented. Post-Cold War threats, though numerous, are likely to be lower in intensity as the threat of nuclear annihilation is diminished. It is argued here that under heightened threat conditions leaders would more likely ignore public opinion if they believed the state’s security was threatened, but when threats are reduced a potential opinion–policy relationship is likely to emerge. 18 A leader’s reliance on public opinion with regard to foreign policy decisions is thus conditional on the external threat environment. It is the security threat, then, that confounds the potential opinion–foreign policy nexus, and it is the removal of such a threat that allows for governments to be responsive to public opinion.
Electoral threats matter
In democracies, leaders have an incentive to respond to public opinion because it influences their ability to retain office. In a democracy, a leader’s primary preference may be to retain power or to be re-elected; they therefore must be ‘vote-maximizing’ or responsive to domestic audience preferences or suffer the consequences at the ballot box. 19 For example, the lesson from history suggests that elected officials may be removed from office in a future electoral cycle in response to unpopular policy decisions. In 1983 the failure of German Chancellor Helmut Schmidt to survive a constructive vote of no confidence after his unpopular decision to allow the US to install Pershing II missiles in Germany is a case in point. 20 More recently, various national parties that supported the Iraq War suffered losses at the polls after reaction to the unpopular policy. 21 Furthermore, leaders and the public might rally against external threats. Why then would leaders ignore public opinion?
Opinion–policy conditionality
While public opinion may determine a democracy’s foreign policy in times of low security threat, the presence of an external threat causes political leaders to downplay public opinion for several reasons. First, elite and mass opinion can differ because of a divergence in ‘values, goals, and interests’ (Page and Barabas, 2000: 362). Security threats may force political officials to override public opinion in the short term in their primary role as defender of the state in order to achieve the greater good of the survival of the state and benefit from the longer-term payoff that the successful policy brings. 22 Therefore, leaders represent not only the interests of their constituents, but also the interests of the state, the primary one, as realists assert, being survival (Waltz, 1979).
When there is an external threat that jeopardizes the state’s autonomy, existence or way of life, the leader must condition his or her behaviour on that external threat. The classical realist framework can account for the leader’s behaviour in choosing the state’s security over public opinion. For example, Machiavelli exhorts the Prince to recognize the ‘prudence’ of putting the state’s security interest above other interests, particularly with regard to its survival ([1515] 1952: 85). Prudence as the ‘weighing of the consequences of alternative political actions’ according to Morgenthau is indeed the ‘supreme virtue in politics; ([1948] 1993: 3–14). Responsivity on the part of a leader to the public is balanced against threat as an act of prudence.
Second, leaders and the public may be ‘asymmetrically informed’. 23 Leaders, having private information, may know more about a particular threat and presumably would be able to make a better judgment about a proper course of action. The public has little access to the reliable sources of information that can allow for an accurate assessment of threat. Better, more accurate information regarding security threats can compel policy decision-makers to override public opinion in order to produce policy outputs that reflect the state’s interest. Moreover, what information the public has is funnelled through various media with their own perspective and interests, leaving the public susceptible to manipulation. DeHaven, for example, notes that during the Cold War the Soviet Union attempted to steer foreign policy in the West by manipulation of the public through the media’s promoting the view that the West’s policies would increase the threat of conflict (1991: 87). To the extent that leaders have more information, they can better understand the ongoing manipulation of opinion and may be willing to discount opinion in the short term.
Third, the publics of the West (ex US) have become largely pacifist and therefore support for hawkish security, which may be viewed as preparation for war, may simply be politically untenable (Kagan, 2003; Mueller, 2004). 24 As previously stated, however, because of differing interests leaders may not necessarily share the public’s pacifist sensibilities and pursue pacifist foreign policy for political expediency. For example, a strong security policy is expensive and often comes at the expense of preferred domestic spending. 25 The US, as guarantor of the West’s security during the Cold War can reasonably be expected to continue its ‘security umbrella’. 26 Therefore, alliance governments may not wish to pay for security services, which will likely be provided in any event, thus establishing opinion–policy congruency. On the other hand, even if leaders share the public’s pacifism, they may be forced into accepting a policy due to alliance and security necessities that may cause a disconnect with their constituents’ preferences. 27
The logic of this account leads to the hypothesis that a leader will have a higher probability of assigning primacy to the security interests of the state over public opinion when an external threat is present. More specifically, congruence between the public’s pacifist preference and the leadership’s support for US foreign policy are conditional on the external environment. External threat, defined as that which jeopardizes the state’s existence or way of life, is the condition under which a potential opinion–policy relationship is annulled. This relationship is expressed in the following hypotheses:
Hypothesis 1: As external threat increases, the effect of public opinion on government foreign policy congruence with US foreign policy decreases.
Hypothesis 2: As external threat decreases, the effect of public opinion on government foreign policy congruence with US foreign policy increases.
Research design
To test the hypotheses, this article analyses panel data for eight major US Cold War allies across a 24-year period. Data limitations require a period between 1973 and 1998 due to cases where there might be opinion data, yet no threat data or threat data and no opinion data. Countries where both opinion and threat data were not available were duly eliminated. Next, the data, variables and model operationalization are discussed.
The dependent variable is a measure of congruence with US foreign policy or Foreign Policy Position. The unit of analysis is each country’s yearly United Nations General Assembly (UNGA) vote similarity between the US and each of the eight countries in the analysis, or government-year, between 1973 and 1998. The US’s policy varies over this period (two Democratic and four Republican administrations) from less hawkish (Détente) to more hawkish (Reagan) and again to the post-Cold War period. However, the concept that is attempted to be measured is not bellicosity or pacifism per se, but rather the specific country–US foreign policy similarity dynamic. While a UN vote does not a policy make, it is a relatively low-cost public expression of a government’s policy preference that signals proximity to both the domestic public and the US preferences.
The policy benchmark is a Left–Right scale where the US is the polar right of the spectrum. This benchmark is useful for several reasons. First, studies have shown that a Left–Right partisan dimension is comparable across most industrialized countries and that there is a direct relationship between political partisanship and policy preferences, which make it the primary determinant of voter choice (Budge et al., 2001: 159). Second, partiality to US policy support in foreign affairs is also generally a bellwether indicator of Left or Right partisanship (Ruehl, 2003; Voeten, 2000). The US pursued a relatively hawkish policy throughout the Cold War as the West’s leader, and can accurately be characterized as the Right polar extreme with regard to the Left polar opposite the Soviet Union. By definition, leftist policy is likely to be more dovish or pacifist with regard to the Soviet Union during the Cold War and diverge from the US policy more than rightist policy.
The general ideological dynamics remain consistent during and after the Cold War despite systemic changes. Voeten (2000), for example, found empirical support for the notion of a US Right pole, not only for the Cold War period, which would seem intuitive, but for the post-Cold War period as well. He concludes ‘post-Cold War voting is mostly one dimensional’ along a linear spectrum of polar extremes, where ‘the United States and its Western allies occupy one extreme pole’ (2000: 186). 28 If the opinion–policy nexus works as suggested in the literature, then foreign policy should correlate to the Left or Right composition of the government and naturally to the median voter who put that government in power. A divergence from the US bellwether in response to public opinion should provide a test of the hypotheses stated above.
The measure is created from aggregated annual country voting data on UNGA resolutions for each state using all adopted resolutions put to a roll call, which are coded for yes or no types of responses for the chosen period. The vote is coded as one if the country case under analysis voted the same as the US, zero otherwise. 29 The mean is then calculated for each year to make indexes for each country case, which would show as a 1, if all votes were perfectly congruent, or a 0, if all votes were non-congruent for a particular year. Since the theory proposed in this article calls for only security-related issues, it is necessary to keep only the security specific votes for the analysis. 30 The author chose not to include abstentions since deducing the motivation behind an abstention is ultimately subjective and there is no preferred method or precedent in the literature. 31 The dependent variable is scaled from zero to one (0 / 1.0) with a mean of 0.59 and standard deviation of 0.30.
The primary explanatory variable for the models is the pacifist public sentiment, Pacifism. Data from the Manifesto Research Group (MRG) and Comparative Manifesto Program (CMP) are begun with, which use a content analysis approach to calculate the number of phrases in support of specific policies, such as pacifism, for each political party’s manifesto, in each of the countries under analysis. The data uses a coding of party statements that can be categorized as Left, Right or neutral. This allows for assessing a political party’s policy space on a Left–Right scale. Elections show the aggregated public support for each party’s policies as a percentage of the vote (Budge et al., 2001). A high percentage of citizens’ support (vote) for parties with a high quantity of expressions for support of pacifism will correspondingly produce a higher measurement of public pacifism. This is multiplied by a negative 1 so that my higher values correspond to Right on an ideological spectrum as this corresponds better to the dependent variable (higher values = rightist policy).
The Pacifism measure is comparable to the dependent variable as they are both measured on a Left–Right spectrum. This dimension, as previously stated, is shown to exist in most industrialized countries; it is comparable across various countries and it is the primary determinant of voter choice. 32 The variable is valid due to its ‘saliency’ measurement of policy preference rather than a pro–con issue contrast approach. 33 Saliency refers to the underlying value preferences which affect policy choices and generally allow issues to be categorized as a coherent set of ideologically consistent partisan policies, left or right, such that ‘there is only one tenable policy on each issue’ (Budge et al., 2001: 85). Factor analyses show the great extent to which various policies, including pacifism, which can be categorized as Left or Right, co-vary with one another. 34
The model also uses Eurobarometer self-placement survey data of Left–Right self-placement and support for defence spending. These surveys aggregate likely voters’ political views and their support along the Left–Right spectrum. The self-placement data are positively correlated with the MRG/CMP data (i.e. r = 0.28 (manifesto pacifist statements support) and r = 0.62 (left–right-partisanship measure). This provides a validity test for the MRG/CMP data and allows for imputing data to fill in the gaps in MRG/CMP data for the years when there are no elections. 35 This process is possible due to the more or less consistent nature of ideology as expressed in Budge’s salience theory.
A country-specific Threat measure is created using two distinct measurements. The first is designed to capture the maximum effect of the before the Cold War (CW coded 1) and after (CW coded 0) periods as a natural experiment. The second measure is more nuanced and designed to capture the level or intensity of threat for each country over the period under analysis using KOSIMO data. KOSIMO uses a qualitative definition of conflict and war with a content analysis approach that includes non-violent cases or ‘latent’ conflict. 36 Wars and conflicts of lesser intensity are classified according to the actual amount of violence observed, and not according to the number of fatalities. 37
Measures of conflict using data based on the onset of militarized crises, such as International Crisis Behavior Project (ICB) and the Militarized Interstate Disputes (MIDs), were found to be inadequate for this study as they may not provide the best way to assess and specify threat intensity and may not actually reflect an accurate picture of the threat perceptions of publics and their governments in this period. For example, the installation of the SS 20 intermediate range missile and the Pershing II installation as a response was certainly perceived as a threat crisis but may not rise to the level of a crisis according to the specification of measures which require a specific number of deaths over definite time criteria such as MIDs, for example. Also, because violent interstate crises between the West and the Warsaw Pact are relatively rare events during the Cold War, these measures do not have the variability needed for an adequate test of the hypotheses specified here. KOSIMO successfully passes validity tests in comparison to alternate measures suggesting that it is a valid and reliable indicator of conflict from latent to military conflicts. 38 As a robustness check, I also use alternative operationalizations of threat in my models. 39 The Threat variable is specified so that higher values represent higher threat levels corresponding to the Foreign Policy Position and the Pacifism variables. 40
While much of these data and methods of operationalizing conflict as used in previous studies are typically coded dyadically, my expectations are monadic rather than dyadic. 41 A one-to-many directed-dyad approach is therefore used. That is, all interstate interactions are included where there are indications of threat in the collapsed country-specific measure for the period in question. The variable for the models reported takes on a value of 1 for all violent events and 0.5 for all non-violent and latent events, with zero being no crisis/conflict events, producing a possible range of 0 to 1.
The theory put forth here requires that the effects of the Threat and Pacifism variables be conditional on each other. A multiplicative interactive term is therefore generated from the public opinion and security threat variables, Threat*Pacifism, in order to account for the conditional effects of the two. It is this variable that is expected to provide the key support of the theory behind the model for this analysis.
Several alternate explanations are controlled for in the respective analyses. For example, Lewin states that ‘politicians, just like voters, are primarily guided by their self-interest’ (1991: 72). If one defines the leader narrowly as an office-seeking ‘vote-maximizer’ primarily interested in re-election (i.e. public–choice approach), then it is coherent to propose the hypothesis that politicians will closely follow public opinion and perhaps use public spending in the run up to an election to help ensure their electoral success (Hibbs, 1977). To account for the occurrence of national elections, the author has constructed an election year and opinion interaction variable, Elections and Elections*Pacifism, respectively.
In separate models, control variables are included to account for partisanship of public opinion and a dichotomous variable for Germany. 42 The Left–Right variable, using data from Budge et al. (2001), allows for assessing whether it is partisanship rather than pacifism that drives the results. To account for the argument of German exceptionalism, the variable Germany is coded 1 for German cases and 0 otherwise. This allows for contrasting analyses that include Germany and with those that exclude Germany to determine if Germany is driving the overall results. If the results are in line with expectations even when accounting for Germany, then it will provide stronger support for the general theory. 43
Expectations
Table 1 highlights the author’s expectations. Given the assumptions of the public’s asymmetric information and proclivity to pacifism, it is expected that the effect of public opinion on foreign policy decision-making will be inverse to the level of external security threat. That is, as the level of security threat increases, the effect of public opinion on foreign policy outputs should decrease, and, as security threat declines, the effect of opinion on foreign policy should increase. A positive beta for Pacifism is also expected under low threat conditions in compliance with the theory put forth here. A negative coefficient is possible for the Pacifism variable under conditions of higher security threat as during the Cold War. This is because governments may be responsive to threat in pursuing a close security foreign policy similarity with the US – a move to the Right. If, however, public opinion consistently reacts to the same threat with a high degree of pacifism – a move to the Left – there will appear to be a negative relationship between public opinion and foreign policy.
Expected results
Since the author expects the beta for Pacifism to be positive and to be confounded by Threat, my beta for Threat*Pacifism must be in the opposite direction as the beta for Pacifism. As Threat increases, it is expected that the beta coefficient for Pacifism will become increasingly less positive, suggesting that the effect of Pacifism on Foreign Policy Position decreases. This dynamic is the result of the interactive effect between Pacifism and Threat, as the explanatory variables are expected to be interdependent, and is the relationship of substantive importance in this analysis.
Because this model requires the interaction of my Pacifism and Threat variables, the specification of multiplicative interaction models must be properly highlighted in order for a substantive interpretation of the regression results to be meaningful. Brambor et al. (2006: 74) state that the correct marginal effect of the primary explanatory variable is calculated as: β1 + β3*β2 (Pacifism + Threat*Pacifism). This means that the coefficient of my Pacifism variable is not necessarily substantively important in and of itself, but only when it is viewed in the context of the conditionality imposed upon it by the other constitutive variable, Threat. It is the coefficient of Pacifism plus the coefficient of Threat*Pacifism that is substantively interesting. This is the result of the interaction of the two variables of substantive import, Pacifism and Threat, and how that interaction reduces the effect of Pacifism on Foreign Policy Position. 44
Estimation
This analysis uses cross-national panel data across a 24-year period. Panel data have three advantages over pure time series of pure cross-sectional data. First, they allow for estimation of generalizable hypotheses common to the group of countries chosen. Second, they allow for testing causal relationships under my assumption of unit homogeneity. 45 Third, this approach has advantages for statistical modelling by increasing sample size. 46 However, in combining spatial and temporal domains, the cross-sectional time-series panel data may create problems of non-stationarity, autocorrelation, heteroskedasticity, multicollinearity or bias in the estimation of standard errors. 47 A lagged dependent variable (LDV) is used on the right-hand side for dealing with serial autocorrelation. The LDV model is also advantageous for this analysis not only because of the focus on several countries over time but also because they are mostly in a common geographic region, some of which are contiguous to several others. 48 Beck and Katz (1995, 1996) recommend panel-corrected standard errors (PCSE) for dealing with potential panel heteroskedasticity despite Achen’s (2000) inflated standard errors warning. 49 Using the constitutive with the interaction terms may increase multicollinearity leading to large standard errors and lack of significance (Type II error). Although Barambor et al. (2006) argue that this problem is often overstated, it is estimated for both an LDV and a non-lagged model under several specification criteria, including PCSE’s, in order to assess these issues and to check the robustness of the results.
Finally, the model uses a panel-estimated approach with random effects that control for country-specific effects that are likely to be present in the error term (Wooldridge, 2001). 50 Test results to determine if the individual effects are not correlated with the other regressors produced insignificant results (Chi2 = 6.69, p = 0.153), and therefore a random effects linear estimator was chosen. 51 The random effects method is also superior to the fixed effects approach in this case because it allows for possible variations within panels, while most fixed effects specifications would not. 52 While admittedly this study has certain spatial as well as small N difficulties, it nevertheless should show the trends and relationships that have taken shape during the before and after cases of the Cold War.
Results
Table 2 presents the regression model results, which are robust to the other estimation techniques discussed. 53 As hypothesized, the lagged security index measure, designed to show unidirectional causation from public opinion to policy, is highly significant, and increases the overall R2 from 0.07 to 0.70, suggesting that the inclusion of this measure explains most of the variation in Foreign Policy Position measure. Nevertheless, it is remarkable that even after controlling for past similarity of the Foreign Policy Position measure (by using a lagged dependent variable in the model), Pacifism is still negative and significant and the interactive term is still positive and significant.
Effects of threat and pacifism on foreign policy position
p <0.05, **p <0.01, ***p <0.001 one-tailed test. Beta coefficients with robust standard errors in parentheses.
Since the coefficient of the Pacifism variable is substantively important not in and of itself, but only when it is viewed in the context of the conditionality imposed upon it by the other constitutive variable, I discuss only generally the coefficients of the individual variables and instead focus on the relationship between Pacifism and Threat. This relationship is estimated by the coefficient of Pacifism plus the coefficient of Threat* Pacifism. This coefficient is the effect of the interaction of the two variables of substantive import and is the substantive marginal effect of Pacifism in this interaction model.
The primary support for the first hypothesis (H1), that public opinion should positively affect policy when Threat is low, is supported. 54 Indeed, a one-unit increase in the value of Pacifism is associated with an increase in the Foreign Policy Position variable of 0.028 (beta coefficient) when Threat is low (held constant at zero). The Pacifism variable, where higher levels indicate security support is both positive and significant as expected. This suggests that lower pacifism or a greater public support of security is associated with a closer congruence between the US and the respective governments’ foreign policies. The results in Model 1 thus confirm the first hypothesis.
The second hypothesis (H2), that threat is a mitigating or confounding factor conditioning the effect of pacifist public opinion on a government’s foreign policy, is provided by the marginal effects of the variable Threat*Pacifism (Model 1). As expected, the coefficient of the product of Threat*Pacifism is negative at −3.7 (s.e. = 0.77) and is highly significant at the 0.001 level. 55 This interactive term, comprising the product of Pacifism and the Threat variables, was designed to express the expected change in the relationship between Pacifism and Foreign Policy Position when the Threat variable increases for each one unit degree from 0 to its highest value 1. In other words, when there is a security threat, the government is less likely to agree with the public.
This relationship is clearly shown in Figure 1. 56 As expected, the marginal effect is positive and statistically significant for all values of Threat greater than zero (see Figure 1). 57 This tells us, substantively, that the effect of the Threat gradually inverted the relationship between public opinion and foreign policy outputs by 0.027. The threat–cohesion claim is not supported here. These results thus confirm the second hypothesis (Figures 2 and 3)

Model 1: Pacifism and Threat Conditionality.

Model 2: Pacifism and Threat Conditionality.

Model 3: Pacifism and Threat Conditionality.
I now turn to the control variables. A one-unit increase in the control variable for election years, holding all other variables constant, resulted in a positive coefficient or increase in Foreign Policy Position of 0.029. This measure was used in an attempt to see if election year politics, such as tailoring policy, a boost in fiscal stimuli or other benefits to the voters, which are often peculiar to election years, might have had some affect. Additionally, events and circumstances, which might be turned to an advantage, and electoral success, such as the Iraq crisis or the German floods in 2002, should be captured in such a variable. This result (Model 1) is not significant, however, and does not support such contentions.
Because public opinion for pacifism and partisanship are saliently correlated, an alternative explanation for the effect of the relationship between this sentiment and security threats on foreign policy outputs is that partisanship, rather than pacifism, is driving foreign policy decisions. Specifically, the alternative hypothesis would suggest pacifism is a function of partisanship. Model 2 examines this hypothesis. The coefficient for left–right placement is statistically significant and indicates that partisan considerations may have effects independent of pacifist public opinion. However, this result is not robust with regard to alternate measures. For example, while the Eurobarometer measure produced positive and significant results, the MRG/CMP measure did not. This is likely due to endogenous effects of the MRG/CMP and Pacifism variables, so I cannot determine whether or not the results are dependent on the specification of the variable. Therefore, the partisanship alternate hypothesis cannot be definitively ruled out nor can it be supported. Additionally, with the inclusion of this variable in the model the dummy variable for elections remains positive but becomes significant.
Is the inclusion of Germany driving the results in the cross-sectional analysis? The coefficient on the Pacifism variable remains positive, but is not significant when controlling for Germany in Model 3. Furthermore, the beta coefficient for the Germany variable is highly significant. This appears to support the alternate hypothesis that Germany is an important factor driving the cross-sectional results. Nevertheless, even when accounting for the presence of Germany, the effect of Threat on the relationship of Pacifism and Foreign Policy Placement remains robust. The inclusion of Germany does not change the nature of the marginal effect of Pacifism on the Foreign Policy Placement variable under conditions of threat. 58 This suggests strong support for hypothesis 1 and hypothesis 2, that the effect of Pacifism on the Foreign Policy Placement variable is conditional on Threat. Therefore, the alternate hypothesis of the uniqueness of Germany cannot be compellingly confirmed and the regression results conform to the expectations of the theory.
Conclusion
The relationship between public opinion and foreign policy formation is more complex than previously estimated. Here, a foreign policy model of the effect of external threats on the relationship between public opinion and the state’s foreign policy formation that clarifies this relationship is offered. Results here show that security threats confound the establishment of an opinion–policy nexus, while relaxation of the threat environment allows for a foreign policy that more closely mirrors public opinion. This has theoretical implications with particular relevance to democratic theory. This finding, in effect, lends support to a ‘democratic frustration’ model, and suggests that there are several impeding factors on the potential opinion–policy nexus. It should be pointed out that security affects may be just one impeding factor. Furthermore, the assumptions of the public’s asymmetric information and proclivity to pacifism were not directly tested. With greater information symmetry under threat conditions, public pacifism may decrease and opinion–policy would likely converge around a hawkish security policy.
Additional research is therefore necessary in order to determine the extent to which public opinion may actually be relevant or even negatively correlated to government foreign policy outputs in periods of threat, if renewed threats can alter opinion–policy congruence once an opinion nexus is established, and in particular how non-state threats like terrorism affect an opinion–policy nexus in the post-9/11 world. The models and new measures developed here should prove useful to researchers who seek to test hypotheses implied by theories of foreign policy, which necessitate demonstrating the dynamic interaction between domestic and systemic influences. The integrated foreign policy model applied in this study provides improved explanatory power over existing approaches and demonstrates that realist analyses need not be limited in scope to state or systemic level variables.
Moreover, the finding has relevance for foreign policy considerations such as alliance-formation. In a world grown accustomed to US security guarantees, there is little threat perception among the publics of Western democracies or incentives to pay for expensive security policy among policy-makers. Unfettered by Cold War threat constraints, the US and allied governments are able to pursue their respective foreign policies with more flexibility. If the voters of the West have become largely pacifist and an opinion–policy nexus is to exist, policy-makers’ support for hawkish security policy is increasingly untenable. As a result, the likelihood of former allies establishing more divergent foreign policy positions, with respect to the US and each other, is inevitable. This presages a new transatlantic relationship between the US and its former Cold War alliance partners, as national policy more closely reflects public opinion.
Footnotes
Funding
This research received no specific grant from any funding agency in the public, commercial of not-for-profit sectors.
