Citizens’ willingness to fight: wealth, values, and threats

In the past decade, a promising avenue of research on the causes of citizens’ bellicosity has emerged, focusing on WVS/EVS data. Numerous studies have homed in on a specific item within this survey data: respondents’ willingness to fight for their country. For instance, existing research has taken citizens’ willingness to fight to reflect how much they value their own life. In line with their theory of value change, Inglehart et al. argue that the willingness to risk one’s life in battle decreases as opportunities for self-realization increase in more affluent societies.Footnote ¹² Because of the wealth–opportunity–value link, life becomes more valuable as opportunities for self-realization increase.Footnote ¹³ While Inglehart et al. focus on between-country differences, other authors have shown that this might apply within societies as well. In countries with high inequality, wealthy individuals are less inclined to make personal sacrifices, i.e., fight for their country.Footnote ¹⁴

The wealth–opportunity–value link covers one important dimension of the willingness to fight item of the WVS/EVS. However, the item is multidimensional, i.e., responses to the question indicate more than just the value attached to one’s own life. For example, they are also an indication of patriotism, i.e., of the value attributed to one’s country.Footnote ¹⁵

Proximate conflict, threat perceptions, and beliefs about warfare

We argue that the interpretation of a declining willingness to fight as part of a broader development towards emancipative values, supported by growing levels of wealth, is incomplete. The willingness to fight is also influenced by armed conflict in a country’s vicinity. The higher the number of armed conflicts and the higher their severity in a country’s neighbourhood, the more likely citizens will indicate their willingness to fight for their country. Although testing the underlying causal mechanisms empirically is beyond the scope of this article, we posit that the proximity of armed conflict could impact on citizen attitudes in the following ways: First, the outbreak and/or escalation of armed conflict close to home serves as a reminder that the recourse to the use of force remains possible, notwithstanding the prohibition in the UN Charter. Occurrences of armed conflict closer to home make it more difficult to view international relations as a Kantian community of interdependent societies, bound together by international law and a dense web of institutions, with little necessity to fight for one’s country, if not (yet) on a global scale then at least on a regional one. Instead, interpretations of international relations as a Hobbesian state of anarchy, where states whose citizens are unwilling to fight for their country run higher risks of being bullied, exploited, or even conquered by other states gain plausibility.

Second, and related, citizens are more likely to feel threatened themselves if they observe states (or armed non-state actors) in their neighbourhood resorting to force. Whether armed conflict close by is caused by conflict over resources, revisionist claims to contested territory, rising (hyper-)nationalism or authoritarianism, citizens will find it harder to dismiss neighbouring leaders’ threatening rhetoric as cheap talk and to believe that their own country is effectively shielded from the spread of conflict in the future. Research has indeed shown that citizens in NATO countries and especially those protected by US troop deployments show a lower willingness to fight for their country,Footnote ¹⁶ while residents of countries engaged in territorial disputes tend to express a higher willingness to fight.Footnote ¹⁷

In both cases, answering the question whether one would be willing to fight for one’s country affirmatively becomes more likely because it appears as a contribution to effective deterrence or defence, rather than a counterproductive signal of resolve that might lead to a spiral of threats and counter-threats and even to an unintended war.

It is important to clarify that our independent variable is not merely the overall level of conflict worldwide. Indeed, proximity of a conflict matters for at least two reasons: First, proximate conflicts receive a disproportionately higher level of media attention.Footnote ¹⁸ Research has shown that media coverage of political violence evokes emotional reactions in viewers, which can influence their policy preferences.Footnote ¹⁹ Yet, existing studies have thus far not explored the relationship between conflict reporting and citizens’ willingness to fight. While we do not test this relationship directly ourselves, we argue that media reporting is one factor that links proximate conflict to respondents’ willingness to fight.

A second avenue that connects proximate conflicts and citizens’ willingness to fight is the influx of refugees, or previous migration which has established ethnic kinship with conflict parties in a nearby war. While some refugee flows outgrow regional boundaries, most often refugees seek shelter in nearby countries.Footnote ²⁰ Consequently, proximate conflicts lead to an increase in the number of refugees and involuntary migrations to neighbouring countries.Footnote ²¹ Similarly, most regular migration occurs regionally as well. Thus, there exists a higher likelihood that countries close to a conflict share extensive ethnic connections to that conflict. In turn, regular and forced migration increases awareness of proximate conflicts and could thereby impact citizens’ willingness to fight.

It is important to note that citizens’ willingness to fight is not only influenced by spatial characteristics but also by temporal ones. In fact, societies with a history of armed conflict tend to harbour a heightened sense of threat and a stronger inclination for defence.Footnote ²² Simultaneously, critical events can lose their potency over long time periods.Footnote ²³ These temporal constraints inform our research: We do not only look at ongoing conflicts but also examine how the recent history of proximate conflicts influences citizens willingness to fight.

Lastly, there is a rich literature on the spill-over effects of conflicts. Research has highlighted several key aspects of this relationship. For example, Buhaug and Gleditsch (2008) argued that a country’s risk of civil conflict is influenced by the civil conflicts of its immediate neighbours. However, the likelihood of conflict spillover is often mitigated by the capacity of the state.Footnote ²⁴ While this body of research, and its proposed causal mechanisms, is informative, it diverges from our own research in two important ways: First, beyond a few exceptions,Footnote ²⁵ the literature has focused on civil wars. However, we believe that both proximate civil and interstate wars shape citizens’ willingness to fight. Second, many existing studies highlight or exclusively identify effects emanating from adjacent states.Footnote ²⁶ In contrast, we argue that the decision to engage in conflict is also influenced by proximate conflicts beyond immediate neighbours, as these shape how bellicose citizens perceive their surroundings and thus, the level of threat they experience.

Taken together, these deliberations result in the following hypothesis:

Data

We evaluate how proximate conflicts affect citizens’ willingness to fight for their country. Our study draws on the WVS/EVS data, an individual-level national survey conducted across the globe. The World Value Survey has been conducted in seven waves from 1981 to 2022. The European Value Study, which by design is compatible with the data obtained in the WVS, has been conducted in five waves from 1981 to 2017. As Figure 3 shows, the question we rely on as our dependent variable, named E012 in the WVS/EVS coding scheme, has been included in almost all conducted surveys, resulting in data from 471,548 individuals in 112 countries.

Figure 3.

Number of countries and waves in which the ‘willingness to fight’ question has been asked.

The map and Table 1 highlight the unequal regional representation in the surveys.Footnote ²⁸ This leads to higher external validity for our findings in regions such as North America, Europe, and Asia, where numerous countries have participated in multiple surveys. In contrast, Latin America is less represented, and the lack of data is most pronounced in North and sub-Saharan Africa.

Table 1.

Number of respondents per region across all waves.

As a result, while our findings are relevant across various regions and contexts, caution is warranted when drawing conclusions about African countries. This is particularly important given the comparably low degree of statehood in sub-Saharan Africa and high importance of ethnic kinship within countries. Thus, responses in this region might differ from other regions.

Dependent variable

The dependent variable is based on the following WVS/EVS item: ‘Of course, we all hope that there will not be another war, but if it were to come to that, would you be willing to fight for your country?’. The question has been asked in all waves of the survey, except the 2008 EVS. The response ‘yes’ is represented by 1, and ‘no’ is represented by 0. We exclude all instances of missing data, along with responses categorized as ‘don’t know’.Footnote ²⁹

There is one frequently named issues with this variable, namely that the question is formulated in vague terms.Footnote ³⁰ Indeed, researchers have pointed out that there might be a big difference between participation in a war of choice and a war of necessity. Footnote ³¹ Our focus group research corroborates these concerns, with all participating students expressing reluctance to fight for their country outside the context of self-defence. However, for our research this is of lesser importance: We aim to determine whether proximate conflicts impact citizens’ perceptions of the necessity of involvement in any form of conflict, regardless of their specific interpretations of the willingness to fight question. In fact, we do not make claims about the nature of conflicts we expect respondents to associate with the question.

As depicted in Figure 4, the overall average of people responding ‘yes’ when asked if they are willing to fight for their country is 70.6 percent. However, noteworthy regional variations exist, with the percentage of ‘yes’ answers ranging from 79 percent in North Africa to 62 percent in Western Europe.

Figure 4.

Average percentage of ‘yes/no’ responses in all WVS/EVS waves by region.

Main independent variables

Our main independent variable is a moving count of proximate conflicts per country. We rely on data from the Uppsala Conflict Data Program (UCDP) Armed Conflict Database to construct it.Footnote ³² More specifically, we consider all types of conflicts coded in the UCDP dataset (extrasystemic, interstate, intrastate, internationalized intrastate) and all intensity levels (minor armed conflict and war). To assign each conflict to a country, we use the location Footnote ³³ variable and then aggregate all conflicts occurring in a country per year. Next, conflict locations are matched with geographical data from the cshapes package in R to calculate the minimum distance between each state dyad in the international system.Footnote ³⁴ This results in a dyadic dataset in which all states a are paired with the number of conflicts occurring in state b per year, except from those conflicts in which state a is actively involved. We then calculate the moving sum of the number of conflicts occurring in a distance of 500 km of each state a for the past ten years, while weighting wars with a factor of 2.Footnote ³⁵ This process collapses the dyadic structure into a monadic dataset.

In Figure 5, we visualize the independent variable on the example of France. The French were interviewed by the World and European Value Survey in 1981, 1990, 1999, 2006, and 2018. In 2006, we subscribe a proximate conflict score of 14 to France. In the decade until 2006, there were two conflicts in France’s 500 km radius: A civil conflict in the United Kingdom in 1998; and a civil conflict in Algeria from 1997 to 2006. Notably, during the years 1997–1999, the Algerian civil conflict intensified, reaching the intensity of a war as defined by the UCDP criteria. This adds to one year of minor conflict in the United Kingdom, and seven years of minor conflict and three years of war in Algeria, resulting in a score of 14 (as noted before, wars are weighted with a factor of 2).

Figure 5.

Example of independent variable – France in 2006.

In Figure 6, we depict the distribution of the independent variable, grouped by region: Countries in Africa or Asia experience a considerably higher number of conflicts, especially wars, within 500 km of their borders than those in Western Europe, the Americas or Oceania.

Figure 6.

Distribution of the aggregate number of conflicts within a 500 km range of a given country over a timespan of ten years; sorted by region.

Noticeable, the average percentage of ‘yes’ responses in the willingness to fight question in Figure 4 corresponds loosely to the distributions shown in Figure 6, i.e., where there is a larger distribution of proximate conflicts there also seems to be more people willing to fight for their country. This is further indicated in Figure 7, where we plot the mean proportion of ‘yes’ responses per country-year against our proximate conflict score which we divide into three quantiles corresponding to the spread of the data: ‘low’, ‘medium’, and ‘high’. As becomes immediately evident, the mean proportion of respondents answering ‘yes’ moves up with the level of conflict both in the median and in the lower and upper quartile.

Figure 7.

Visualizing the distribution of the average percentage of ‘yes’ to willingness to fight by country-year per proximate conflict level (500 km, ten-year moving sum).

Lastly, we visualize the relationship between the willingness to fight and the sum of proximate conflict directly in a scatterplot in Figure 8. We plot the mean willingness to fight per country-year against the sum of proximate conflict the country experienced in the year in question. Both the Loess fitted line, and a linear regression line indicate an overall positive relationship between these two variables, i.e., moving from low to high averages in the willingness to fight corresponds with a higher level of proximate conflict. In sum, our descriptive analysis suggests that there might indeed be a positive relation between citizens’ willingness to fight and proximate conflict.

Figure 8.

Scatter plot, visualizing the relationship between average willingness to fight and the sum of proximate conflict per country-year.

In addition to our main independent variable, we employ a second, more complex, measure. Here we use a dynamic calculation of distance to construct the variable. We take the inverted distance divided by the sum of all distances to normalize the outcome, and multiply it with the intensity level of conflict, thereby automatically applying a factor of 2 to wars. In mathematical terms, it can be depicted as follows:

(1)\begin{equation}{C_j}\, \times {w_{ij}}\end{equation}

where ${C_j}\, \in \left\{ {0,1,2} \right\}$ with $0$ representing no conflict, $1$ minor conflict and $2\,$war, is the conflict occurring in state $j$ and ${w_{ij}}$ is a distance weight calculated as

(2)\begin{equation}{w_{ij}} = \,\frac{{1/{d_{ij}}}}{{\mathop \sum \nolimits_{j = 1}^n 1/{d_{ij}}}}\end{equation}

This results in a second independent variable which dynamically penalizes conflicts that are further away. This approach is taken from Buhaug and Gleditsch.Footnote ³⁶ We then again calculate a moving sum of the past decade (see Appendix for corresponding descriptive statistics).

Control variables

Next to our main independent variable, we add a range of control variables on the individual- and country-level. Existing research has pointed to three broad dimensions that influence citizens’ willingness to fight: the value one places on their own life, the value attributed to the country and a country’s defence policy environment. We account for a range of standard control variables, including Gross Domestic Product (GDP) and age, as well as factors related to these three dimensions.

Individual level. On the individual level, we rely on data provided by the WVS/EVS. As is common practice in survey data analysis, we include a variable for sex and age in our analyses. Indeed, women often exhibit a greater aversion to violence.Footnote ³⁷ Moreover, as age increases, the willingness to fight is likely to decrease, not least due to the decreasing feasibility of participation in warfare. In addition, we include a measure of income that asks respondents to indicate their income from lowest to highest group (0–10). Earlier research has shown that income levels can influence respondents’ willingness to fight. Furthermore, military recruitment often targets low-income individuals, which might affect their willingness to fight.Footnote ³⁸ Lastly, we add a measure of confidence in the national military. We believe it is important to account for this aspect as citizens might not be willing to fight, even in the face of threats, if they do not generally trust their own military. In fact, previous studies have found this variable to be positively associated with citizens’ willingness to fight.Footnote ³⁹

We conduct a principal component analysis (PCA)Footnote ⁴⁰ of the three pro-choice itemsFootnote ⁴¹ used by Inglehart et al., as they have shown that higher pro-choice attitudes can lower citizens willingness to fight. The three pro-choice items reflect a wealth-opportunity-value link, which aligns with the dimension concerning the value one places on their own life. To control for the value respondents directly place on their own country, we also add a variable measuring national pride, which might positively affect citizens’ willingness to fight.

Country level. As is common practice we add the natural logarithm of GDP per capita (constant 2010 US$) taken from the World Bank through the wbstats package in R.Footnote ⁴² A nation’s wealth might correspond to the average value people place on their own life (wealth–value–opportunity link). We then add the Polyarchy democracy scale which we take from the Quality of Government dataset based on the VDem dataset.Footnote ⁴³ This could positively correlate with the significance respondents place on their nation, potentially heightening their willingness to fight. However, this is not certain. In fact, some prior studies have shown that individuals in democratic nations tend to exhibit a lower willingness to fight.Footnote ⁴⁴

Next, we control for a range of additional factors that touch upon the country’ legacy of conflict and defence policy. We add a variable on each country’s history of war using the UCDP Armed Conflict Database, which is coded from 0 to 1, where recent wars are weighted more heavily, and a value of 1 indicates current involvement. In addition, we include a dummy for military conscription.Footnote ⁴⁵ All these variables are lagged by one year to provide us with data until 2022, corresponding to the last WVS/EVS survey year. Lastly, studies have shown that a nation’s readiness to engage in conflict may diminish when under the protective umbrella of an ally. Much of this research has specifically focused on the United States.Footnote ⁴⁶ Accordingly, we have incorporated a binary variable to represent the status of being an US ally. However, we think that this dynamic could similarly influence nations that benefit from defence commitments from any country, not solely the United States. To account for such circumstances, we introduce an additional control variable. Both variables are constructed using the Alliance Treaty Obligations and Provisions database, which we have updated to include data up through 2022.Footnote ⁴⁷

Data analysis

Considering the nested structure of our data, which includes individual-level data alongside country-specific variables, we employ multilevel models. Due to the dichotomous nature of our dependent variable, our chosen method is a logit estimator with country-specific random intercepts. Significantly, these random effects account for all constant confounding factors. In addition, we employ year fixed-effects to assure taking account of temporal confounders as well. As previously noted, some conflicts have a global impact, extending their influence well beyond their immediate region. However, if these conflicts truly have global effects, incorporating year-fixed effects in our analysis would help control for these influences.

We centered all non-dichotomous independent variables around a mean of zero. We run four models: Model 1 only controls for sex and age. In Model 2, we add additional individual- and country-level control variables. Models 3 and 4 are the same as Models 1 and 2 except that we replace our main independent variable (Log Conflict 500 km) with our dynamic independent variable (Log Conflict Dynamic).