Allied commitments and public support for military interventions: A cross-national experiment

Introduction

In world politics, states often seek to enhance their security by entering military alliances based on collective defense agreements (Berkemeier and Fuhrmann, 2018; Leeds et al., 2000, 2002). Recent survey experiments have provided new evidence on the micro-foundations of the credibility of allied commitments: the public is more supportive of intervening militarily on behalf of formal allies than non-allies, largely due to concerns about reputational costs of non-intervention (Tomz et al., 2023; Tomz and Weeks, 2021).

There is, however, an important limit to these latest experimental studies: so far, they have been conducted exclusively in the United States and other NATO member states. Consequently, we lack evidence on whether the effects of alliance treaties generalize beyond the Western context and the NATO alliance. Given the contemporary debates over multipolarity and power shifts in the world order, it is arguably crucial to understand how publics in non-Western regional powers perceive allied commitments (Acharya, 2025; Cooley and Nexon, 2020).

To address this gap, we preregistered a cross-national survey experiment. ¹ Building on the design of a foundational study by Tomz and Weeks (2021), the participants were randomly assigned one version of a vignette describing a fictional scenario of armed aggression against an unnamed country. We experimentally varied whether the victim of the attack had previously signed an alliance treaty with the participant’s own country. In turn, we asked participants whether they favored or opposed sending their country’s military to stop the invasion. This approach allowed us to investigate whether formal allied commitments would increase participants’ willingness to support military intervention to defend the attacked country.

We fielded the survey experiment to nationally representative samples in six countries (N = 7200). We selected Russia, China, India, and Brazil as the four prominent non-Western regional powers. For a cross-national comparison with Western, NATO countries, we additionally included the United States and the United Kingdom.

Our results show that while allied commitments increase public support for military interventions globally, their effect is comparatively weaker in non-Western, non-NATO countries. We also identified potential heterogeneity within the non-Western camp, with the effect of formal military alliances being somewhat stronger in Russia and, to a lesser extent, in China, than in Brazil and India. The cross-national differences, however, were substantively rather small, and some of the formal tests of our preregistered hypotheses therefore failed to reach the threshold of statistical significance.

We proceed as follows. First, we outline our theoretical expectations and hypotheses. Second, we describe the design of our cross-national experiment. Third, we present the results of our preregistered analyses. We conclude by discussing the broader implications of our findings and directions for future research.

Hypotheses

In their original study conducted in the United States, Tomz and Weeks (2021) found empirical support for the claim that military alliances increase public support for war to defend victims of aggression. We can formalize this as hypothesis H ₁ that the public is more likely to support a military intervention on behalf of a formal ally than a non-allied country. To investigate whether this claim generalizes beyond the specific U.S. context, we will follow the approach of Bassan-Nygate et al. (2024) and test H ₁ empirically in:

(a) Pooled sample including participants from all countries in our experiment (cross-national meta-analysis) and

Our study includes six nationally representative samples: two Western/NATO countries (the United States and the United Kingdom) and four non-Western regional powers (Russia, China, India, and Brazil). We expect that, even if military alliances exert an observable “pull effect” across countries, publics in the United States and other NATO countries have internalized the collective defense obligations more deeply. This is primarily due to the extent to which NATO and other formal alliances have been central to the national security strategies of these countries since the early Cold War until today, with high levels of formal institutionalization, active promotion of distinct norms, and constitutive effects on those countries’ identities (Gheciu, 2005; Risse-Kappen, 1996). As such, we preregistered hypothesis H ₂ that the effect of formal alliances on public support for military intervention is stronger in NATO countries than in non-NATO countries.

One might, however, argue that among the remaining four non-Western countries in our case selection, Russia also has a distinct experience with formal alliance treaties. While the Warsaw Pact was a predominantly top-down security arrangement directed from Moscow and dissolved at the end of the Cold War (Kramer, 2025), Russia is a founding member of the Collective Security Treaty Organization (CSTO), a military alliance of post-Soviet states with a collective defense pledge. In contrast, China, India, and Brazil have little to no experience with membership in formal collective defense arrangements. As such, we also preregistered hypothesis H ₃ that the effect of formal alliances on public support for military intervention is stronger in Russia than in China, India, and Brazil.

The final theoretical expectation concerns the heterogeneity within the Western camp. Throughout both Trump administrations, we have observed repeated contestation of the U.S. commitment to NATO’s Article 5 (Schuette, 2021; Sukin et al., 2025). It is conceivable that at least part of the American electorate will now question the value and relevance of military alliances to U.S. foreign policy and national security. We have, however, not observed comparable developments in British domestic politics. As such, we formulated hypothesis H ₄ that the effect of formal alliances on public support for military intervention is stronger in the United Kingdom than in the United States. ²

Experimental design

The design of our cross-national survey experiment employs a text-based vignette originally used in the U.S. study by Tomz and Weeks (2021). ³ After participants provided consent for the study and responded to a battery of sociodemographic questions, we randomly assigned them one version of a fictional scenario describing armed aggression against an unnamed country. We experimentally manipulated four vignette attributes in a 2×2×2×2 full-factorial design: ⁴

(A) whether the victim of the aggression had a military alliance agreement with the participant’s country (no alliance/alliance);

After reading the scenario, participants expressed their level of support for sending their country’s military to stop the invasion on a five-point Likert scale. Responses to this item serve as the main dependent variable in our study (support for military intervention). ⁵

We formally preregistered our analysis plan prior to collecting the data (see Appendix 1) and obtained ethics approval from the Research Ethics Committee of the Faculty of Social Sciences, Charles University.

We fielded the experiment to nationally representative population samples between June and August 2025. Due to the difficulty of conducting academic surveys on sensitive political issues in some of the countries in our case selection (particularly Russia and China, but to some extent also India), we were unable to find a single pollster who could administer the survey across all selected countries. Consequently, our preregistered sampling plan envisioned working with several different internationally recognized polling companies with access to suitable survey participants in the selected countries: IPSOS (the United States and the United Kingdom), Dynata (India and Brazil), Qualtrics (China), and Levada Center (Russia). Our final pooled sample consisted of 7200 observations with a minimum of 1000 participants per country. ⁶

Results

Following our preregistered plan, we first estimated the average alliance effect in the pooled cross-national sample. We conducted an ordinal regression analysis with support for military intervention as the dependent variable, alliance as the independent variable, and participants’ gender, age, education, and nationality as control variables.

As shown in Figure 1, whether the victim of an armed aggression had a collective defense agreement with the participant’s country was positively associated with the willingness to support military intervention. Consequently, in a cross-national meta-analysis, we found clear empirical support for hypothesis H ₁ that the public is more likely to support a military intervention on behalf of a formal ally than a non-allied country.OPEN IN VIEWER

Figure 1 also shows the results of H ₁ tests in individual national samples. ⁷ The observed effect of alliance agreements was consistently positive across countries, providing support for the cross-national sign-generalizability of the original claim. However, note that the effect size is smaller in China, India, and Brazil than in the United Kingdom, United States, and Russia, and it is not significantly distinguishable from zero among participants from India and Brazil. The visualization in the upper panel of Figure 1 nevertheless suggests that, at least for China and India, the limited impact of allied commitments is not necessarily a function of low willingness to assist allies, but rather a relatively high willingness to intervene militarily even in the “no-alliance” scenario.

Following our preregistered plan, we subsequently built another model with the interaction between alliance and the distinction between participants from NATO and non-NATO countries as a predictor. Figure 2 shows a statistically significant interaction: participants from NATO member states were relatively more likely to support military intervention on behalf of a formal ally than of a non-allied country. As such, we found empirical support for hypothesis H ₂ that the effect of formal alliances on public support for military intervention is stronger in NATO countries than in non-NATO countries.OPEN IN VIEWER

Similarly, for the respective national sub-samples, we used the interaction between alliance and the distinction between participants from Russia and other non-Western countries (China, India, and Brazil) as a predictor. The aim of this model was to test hypothesis H ₃ that the effect of formal alliances on public support for military intervention is stronger in Russia than in China, India, and Brazil. Figure 2 shows the interaction effect in the predicted direction, with the attitudes of Russian participants toward military intervention being more strongly affected by an alliance agreement. However, the statistical test did not reach the threshold of statistical significance. It is also worth noting that in the Russian sample, we observed the lowest baseline level of support for the intervention in our study. This arguably provided more space for the allied commitments to increase participants’ attitudes than in the Chinese and Indian samples, where participants displayed relatively high support for intervening even in the absence of prior allied commitments.

The final comparison, according to our preregistered plan, concerned the NATO sub-sample of American and British participants. This test concerned the hypothesis H ₄ that the effect of formal alliances on public support for military intervention is stronger in the United Kingdom than in the United States. Although we found the positive effect in line with our theoretical expectations (see Figure 2), the difference in point estimates was not statistically significant.

We provide the detailed results of all the preregistered statistical analyses in the Supplemental Appendix 5 (hypothesis H ₁ ) and 6 (hypotheses H ₂ –H ₄ ). In Appendix 10, we additionally provide a power analysis for all preregistered statistical tests and discussed the likelihood that the failure to find a statistically significant interaction for H ₃ and H ₄ is due to a sample size that is too limited to detect a substantively small effect. Finally, in Appendix 9, we analyze and discuss the differences in effect sizes between our experiment and the original study by Tomz and Weeks (2021).

Discussion and conclusions

In this paper, we provided the first experimental evidence that formal alliance commitments shape public support for military interventions globally. At the same time, our study shows that in line with our theoretical expectations, the effects of military alliances on public attitudes were weaker in non-Western, non-NATO countries.

Although more research is needed to investigate the causes of this phenomenon, we speculate that this is primarily due to the extent to which NATO and other formal alliances have long been central to the national security strategies of Western countries. Unlike the CSTO or other existing collective defense treaties, NATO exhibits high levels of formal institutionalization, actively communicates its mission to the Western publics, and is often closely linked to the identity of its member states (Gheciu, 2005; Risse-Kappen, 1996). Another possible explanation is the domestic institutional design of liberal democracies, which makes reneging on international agreements more difficult and increases leaders’ accountability for broken promises. Both these factors may contribute to earlier findings of observational studies that democracies are more reliable allies than non-democratic countries (Leeds, 2003).

However, we underscore that our cross-national study shows a more nuanced pattern of heterogeneity than the simple NATO/non-NATO or Western/non-Western divide. Differences in estimated effect sizes between countries are modest (and often not statistically distinguishable), but they follow a clear ordering: the largest alliance effect appears in the United Kingdom, followed by the United States, Russia, and China, with the smallest effects in India and Brazil. These cross-national differences may reflect, among other factors, the recent shifts in the U.S. approach to NATO (Pinto and da Silva, 2025; Sukin et al., 2025) and the greater salience of Russia and China in contemporary geopolitics relative to India and Brazil. Furthermore, we would like to highlight meaningful differences in baseline support for military interventions across countries, with Russian citizens particularly averse to defending the attacked country under any condition, and Chinese and Indian citizens relatively supportive of military action even when the victim was not a formal ally. Future researchers could build on these findings to investigate the sources of this cross-national heterogeneity.

Our study complements and further extends the new wave of IR scholarship that employs survey experiments on population samples to investigate the microfoundations of military alliances and collective defense (Alley, 2022; Tomz et al., 2023; Tomz and Weeks, 2021; Weckman, 2023; Xu et al., 2023). ⁸ Building on the literature on domestic audience costs (Fearon, 1994; Tomz, 2007) and elite responsiveness (Chu and Recchia, 2022; Lin-Greenberg, 2021; Smetana et al., 2025; Tomz et al., 2020), this body of IR scholarship convincingly argues that understanding public preferences regarding the use of force is crucial for our understanding of elite decision-making in collective defense scenarios.

At this point, the reader might wonder to what extent these theoretical claims about the role of public opinion apply to the two countries in our study pool that do not pass as democratic states: Russia and China. We contend that the corresponding role of the public in autocratic settings is severely limited, particularly due to the inability to express disapproval with foreign policy in free and fair elections. However, the latest research on autocratic audience costs (Li and Chen, 2021; Quek and Johnston, 2018; Smetana, 2025; Weiss and Dafoe, 2019) and autocratic accountability (Chen et al., 2016; Horne, 2012; Kinne, 2005; Meng et al., 2017; Truex, 2016) shows that even in non-democracies, the public is willing to express disapproval of unpopular foreign policy and autocrats are surprisingly responsive to public preferences in this domain.

More generally, our study contributes to the broader scholarly efforts to investigate the cross-national generalizability of IR experiments, which are still predominantly fielded to U.S. population samples (Bassan-Nygate et al., 2024; Levin and Trager, 2019). Future research could further extend our work by surveying the populations in regions where military alliances could play a prominent role in regional security affairs, such as the post-Soviet space, Indo-Pacific, and the Middle East.

Supplemental material