Are ideological and partisan affinities determining voters’ support of arms deliveries? Insights from a large-scale survey experiment in France and Germany

Abstract

The 2022 Russian invasion of Ukraine raised for many parties the question of how to position themselves in view of urgently requested arms deliveries. Since, the topic of arms trade, which has hitherto rarely been addressed, has become a heavily politicized and divisive issue and partly even polarized public opinion. A major prerequisite for parties’ position-taking is to anticipate how voters react to such arms transfers and, more specifically, whether their respective attitudes are structured along the predominant left-right axis. Based on a large-scale survey experiment with French and German voters ($N = 6617$) in the year before the Russian invasion, we are able to focus on the relationship between ideological predispositions, vote intentions, and issue attitudes in a non-politicized period. Using both vignette and conjoint experiments, we demonstrate that voters’ attitudes on military transfers can be subsumed remarkably well under the left-right scale. Differentiating the impact of normative and economic considerations, the former is stronger among the left, while the latter also affects the attitudes of rightist citizens. However, normative considerations are the most important concern along the whole political spectrum. The turn of the German Green Party in 2022 to assist countries that are being aggressively attacked (because of the Responsibility to Protect), was not reflected in our data.

Introduction

Transfers of major conventional weapons are only one part of governments’ manifold policies, at least if the country is able to produce for the wider world market and to export or provide military aid. Regularly, transfer decisions go unnoticed by the wider democratic public: their overall macroeconomic relevance is usually relatively small, while governments keep decisions low profile due to their oftentimes classified nature and in order to avoid embarrassing discussions on conflicting normative implications. Particularly, if receiver countries have a dubious reputation (e.g., violating human rights, being actively involved in a conflict) elected governments try to downplay arms exports or keep them completely under the radar of public debate. Nevertheless, there are events that tear up the veils of discretion suffusing arms transfers. For example, the infamous murder of the journalist Jamal Khashoggi in 2018, presumably commissioned by the, at that time, Saudi Arabian Minister of Defence Mohammed bin Salman, led to widespread protest against follow-up arms transfers to this regime in a series of Western provider countries. However, the actual governmental behavior in Germany on the one hand and in France and the United Kingdom (UK) on the other hand then contrasted quite heavily: Germany banned any delivery to Saudi Arabia, while the UK and France continued to supply and pressured Germany to provide product components for fighters destined to Saudi Arabia.¹ Even more prominently, the Russian invasion of Ukraine in 2022 with the successive military support by many countries put the topic of arms exports high up on the agenda since then, raising pro-support as well as protest campaigns against weapons supply in Western countries.² In Germany, the governing Green Party, originally being known as the issue owner of anti-militarism, now took a strong stance for supporting Ukraine, together with its coalition partner, the liberal Free Democratic Party (FDP). Contrarily, the Social Democratic Party (SPD) with its chancellor Olaf Scholz acted quite reserved due to conflicting intraparty factions.³ As of 2023, parties that position against supplies in Germany consist of an alliance between the far-right party Alternative for Germany (AfD) and the far-left party The Left – which has even had a split, among other reasons because of a vocal anti-supply faction.⁴ A similar picture can be observed in France, where the party leader of the right-wing Rassemblement National (RN), Marine Le Pen, is strictly against weapons supplies to Ukraine,⁵ just as the far-left Communists and La France Insoumise.⁶ In the USA, parts of the usually arms trade supporting Republican Party meanwhile openly criticize military assistance to Ukraine.⁷ Such (anecdotal) evidence indicates that (a) public opinion reacts stronger in some countries (e.g., Germany) as compared to others (e.g., France, Spain, USA, UK), (b) that there appear quite surprising coalitions of far-right and far-left parties/candidates, (c) that parties originally taking strong anti-militaristic stances now support military assistance to Ukraine, (d) that factions within parties even split off partly based on this issue. This raises the question to which extent voters’ and parties’ positions converge or diverge on arms trade, how voters along the political spectrum form preferences on arms trade, and how important the issue is for their vote choice. It is usually argued that parties learn from elections and surveys about voters’ issue preferences (Wlezien and Soroka 2007; Soroka and Wlezien 2022). However, before the war in Ukraine, there were near to no existing surveys, even less time series, or survey experiments (a rare exception is, e.g., Efrat and Yair (2024) for an Israeli sample) on attitudes toward arms transfers.⁸ Only with the looming onset of intensified war in Ukraine in 2022, numerous surveys investigating attitudes toward arms transfers emerged (Hoffmann and Schmidt 2024). The lack of such surveys is not a sufficient but at least a necessary indicator that there was no party competition on this issue. This corroborates our initial statement about the absence of this topic on the established issue agenda,⁹ even in top-exporting democracies. In such a setting it is therefore highly important for parties and candidates to integrate newly appearing issues into pre-existing ideological dimensions – as these facilitate political debates in the sense of informative heuristics, which cognitive misers can rely on (see Lupia 1994). The research questions we pose in this article therefore are: do citizens on the political left and right show different attitudes toward arms exports? Why is this the case, that is, how do voters perceive and weigh the economic, normative, and strategic implications of arms exports? And is the issue of arms trade relevant to citizen’s voting decisions?

Empirically, we use data ( $N = 6617$ ) from a quota-representative sample of French and German citizens just before the Russian invasion of Ukraine. Hence, our paper contributes from the perspective of a non-polarized situation. For this time period, we investigate to what extent attitudes of the electorate toward arms transfers consistently relate to established ideological predispositions and party affiliations, or whether this new issue runs counter to well-known linkages. More concretely, are voters to the right more inclined toward arms transfer and trade, as compared to leftist voters? And second, are anti-militaristic left parties consistently attracting those voters refusing arms transfers and rightist parties consistently attracting those voters strongly backing these transfers? Or do we observe a heterogeneous composition of issue attitudes within parties? In the following we provide a theoretical rationale for contrasting experimentally elicited arms transfer attitudes, the left-right dimension, and vote intentions/party identifications (PID). Then, we introduce our data and research design before presenting and discussing our results.

Theoretical considerations

What are the attitudes of voters of different parties and in different countries regarding the policy area of arms trade? Which specific aspects do they consider? With the onset of the Russian aggression in Ukraine, weapons transfers became prominent on the agenda. But given their wider economic and strategic implications, and in view of an emerging European Union defense sector and joint arms proliferation policies, it is of high practical relevance for parties and candidates to know how voters react to such policies more generally. The complicated amalgamation of sometimes contradictory considerations makes it difficult for parties to understand the reactions of voters. Do they respond in accordance with prefabricated ideological cues like the predominant left-right axis (Downs 1957; Lupia 1994)? Or do we see diverging ideological and partisan alignments that would point to an alleged independent dimension like cosmopolitanism (Kriesi et al. 2006; Rovny and Polk 2019)? Therefore, we propose to investigate for the first time systematically whether the ideological left-right positions of voters are related to their opinion on weapons transfers.¹⁰ We also analyze whether these attitudes differ between adherents of different parties in a non-mobilized context where the issue was not acutely on the agenda, namely, in 2021 in Germany and France. We focus on this time period in order to mimic as closely as possible the uncertainty of parties when having to take positions and to explain their positions.

We start with the observation that the issue of the arms trade is an upcoming new issue, even a complicated hard issue (Carmines and Stimson 1980), characterized by subdimensions. Hard issues are not permanently on the agenda, they require information and intellectual examination. Arms transfers are such a hard issue: before the war between Russia and Ukraine, they surfaced prominently only from time to time. And they are complicated insofar as they combine subdimensions: easy-to-assess economic considerations (jobs and national income) and complicated geostrategic reasoning (the transferred weapons may lead to less security and stability and in the end damage one’s country’s interests).¹¹ Especially, geostrategic arguments are often connected to legal and regulatory considerations. For example, arms transfers to repressive systems or to countries involved in domestic or international conflict are forbidden under several national and supranational stipulations (e.g., the German War Weapons Control Act, the EU Council Decision (CFSP) 2019/1560, the UN Arms Trade Treaty). Such normative considerations then allow to insert complicated geostrategic reasoning into easy-to-grasp aspects and are therefore frequently mobilized, for example, in German public debates.

Hence, our theoretical starting points are threefold: first, we take the perspective of the neo-Downsian spatial theory of ideology (Hinich and Munger 1994; Enelow and Hinich 1984) and combine it with factorial experiments. The neo-Downsian theory of ideology posits that the diverse concrete issue positions of candidates/parties map onto one or multiple ideological dimension(s). Despite us following their general theoretical conception of ideology, consisting of a coherent portfolio of policy positions, we emphasize a different challenge for parties, namely, to come to coherent positions within issues. As outlined above, complex new issues often consist of a combination of subissues. To find a consistent position on such composite issues can constitute a major problem for parties. For example, when parties strategically introduce new issues (Hinich and Munger 2008; Baumgartner and Jones 1993), it is unclear which sub-aspects voters consider and which positions therein they prefer. Therefore, we propose to focus on the within-issue trade-offs and contrast them with the existing predominant ideological left-right dimension. For this, we apply a multi-attribute factorial design. By definition, these combine several dimensions or, as in our case, subdimensions, which have to be related by individuals to some higher-order choice dimension (for multi-attributive utility theory and multi-criteria decision-making, see (Keeney and Raiffa 1993)). The subdimensions of our factorial design correspond to more or less contested within-issue aspects.

Second, we propose that transfers and exports of major conventional weapons constitute public policies that are usually not politicized and therefore are not part of established issue agendas and issue competition in political systems. A major corroboration of this assertion can be found in the Manifesto Project, which codes quasi-sentences in party and electoral manifestos worldwide. Actually, there is no explicit coding of weapons transfers or exports per se, but rather of general attitudes to the military and peace.¹² Zooming onto our country cases of Germany and France, we provide a more systematic assessment (for 2013 to 2021). In the case of French party manifestos, we observe a lack of agenda prominence of these issues almost among all parties (one relevant exception: French Green Party). For Germany, some left-leaning parties (Green Party, Left, SPD) allocate lengthy passages to the issue, opposing arms exports. On the contrary, parties from the right political spectrum (the conservative Christian Democrats (CDU/CSU, jointly called Union), FDP, and AfD) either ignore the topic or only selectively highlight specific aspects. The online Appendix Section A.1.2 provides more details. In sum, the issue is rarely dealt with, and if so, it is rather in leftist parties. They then position against armaments and the proliferation of arms – in seeming accordance with their anti-militaristic reputation.

This last observation leads us, third, to expect a seamless and consistent insertion of anti-export attitudes to the left-right dimension, once these are emphasized. Note that this insertion should be more prominent in Germany compared to France for two reasons. First, the topic is dealt with more frequently and with more attention attributed to the former case. Second, the French party system experienced relevant reshuffling running counter to the traditional left-right dimension. Therefore left-right meanings may be not as helpful as an anchoring device compared to Germany (Angenendt and Kinski 2022; Dostal 2021).¹³

An important follow-up question is how such attitudes translate or relate to voting intentions. In a situation of non-politicization, voters cannot reason about the explicit position-taking of parties. Therefore, at the time of our survey, they reveal party affiliations without factoring in the salience of the issue and the parties’ stances. Our analysis should thus give us an impression about possible mismatches between voters’ attitudes and parties’ postwar position-taking.

Finally, in comparison with prominent studies focusing on the use of military force (e.g., Dill and Schubiger 2021; Tomz et al. 2020; Kertzer and Zeitzoff 2017), arms transfers constitute much less abstract and hypothetical situations for citizens – at least for citizens in most democratic middle powers that our case selection exemplifies, that is, publics outside the USA and outside the few democracies in active conflict, such as Israel. So on the one hand, our case directly relates to the potential use of force and hence ties to a broad literature that investigates whether and how citizens constrain governments’ high politics. But on the other hand, we select an issue that occurs with relevant frequency and is, at least at times, politically contested. This contributes to these questions being directly and intuitively comprehensible by citizens, increasing the external validity of our results (Findley et al. 2021).

Research design

The following section presents our research design. A pre-registration for the research design of this study can be accessed at https://osf.io/uewrt/ , and the questionnaire can be accessed at https://osf.io/kn59j/.¹⁴

Survey

We fielded a population-representative quota survey with 6617 respondents from Germany ( $N = 3250$ ) and France ( $N = 3367$ ) (field time 10/2020–01/2021) via the survey company Kantar.¹⁵ Given that speeding and satisficing are relevant concerns in such panels, Kantar excluded respondents with a survey time lower than 40% of the survey time median during data collection. To prevent that inattentiveness biases results, we excluded additional problematic cases ( $N = 514$ ), that is, straight-liners and super-speeders in standard item batteries and super-speeders in all conjoint tasks.¹⁶ Our sample is quota-representative for the German and French adult population (voting-eligible citizens $ \gt 18$ ) with respect to age, education, gender, and region (German Bundesländer/French régions métropolitaines). Deviations from quota targets were minimal.¹⁷ This makes us confident that our respondents more generally represent the German/French population and that our results convey external validity. Note that, in addition, we have relevant shares of left and right-wing voters from all major parties in our survey.¹⁸ Finally, the procedure and survey instrument were approved by the Ethics Commission of the Social Science Faculty of LMU Munich.

Experiment

At the core, our survey consists of two vignette experiments (Mutz 2011) and a conjoint survey experiment (Hainmueller et al. 2014). These serve to capture the nuanced preferences of respondents vis-à-vis arms exports and their vote intentions.¹⁹

Vignette experiment on arms trade policy: In view of the rare public debates on arms exports in the pre-Ukraine war period, we expect attitudes on arms exports to depend mainly on the normative context of the recipient country and the economic value for the sending country. To capture first-order preferences along these two dimensions and inquire why respondents voice these preferences, we provide for a stylized vignette experiment. Table 1 presents the wording of both dimensions.

Table 1. Wording of the two arms trade policy vignette dimensions (translated to English)

Dimensions were displayed in uniform random combinations to respondents. The sequence of the two vignette sentences was randomized, also, to preclude any ordering bias. Subsequent to the vignette display, respondents were presented with a binary choice whether they “agree” or “not agree” with such an arms transfer to take place.²⁰

While the latter question allows us to capture preferences on whether the economic and normative features of an arms transfer affect citizen’s agreement, the core of our research design is to inquire why respondents voice these preferences. To this end, we fielded questions tailored to investigate the reasons why respondents agree/disagree with an arms transfer. On a 7-point Likert scale (1: not agree at all; 7: agree strongly), we asked a battery of mechanism questions. These asked respondents to rate the arms transfer displayed with the vignette as (not) being:

• bad for moral reasons.

• bad for the security of [Germany/France].

• bad for the security of the world.

• good for the people in [Germany/France].

• good for the people in the recipient state.

This combination of a two-dimensional factorial survey experiment with subsequent questions to understand mechanisms for choice was inspired by Rudolph et al. (2023).

Conjoint experiment on arms trade policy: To capture the more nuanced preferences of respondents, respondents also replied to a conjoint experimental task. Here, just as with the vignette experiment, we introduced respondents to a scenario in which they were to assess hypothetical arms transfers, taking care to avoid any positive or negative framing effects when introducing the topic.²¹ We then confronted respondents with six paired arms export profiles, on which they performed choice and rating tasks (using a 7-point scale). Profiles contained nine attributes with 2–5 uniformly randomized attribute levels. Online Appendix Figure A.1 exemplifies such a choice task. With $6$ conjoint tasks $K$ consisting of two profiles $J$ we have $N \cdot K \cdot J = 79,404$ observations on the profile level. Extensive pre-registered power simulations motivated this sample size.

The arms exports presented to respondents differed in nuanced ways for economic implications for the home country (domestic welfare; employment; economic trade partnership), the normative context in the recipient country (levels of democracy; human rights situation; ongoing conflicts), as well as additional contextual characteristics (alliance status with the recipient; type of arms exported; arms trade behavior of third countries). Table 2 presents an overview and online Appendix Section A.2.2 gives additional explanations on the attributes and their levels.²²

Table 2. Attribute and level wording of conjoint experiment

Note: English translation of the conjoint attributes and levels. For original wording, see the master questionnaire (pre-registered at https://osf.io/kn59j/).

Vignette experiment on voting behavior: Last, we included a split-sample vignette experiment to understand how voters perceive the positions of parties on the issue of arms trade and on their perception of how important a party’s arms trade policy is for their voting decision. First, we inquired importance for voting – “Besides arms exports, there are many other political issues. How important or unimportant is it for you, when making your voting decision, what the stance of party [party name vignette] on arms deliveries is?” (response scale 1 (not important at all) to 7 (very important)). Second, we inquired about perceived party positions on arms trade – “Do you think the party [party name vignette] wants to limit arms deliveries more or less compared to today?” (response scale 1 (restrict much more) to 7 (restrict much less). Importantly, we randomized the naming of parties in this question; hence, some respondents assessed the party they actually intended to vote for, while others did not. This corresponds to 6 (Germany) and 8 (France) one-dimensional vignettes. Online Appendix Tables A.5 and A.6 contain the corresponding distributions of respondents’ party preferences and party names administered in the survey questions.

Randomizing the party names serves two purposes: first, concerning the perceived policy positions, we reduce the response burden as respondents only have to assess one as compared to multiple parties. Second, we can directly compare the experimental subgroup that assessed their preferred party, compared to all other respondents. This gives us a causal indication of whether partisans show different evaluations compared to the rest of the citizenry, as our approach prevents survey-design-induced spillovers of assessments of one party to the other and consistency bias regarding the importance of policy positions for voting, akin to social desirability. Hence, compared to a standard approach asking respondents to rate all parties, respondents could here not anchor their responses for party $B$ in relation to party $A$ (and correspondingly for parties $C,D, \ldots $ ) and estimates derive from between-respondent variation only, which implies that there can, by explicit design, be no spillovers. As we use the information on respondents’ PID to let a random 50% rate their own and the remaining 50% rate another party, we have sufficient statistical power to assess whether a party is assessed differently by partisans compared to nonpartisans for most parties (we exclude parties that received less than 5% of the vote due to sample size considerations).

Political ideology and party vote intentions: Left-right ideology was captured with a standard 11-point Likert scale. This scale was split into tertiles of left, center, and right for ease of analysis. 38.0% of respondents perceived themselves on the political left (rating 1–5), 34.1% of respondents in the center (rating 6), and 27.9% of respondents on the political right (rating 7–11). Shares of left- and center-leaning respondents, compared to right-leaning respondents, are slightly higher in Germany (with 40.5% scoring left, 37.3% center, and 22.2% right) compared to France (with 35.5% scoring left, 31.1% center, and 33.4% right).²³ Party vote intention was captured via a hypothetical election question if legislative elections (first round for France) were held next Sunday.

Strengths and limitations of the design: The survey-experimental approach we apply allows us to relate experimenter-induced variation in question wording to differences in answering behavior in a causal way (Hainmueller et al. 2014; Mutz 2011). This improves over a number of known concerns from surveys eliciting population preferences (Stantcheva 2023). However, first, our factorial experiments are stylized miniatures of real-world scenarios. Hence, respondents could read unmentioned features into certain attribute expressions (“masking”) (Dafoe et al. 2018). We explicitly approach this challenge in the nine-dimensional conjoint experiment, preventing the masking of the attributes on democracy/human rights compliance and monetary value/job creation by, for example, conflict situation, alliance status, or weapon harm potential. We thereby also aimed to induce the mindset of an “average scenario” for respondents when they subsequently entered the vignette experiment. Online Appendix Table A.3 shows compelling evidence that this is the case: comparing respondent decisions in the vignette tasks to decisions in the conjoint task that exactly mimic the vignette scenario, while randomly varying all other attributes, leads to nearly identical choice behavior.

Second, factorial experiments can exhibit social desirability bias by emphasizing obtrusive attributes (in our case, e.g., support for autocracies for monetary gain) (Mutz 2011). However, recent research (Horiuchi et al. 2022; Auspurg et al. 2014) shows that social desirability bias is likely of low concern in factorial experiments, at least compared to standard survey items, and even for groups most likely to exhibit these biases. We might be most worried that the vignette experiment could suffer from such bias given it exhibits only two prominent and salient dimensions. However, also here, the comparability of conjoint and vignette results alleviates these concerns (see online Appendix Table A.3).

Last, a relevant limitation of survey experiments concerns their “ecological validity,” that is, the question of whether respondents’ hypothetical choices in a low-consequence environment travel to their real-world decision-making. However, related research by Hainmueller et al. (2015) shows that survey-experimental results can mimic consequential real-world choices, in their case also for a topic prone to social desirability in standard survey questions. Additional details on this discussion are presented in online Appendix Section A.2.4.

Estimation: We base our analysis of both the vignette and conjoint experiment on average marginal component effects (AMCEs, Hainmueller et al. (2014)) and marginal means (MMs, Leeper et al. (2020)). AMCEs and MMs are estimated based on saturated linear regression models, with robust/respondent-clustered standard errors. We capture interactions with political ideology/party affiliation by subgroup analysis or by including interaction terms (i.e., average marginal component interaction effects or differences in marginal means). We estimate AMCEs for the vignette and for the conjoint choice task for all respondents. This allows us to arrive at population-level marginal shifts in public opinion within subgroups of ideology with changes in attribute levels.

As our analysis is based on experimental variation, we report causal estimates for average differences in preferences with different vignette or conjoint attribute expressions, overall or within subgroups of political ideology. When speaking of the statistical significance of differences between treatment groups, we follow the current convention of labeling findings with $p \lt 0.05$ as “statistically significant,” being aware of the shortcomings of such thresholds (Wasserstein and Lazar 2016; Gill 1999), where prominent calls have been made for much stricter cut-offs (Benjamin et al. 2018). Given recent suggestions of a substantive interpretation of coefficient differences based on a continuous understanding of statistical significance (McShane et al. 2019; Greenland et al. 2016), we therefore also report absolute p-values, to allow for maximum transparency.

Results

Vignette experiment on arms trade policy

Normative and economic aspects structure preferences toward arms export for the political left and right: In the first step, we assess how the attributes of democratic versus autocratic regime type and high versus low economic value affect approval for arms exports on the political left, center, and right. Figure 1 presents the corresponding subgroup means of support for an arms trade by political ideology and vignette conditions and online Appendix Figure A.2 the respective regression coefficients (AMCEs).

Figure 1. Group means for support for an arms trade by political ideology (right vs. center vs. left) and vignette dimensions. Predictions based on regression of arms trade approval on triple interaction of respondent ideology (left, center, and right panel of the figure), vignette dimension autocratic (0) versus democratic (1) recipient country context (x-axis) and vignette dimension trade with low (green coefficients) versus high economic value (orange coefficients). 95% confidence intervals from robust standard errors shown. See online Appendix Figure A.2 for regression coefficients.

First, as can be seen from the comparison along the x-axes between autocratic and democratic contexts, the arms trade is on average strongly rejected for the former and approved for the latter contexts by a majority of citizens, irrespective of ideology. Notably, however, baseline support for trade is much higher for respondents on the political right compared to respondents with centrist or, even more, leftist leanings. This differential is larger for democratic contexts (around 80%, compared to around 60%, so a left-right differential of 20 percentage points) compared to autocratic contexts (around 25%, compared to around 13%, so a left-right differential of only around 15 percentage points). Hence, respondents on the political right react, despite higher baseline support for trade with autocratic contexts, more strongly to the democratic vignette. The vignette conditions communicating a high economic value of the deal do not move respondents on the political left – whereas support increases for respondents on the political right and substantially so (only) for trade with autocratic contexts (by 8 percentage points, difference significant at the 1% level). Hence, citizens on the political right seemingly perceive a trade-off by which high economic value can overcome the inclination to support arms trade with autocratic regimes. For respondents with a centrist leaning, we observe a nuanced picture by which support increases with high-value deals to autocracies (by 6.6 percentage points, difference significant at the 1% level) while decreasing for trade with democracies (by 5.5 percentage points, though this difference is insignificantly estimated with $p = 0.0501$ ).

So overall, and first, support for arms trade is around 15–20 percentage points higher for respondents on the political right compared to the left. Second, the normative context is strongly structuring attitudes, irrespective of political ideology, with support increasing substantively with trade to democracies. Third, economic considerations resonate with the political right and center, but not with the political left.²⁴

Material and immaterial consequences of arms trades are weighted differently but also perceived differently along the political spectrum: In this section, we exploit a battery of questions on the consequences of arms trades to investigate why citizens attitudes toward arms transfers and vignette conditions are structured differently along the spectrum of political ideology. To this end, we inquired how arms trades are perceived along three clusters of consequences (moral repulsion; security repercussions for home country and world; welfare repercussions for home country and receiving country). As presented in Figure 2, reporting marginal means overall and by ideology, we see, as expected, that respondents on the political left indicate higher moral repulsion, higher worry for security implications, and worse welfare to any arms trade compared to those on the political right – reflecting their generally higher skepticism regarding arms transfers in general.

Figure 2. Mechanism battery overall and by political ideology. Marginal means calculated from predictions from linear regression of agreement with statement in panel header (on a 7-point scale; panels 4 and 5 reverse coded) on two vignette dimensions (recipient country autocracy/democracy; arms trade of low/high value). Corresponding AMCEs are displayed in the online Appendix Figure A.4.

In a next step, we investigate differences in these marginal means, that is, changes in the spread of marginal means, under differing vignette conditions. We use trades with autocracies and low-value deals as baseline. Online Appendix Figure A.4 displays these differences overall and within subgroups of ideology.

First, how do consequences that citizens expect from arms exports differ by vignette attributes? Here, interpreting the black coefficients (circle) for the pooled sample, we find that exports to autocratic contexts are perceived as more morally repulsive, worse for own country and world security, and worse for own country and other country populations. Given the substantively large differences between vignette conditions, of in-between 0.9 and 1.3 scale points, this explains well the strong shift in approval rates for democratic compared to autocratic settings. Also, high (compared to low) economic value leads to, on average, higher scoring for moral repulsion and higher scoring for world (but not national) security (note that this finding is insignificantly estimated, at $p = 0.095$ ), as well as higher perceived benefits for the own (but not recipient) country population. Changes in perceptions are much less pronounced in absolute terms for the economic dimension of the vignette, however, with only perceived benefits for the domestic population being of substantive significance (with a shift of about 0.4 scale points).

Second, are these consequences perceived differently along the political spectrum, that is, when comparing blue to turquoise and brown coefficients? Here, we again find notable differences between subgroups. First, right-leaning respondents (compared to left and centrist respondents) show a stronger increase in moral repulsion but also more negative domestic and global security repercussions for arms trades with autocracies (while starting at a lower baseline). Second, right-leaning respondents do not perceive high-value deals as morally worse compared to low-value deals – while left-leaning respondents perceive high-value deals as reducing domestic security. Third, ideology does not moderate assessments of domestic or recipient country welfare.

Taken together, we found in the previous section that trades with democracies are approved more irrespective of ideology – this links to a pattern of perceived superiority for all types of consequences with trade to these contexts – less moral repulsion, higher security, and more welfare. The stronger reaction of right-leaning compared to left-leaning respondents in supporting arms trades with democracies resonates in right-leaning respondents perceiving trade with democracies being even less morally repulsive and even better for own and world security compared to left-leaning respondents. These findings also allows us to explain why respondents on the political right react to high-value deals with more approval: they do not perceive such deals as morally different to low-value deals or as having different security implications for the own country – while attaching higher domestic welfare benefits to high-value deals.

Overall, we find that differences in policy positions, therefore, not only stem from a different weight of economic versus normative and home country versus foreign country benefits among left- and right-leaning respondents; they also stem from differences in the perceived consequences of these policies.

Conjoint experiment on arms trade policy: Citizens on the political left and right place differing emphasis on six out of nine attributes of arms exports

Next, we investigate more complex reactions to nine attributes of arms exports on the political left and right. To this end, we present respondents with nuanced arms trade scenarios, where the normative and economic dimensions are disaggregated into three subdimensions each and where we also present further contextual factors.

Figure 3 presents marginal means for respondents on the political left and right (left panel), and their difference (right panel). Reactions of centrist respondents are situated in-between these two coefficients (not reported here for ease of presentation).

Figure 3. Marginal means from linear regression of binary choice variable on conjoint dimensions for subgroups of respondents with rightist versus leftist ideology (left panel) and their difference (right panel). 95% confidence intervals from respondent-clustered standard errors shown. Online Appendix Figure A.5 includes centrist respondents.

From Figure 3, we observe, first of all, that irrespective of political ideology, respondents react to normatively repulsive contexts (e.g., autocracy; human rights abuses; ongoing conflict) and to lower economic value (low monetary or employment benefits) with lower choice probabilities when forced to choose among two profiles. This reflects the finding from the vignette experiment that trade with democracies is much preferred to trade with autocratic contexts. Additionally, as expected, weapons with larger harm potential, lacking security partnerships, and trade by geopolitical adversaries see lower choice probabilities.²⁵

Second, citizens on the political left and right show relevant differences between the factors that they most prominently base their decisions on. On the one hand, citizens on the political left place a much stronger emphasis on normative aspects. They deselect arms trades to autocratic contexts, arms trades to human rights abusive regimes, and arms trades with ongoing conflict more – one notable exception for the latter case is that we observe higher choice probabilities for arms trade with just war scenarios (“country under attack”) for citizens on the political left.

Conversely, citizens on the political right react more strongly to the economic value of a deal. Here, we observe much higher choice probabilities for high-value and high-employment trades (though no differential reaction for trading partnerships). Just as with the vignette experiment (see above) citizens on the political left react not at all (to employment benefits) or only marginally (to economic value), that is, only citizens on the political right consider the economic benefits of an arms trade in their political decision-making.

Concerning the contextual attributes we fielded, we see neither differences for the evaluation of strategic partnerships nor for other countries trading – however, weapons with larger harm potential see lower choice probabilities for citizens on the political left.

Taken together, citizens on the political left place considerably stronger emphasis on the foreign country context with which arms are to be traded, while citizens on the political right place a stronger emphasis on the economic benefits that such a transfer has. Notably, the conjoint experiment allows us to provide evidence that this also holds for contexts of ongoing conflict, which are more strongly deselected by citizens on the political left, except for just war scenarios.

Vignette experiment on arms trade policy disaggregated by party affiliation

In this section, we extend the analyses above to respondents’ party affiliation. To this end, we differentiate results by respondents’ vote intention for the main parties of the German and the French party systems (scoring above 5% in our sample). Results are presented in Figures 4 and 5. The display order of parties reflects their general left-right score according to the Comparative Manifestos Project.²⁶

Figure 4. Vignette by party vote intention, German respondents. Coefficients based on linear regression of agreement with statement in panel header (on a 7-point scale; panels 4 and 5 reverse coded) on two vignette dimensions (recipient country autocracy/democracy; arms trade of low/high value) by subgroups of respondents intending to vote for party indicated in legend. 95% confidence intervals from robust standard errors shown.

Figure 5. Vignette by vote intention, French respondents. Coefficients based on linear regression of agreement with statement in panel header (on a 7-point scale; panels 4 and 5 reverse coded) on two vignette dimensions (recipient country autocracy/democracy; arms trade of low/high value) by subgroups of respondents intending to vote for party indicated in legend. 95% confidence intervals from robust standard errors shown.

For Germany, results indicate, first, that the party-level left-right axis works well with respect to the ordering of citizens with respective vote intentions at baseline: baseline support (the constant) for arms trade increases from left to right – respondents intending to vote for The Left score lowest, those intending to vote for the AfD highest at baseline. Second, reactions to the democratic regime type attribute expression give indications of a nonlinear reaction, where respondents intending to vote for centrist parties react stronger than both those intending to vote for The Left and the AfD. Concerning the economic value attribute levels, respondents with different vote intentions do not differ statistically significantly from each other, but interestingly, respondents intending to vote for The Left attach significantly higher support to high-value deals compared to low-value deals (contrary to the majority of other supporters of left-leaning parties).

For France, the general left-right axis also structures respondents’ preferences toward arms exports, though in a much less clear pattern. At baseline (constant), point estimates order as expected. However, differences are substantively much smaller compared to Germany, and there are no statistically significant differences between parties. Concerning the economic attribute, it is (as expected) those intending to vote for parties at the right of the political spectrum who show a significantly positive reaction to high-value arms trades (clearly those with a vote intention for the right-wing populist RN). With respect to the democratic as compared to the autocratic vignette attribute level expression, it is respondents intending to vote for Parti Socialiste and La République En Marche who react strongest, while those intending to vote for the Green Party react least. These patterns do not align clearly with the left-right continuum on the party level.

Vignette experiment on voting behavior: are party differences relevant for voting?

In a last step, we inquire whether citizens deem the behavior and positioning of parties with respect to arms exports as important for their own voting decision. To this end, we fielded two questions on the perceived issue position of the party and on the importance of a party’s arms trade policy for the respondent’s voting decision. Importantly, the naming of the party is inserted as a randomized ${1}\times{5}/{1}\times{7}$ vignette; hence, respondents assessed only one party, preventing any anchoring and spillovers. Also, this allows us to compare respondents who assessed the party they actually intend to vote for (i.e., their preferred party), while others do not. Again, estimated coefficients for parties are ordered according to their Manifesto Project left-right scores to allow for easy comparisons.

Figure 6 displays the results for Germany. Two core results emerge: first, as can be seen from the constant in the left panel, at baseline, respondents order the position of parties well along the left-right continuum from most restrictive to arms exports (The Left, Green Party) to most lenient (Union, AfD). Also in the left panel, we can see that respondents from all parties perceive the party they themselves identify with as more restrictive compared to all other respondents; this is most strongly expressed for respondents identifying with The Left and the AfD. Second, as can be seen from the right panel, arms trade policy is assessed of medium importance for voters when they assess parties other than their PID party (rating of 3.6, a score below the midpoint of the 7-point scale, on average). This increases markedly when they assess their PID party (to 4.2), mostly so for voters on the political left (Greens, Left Party), but, notably, also the AfD. If this result was driven by social desirability only, we would not expect to see this strong pattern of the party of the own vote intention being rated consistently higher compared to all other parties. We can also see that importance is assessed higher for parties in government (at the time of the survey: Union, SPD) by all voters.

Figure 6. Assessment of party policy (“Do you think the party [party name vignette] wants to limit arms deliveries more (1) or less (7) compared to today,” left panel) and importance (“How unimportant (1) or important (7) is it for you, when making your voting decision, what the stance of party [party name vignette] on arms deliveries is,” right panel) by whether respondent assesses the party they intend to vote for (“PID party of respondent”) or not for German respondents. Coefficients from linear regression of dependent variable in panel header on vignette matching assessed party with displayed party (“PID party of respondent”) by subgroups of respondents intending to vote for the party indicated in the legend. Parties $ \gt 5\% $ of votes shown. 95% confidence intervals from robust standard errors displayed.

Figure 7 displays the results for French respondents. Again, the left-right axis structures the assessment of parties’ policies (left panel, constant) – with the exception of the Green Party, which is perceived as more restrictive on arms trade compared to its position on the left-right axis. Notably, parties are perceived as more positive on average (compared to German respondents), there is less perceived spread between parties (compared to German respondents) and respondents show no bias in perception with respect to their own party (compared to German respondents). At the same time, respondents with a vote intention more on the left (La France Insoumise, Parti Socialiste and Greens) attach higher importance to the policy when assessing their own party compared to those more on the right (En Marche, Rassemblement National and Republicans) (as in Germany); also, starting from a lower baseline (compared to Germany), respondents rate importance for the governing party (En Marche) slightly higher (as in Germany).

Taken together, we observe with respect to party positions that party-level left-right placement corresponds to perceived party positions of voters. In 2020, voters could not be expected to be aware of nuanced party positions on arms trade (given the issue was rarely politicized before the Ukraine-Russian war) – in such situations, voters can be expected to either apply heuristics – such as standard left-right – which will allow them to place parties; or they will provide for inconsistent placements of parties if they perceive the issue as orthogonal to standard party competition. Our results provide tentative evidence for the former.

With respect to the importance of policy positions for voting, respondents leaning toward a party on the political left attach a higher importance to parties’ policy positions when assessing the party they intend to vote for, indicating these voters are more attentive to arms trade policy. Notably, ratings of 4.1 on average (scale of 1–7) for the PID-party positions’ importance suggest that the issue of arms trade is perceived of medium importance overall by respondents in late 2020.

Finally, comparing the French and German context, respondents perceive much less differences between the policy positions of parties in France compared to Germany. This is in line with our analysis of manifestos and the identified near absence of the topic in France. It corroborates our expectations that the arms trade issue should therefore be less polarizing in this context. Also, with respect to voting decisions, German respondents attach a higher overall importance to the arms trade issue, particularly so among respondents leaning toward left-wing parties, compared to French respondents. This indicates that the German government and particularly parties on the political left are much more closely watched by their respective partisan supporters on their arms trade policy.

Figure 7. Assessment of party policy (“Do you think the party [party name vignette] wants to limit arms deliveries more (1) or less (7) compared to today,” left panel) and importance (“How unimportant (1) or important (7) is it for you, when making your voting decision, what the stance of party [party name vignette] on arms deliveries is” right panel) by whether respondent assesses the party they intend to vote for (“PID party of respondent”) or not for French respondents. Coefficients from linear regression of dependent variable in panel header on vignette matching assessed party with displayed party (“PID party of respondent”) by subgroups of respondents intending to vote for the party indicated in the legend. Parties $ \gt 5\% $ of votes shown. 95% confidence intervals from robust standard errors displayed.

Conclusion

We provide the first empirical analysis of the relationship between citizen attitudes toward arms trade and political ideology as well as vote intentions. Our insights from two general vignette and a nuanced conjoint experiment provide a robust pattern. First, and as expected, left-leaning respondents are much more skeptical regarding arms exports at baseline compared to centrist and right-leaning respondents. Second, respondents along the whole ideological spectrum react with substantively increased support for arms trade with democracies (compared to autocracies), even though this increase is substantively stronger for right-leaning respondents. Third, economic considerations do only marginally resonate on the political left but substantively on the political right – there, they can considerably increase support for trade with autocratic contexts. These results similarly reflect in an accompanying conjoint experiment with much more nuanced dimensions. Fourth, with specific mechanism tests, we can show that these differences in preferences toward arms trade policy relate not only to a different weight of moral, as opposed to strategic and economic, consequences of arms trade – but also to differences in perceived consequences. For example, left-leaning respondents do perceive arms trade as more morally repulsive compared to right-leaning respondents. Additionally, while trade with democracies is perceived as less repulsive irrespective of ideology, this easing of moral concern is less pronounced on the political left compared to the political right – while the converse picture is true regarding trade of high compared to low economic value (which are more strongly judged as morally bad on the political left). Corresponding patterns emerge for security and welfare considerations, explaining the policy choices we observed.

These results are in line with related work that indicates that political ideology is a core moderator of both citizen attitudes toward policy and party behavior in the realm of foreign policy. While in related areas this relationship shows in a complex, and partly nonlinear manner (e.g., for trade, see Rudolph et al. (2022); for security policy, see Haesebrouck and Mello (2020)), citizen preferences for arms trade align remarkably well with the left-right axis. This is striking as the general Manifesto coding scheme does not even mention this issue. Our results also link to findings in related policy areas that indicate that normative concerns, as expressed in the overwhelming skepticism of respondents to export arms to morally repulsive contexts, are important for preference formation, as shown for military interventions (Tomz et al. 2020; Dill and Schubiger 2021), foreign aid provision (Heinrich and Kobayashi 2020), trade in commercial goods (Lechner 2016), or supply chain management (Rudolph et al. 2023) – our findings add a more nuanced picture here, however, indicating that while an autocratic foreign country context provides for a red line among leftist and rightist respondents, political ideology still is directly related to how much emphasis citizens place on normative compared to economic considerations.

Fifth and last, the left-right ordering of parties subsequent to the Manifesto Project aligns well with respondents’ perceptions of party positions on the issue of the arms trade – whereas in Germany, these are more strongly differentiated, pointing to a stronger contestation of the issue in day-to-day politics. This reflects in a large stated importance of the issue for their voting decisions, particularly so among respondents leaning toward left-wing parties. This indicates that particularly German parties on the political left are closely watched by their respective partisan base on arms trade policy.

While the older literature on public opinion and foreign policy mainly supposed an ignorant public (see Lippmann 1922; Almond 1950) or a public following elite cues (see Berinsky 2007; Foyle 2017), our research is here much in line with recent findings that the scope and form of political mandates for government action are influenced by public preferences. In this line of reasoning, public opinion directly affects foreign policy via audience cost arguments (Kertzer and Brutger 2016), arguments centered on selection and accountability (Tomz et al. 2020), or elite preference formation (Lin-Greenberg 2021; Opitz et al. 2022). Tomz et al. (2020) provide compelling evidence of how elected representatives align foreign policy with public preferences. Our results on both the stated importance of parties’ arms trade policy for voting decisions and of cognizant perceptions of party positioning are in line with recent findings that citizens are as vigilant regarding foreign policy issues as they are for domestic politics (Kertzer and Zeitzoff 2017).

What are the practical implications of our results? First, our conjoint and vignette experiments are built by arguments usually put forward in public discourse – by political and economic stakeholders but also by citizens. Insofar, we simulate the public debate and assess how citizens make sense of and react to these complex questions. This is highly important for political practitioners as they notoriously “speak in the dark,” that is, communicate under incomplete information regarding the domestic audience. Second, relating these reactions to the left-right dimension, as we do here for the first time, is also of direct practical relevance: left-right labels are a communication device for parties and citizens alike (Downs 1957; Enelow and Hinich 1984; Hinich and Munger 1994). Therefore, if preferences on arms transfers could be consistently subsumed under a left-right dimension, this would relieve parties from complex reasoning and citizens from a cognitive burden. We demonstrate that this is only partly the case: on the one hand, the left-right dimension is effective in spanning the overall structure of the discourse on arms transfers. In this light, and in accordance with insights of Rudolph et al. (2024), we observe a considerable proportion of rejection of arms trade even under favorable context conditions (in line with pacifist preferences), with strongholds among the political left. Targeting nuanced campaign messages to such an audience can hardly be expected to be fruitful, while messages focusing on principled, that is, unconditional refusal of any transfer of weapons (as with the recent foundation of a new party in Germany) could resonate with this subset. Moreover, our results indicate that morality-based anti-trade arguments trump economy-based pro-trade arguments along the whole ideological spectrum. Thus, usual conservative and liberal arguments related to welfare gains or labor protection should resonate only among a very small, mostly centrist or right-leaning subset of the population. In view of this intriguing insight, future research could investigate how competing moral aspects are received. A compelling example in this regard is the, anecdotally, successful communication strategy of the German Green Party regarding arms transfers to Ukraine – with the party being rooted in a long tradition of anti-militarism out of pacifist motivations, currently emphasizing arms transfers for the right to protect. From this, a practical, admittedly sharpened, insight for campaigners would be that counteracting moral with economic arguments will be less effective compared to counteracting moral with moral arguments.

As of now, there exists little research regarding arms trade from the perspective of public opinion (for rare exceptions, see Rudolph et al. (2024) or Efrat and Yair (2024)). To identify whether the patterns we identified replicate in other country contexts and also in the, as of 2024, highly politicized times, more studies are needed.