Introduction
What are the political consequences of wars and conflicts? Existing literature has predominantly examined this question in warring countries and found significant ramifications for their domestic politics and public opinion, most notably in the form of ‘rally-round-the-flag effects’ (Mueller Reference Mueller1970), where citizens rally behind their governments, fostering increased political trust, collective action, and political participation (Baker and Oneal Reference Baker and Oneal2001; Hetherington and Nelson, Reference Hetherington and Nelson2003; Koch Reference Koch2011; Gilligan, Pasquale and Samii Reference Gilligan, Pasquale and Samii2014). However, others have questioned whether political gains are always present, and some even document negative effects (Kijewski and Freitag Reference Kijewski and Freitag2018; Burchard Reference Burchard2015), including on trust and support for the incumbent (Karol and Miguel Reference Karol and Miguel2007; Croco Reference Croco2011). Yet, the focus has predominantly been on countries directly involved in conflicts,Footnote 1 although there are good reasons to suspect that other countries may also be affected. In this article, we aim to explore the effect of wars abroad on the political trust of people living in countries that are not directly engaged in the conflict. Studying this question is important because neglecting the indirect effects of wars may underestimate their aggregate impact on trust, which has been declining across advanced democracies, with far-reaching social, economic, and political consequences.Footnote 2
In addition to having received no attention to date, the theoretical expectations regarding the significance and direction of the effect of wars abroad on attitudes in uninvolved countries are a priori mixed. On the one hand, wars abroad could change the real (or perceived) degree of external threat even to countries that are not directly involved, prompting a ‘rally-around-the-flag’ response, where citizens support domestic political actors in times of crisis (Hetherington and Nelson Reference Hetherington and Nelson2003). This assumes crucially that threat perceptions are altered by conflicts even when countries are not directly involved. In addition, disagreements over the management of the crisis and evaluations of the policy response may over time emerge, hence limiting the duration of the effect (Hetherington and Nelson Reference Hetherington and Nelson2003). On the other hand, conflicts in neighbouring countries could lead to negative economic spill-over effects or spur unpopular policy decisions, both potentially lowering political trust (Murtin et al. Reference Murtin, Fleischer, Siegerink, Aassve, Algan, Boarini, González, Lonti, Grimalda, Vallve, Kim, Lee, Putterman and Smith2018; Levi and Stoker Reference Levi and Stoker2000).
In this short empirical article, our aim is not to develop a new fully-fledged theory but instead to test whether existing expectations from previous literature also hold when countries are not directly involved in the conflict. Finding such an effect would imply that our current understanding of the political consequences of wars may not capture their total aggregate effects, once we consider both its direct impact on the relatively small number of warring countries and its indirect effects on the relatively more numerous uninvolved countries. While we provide a comprehensive overview of the relevance, determinants and consequences of trust in section J.1, and of previous studies on conflicts and attitudes in section J.2 in the appendix, our main aim is to provide causal evidence for this previously untested empirical question. Nevertheless, by building on existing literature, we theorize and explore empirically different potential mechanisms linking wars abroad to political trust at home in section J.3 in the appendix, where we present a range of descriptive and causal evidence for these mechanisms.
To test whether foreign wars affect trust in indirectly involvedFootnote 3 countries, we analyse the impact of the Russian invasion of Ukraine on the political trust of respondents in several European democracies that are geographically close to the conflict but not directly involved. Russia’s invasion started one of the largest European conflicts since the Second World War (Smit et al. Reference Smit, Martin, Buehler, White, Greenberg, Mysore and Chewning2022; Simchi-Levi and Haren Reference Simchi-Levi and Haren2022) and has led to significant economic (Boungou and Yatié Reference Boungou and Yatié2022) and geopolitical consequences for Europe, causing food price rises and an energy crisis (Osendarp et al. Reference Osendarp, Verburg, Bhutta, Black, de Pee, Fabrizio, Headey, Heidkamp, Laborde and Truel2022; EU Council 2022). There is already strong evidence that the invasion had large political and economic consequences for the two countries directly involved in the conflict (Gehring Reference Gehring2022; Steiner et al. Reference Steiner, Berlinschi, Farvaque, Fidrmuc, Harms, Mihailov, Neugart and Stanek2022). Yet, we do not know whether and how the invasion has changed citizens’ political trust abroad. To address this question, we employ the ‘Unexpected Event during Survey Design’ on the tenth wave of the European Social Survey (ESS), which includes questions about trust in politicians, political parties, and national and European parliaments, respectively. Since the timing of the Russian invasion overlapped with the fieldwork of the ESS in eleven countries, our quasi-experimental research design relies on respondents’ interview date, which is fixed in advance, to compare respondents’ preferences and views before and after the invasion during the same survey period.
Our findings show that Russia’s invasion increased trust towards both political actors and institutions,Footnote 4 aligning with the ‘patriotism school’, which posits a rallying around political entities during security crises (Hetherington and Nelson Reference Hetherington and Nelson2003). The findings remain robust across a wide range of control variables and time bandwidths, the inclusion of trends, jack-knife country exclusion, estimation methods and standard errors, including different types of clustering. In section J.3 of in the appendix, we provide some descriptive evidence consistent with the notion that European citizens experienced the war as a threat to their domestic institutions and political system, fearing military escalation, and further causal evidence that the invasion made people more satisfied with their government and democracy. Consistent with the positive effect of the invasion on political trust that we find, we further show there was no evidence for a negative insecurity mechanism: the invasion had no causal impact on subjective income insecurity and even increased satisfaction with the national economy.
We interpret these findings as consistent with the presence of solidarity and interdependence between democracies posited in democratic peace theory: the war has affected the attitudes of people living in democracies that are not directly attacked, but nevertheless close to a conflict where a democracy was attacked by a threatening autocracy. We present additional causal evidence showing that previous US interventions in Syria and Iraq, respectively, had no statistically significant effect on political trust in Europe, while the current Russian invasion did not impact political trust in two countries further away from the conflict.Footnote 5 Future research should further identify and analyse the scope conditions for wars abroad to affect attitudes in non-involved countries.
Data and Method
We rely on the second release of the tenth wave of the ESS, which has been carried out every two years since 2002.Footnote 6 Crucially, for our purposes, the precise time at which respondents are interviewed is set in advance, not altered as a function of events, and recorded in the survey. The start of the full-scale Russian invasion in the early hours of 24th February 2022 represents an exogenous shock to respondents being surveyed thereafter, which allows us to use the timing of the invasion that occurred during the ESS fieldwork period to assign respondents to ‘control’ and ‘treated’ groups. Our sample encompasses eleven countries where respondents were surveyed both before and after the invasion: Belgium, Great Britain, Greece, Italy, Portugal, Norway, Montenegro, Macedonia, Norway, Netherlands, and Switzerland. As can be seen in Figure A.3 in the appendix, a large number of respondents were interviewed prior to and during the war. We estimate the effect of the invasion of Ukraine using the following specification:
$${y_{ic}} = \alpha + \beta \,{\rm{Invasio}}{{\rm{n}}_{ic}} + \gamma '{X_{ic}} + {\gamma _c} + {\varepsilon _{ic}}$$
First, our outcome is captured by y ic , which represents respondent i’s levels of political trust in a country c. Specifically, the ESS team asks each respondent how much they trust their politicians, their political parties, as well as their national and European parliament, respectively, capturing both trust in ‘actor-centred recipients’ and in ‘institutional recipients’ (Petrarca, Giebler and Weßels Reference Petrarca, Giebler and Weßels2022; Easton Reference Easton1975).Footnote 7 The scale ranges from zero, if they have no trust at all, to ten if they have complete trust. The precise wording for each dependent variable is shown in Figure A2, while summary statistics are presented in Table A.3 in the appendix.
Second, the effect of the treatment is captured by β. The exact timing of the war is as good as random with respect to the date during which interviews are scheduled, which is decided at the sampling stage and, according to the ESS sampling procedures, is never changed. Hence, this allows us to causally identify the effect of the invasion on supporting various dimensions of trust.
Third, X ic is a set of baseline individual-level control variables: the age and gender of respondents; their years of education; their subjective income insecurity (higher values indicate higher insecurity); and the source of their income (wages, unemployment, social assistance, pensions, or investments). Descriptive statistics for all variables are shown in Table A.3 in the appendix. We also include γ c country-specific fixed effects while ε ic is the error term. We estimate our specification using the Ordinary Least Squares method and report robust standard errors, but also check for robustness to different error structures, fixed effects, and inclusion of trends and controls.
There are several potential threats to the identification of our design (Muñoz, Falcó-Gimeno and Hernández Reference Muñoz, Falcó-Gimeno and Hernández2020).Footnote 8 The first threat is non-compliance, which arises if the respondents are unaware of the treatment and could therefore be wrongly assigned to the treatment group. This is unlikely to be an issue in our case since the Russian invasion of Ukraine was admittedly one of the largest events of 2022. Ukraine was the world’s third most searched term on Google overall and number one in the news category. Furthermore, Russia attacked Ukraine at about 5:00 am CET, which is prior to the first ESS interview on 24 February 2022, and the invasion was widely reported in the morning news across European countries.
Second, to ensure the survey interview outcome is unaffected by the timing of the treatment, except through the event of interest, we conduct placebo and falsification tests. Specifically, we carry out placebo treatments at different dates of the control group, falsification tests on other outcome variables, and also examine whether other events had an effect; for instance, the Russian military troops arriving at the Belarus border and the escalation of the conflict in the Donbas region weeks before the invasion.
Finally, while the treatment is exogenous to respondents’ behaviours or views, it could be that for other reasons the treatment status of the respondent is correlated with our outcome variables. While the ESS ensures that the samples are representative and balanced overall in each country, individuals who are randomly interviewed before or after the Russian invasion might have different characteristics, which may themselves in turn affect responses to questions captured in our dependent variables. The following ignorability tests are recommended (Muñoz, Falcó-Gimeno and Hernández Reference Muñoz, Falcó-Gimeno and Hernández2020): balance tests, covariate adjustments, multiple bandwidths, and analysis of non-responses. We carry out balance tests and apply entropy balancing to achieve covariate symmetry between our treatment and control groups (Hainmueller Reference Hainmueller2012). The differences between control and treatment groups in our outcome variables are further investigated for eight different bandwidths, 10, 14, 21, 31, 40, 50, and 60 days, after the start of the invasion, respectively. Attrition could potentially bias our findings but survey collection rates are not themselves affected by the war, which minimizes the risk of attrition. We test whether nonresponse affects our results in Table F.1 in the appendix.
Results
Table 1 presents our baseline results using a 14-day bandwidth for four measures of political trust.Footnote 9 We find a statistically significant and positive effect of the Russian invasion on trust in politicians and political parties. However, with this bandwidth, there is no effect on trust in the national parliament or in the European Parliament. These results hold regardless of whether we include or exclude country fixed effects or controls, as shown by columns (1) to (6) in the table. When both country fixed effects and controls are included, the effect of the invasion on trust in politicians is approximately 0.207, equivalent to about 8.6 per cent of the standard deviation of this variable in the pre-treatment control group. Similarly, trust in political parties increases by 0.226, or about 9.5 per cent of one standard deviation of this variable. The effect of the invasion on these two variables remains statistically significant throughout and peaks at thirty-one days, whereas trust in the national parliament only becomes statistically significant for bandwidths of several weeks or more (Figure 1). By contrast, the effect on trust in the European Parliament stays non-significant throughout.Footnote 10
The effect of the invasion on trust
Note: We use a 14-day bandwidth in all specifications presented in this table. We apply entropy balancing for the control group in regressions presented in columns (2), (3), (5), and (6). Specifications (2), (3), (5), and (6) include country fixed effects, while we also apply the full set of controls in specifications (3) and (6). Control variables include age, gender, education, marital status, income difficulties, and source of income. Robust standard errors are adjusted at the individual level. Coefficients that are significantly different from zero are denoted by the following system: * p < 0.10; ** p < 0.05; *** p < 0.01. Full results are presented in tables B.1 and B.2 in the appendix.
The effect of the invasion on political trust across time.
Note: This figure plots the estimates of the effect of the Russian invasion on various dimensions of trust. Circles are OLS coefficient estimates from distinct regressions of each dependent variable on a dummy variable taking value one if respondents were interviewed after the start of the invasion, and zero otherwise, for different time bandwidths. All regressions include controls, country fixed effects, and entropy weights (see appendix for further description and summary statistics). Vertical bars are 90 per cent confidence intervals calculated with robust standard errors. Full results are presented in tables C.1, C.2, C.3, and C.4 in the online appendix.
We implement a range of tests to rule out threats to identification and carry out several robustness checks. First, responses to the trust questions could themselves pick up longer-term trends; for instance, if people over time become more trusting of politicians and political parties, irrespective of the start of the invasion. However, we show that the results are not affected by the inclusion of a trend (Table F2), while various placebo and falsification tests are reported in section F.1 in the appendix.
Second, we replicate our analyses for different bandwidths varying the number of days since the start of the Russian invasion. As can be seen in Figure 1 (full results are reported in C.1, C.2, C.3, and C.4 in the appendix), the effect of the invasion on trust in politicians and political parties is positive and statistically significant for all eight bandwidths, although it appears to peak after one month. This is consistent with the ‘opinion leadership school’ positing that the initial reluctance of the political and media elites to criticize political elites and institutions in times of crisis is short-lived, and over time, disagreements over the management of the crisis and evaluations of the policy response resume, hence shaping the duration of the effect (Hetherington and Nelson, Reference Hetherington and Nelson2003). By contrast, the effect of the invasion on trust in the national parliament has a similar inverted U-shape but is only statistically significant for the bandwidths 28, 31, 40, and 50 days (and borderline for 21 days).
Third, the main results do not depend on the fixed effect structure, the use of weights, which control sets are included, the estimation methods or standard error adjustment, or the exclusion of any one country. Results for different country fixed effects, weights and controls are shown in Tables B.3 to B.6 in the appendix. While we relied on the Ordinary Least Squares approach for all our estimations so far, the results are unchanged if we use instead Ordinal Logistic Models that could be argued to more closely match the ordinal scale of our dependent (Tables D.1 to D.4 in the appendix). In addition, all our baseline results report robust standard errors because clustering errors at the country level would not be asymptotically justified. As robustness checks, we opt for alternative clusters at the date (column 3 in Table F.3 in the appendix) and country-date levels (column 4). We also report results when using wild cluster bootstrapped t-statistics (column 5) following Cameron, Gelbach and Miller (Reference Cameron, Gelbach and Miller2008). The results are statistically significant throughout.
Fourth, we investigate the heterogeneity of our effect. At the country level, all our results so far included country fixed effects to absorb any unobservable time-invariant country heterogeneity. We nevertheless carry out a fully-fledged jack-knife stepwise country exclusion analysis for all bandwidths (see Table E.1 in the appendix). The effect of the Russian invasion on political trust is always significant except in the following exceptions: trust in the national parliament for the bandwidths of 14 and 21 days (which were also insignificant in the baseline using the full sample) when excluding Greece; and for the bandwidth of 60 days when excluding Macedonia, and trust in parties and politicians for the bandwidths of 14 and 21 days when excluding Greece. At the individual level, heterogeneity analyses are useful to help us rule out that the average effect is driven by small groups of individuals, and the presence of heterogeneity can shed light on causal mechanisms. Because it is important to ensure that the moderators cannot themselves be affected by the invasion (Muñoz, Falcó-Gimeno and Hernández, Reference Muñoz, Falcó-Gimeno and Hernández2020), we select age, gender, and income source (Tables H.1, H.2, H.3, H.4 in the appendix).
Overall, our findings provide strong and robust evidence of the invasion’s positive impact on trust in politicians, political parties, and in the national parliament. In section J.3 in the appendix, we explore different potential mechanisms. Using other surveys, google searches, and media mentions, we present descriptive evidence that the invasion increased threat perceptions, a mechanism linking conflicts to political trust. In addition, we find causal evidence that the invasion led to an increase in satisfaction with the government and with democracy. By contrast, despite potential negative economic shocks and refugee inflows, the invasion had no significant impact on subjective economic insecurity nor on perceptions of immigration’s impact on the economy, and instead led to higher satisfaction with the national economy.
Conclusion
The Russian invasion of Ukraine had significant, indirect economic and security repercussions across Europe, constituting a bundled treatment encompassing the initial attack, subsequent economic impacts, and governments’ policy responses. In this letter, we used the ‘Unexpected Event during Survey Design’ to test whether the Russian invasion of Ukraine had an effect on political trust. We find that this shock impacted political trust in European democracies that were close to the conflict. Although it is challenging to precisely quantify and systematically test the underlying mechanisms, a mix of descriptive and causal evidence using other surveys, Google searches, and media mentions all indicate heightened threat perceptions among the European public post-invasion.
Our results that the war increased trust in national politicians, political parties, and, in the medium term in the national parliament, are consistent with the ‘patriotism school’ where times of crisis and greater (perceived or real) threats lead people to rally around political actors and institutions in a show of national unity (Hetherington and Nelson Reference Hetherington and Nelson2003). Whereas the role of patriotism typically operates at the level of political community (Easton Reference Easton1975), which leads to higher support for the executive in the American politics literature, we find that this dynamic also applies to regime institutions; that is, at a more intermediate level of abstraction. However, the effect then culminated in the medium term, consistent with the ‘opinion leadership school’, which contends that political and media elites are initially reluctant to criticize political elites and institutions at times of crisis, but this unity is short-lived, and over time, disagreements over the management of the crisis and evaluations of the policy response resume, hence shaping the duration of the effect (Hetherington and Nelson Reference Hetherington and Nelson2003).
Whereas previous research mainly focused on belligerent countries, our article provides causal evidence that external threats also increase political trust in European countries that were not directly involved. This contributes to long-standing debates about the effects of wars, conflicts, and violence on domestic politics, most notably trust in, and support for, the incumbent, by showing that an attack on a geographically close democracy may raise political trust in other democracies, even if they are not directly involved.
Morever, our results raise the wider question of whether the political regime of the attacker and the attacked country, respectively, as well as the geographical proximity to the conflict, are necessary conditions to observe an effect on trust. In section K of the appendix, we explore this question in two ways. First, using previous waves of the ESS, we show that US past military interventions in Iraq and Syria, respectively, had no significant effect on trust in Europe. Second, we searched for other surveys where the fieldwork overlapped with the timing of the Russian invasion and found only two other surveys that met this criterion: using the world values survey, we find that the invasion had no effect on political trust in Uruguay; and using the Arab Barometer, we find it had no effect on trust in Jordan. These additional analyses corroborate the possibility that the effects of conflicts abroad can only spill over across borders to non-warring countries if they are geographically close enough and share a similar political regime to the country that is attacked. Taken together, these findings contribute to our understanding of the complex and indirect effects of conflicts on domestic political trust.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123424000838.
Data availability statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/ZNAECE.
Acknowledgements
For excellent comments and constructive suggestions, we are grateful to the five reviewers and the editor of the British Journal of Political Science.
Financial support
The authors declare none.
Competing interests
The authors declare none.
Open access
