Theory: From the Gender Gap to the Suffragist Peace

How might the extension of suffrage to women affect theoretical arguments about the democratic peace, that is, the probability of war between democracies? As Sjoberg argues, theories of the democratic peace, like many other theories of war, have been largely “genderless.”Footnote ⁸ Addressing the effect of women's suffrage on the democratic peace requires not only theorizing how women's preferences influence a democratic state's willingness to fight once women get the vote, but also how that state interacts with other states strategically in the international system. Our theory thus engages aspects of women's preferences at the individual level, how those preferences gain political voice in democratic government, and how those governments’ increased political costs for war affect interstate bargaining and the likelihood of conflict. These arguments lead to predictions about the likelihood that suffrage democracies (democracies with female suffrage) will initiate conflict or end up in wars with different types of states (such as other suffrage democracies). We first consider theories of the democratic peace and how women's preferences for war might affect these arguments before building our own theoretical argument for how these gender differences lead to a suffragist peace at the international level.

Although there is broad empirical acceptance of the proposition that democracies do not fight each other, there remains theoretical and empirical disagreement over the nature and source of this empirical regularity.Footnote ⁹ The “democratic peace” is often formulated in explicitly dyadic terms: democracies do not fight each other but are still prone to fight with autocracies.Footnote ¹⁰ Alternatively, the peaceful nature of democracies might lead to a monadic effect, where democracies are less prone to conflict with other states, regardless of regime type.Footnote ¹¹ As Debs and Goemans point out, scholars still do not agree on whether this is a “cats-and-dogs” effect where democracies, and perhaps also autocracies, tend not to fight their own kind but mixed dyads are more war prone.Footnote ¹²

There is also a lengthy debate over the source of the democratic peace—one which has recently seen a renewed focus on public opinion.Footnote ¹³ Of course, scholarship on public opinion and foreign policy in democracies suggests that elite cues are important for activating issue preferences.Footnote ¹⁴ But public opinion serves as an important constraint on democratic leaders.Footnote ¹⁵ It remains unclear, however, whether the public constrains democratic leaders to be more peaceful overall, as suggested by Kant's argument that citizens will be more hesitant to go to war (i.e., monadic democratic peace); or whether the public specifically prefers to avoid war with other democracies (i.e., dyadic democratic peace), as Tomz and Weeks contend;Footnote ¹⁶ or both.

If public opinion is an important driver of the democratic peace, then changes in the composition of the electorate, such as the extension of suffrage to women, matter to the extent that newly enfranchised groups have systematically different preferences. We argue that adding a large pool of voters with structurally different preferences is likely to change the calculus of political leaders as they contemplate the use of force. As Caprioli and Boyer put it, “by gaining political influence through voting, women's values should influence leaders’ decisions.”Footnote ¹⁷ Indeed, Oneal and Russett note that differences in the franchise, especially for women, are a complicating factor in measuring democracy and that

The extension of suffrage to women may thus restrain states from using force. This connection is made even more plausible by the growing number of studies arguing that women's suffrage had significant effects on government in other areas. Lott and Kenny show that the size of the government increased dramatically in US states that granted women suffrage because the inclusion of women shifted voting populations’ preferences to the left.Footnote ¹⁹ Significantly, the estimated effects of suffrage are similar in states that extended the vote voluntarily and those that were forced to do so even though they did not ratify the nineteenth amendment, suggesting that if a confounding variable causes both suffrage and the size of government, its effects are small.Footnote ²⁰ Abrams and Settle find remarkably similar effects in a different time and place—when Switzerland finally extended voting rights to women in 1971.Footnote ²¹

Thus far, however, few studies have investigated the link between suffrage and peace. The effect of women's representation in legislatures on international affairs has received more attention.Footnote ²² In a study of female leaders and conflict intensity, Caprioli and Boyer include suffrage in their study of crisis severity, but explicitly treat it as a control, focusing instead on female representation in the legislature as a measure of gender equality.Footnote ²³ In a study of militarism, Caprioli examines suffrage more directly as a measure of political equality, but again in tandem with parliamentary representation, and for a more limited time period (1960–1992) than we examine here.Footnote ²⁴ Hudson and colleagues find that suffrage is one of many factors connecting gender and peace.Footnote ²⁵

To theorize the effect of extending the franchise to women on the democratic peace we need to understand both the specific contours of women's preferences as well as the group's ability to express those preferences. We first consider individual-level preferences, and then turn to their expression in democratic politics and, in turn, how these preferences affect interstate bargaining and the consequences for understanding the democratic peace.

The Nature of the Gender Gap in Public Opinion on the Use of Force

We turn first to the individual level. There are strong theoretical reasons to believe that women have more peaceful preferences than men, although there are important nuances relevant to the politics of war and peace. In terms of possible influences on the democratic peace, three aspects of the gender gap are relevant.

First, theory and evidence suggest that differences between men's and women's preferences about the use of force are general and persistent across space and time, but also subject to context-specific factors. The gender gap in public opinion on the use of force is well-established in the United States, where much research has focused. As Conover and Sapiro note, “although American women as a group can hardly be classified as pacifist, they certainly appear less militaristic than American men.”Footnote ²⁶ In an analysis of gender differences in public attitudes toward the US use of military force in twenty-four cases from 1982 to 2013, Eichenberg finds that women are less supportive of using force across time, the size of interventions, and the partisanship of the president, although the magnitude of the gender gap varies.Footnote ²⁷ Berinsky notes that the gender gap in preferences on various questions surrounding US involvement in World War II persisted even after Pearl Harbor, although the preferences of both genders generally moved together.Footnote ²⁸

These findings comport with the few studies that explore the gender gap in attitudes toward war outside the United States.Footnote ²⁹ Eichenberg finds that the gender gap appears to hold crossnationally, though its size and nature varies by country and context.Footnote ³⁰ Country-specific evidence is also supportive. In 2014, 47 percent of male versus 39 percent of female Kenyans preferred that the Kenyan Defense Forces should continue their activities in neighboring Somalia.Footnote ³¹ In Japan in 2016, 49 percent of men versus 35 percent of women were in favor of changing the constitution to permit Japan to have a military.Footnote ³²

While explaining the gap is beyond the scope of this article, we note that scholars have pointed to a mix of biological and social factors. The gap is both persistent and varying in size by context. Beginning at early ages and spanning cultural contexts, males are more aggressive than females, across a range of measures.Footnote ³³ In a wide-ranging study of the “war system,” Goldstein notes the puzzle that “despite the diversity of gender and of war separately, gender roles in war are very consistent across all known human societies.”Footnote ³⁴ He concludes that “the gendering of war appears to result from a combination of factors,” including average biological differences between men and women, and “culturally constructed gender roles.”Footnote ³⁵ In the context of the first Gulf War, Conover and Sapiro categorize explanations for the gender gap into those that focus on biological difference or early childhood socialization (“sex and gender”), those that focus on the experience of motherhood (“maternalism”), and those that emphasize political activation of gendered views of war (“civic feminism”).Footnote ³⁶ As they note, many studies moved away from biological determinism and emphasized instead “early differential socialization and experience.”Footnote ³⁷ Whether biological or social, these explanations “share an important implication,” namely that “differences between women's and men's orientations to war are founded in childhood or earlier and cannot be explained away by any other aspects of their lives.”Footnote ³⁸ The gender gap has persisted in polling data across decades, and therefore it cannot be entirely explained by characteristics that have varied substantially over time.Footnote ³⁹

A second individual-level consideration concerns whether the gender gap interacts with the regime type of the target. Tomz and Weeks found that democratic citizens perceive other democracies as less threatening than autocracies.Footnote ⁴⁰ If the reasons driving Tomz and Weeks's result are more important for women than men, then we should see the gender gap (among democratic citizens) amplified for considerations of using force against democratic targets. To date, few studies have investigated this at the individual level. One partial exception is Brooks and Valentino, who draw on research showing that “women are more likely to favor cooperation and compromise within groups over aggression as a means for settling disagreements,” and theorize that this “consensus orientation” leads women to be more likely to approve military interventions with multilateral and especially UN approval.Footnote ⁴¹ But they find only modest support for this hypothesis. In a national-level argument more specifically focused on regime type, Risse-Kappen argues that the dyadic logic of democracies fighting autocracies but not other democracies draws on in-group/out-group dynamics.Footnote ⁴² But as Goldstein describes, there is little evidence that out-group psychology is gendered.Footnote ⁴³ The gender gap appears to stem from views of violence rather than views of regime type. Although women are in general more likely to perceive many types of threats, they are less likely to favor violent responses; this may be due in part to women being more likely to feel anxiety, which increases risk aversion.Footnote ⁴⁴ Eichenberg finds that the data “point to the centrality of military force, violence, and war as the most important variables affecting gender difference,” at both the individual and aggregate levels.Footnote ⁴⁵

A third individual-level feature worth noting is that, despite the persistence of the gap across country and context, women are not automatic pacifists; scholars have shown that the gender gap can be moderated by politics or context, such as whether the war stakes are humanitarian rather than strategic.Footnote ⁴⁶ On important security issues such as defense spending, women and men seem to respond similarly to changing conditions or information.Footnote ⁴⁷ While women seem to have a more baseline dovish preference, they appear to respond to information and context in a similar way as men.Footnote ⁴⁸

Importantly, as McDermott argues, men and women have different tolerance and motivations for fighting, but share “a similar desire to protect and defend their territory, their material resources, or the health and welfare of their children.”Footnote ⁴⁹ That women may not want to pick or get involved in fights overseas, but are willing to defend territory when attacked, has implications for both monadic democratic peace and audience cost arguments.

Most of the individual-level opinion findings to date—with the notable exception of Brooks and Valentino—are based on observational surveys with relatively simple “for-or-against” questions.Footnote ⁵⁰ These data admit to multiple interpretations because the counterfactual policies, then and earlier in the crisis, are not always clear. At a particular stage of a crisis, a respondent may support escalating to war, instead of backing down or staying out, even while they might have more strongly preferred other pacific policies. For example, respondents may support the use of force, given that their president has publicly threatened it,Footnote ⁵¹ while disapproving of their president's making the threat in the first place.Footnote ⁵² Our meta-analysis of survey experiments in part addresses this shortcoming.

How Women's Preferences Matter

The impact of a group's preferences on state behavior depends on the group's electoral power and political voice, as channeled through democratic institutions. To a first approximation, the level of democracy determines the extent of public influence, and the extension of suffrage determines whether the voting public includes women; thus it is the two together that jointly determine whether mass-level female preferences influence state behavior. Democratic constraint on the use of force is not automatic and depends on features such as political parties and access to free media that provide citizens with sufficient information to exercise their ability to hold democratic leaders accountable.Footnote ⁵³

We argue that institutions enable two pathways through which women's suffrage influences elite decision makers. The first operates by changing the incentives of elected politicians, making them more responsive to issues that they perceive women care about. Thus, Morgan demonstrates that existing politicians became more open to policies designed to help working women, such as parental leave laws.Footnote ⁵⁴ Following the extension of suffrage in Britain, many conservatives, who feared and did not welcome women's suffrage, acknowledged that their party would, “of necessity, be compelled to attract them.”Footnote ⁵⁵ The further extension of women's suffrage in the United Kingdom was a striking example of male elite deference to the views of the electorate. Following partial suffrage in 1918, most conservative elites did not welcome full suffrage, but they understood that obstructing it would alienate a large portion of the new electorate.Footnote ⁵⁶ Following suffrage in the United States, women become more politically engaged over time, on average, though their political mobilization varied over time and space.Footnote ⁵⁷

The second route to influence is through choosing elites who better represent public preferences. Observation of gender differences in voting behavior along the lines predicted would be supporting evidence of this mechanism. However, such an observation is not a necessary consequence of female voters influencing policy: along the lines of the median voter theorem, female suffrage could counterfactually change multiple candidates in the direction of female voters’ preferences, without opening up this policy dimension as a basis for choice between candidates.

Women's Suffrage and Peace: Hypotheses

We are now in a position to theorize how women's suffrage might affect international conflict and, in particular, the democratic peace. We have argued that the level of democracy and the existence of suffrage jointly determine whether mass-level female preferences shape policymakers’ preferences over foreign policy. But whether and how policymaker preferences affect interstate conflict behavior depends on strategic context. Does a shift toward pacific preferences get absorbed by a compensating increase in aggression by others? Do these more pacific suffrage democracies find ways to avoid conflicts with all other countries (a monadic suffragist peace)? Or, does it take two to avoid conflict, thus leading only to a dyadic suffragist peace?

To reiterate: the public opinion evidence suggests that women are more likely than men to oppose the use of force overall, not just against other democracies. Women will not oppose defending against an attack, and are responsive to context and justification—women and men respond to many of the same contextual factors. Thus, electorates that include women have a lower baseline level of support for conflict but are otherwise largely similar to other electorates. We discuss the implications of this conceptualization here, and formalize these ideas in two models in the appendix. To be clear, this is not the only path through which gender might affect the democratic peace, but it is the most direct route for the large shift in public preferences associated with the advent of women's suffrage.Footnote ⁵⁸

The bargaining model framework associated with Fearon's work provides a starting point for understanding the effects of such changes in the political costs of conflict.Footnote ⁵⁹ In general, increases in these costs decrease the likelihood of both conflict initiation and war. One reason is that increases in costs increase the size of bargaining ranges. This implies a reduction in the set of status quos which, in the first place, require renegotiation to avoid war; all else equal, this reduces the likelihood of war. A wider bargaining range can also reduce the need for war as a result of commitment problems. To see this, consider the commitment problem that occurs because power is shifting in favor of one side. War occurs if the side that is losing in power prefers war today to the best bargain it could strike later, once power has shifted. Increases in war costs to the side that is losing in power make it less willing to fight today; increases in war costs to the side that is gaining in power improve the quality of the bargain that the declining state will be able to strike later. Thus, both factors undermine the commitment problem such that the declining state may prefer negotiated settlements to war.

The case of conflicts that result from incomplete information is similar, but with a wrinkle. On the one hand, increasing costs of conflict make states less willing to fight. On the other, rising costs may entice opponents to make greater demands. Which of these two factors has the greater effect on the probability of conflict depends on the specific model under consideration and in particular on the ways that rising costs influence actors’ domains of uncertainty. In many models of international conflict, increases in the costs of war—or in the benefits of peace, which are modeled as mathematically equivalent—decrease the likelihood of conflict initiation for that actor and the probability of war overall.Footnote ⁶⁰ In a few others, however, increasing costs can increase the probability of war for some parameter values.Footnote ⁶¹

In spite of this complexity, all of these models imply that states are less likely to force a change in the status quo by initiating a crisis when their own costs of conflict increase relative to the benefits of peace.Footnote ⁶² Thus, we expect suffrage democracies to be less likely to initiate crises—a monadic democratic peace with respect to crisis initiation. As Rousseau and colleagues argue, “in comparing the monadic and dyadic propositions, it is crucial to distinguish between the use of force in response to force (such as defending against an invasion) and the initiation of force (such as launching an invasion) … public opinion may be averse to using force but recognizes the necessity of it when faced with an attack.”Footnote ⁶³ Although a suffrage democracy will issue fewer threats, if attacked it would still have good reason to defend itself, as we discussed earlier. Thus, our monadic hypothesis refers to the likelihood of crisis initiation.

The analysis of suffrage democracy's effect on the overall probability of conflict is more complicated because of the possibility that other states will increase their demands on states who are less willing to fight. A state with more pacifist preferences as a result of the adoption of suffrage may not be able to avoid conflicts with partners who do not feel the same. After all, as McDermott notes, women are likely to support defending against aggression.Footnote ⁶⁴ If war is a strategic imperative because of foreign aggressors, marginal shifts in preferences will not influence the overall likelihood of conflict. Thus, it may be that some long-running rivalries, such as the one between France and Germany, may have required both sides to adopt democratic institutions and women's suffrage before the rivalry could be ended. Additionally, if suffrage reduces the probability of initiation but suffrage democracies still fight back if attacked, states of all regime types should have less concern that democracies with suffrage will fear exploitation and thus take aggressive action against them.Footnote ⁶⁵ When both sides are more pacific, neither need fear exploitation, trust will increase, and we expect the overall likelihood of conflict to decrease.

In the context of traditional bargaining models where the sides have incomplete information about each other's resolve, mutual increases in conflict costs—as opposed to cost increases on only one side—generally decrease the likelihood of conflict. When costs to the two sides increase in tandem, the scope of the demand from the challenging state on the other tends to remain unchanged, as we show formally in Appendix E, but the probability that the other state is willing to fight for any given bargaining outcome declines with the result that the overall likelihood of conflict declines. Thus, whether we conceive of international conflicts as resulting from deficits of trust that cooperation will be reciprocated or uncertainty about actors’ resolve, dyadic increases in the costs of conflict lead to a reduction in the likelihood of conflict. We state this as the Dyadic Suffragist Peace Hypothesis. Appendix E formalizes these arguments in two models, one of trust and one of incomplete information about resolve. Notably, this argument does not mean that women are more likely to perceive democracies as less threatening more generally, as in traditional dyadic democratic peace logic. Rather, it is specifically that the extension of suffrage to women, with their generally more peaceful preferences, alters the preferences and aggressive tendencies of both countries, and likely how each perceives the other.

Research Design

Our arguments move from the individual level to the international outcome level, and so our research design accordingly uses data from both levels. First, at the individual level, we conduct a meta-analysis of recent survey experiments in the field, breaking respondents down by sex. Here, our primary aim is to show that women are, on average, less supportive of using force across a variety of controlled comparisons in different scenarios and varying national and international contexts. Although we are limited in the scope of our discussion of the individual-level data here, we report further results in the appendix. We also briefly note results from experiments on audience costs and the democratic peace that bear on our theoretical discussion. These experiments provide a unique source for drawing inferences about the effect of sex (again, using this term pragmatically because available data do not allow for more nuanced exploration of gender differences). They are designed to isolate factors that researchers in the field consider to be important determinants of conflict. To the extent that women are less supportive of using force (and more supportive of peaceful solutions) across these different contexts, we will have individual-level evidence that women represent a pool of voters with more pacific preferences.

We then use interstate conflict data to assess the nature of the suffragist peace, and specifically to evaluate the monadic and dyadic hypotheses. Our dyadic hypothesis and tests correspond closely to the dyadic analyses of the democratic peace that are prevalent in the literature. Our analysis therefore serves as a direct test of the suffragist peace against the most prominent alternatives.

Meta-Analysis of Differences Between Male and Female Preferences on the Use of Force

For the meta-analysis, we use what we believe to be the universe of experiments for which data were available through 2016 conducted by international relations scholars on representative samples of national populations that describe a potential or actual international conflict and ask respondents about their preferences over crisis bargaining outcomes.Footnote ⁶⁶ We also conducted four original experiments in four different countries. In total, we analyze seventeen studies comprising more than 20,000 respondents from six countries and four continents.

The survey vignettes of these studies offer diverse reasons for the use of force across a range of contexts, including the response of British and US populations to using force to prevent nuclear weapons acquisition;Footnote ⁶⁷ protecting an ally, humanitarian intervention, and regime change;Footnote ⁶⁸ Japanese support for the use of force alongside the US in the service of democratic regime change in the Middle East;Footnote ⁶⁹ military action in East Timor;Footnote ⁷⁰ conflict between the United States and with Russia in the Arctic;Footnote ⁷¹ and an attack on an Al Qaeda nuclear weapons lab in Syria.Footnote ⁷² The original data we collect come from experiments in Egypt, Israel, Turkey, and the United States. Each of these experiments ask similar questions about a resource conflict in the Mediterranean or, in the case of the US, in the Arctic.Footnote ⁷³

To perform the meta-analysis, we restrict attention to vignettes that describe either a use of force in the future or a successful use of force in the past. Some of these experiments included a description of a conflict as a treatment alongside other treatments describing peaceful outcomes while others included only a question about a choice to engage in conflict following a description of an international context. We exclude vignettes describing unsuccessful uses of force because reactions to these may conflate reactions about the use of force itself and about the defeat. When participants were asked if force should be used or should be used only under a more restrictive set of conditions, we coded only the unrestricted-use-of-force option as advocating the use of force.

The main result of this meta-analysis is unequivocal. In every study, support for the use of force is higher among men than among women. The average level of support among men is just over 50 percent while the average support among women is only 38 percent, a thirteen-percentage-point difference which is highly significant (p < 10⁻⁶⁸). These effects span the globe to a remarkable degree.Footnote ⁷⁴ The difference between the sexes in approval of the use of force over Mediterranean resources is 14 percent in Egypt, for instance, and 12 percent in Turkey (p < .01). Each of the US and UK studies are individually significant at conventional levels, and the pooled difference of 12 percent has p < 10⁻²⁴. In the two studies from Japan, the difference in approval levels by sex is a remarkable twenty-three percentage points (p < 10⁻²⁷). In fact, the difference between male and female support is highly significant in each study of at least 500 respondents with the sole exceptions of our original study conducted in Israel (p = .1), which is consistent with Tessler and Warriner's findings, and the Press, Sagan, and Valentino study in which approval of an attack on an al Qaeda nuclear weapons lab is high for both sexes (p = .15).Footnote ⁷⁵ These results are presented in Table 1.

Table 1. Conflict preferences by sex in seventeen experiments

Egypt, Israel, and Turkey original data indicate percent of respondents marking above 5 on a 10-point scale. * p < .05; ** p < .01; *** p < .001.

† Indicates number of respondents asked about the use of force or administered a use-of-force treatment, which is often less than the total respondents in the study.

In every region we examined—North America, Europe, the Middle East, and Asia—the difference in preferences is statistically significant, often at extreme levels. The results of the meta-analysis also suggest differences across regions. The effect appears largest in Japan and smallest in Israel, for instance. But, across cultural contexts and regions, women are less approving of the use of force vis-à-vis a peaceful alternative.

The results also emphasize, however, that women are neither uniformly nor unalterably opposed to the use of force. Indeed, the weighted average of approval for force in all fifteen studies is 38 percent among women, with many studies showing support in the 30 to 40 percent range. Furthermore, prior experiments that explicitly explore the gender gap on war (e.g., Brooks and Valentino) as well as observational data (e.g., Eichenberg) suggest that women support the use of force in particular contexts.Footnote ⁷⁶

Although a full exploration of results at different stages of conflict processes is beyond the scope of this paper, we note two other relevant sets of results here (and in Part F of the online appendix, we present results from other points on the escalatory ladder). First, and related to the initiation of crises, several studies allow us to examine how men and women evaluate leaders who back down from a threat versus those who remain out of a conflict entirely, which is known as the audience cost.Footnote ⁷⁷ We examine four studies conducted on representative samples of the US population and comprising over 7,000 respondents in total. Table 2 presents the results. It is apparent that both men and women impose audience costs: approval for both is much higher for leaders who remain out of a conflict and do not threaten force than for leaders who make a threat on which they do not follow through. In each of the studies, the audience cost imposed by the women is higher than that imposed by the male population, but in no study considered in isolation is the difference between the sexes statistically significant. When we pool the data from all four studies, however, we see that the sex difference in audience costs is highly significant at the p < .001 level.

Table 2. Sex differences in staying out versus backing down

* p < .05; ** p < .01; *** p < .001.

† Indicates the number of respondents given the Stay Out or Back Down treatments. ‡ Includes all audience cost conditions that do not involve new information arising during the crisis. § Includes Back Down conditions that do not involve the use of force since those that do could be catagerized as Unsuccessful War. §§ Pools data for studies 1 and 2 from this article.

But these results mask an important difference: women tend to impose these audience costs for a different reason than men. Following the procedure used by Kertzer and Brutger, we used the data from this study to decompose audience costs into a cost for saying one thing and doing another (inconsistency) and a cost for making a threat in the first place (belligerency).Footnote ⁷⁸ The results are striking: while most men impose audience costs because of inconsistency and the threat to reputation that this implies, most women impose audience costs for belligerency. The results support the theoretical expectation that women are less supportive of initiating threats and crises that might lead to using force. Part F.3 of the online appendix presents this analysis.

Second, and more directly related to the democratic peace, men and women appear to perceive regime types similarly even as women's lower baseline preference for force affects their willingness to strike both democratic and autocratic targets. Here, it is particularly useful to consider the Johns and Davies and the Tomz and Weeks experiments on public opinion and the democratic peace, which examine public support for military strikes against a democracy or an otherwise identical autocracy and are thus the individual-level studies of most direct relevance to this paper (see Table 3).Footnote ⁷⁹

Table 3. Analysis of democratic peace experiments by sex

Note that the numbers for the Johns and Davies UK experiments are slightly different from those reported in the article because we use their whole sample rather than the subset on which the article focuses. † p < .1. * p < .05; ** p < .01; *** p < .001.

Both experiments include US and UK samples. In the US, more than half of respondents support a strike against an autocracy while fewer than half support a strike if the target is a democracy. Breaking this result down by sex, however, reveals a striking pattern. Among men, 57 percent support a strike against an autocracy, versus 48 percent against a democracy. But among women, 45 percent support using force against an autocracy, versus 38 percent against a democracy. Thus, more than half of men are either in favor of a strike or nearly so, whatever the target's regime type, whereas a majority of women never are. The differences between men and women are highly significant whether the target is a democracy or an autocracy. Although results for the United Kingdom show treatment effects of very similar magnitude in terms of regime type across sex, in the UK both sexes have lower baseline support for war across regime types, with minorities of both sexes favoring a strike. In both countries, expanding suffrage to women would thus decrease support for using force against both democracies and autocracies (p < .00000001 for both regime types).

Overall, the meta-analysis supports our theoretical discussion: women have a lower baseline preference for using force across many contexts and regions; women do not perceive democratic regimes differently than men; and women are less supportive of initiating crises, but are overall far from pure pacifists. These findings point to not only a dyadic suffragist peace, but also a monadic suffragist peace with respect to initiation. We now turn to the analysis of international conflict behavior.

Statistical Analysis of Suffrage-Democracy and Conflict

We first examine crossnational evidence for the monadic initiation hypothesis over the period 1816 to 2010 using directed-dyad initiation models, where the unit of observation is the directed dyad and the outcome variable is the initiation of a militarized dispute. This enables us to account for well-established correlates of conflict including relative capabilities, contiguity, distance, and alliances. Models assessing the likelihood of initiation using country-year data are in Part B.3 of the appendix. We then assess the evidence for the dyadic suffragist peace using the standard nondirected dyad framework generally employed for assessments of the democratic peace. In keeping with prior approaches, we omit all but the first years of World War I and World War II and all but the first year of ongoing, multi-year disputes. All data on dispute involvement are taken from the Dyadic Militarized Interstate Dispute data set version 3.1. To ensure consistency across measures of the democratic peace, data on regime type are taken from the Polity IV data set.Footnote ⁸⁰ Models using alternative measures of democracy and alternative measures of interstate disputes coded by Gibler, Miller, and Little are reported in the appendix, Part B.Footnote ⁸¹ Those models do not differ significantly from the results we report. Unless noted, variables are coded as described by Oneal and Russett.Footnote ⁸²

Independent Variable: Women's Suffrage

Data on women's suffrage were collected primarily using the United Nations report on the Progress of the World's Women which lists the voting status of women by year in all 196 countries. In 160 countries, women's suffrage was granted to all women within the country at the same time. In thirteen other countries, the right to vote was granted to women in stages according to different sets of conditions. Some constraints were unique. In Belgium, for instance, war widows, mothers of those killed in war, and female political prisoners were granted the right to vote in 1919 while all other Belgian women of an equivalent voting age of men were granted the vote in 1948. Other states—such as Australia and Canada—first granted women suffrage on the basis of race, while some states—such as Bolivia, Ireland, Romania, and the United Kingdom—first granted women suffrage on the basis of literacy, property rights, or education level, and only later adopted legislation allowing full women's suffrage at an equivalent age as men. Of course, the right to vote did not always translate into actual votes—for example, Black women in the Jim Crow South waited decades before they could exercise their voting rights. Nevertheless, we estimate that within all but six of these thirteen countries that adopted suffrage in a piecemeal fashion, the first wave of women's suffrage created voting populations in which women constituted at least 40 percent of all eligible voters.Footnote ⁸⁴ The first wave of women's suffrage therefore represents the largest overall shift in the gender balance of the electorate within all but six of the 196 states included within the analysis. Although the effect of this substantial shift is our primary interest for this study, we also assess the effect of other measures of suffrage, including universal women's suffrage and whether a national election has been held in which women were eligible to vote, within Part B.10 of the appendix.

Figure 1 illustrates the number of democracies that did and did not allow women's suffrage within each year between 1815 and 1975.Footnote ⁸⁵ The figure shows that while the number of democracies grew slowly but gradually over the nineteenth century, no woman held the right to vote in national elections until 1893 when New Zealand became the first state to grant universal women's suffrage, followed by Australia in 1902, Finland in 1906, and Norway in 1913. The graph shows a sharp increase in the number of democracies both with and without women's suffrage in the period 1915 to 1922. Denmark, Canada, Austria, Germany, Ireland, the United Kingdom, and the United States, among others, adopted women's suffrage during this period. By 1955, all but two democracies, Sudan and Switzerland, had granted women the vote.Footnote ⁸⁶ Because of the relative lack of variation prior to 1893 and after 1955, the analysis focuses in detail on the intervening years.

Figure 1. Democracies with and without women's suffrage

Perhaps surprisingly, women have been officially granted the right to vote in the vast majority of countries, including those that do not actively hold national elections. Only five states—Brunei, Saudi Arabia, Oman, Qatar, and the UAE—had not granted women's suffrage by 1999. Given our hypothesis that the interaction between women's suffrage and a state's level of democracy will determine the degree to which female preferences on the use of force constrain state behavior, we do not expect states that have legally granted suffrage but are not sufficiently democratic to behave more peacefully than other states. The coding of our primary variable women's suffrage therefore depends upon two components. First, it considers whether any women in the state are able to vote in national elections. Second, it incorporates a baseline measure of democratic institutions below which we would not expect suffrage to affect a state's behavior. Because we have no theoretical reason to believe that women's votes should affect the conflict propensity of autocratic states, the variable is coded 1 if suffrage for national elections has been extended to women in a state and the state's Polity score is 1 or higher. The variable is otherwise coded as 0. To ensure that our results are not dependent on this specific operationalization, in robustness analyses we also operationalize democratic institutions either (1) using a polity threshold of 6, or (2) using the specific polity indicators for open and competitive elections (i.e., if the state's democracy, not polity, score is 6 or higher. See Part B.12 of the appendix for further discussion of these robustness analyses). We then created the dyadic variable joint women's suffrage democracy, coded 1 if women's suffrage equals 1 for both states in the dyad.

To assess the potential monadic effects of women's suffrage on the likelihood that a democratic state initiates conflict, we created two dichotomous variables. The first variable, democracy without women's suffrage, is constructed using our suffrage variable and the standard coding of democracy and is coded 1 if the state has a Polity score of 6 or above and if women are not allowed to vote in national elections, and is otherwise coded 0. The second binary variable, autocracy, utilizes the standard coding of autocracy and is coded 1 if the state's Polity score is 5 or below and otherwise coded as 0. Inclusion of these two variables within our core monadic analyses enables us to directly compare the probabilities of initiation for autocracies and democracies without women's suffrage to the baseline probability of initiation for democracies with women's suffrage. To investigate whether the pacifying monadic effects of women's suffrage increase as the state becomes more democratic, we also created a continuous measure of democracy and women's suffrage. This variable, polity * women's suffrage, interacts the state's democracy score with the variable women's suffrage.Footnote ⁸⁷

To assess the effect of women's suffrage within a dyadic setting, we employed two very similar approaches. The first approach utilizes binary variables similar to those described for the monadic analysis. First, we created the variables joint autocracy, coded 1 if both states within the dyad have polity scores of 5 or less and otherwise as 0, and joint democracy without women's suffrage, coded 1 if both states have polity scores of 6 or higher and if neither state allows women to vote and is otherwise coded 0. If our monadic hypothesis is correct, we should expect significant differences in the conflict propensity of mixed dyads. For instance, a mixed dyad containing one democracy with women's suffrage and one democracy without women's suffrage may be significantly less conflict prone than a dyad containing one democracy without women's suffrage and one autocracy. To account for these potential differences, we include three additional variables to account for types of mixed dyads. The first, democracy without women's suffrage / autocracy, is coded 1 if one state in the dyad has a polity score of 6 or greater and does not allow women the vote and the other state has a polity score less than 6. The second, democracy without women's suffrage / democracy with women's suffrage is coded 1 if one state in the dyad has a polity score of 6 or greater and does not allow women the vote and the other state has a polity score greater than 6 and allows women to vote. The final variable, democracy with women's suffrage / autocracy, is coded 1 if one state in the dyad has a polity score of 6 or higher and allows women the vote and the other state has a polity score of 5 or less. The inclusion of these variables ensures that the baseline for comparison within the dichotomous dyadic models is the likelihood of dispute involvement between joint women's suffrage democracies.

The second dyadic approach for analyzing suffrage's dyadic effects employs continuous variables for democracy. We utilize the standard DEM_L variable which is obtained by first calculating the Polity score of each state, a value which ranges from -10 for extreme autocracies to 10 for the most democratic states, and then taking the lower of these two calculations within each dyad. For our primary variable of interest, we interact our variable assessing joint women's suffrage with the lowest democracy score to create dem_L * joint women's suffrage.

In keeping with the standard empirical approach, we also include the variable dem_H, which lists the highest Polity score within the dyad. Prior analysis of the democratic peace has shown that increasing the highest democracy score while holding the lowest democracy score constant correlates with an increase in conflict propensity. This finding is credited to the idea that states with greater differences in regime type are more likely to fight each other.Footnote ⁸⁸ We also include the variable at least 1 women's suffrage, coded 1 if one or both states within the dyad allow women's suffrage and have a polity score higher than 5 and 0 otherwise. Finally, we include the analogous variable suffrage-democracy_H = max((women's suffrage_StateA * polity_StateA), (women's suffrage_StateB * polity_StateB)).

Dyadic Findings

We now assess our dyadic hypothesis that dyads in which both states are democracies with women's suffrage are less likely to fight than joint autocratic dyads and joint democratic dyads without women's suffrage. We estimated seven models of the relationship between suffrage and dispute involvement using logistic regression on nondirected dyad data from 1816 to 2010, presented in Table 5.Footnote ⁹³ Model 1 assesses the relationship between dispute propensity and the five dichotomous measures accounting for dyadic regime type over the period from 1816 to 2010 and column 2 presents analysis of this model between the years 1893 and 1955. The coefficients in both models indicate that all dyadic types, except mixed democratic dyads in which one state has women's suffrage, are significantly more likely to engage in conflict than joint democratic dyads with women's suffrage. The predicted probabilities of conflict propensity by dyad type, generated using model 1, are presented in Figure 4.Footnote ⁹⁴ The figure illustrates that the predicted probability of disputes within joint women's suffrage dyads is significantly lower than the probability within joint democratic dyads without women's suffrage.Footnote ⁹⁵

Figure 4. Predicted probability of conflict by dyad type, 1816–2010

Table 5. Dyadic models of dispute involvement, 1816–2010

Notes: Robust standard errors clustered by dyad in parentheses beside. * p < .05; ** p < .01; *** p < .001.

The same is true of the relationship with joint autocracies, as the dyadic hypothesis predicts. The probability of dispute involvement within joint democracies without women's suffrage cannot, however, be distinguished from the conflict propensity of joint autocracies. The dispute propensity of joint democratic dyads in which one state grants women the vote and the other state does not is significantly lower than the likelihood of dispute between two democracies without women's suffrage.Footnote ⁹⁶ The likelihood of dispute involvement between one autocracy and one women's suffrage democracy cannot, however, be distinguished from the likelihood within joint autocratic or joint democratic dyads without women's suffrage.Footnote ⁹⁷

Models 3 through 7 in Table 5 analyze the relationship between the continuous measures of suffrage and dispute propensity. Model 3 replicates the standard model of the democratic peace from Oneal and Russett, with the inclusion of variables accounting for nuclear capabilities.Footnote ⁹⁸ In keeping with prior results, the model shows that dispute involvement and the lowest democracy score are negatively and significantly correlated while high democracy score is positively and significantly correlated with dispute activity. Model 4 includes within this standard model the interaction variable dem_L * joint women's suffrage as well as its component variables, the monadic measure of women's suffrage and our continuous variable suffrage-democracy_H. Because our primary variable is an interaction, the size of the coefficient cannot be interpreted directly. We do see, however, that it is negative and significant within this model, which also includes the variable accounting for women's civil liberties, and within model 7 which includes dyad fixed effects.Footnote ⁹⁹

Figures 5 and 6 present the substantive interpretation of the relationship between dyadic suffrage and conflict propensity as estimated using model 5. The three graphs in Figure 5 each depict the probability of disputes in dyads in which both states share the same Polity score. As the graphs show, the effect of increasing dyadic levels of democracy on dispute propensity appears to depend significantly on the gender of eligible voters within the two states. An increase in dyadic democracy within male-only voting populations is associated with a potential very slight decrease in the likelihood of conflict. An increase of dyadic democracy within joint women's suffrage dyads is, by contrast, associated with a significant decline in conflict propensity.

Figure 5. The probability of dispute as a function of dyadic suffrage, 1816–2010

Figure 6. Change in dispute propensity as a result of joint women's suffrage

Figure 6 presents the percentage difference in the predicted probability of conflict between joint women's suffrage dyads and democratic dyads in which neither state grants women the vote. The figure illustrates the significant decline in dispute involvement that correlates with the joint adoption of women's suffrage within increasingly democratic societies. The probability of disputes within dyads in which both states possess Polity scores of 7 and in which neither state has granted women the right to vote is, for instance, 39 percent higher than in otherwise-identical dyads with joint women's suffrage. The pacifying effects corresponding within dyadic women's suffrage become even more prominent as dyads become more democratic. Substantively, a one-standard-deviation increase in women's civil liberties similarly correlates with a 28.7 percent decline in dispute propensity (p < .001).

To further probe the dyadic effects of gender on democratization, we also analyzed the standard model of the democratic peace (model 3) over the period 1816 to 1892 when all democratic states prohibited women from voting. We find that an increase in the low dyadic democracy score from 0 to 10 during that period was associated with a 38.8 percent increase in dispute propensity (p < .05).Footnote ¹⁰⁰ The aggressive behavior of the United States, France, and Great Britain, each of which had a democracy score of 6 or higher during this period, likely explains a significant proportion of this increase. Omission of those three powerful states from this analysis does not, however, reduce the size or direction of the coefficient (p = .11).

We also analyzed the effects of democracy and women's suffrage during the specific period of 1890 to 1930, a period in which roughly two-thirds of democracies granted women the vote, and thus half of the number of democratic dyads had joint women's suffrage and half of democratic dyads did not. While the relative paucity of data during this period renders it difficult to draw confident inferences, the estimated conflict trends during this period roughly mimic those presented in Figure 5, as shown in Part B.8 of the appendix. A one-tailed t-test of dispute propensity during this period also indicates the following. While jointly democratic dyads during this time were 38.9 percent less likely to experience a dispute than jointly nondemocratic dyads (p = .0448), democratic dyads in which neither state allowed women's suffrage were 189 percent more likely to experience a dispute than a joint-autocratic dyad (p = .0005). Democratic dyads with joint suffrage were, by contrast, 93.2 percent less likely to experience a dispute than democratic dyads in which neither or only one state allowed women's suffrage (p = .0003).

Additional analyses in Parts B.7 to B.9 of the appendix indicate that the dyadic results presented here are robust to alternative specifications. First, we evaluated the possibility that results were driven by dyad-period-specific effects. Given the bias-variance trade-offs of each specification, and the principle of avoiding arbitrary specification choices, we analyzed five specifications, systematically varying the width of the time window. We included fixed effects for every dyad-fifty, dyad-forty, dyad-thirty, dyad-twenty, and dyad-ten years within model 5 in Table 5. The coefficient is negative in all models, and significant in the dyad-forty, dyad-thirty, and dyad-twenty models. Second, to evaluate the possibility that our result is driven by a particular historical period, we reran our analysis truncating the data (1) to the period prior to 1930, (2) the interwar period, and (3) to the periods before and after the interwar period, and (4) to the period before and after the Cold War. Third, we examined whether dyadic women's suffrage predicts a decline in the use of force; it does.

Alternative Hypotheses

We examined the validity of a number of alternative hypotheses that might explain the negative correlation between dyadic women's suffrage and dispute involvement. It is possible, for instance, that the timeline of women's suffrage serves as a proxy for the advent and institutionalization of more cooperative international norms during the interwar period and after World War II. As we suggested, however, analysis of the relationship between time and women's suffrage indicates that women's suffrage is not an artifact of historical time period.Footnote ¹⁰¹ First, all models here include a control for year—a simple calendar year variable—which would pick up some temporal trends (particularly those that most closely correspond to a linear change in log-odds over time). Second, the inclusion of twenty-year-dyad fixed effects, for instance, does not alter the relationship between suffrage and lower dispute propensity. Third, as stated, both the monadic and dyadic results are robust, for instance, to analysis from 1893 up through the second World War and within the periods from 1893 to 2010 excluding the Cold War, and from 1930 to 2010, among many others. The dyadic results are robust to analysis within the time periods 1893 to 1930 and during the interwar period, as well as additional periods reported in the appendix.Footnote ¹⁰²

We also considered the possibility that women's suffrage is conflated in one of two ways with the strategic context of a state.Footnote ¹⁰³ On the one hand, states may be, for various reasons, more likely to adopt suffrage following periods of conflict.Footnote ¹⁰⁴ The results we described, in such a case, might simply be capturing war weariness on the part of recent adopters. Numerous pieces of evidence speak against this hypothesis. First, analysis of the interwar period indicates that significant differences in the dispute propensity of suffragist and nonsuffragist democracies existed during this period, suggesting that the results are not explained by differences between pre- and post-World War behavior. McConnaughy also notes that many US states had already extended suffrage (some as early as the 1890s) by the time federal ratification occurred in 1920.Footnote ¹⁰⁵ Similarly, Teele argues that World War I did not cause British suffrage.Footnote ¹⁰⁶ Also, we find that the results presented in Figure 6 hold when omitting those states that adopted suffrage within three years following major war.Footnote ¹⁰⁷ Moreover, in the within-country analysis described in Part B.5 of the appendix, a state's average rate of dispute involvement in years t-20 to t-10 before the adoption of women's suffrage is 17.8 percent higher than in years t+10 to t+20 after women's suffrage (p = .03).

Within-country evidence also calls into question the converse hypothesis—that states with fewer strategic threats may be more likely to adopt women's suffrage than states facing heightened security threats. In such a case, peace, rather than arising as a product of women's suffrage, would instead facilitate the extension of suffrage. However, as we pointed out, states typically experience significantly higher rates of conflict in the decades immediately before the adoption of suffrage than they do after suffrage.

Finally, it is possible that what affects conflict propensity is not the extension of suffrage by gender but rather the extension of suffrage more generally by class. Such extensions by class have typically involved the elimination of literacy, property, or wealth requirements, or the removal of racial barriers to voting. To assess the validity of this hypothesis, we performed a within-country analysis on the effects of extending suffrage by class among male voters using data on suffrage extensions presented by Przeworski.Footnote ¹⁰⁸ As Table 6 shows, these other extensions of suffrage are not associated with greater peace. In fact, they are associated with more conflict, with effect estimates of +10/+23/+27 percent for various temporal windows. While these within-country results are based on a relatively small sample size, they provide little reason to think that our results are being driven by extension of suffrage in general or by class, rather than by gender.

Table 6. Comparison of disputes before and after granting suffrage