Principal Phenomenon under Investigation

The definition and measurement of democracy is one of the most extensively debated subjects in political science,Footnote ³ and it is not our purpose here to engage in these debates. Our focus is on the perceived desirability of democracy as a political system, one of the principal aspects of diffuse support reflecting Lipset’s (Reference Lipset1959, 83) concept of political legitimacy, a “belief that existing political institutions are the most appropriate or proper ones for the society.” A study of the panel data of 2,435 nationally aggregated estimates of support for democracy from 135 countries finds that diffuse support, when compared to other measures, is the most significant correlate of democracy’s survival and endurance (Claassen Reference Claassen2020). Corroborating those findings at the individual level, quantitative analysis of focus group narratives in Chile and Argentina indicate that desirability of democracy as a system, compared to 48 other measures, generated the most consistent variation in strongly held positive and negative orientations across multi-item democracy support profiles in both countries (Carlin Reference Carlin2018). These studies indicate that focusing on this measure is theoretically and substantively warranted and important—in no small part because it has proved to be diagnostic of real-world democratic futures regardless of what individuals specifically mean by democracy when answering survey questions.

We ask how a society’s exposure to war—focusing on ordinary civilians rather than the combatants—may affect this diffuse support for democracy at the individual level.

Theoretical Puzzle

Although we lack cumulative research on individual-level support for democracy in cases like Ukraine during Russia’s invasion, two major literatures and debates therein offer relevant insights.

First, we have research linking political preferences to generalizable psychological factors due to war conditions, notably death and destruction, which Ukraine has continually experienced since 2014 and particularly since 2022. But the assessments of specific attitudinal effects diverge. On the one hand, extensive studies show how war eroded support for democracy in different countries due to “shattered assumptions” of personal invulnerability (Janoff-Bulman Reference Janoff-Bulman1992), “the ball of rage” syndrome (Chemtob et al. Reference Chemtob, Novaco, Hamada and Gross1997), “mortality salience” (Rosenblatt et al. Reference Rosenblatt, Greenberg, Solomon, Pyszczynski and Lyon1989), personal loss and traumas (Canetti-Nisim et al. Reference Canetti-Nisim, Halperin, Sharvit and Hobfoll2009; Dyrstad Reference Dyrstad2013), social intolerance (Bakke, O’Loughlin, and Ward Reference Kristin, John and and Michael2009; Tir and Singh Reference Tir and Singh2015), the prioritizing of control over freedom (Davis and Silver Reference Davis and Silver2004), and “authoritarian thinking” (Hetherington and Suhay Reference Hetherington and Suhay2011).

On the other hand, we have research indicating that the impacts of war loss and trauma may be muted or may lay the ground for stronger support for democracy. Individuals may interpret even grave threats of violence as manageable “perturbations” that are followed by the resetting of democratic norms (Sniderman et al. Reference Sniderman, Petersen, Slothuus, Stubager and Petrov2019). The human tendency to become desensitized to repeated stimuli means political attitudes may change less as violence persists (Nussio Reference Nussio2020). Individual post-traumatic growth in response to life-threatening events may boost prodemocratic attitudes (Tedeschi and Calhoun Reference Tedeschi and Calhoun2004) and political tolerance (Carmil and Breznitz Reference Carmil and Breznitz1991).

The second category of relevant research points to generalizable social factors affecting political preferences in wartime, notably group mobilization. Here, too, conclusions diverge. On the one hand, war arguably raises public support for personalist centralized rule. External threats have been shown to bolster “rallying-’round-the-flag” behavior (Mueller Reference Mueller1970) in different contexts, with the public more willing to delegate authority to individual leaders and the executive over other institutions (Frye Reference Frye2019; Lai and Reiter Reference Lai and Reiter2005; Lambert et al. Reference Lambert, Scherer, Schott, Olson, Andrews, O’Brien and Zisser2010; Tudor and Slater Reference Tudor and Slater2021), mute political opposition (Baum and Groeling Reference Baum and Groeling2008; Brody and Shapiro Reference Brody, Shapiro and Brody1991; Oneal, Lian, and Joyner Reference Oneal, Lian and Joyner1996), and bolster social intolerance (Godefroidt Reference Godefroidt2023; Grosjean Reference Grosjean2014). Moreover, rallying for the incumbents has been shown to be more durable and intense under conditions similar to Ukraine’s, such as during major uses of military force, severe hostility between parties, and fighting over territory (Baker and Oneal Reference Baker and Oneal2001; Oneal, Lian, and Joyner Reference Oneal, Lian and Joyner1996, 265; Tir Reference Tir2012)—particularly territory of symbolic significance, such as Ukraine’s Crimea (Alexseev and Hale Reference Alexseev and Hale2019).

On the other hand, we have evidence that the same quest for group unity while under threat, often in the same or similar countries, gives rise to grassroots civic mobilization, favoring social inclusivity within a nation and strengthening liberal democratic values (Berinsky Reference Berinsky2009). Skocpol (Reference Skocpol2002) noted a 45% rise in public trust in the US government in response to the September 11 terrorist attacks and cited historical evidence that the surge in trust was even more significant during World War II, when Americans needed to “roll bandages for GIs or collect scrap metal to make airplanes.” Similar conclusions were drawn about September 11 (Gaines Reference Gaines2002; Woods Reference Woods2011), and about responses to terrorist attacks in Spain (Dinesen and Jæger Reference Dinesen and Jæger2013) and France (Van Hauwaert and Huber Reference Van Hauwaert and Huber2020). In Ukraine, Darkovich, Savisko, and Rabinovych (Reference Darkovich, Savisko and Rabinovych2023) document how Russia’s full-scale invasion boosted political decentralization as citizens mobilized predominantly around local communities (hromady), and President Volodymyr Zelensky’s popularity surged along with—and not at the expense of—support for democratic institutions and values (Onuch and Hale Reference Onuch and Hale2022).

While showing that war exposure and group mobilization are important to factor into any analysis of war effects on political attitudes, divergent findings point to a theoretical puzzle: Why would the same psychological and social factors associated with war violence affect polity preferences differently? If war brings with it insecurity and the quest for group unity across societies, why may support for democracy in any given society weaken or strengthen in wartime?

While this partly reflects the problem of drawing generalizable inferences from studies that use different variable specifications in different contexts with different data types and methods, the puzzle points to two knowledge gaps. The first one is war type: almost all the individual-level findings on how violence affects support for democracy come from cases of civil wars, insurgencies, terrorist attacks, or militarized interventions in other states.Footnote ⁴ However, econometric analysis (Besley and Persson Reference Besley and Persson2010) and survey research (Grosjean Reference Grosjean2014) indicate that war type may significantly affect at least some political preferences, with exposure to internal war more likely to drive social antagonisms, but exposure to interstate war more likely to favor social inclusivity. Building on these insights, we further examine how individual-level exposure to interstate wars might affect diffuse support for democracy, particularly under the threat of territorial conquest.

The second factor has to do with society and national identity mobilization. Most of the relevant rallying research focuses on the “shifts in attitudes toward the American president” (Feinstein Reference Feinstein2022; Lambert et al. Reference Lambert, Scherer, Schott, Olson, Andrews, O’Brien and Zisser2010, 886) and on leadership approval elsewhere (Feinstein Reference Feinstein2018, Frye Reference Frye2019; Hale Reference Hale2021; Lai and Reiter Reference Lai and Reiter2005). However, Godefroidt’s (Reference Godefroidt2023) meta-analysis of political attitudes under the threat of terrorism in 30 countries over 35 years (drawing on 326 studies of 400,000 individuals) showed that rallying is contingent on a country’s context and its issues. A significant “rally around President Bush” effect in response to September 11 was not observed in many other countries, particularly in relation to non-Islamist threats. Indications have emerged that rallying is multifaceted. Societies may rally not only for leaders, but also for institutions (Hetherington and Nelson Reference Hetherington and Nelson2003) and military operations (Feinstein Reference Feinstein2018). The growing salience of civic national identity in Ukraine within a year of the onset of the Donbas war in 2014 (Kulyk Reference Kulyk2016; Pop-Eleches and Robertson Reference Pop-Eleches and Robertson2018), in contrast to a drop in then president Petro Poroshenko’s support, implies that societies may literally rally around the flag and other state symbols, for nation and polity, regardless of declining support for the leader.Footnote ⁵ As a next step, we examine how civic identity mobilization in wartime may translate into support for democracy as a political system.

Geosocietal Support for Democracy: The Logic and the Hypotheses

To develop a generalizable explanation of Ukraine’s democratic resilience, we propose that support for democracy in a state at war can be geosocietal, in the sense that it is likely to be significantly conditioned by both the public salience of geopolitical (external) threats to democracy and the society’s capacity for civic national mobilization. The geosocietal logic indicates that the international system-level context (notably, war type, the nature of the aggressor and target states, and international alliances) would set the stage for whether civic national mobilization may affect support for democracy positively or negatively and reinforce these effects. At the same time, the society’s intrinsic capacity for civic mobilization, based in domestic conditions, will contribute significantly toward the total effect as a fundamental underlying condition.

We know from well-established literatures about the significant impacts of international-level factors—such as power balances, institutions, and norms—on the domestic structures and politics of states (known as the “second image reversed,” as described by Gourevitch [Reference Gourevitch1978]), and more specifically about the importance of war in the formation of states and national identities (Greenfeld Reference Greenfeld1993; Tilly Reference Tilly1992). Expanding on these insights, we recognize that national identity is not only intrinsic (rooted in a sense of common citizenship or traditions and amplified by common effort and interests) but also relational, grounded in the human tendency to “cleave and compare” across groups (Horowitz Reference Horowitz2001; Nair and Sambanis Reference Nair and Sambanis2019). This means national identity must entail a sense of where one’s nation-state stands in relation to others—of its distinctiveness and commonalities, its strengths and weaknesses, and the costs and benefits of its external interactions. We refer to this identity dimension as geopolitical. Factoring this dimension into the analysis of democracy support is essential to our contribution.

Thе extent to which a threat is external is a critical factor. The fundamental insight of social identity theory (SIT) (Tajfel and Turner Reference Tajfel, Turner, Worchel and Austin1986) is that invasive external (out-group) threats play a particularly strong part in identity mobilization by amplifying intergroup differences across multiple categories. The perceived degree to which the threat of war comes from another state or external actor is likely to amplify the perceived differences between the polity of the aggressor and that of the target state. Intergroup bias—a combination of out-group hostility and in-group solidarity—would then translate into individuals casting the opposing state’s polity in a more negative light, and their own polity in a more positive light, than they would in the absence of the external war threat.

The SIT logic also implies that sensitivity to external threats would depend on how salient civic national (in-group) identity is within a society before it comes under external attack. In theory, the stronger the salience of civic national identity (i.e., the degree to which citizens feel pride in their citizenship) in a democratic or democratizing state before such an invasion, the more there is at stake to defend. Consequently, we would expect a stronger response in support of democracy. A relevant indication comes from a meta-analysis by Bauer and colleagues (Reference Bauer, Blattman, Chytilová, Henrich, Miguel and Mitts2016, 268), who note that war violence affects individual political and social attitudes such as trust differently depending on whether the aggressor is internal or external and whether a population under attack “already possesses a national identity.” Though Bauer and colleagues do not examine whether this logic applies to diffuse support for democracy, their analysis suggests it is plausible through evolution of parochial altruism. On these considerations, we may expect to observe both direct and interaction effects of external attack and national identity salience on support for democracy.

A derivation of SIT known as the common in-group identity model (CIIM) (Gaertner and Dovidio Reference Gaertner and Dovidio2000) also indicates that when individuals feel their whole nation or country is under external attack they are more likely to strengthen their civic national identity (i.e., identify more strongly as Americans, French, or Ukrainians) and group cohesion at the national level, while reducing the salience of ethnic, religious, or other subnational group identities within the state (Rodriguez-Carballeira and Javaloy Reference Rodriguez-Carballeira and Javaloy2005) and boosting social capital and trust in institutions (Besley and Persson Reference Besley and Persson2010; Grosjean Reference Grosjean2014). By extension, this logic suggests that when geopolitical threats come from an authoritarian state, the salience of sociodemographic characteristics underlying these subgroup identities in a democratizing state would decline and they would have a less significant impact on support for democracy than civic national identity mobilization and war threat.

SIT also prompts us to recognize that because in-/out-group categorization at the interstate (geopolitical) level is likely to compound intergroup bias, then support for democracy could be mobilized not only by external attack, but also by perceived external (interstate) threats of different types. For example, an external actor may not only threaten to undermine or suppress democracy in the target state by direct force but may also (or instead) undermine the state’s prospects of membership in international democratic coalitions by waging or threatening limited wars or incursions, or by supporting insurgencies, in the hope of depleting the target state’s resources and demoralizing its population. If the external actor’s aggression is indirect and/or obfuscated, (which could well be the case in internationalized hybrid conflicts), the target state’s mobilization of geopolitical identity will reflect the extent to which the target state’s citizens blame the external actor for war. If such aggression is direct and overt, we would expect the war onset to mobilize out-group hostility directed at the external authoritarian aggressor and bolster in-group solidarity in support of democracy in a society. Shortly after Russia’s full-scale invasion in February 2022, one Kyiv resident vividly summarized this logic of geosocietal mobilization: “Don’t you see what kind of monster is attacking us? Of course, we fight for democracy!”Footnote ⁶

Representing our theoretical propositions in the order laid out above, we formulate the following hypotheses on geosocietal support for democracy applicable to Ukraine (i.e., within a democratic or democratizing state under military attack):

Principal Variables of Interest

For the dependent variable we use the UNASIS survey question closely related to diffuse support for democracy, which has been shown to be a significant predictor of democracy survival (Claassen Reference Claassen2020) and internally consistent across cultural settings (Carlin Reference Carlin2018). We asked: “How important do you find the following personally to you: the democratic development of our country.” The answers were on a five-point Likert scale.Footnote ¹² We use two specifications of this measure: (1) the original scale to estimate the degree of democracy supportFootnote ¹³ and (2) a dummy variable with “important” and “very important” coded “1” and all other responses coded “0” to estimate the probability of explicit support, rather than differences on a scale.Footnote ¹⁴ With other surveys showing that most Ukrainians define democracy as freedom of expression (National Democratic Institute 2022; Szostek and Orlova Reference Szostek and Orlova2022), we use the importance of free speech on the same response scale in robustness checks.

For the civic national identity dimension of geosocietal mobilization, we use National Pride based on the question: “How proud or not proud are you to be a citizen of Ukraine?” Responses were on a five-point Likert scale. For the geopolitical threat dimension, we use Russia Aggressor, based on the question: “Against whom do you believe the Ukrainian forces (the army, the national guard, police, the state security service, and others) mostly fight in the Donbas? (Please select one option from the list.)”Footnote ¹⁵ Respondents who stated that Ukraine was primarily fighting against “local rebels financed, armed, and run by Russia,” or “mercenaries (local or foreign fighting for money) supported by Russia,” or “the regular Russian armed forces” were coded “1.” The rest (who picked “local insurgents who act independently,” or “local insurgents and foreign volunteers who act independently,” or “hard to say”) were coded “0.”Footnote ¹⁶ In the 2021–22 panel study, War Year serves as a proxy for geosocietal mobilization, given practically no variation on pride and threat after Russia’s full-scale attack.

Control Variables

We selected the following control variables to test for established alternative explanations and, in doing so, to minimize the possibility that some unobserved value may drive the results.

Media Effects

We run robustness checks for reliance on Russian versus Ukrainian news sources (see appendix table 1A.8).

Table 1 War Impact on Democracy Support, Nov. 2021–Jul. 2022: Linear Mixed Effects Model

Note: Estimate of fixed effects; ***p < 0.001; **p < 0.01; *p < 0.05.

Whereas no statistical model can rule out all unobserved value effects, our single-country/region large-N design with theoretically grounded, context-specific control variables provides a strong guardrail.Footnote ²¹ Reverse causality is controlled with the inclusion of the pre/post-war-onset panel data tracking the same individuals. In cross-sectional tests, for reversed causality to drive the findings would imply that decreases of support for democracy in wartime would systematically depress both national pride and perceived external threat, contradicting the evidence.

The Donbas Regional Study

War Exposure

We tasked UNASIS to oversample from the Donbas GCAs in their 2017 annual national survey, resulting in 399 new respondents for a total regional sample of 565. That gives us enough power to identify statistically significant relationships of interest. The Donbas setting provides three design benefits. First, we control for region. This is important because regional differences capture complex interactions of ethnic, linguistic, socioeconomic, cultural, institutional, and other factors that remain subject to divergent interpretations and scholarly debate (Arel Reference Arel2002; Darden Reference Darden, Colton, Frye and Legvold2010; Frye Reference Frye2015; Kudelia and van Zyl Reference Kudelia and van Zyl2019; Kulyk Reference Kulyk2011; Kuzio Reference Kuzio2017; O’Loughlin Reference O’Loughlin2001; Sasse Reference Sasse2010). In the Donbas these effects have arguably accounted for significant regional specificity in Ukrainian politics, relations with Russia, governance, social identity, and warlordism (Driscoll Reference Driscoll2015; Kudelia Reference Kudelia2017; Osipian and Osipian Reference Osipian and Osipian2006; Reference Osipian and Osipian2012).

Second, we can assess the impacts of war exposure more rigorously. Our Warzone binary variable distinguishes between respondents who reported living in the “zone of combat operations” as “1” (N = 203) and others as “0” (N = 350).Footnote ²² On theoretical, methodological, and empirical grounds, physical proximity to war violence could be an important variable in its own right, as it may capture unanticipated or hard-to-operationalize effects aside from loss and stress (Bakke, O’Loughlin, and Ward 2009; Dyrstad Reference Dyrstad2013).

Using respondents’ self-categorization has significant advantages in the context of the 2014 Donbas war. To begin with, this measure identifies individuals to whom war exposure has been salient enough to report (thus controlling for otherwise hard-to-estimate individual-level salience and the effect duration of any discrete war event that count measures, typically based on media or observer reports, cannot control). It also captures views of respondents in areas that did experience war damage, but whose specific personal experiences may not have been significant enough to register in media or observer reports (notably, in our sample, the Sartana settlement).

At the same time, this measure captures respondents who experienced both principal types of war exposure in the Donbas in 2014–17: (1) in settlements seized by Russia’s proxy forces but retaken by Ukrainian forces in 2014, mostly after fierce battles (Melnyk et al. Reference Melnyk, Pavlyuk, Petrov and Glushko2019, 7–13), and (2) in settlements with sustained war fighting of variable intensity along a 250-mile “line of control” cutting through the Donbas from north to south. The line changed little from mid-2015 through survey time in mid-2017. Figure 3 illustrates how the distribution of reported war exposure in our survey broadly reflected the uneven patterns of fighting in the Donbas war.

Figure 3 Respondent’s Warzone Experience in the 2017 Donbas Survey

Note: N = 565.

We see larger proportions of respondents reporting war fighting in the region’s center where Russia-backed forces made the most advances in 2014; yet within that area we see more respondents reporting warzone experience in places more intensely contested (Lyman [75%], Kramatorsk [100%], Slov’yansk [70%], and Druzhkivka [60%]) than many settlements closer to the front lines—for example, Chasiv Yar (26%), Bakhmut (27%), and Pokrovs’ke (10%). In the south, we see respondents reporting war exposure in Mariupol (17%) and Pokrovs’ke (20%), where fighting took place in the spring of 2014 as local residents and Ukrainian forces overturned the Russia-backed takeovers, but was less intense than in Kramatorsk and Slov’yansk.

Coding respondents’ reports also captures indirect war experiences that event counts typically miss. According to our Donbas focus groups (in Druzhkivka and Mariupol, 2017), those experiences included witnessing troop movements; encountering and commiserating with war victims; fleeing the horrors of war and not knowing if one might return; stressing out when discovering unexploded bombs at children’s playgrounds; or knowing people whose hair turned gray overnight or who had been wounded, had disappeared, or had died (illustrative citations in the appendix, section A4).

Ordinary Donbas residents in our survey did not choose war experiences, war experiences chose them. Since most intense battles in 2014–17 were for cities, significantly more urban than rural residents in our sample reported living through war fighting (reflected in Pearson Rs for Warzone of 0.183** with urban residency; 0.107* with education; and 0.119* with income). In contrast, factors uncorrelated with the urban–rural divide—age, gender, and language use—were also uncorrelated with Warzone. Further validating our measure, we found that reporting war exposure significantly correlated with Loss (R = 0.338***) and Stress (R = 0.146**).Footnote ²³ Yet, we also found Warzone residents to be 10–12% more likely to view democracy and free speech as important compared to other Donbas residents (both binary relationships being significant).

Statistical Analysis

To test our hypothesis in the context of the Donbas war, we estimated full effects of each independent variable while holding all other predictors constant using multiple linear regression with robust standard errors for Democracy Importance Scale and logistic regression for Democracy Importance Probability. Footnote ²⁴

The results show strong support for H1 through H4. Figure 4a reports the estimated percentage-point change in Democracy Importance Scale from minimum to maximum value for each predictor (all rescaled from 0 to 1). National Pride and Russia Aggressor both have highly significant positive full effects (at p < 0.001, there is less than a 0.1% probability that these relationships were due to chance). Respondents who were very proud of their Ukrainian citizenship ranked the importance of democracy about 21% higher than respondents who were not proud at all. Respondents who saw Russia as the aggressor in the Donbas war ranked it 12% higher than others. Confidence intervals indicate, however, that Russia Aggressor effects are less prone to variation (about half of respondents who saw Russia as the aggressor were likely to rate the importance of democracy about the same or higher than about a quarter of the respondents who were very proud of their Ukrainian citizenship).

Figure 4 Donbas GCAs, 2017: Geosocietal Factors Consistently Predict Democracy Support

Figure 4b shows that the average probability of a respondent considering democracy to be important increases by nearly 24% when National Pride changes from “not proud” to “very proud,” and by over 21% when Russia Aggressor changes from “no” to “yes” with all other predictors held constant.Footnote ²⁵ Once again, the Russia Aggressor factor is more reliable in the sense that the probability of support for democracy among 95% of respondents who considered Russia to be an aggressor is higher than among about one-fifth of respondents at the highest National Pride level. This is reflected in the higher significance level for Russia Aggressor (p < 0.001 compared to p = 0.005 for Pride).

H3 is supported in that (1) most social cleavage markers are not significant, and (2) for those that are, the significance levels are lower than for Pride and Russia Aggressor (cf. above, with p = 0.003 for education and p = 0.014 for age on the democracy support scale, and p = 0.012 for income and p = 0.019 for age). The sole exception was the effect of education on support probability (p = 0.002) being slightly more significant than Pride (p = 0.005).

We also considered that demographic variables may have indirect effects on democracy support by affecting related attitudes, and estimated them with a procedure known as seemingly unrelated regression (sureg in Stata). It shows that the proportion of Pride and Russia Aggressor total effects on the democracy support scale—mediated by education, income, age, rural residency, language, and female gender taken together—while statistically significant (p = 0.005 for Pride and p = 0.002 for Russia Aggressor), was substantively minor (13% and 13.2%, respectively). While supporting H3, these findings indicate that pride and threat do not displace well-established sociodemographic factors, but act alongside them, albeit more strongly and consistently.

To test H4 we estimated marginal effects of the interaction term between Pride and Russia Aggressor on Democracy Importance Scale using Stata. Figure 5 shows the effects of pride and the Russian threat on democracy support while controlling for all other variables, with the estimates from the same regression model as reported above. Based on the visual inference rule (Cumming and Finch Reference Cumming and Finch2005) for error bar overlaps, the interaction effect of Pride with Russia Aggressor was statistically significant except for those respondents who were not at all proud of their Ukrainian citizenship. That total interaction effect (the difference between the leftmost bottom dot and the rightmost top dot) is about 1.5 on a five-point scale—that is, 50% higher than the effect of Pride while holding Russia Aggressor constant (about a one-point total difference from left to right on either line), and three times as high as the effect of Russia Aggressor while holding Pride constant at any level (a 0.5-point vertical gap between the lines). This upholds H4.Footnote ²⁶

Figure 5 Pride and Threat Interaction on the Democracy Importance Scale (Donbas GCAs, 2017)

With free speech importance as the dependent variable, we got near-identical results, with the main difference being that Pride related to the probability of free speech support at 2.4 percentage points below the conventional 95% significance level (at p = 0.074) (see appendix section A5).

Finally, we observe that while macrolevel geosocietal context measures (pride and threat) are statistically significant, none of the direct microlevel measures of war exposure (Loss, Stress, and Warzone) are. Warzone’s relationship to democracy importance (significant at the bivariate level, R = 0.144**) is no longer significant in the multiple regression. Sureg tests indicate this probably happens because about 41% of the Warzone effect is mediated by Russia Aggressor.

Out-of-Initial-Sample Tests: Ukraine 2017 and 2018

Our Donbas results held up across the rest of Ukraine in the same year and the year after with the same regression tests (the only two UNASIS annual surveys when all measures of interest to us were available).Footnote ²⁷ As figure 6 shows, both National Pride and Russia Aggressor are significant predictors of democracy support in all four tests (at p < 0.001, except for Russia on Democracy Importance Scale in 2017 at p = 0.003).

Figure 6 Pride and Threat Boost Democracy Support across Ukraine outside Donbas (2017–18)

Note: STRESS is a measure of anxiety, tension, and nightmares; STRESS1 is a measure of anxiety and tension only. See appendix section A8 for more details.

Full effects are substantively meaningful. Respondents who were very proud to be Ukrainian citizens valued democracy 17% and 14% higher (for 2017 and 2018, respectively) and were 24% and 21% more likely to view democracy as important than respondents who were not at all proud. For respondents who saw Russia as the aggressor compared to those who did not, those effects were 10% and 5% and 18% and 9%, respectively. This upholds H1 and H2.

H3 is supported given that all other social subgroup markers are shown as either insignificant or less significant than Pride and Russia Aggressor or contingent on other sociodemographic characteristics (e.g., age having negative effects in a historically stronger Sovietized Donbas, but being insignificant or positively related to democracy support elsewhere). The proportion of total effects of Pride and Russia Aggressor on Democracy Importance Scale mediated by sociodemographic variables in the sureg test was insignificant in both years (see appendix section A6).

Replicating our Donbas test of marginal effects for the interaction term between Pride and Russia Aggressor, we find strong support for H4 in 2017 and partial support in 2018. While smaller than in the Donbas, the full interaction effect in 2017 is about twice as strong as the effect of Pride controlling for Russia and of Russia controlling for Pride (figure 7a). The results are mixed for 2018 (figure 7b). Russia Aggressor still gives a boost to Pride effects on the democracy support scale, yet for all levels of Pride except “mostly proud” it is not statistically significant at p < 0.05. However, that exception is substantively important, because the “mostly proud” group comprises 44% of all respondents. It means that for nearly half of Ukraine’s respondents the interaction of Pride and Russia was significant.

Figure 7 Pride and Threat Interaction on the Democracy Importance Scale in Ukraine outside Donbas

The decline of Russia Aggressor’s boost of Pride in 2018 reflects the reduction of Russia Aggressor’s direct effect on democracy support in 2018 compared to 2017 (figure 6). Macrolevel context offers a plausible explanation in that the battle death rates in the Donbas war dropped fivefold after mid-2016 and the front line stabilized, leading to its decreasing salience over time. Another indication is that in 2017 the combined effect size of Pride–Russia Aggressor within the Donbas, a region more affected by the war, was substantially larger than outside the Donbas.

Checking for free speech support, we again got near-identical results supporting H1–H4, the exception being that Russian threat was not a significant predictor of free speech importance scale in 2018 and consequently not a significant booster of Pride’s effect, although it was a highly significant predictor that same year of probability of support for free speech (p = 0.006).

Finally, turning to the microlevel war exposure factors, War Loss has no significant association with democracy support and the effects of Stress are much weaker than the macrolevel (Pride and Russia Aggressor) factors. However, we see that outside the Donbas, Stress reduced democracy support in 2017–18. While these effects are small (between 2.7% and 5%), they remind us of the severe strains and grueling trials of war that Ukrainians are overcoming in pursuit of their democratic aspirations.Footnote ²⁸

Russia’s Full-Scale Invasion Effects: Ukraine 2021–22 Pre/Post War-Onset Panel Study

The geosocietal identity mobilization we observed in 2017–18 paved the way for a massive, spirited surge of democracy support across Ukraine in response to Russia’s full-scale attack in February 2022. Our November 2021–June/July 2022 panel survey data enables a longitudinal pre-post (repeated measures) design with a continuous treatment condition (war), where prewar baseline data represents a de facto control group, and using the same subjects controls for confounders as the treatment continues.Footnote ²⁹ In our case we have two repeated measures of the same subjects with an intervening “treatment” event (war) continuously affecting all subjects at the time of the repeated measurement. This design has significantly more leverage than (and should not be confused with) pretest-posttest one-group designs. Critically, the continuation of the treatment/intervening condition through the second measurement minimizes or removes common shortcomings of pretest-posttest designs, particularly intervening posttreatment effects and competing developments at treatment time. The scale and brutality of Russia’s war on Ukraine generated conditions overshadowing any other putative intervening event at the time of the repeat survey, and it also minimized the likelihood of instrument reactivity and Hawthorne effects (Spector Reference Spector1981, 28–30).

Russia’s full-scale invasion in 2022 also erased the ambiguity about enemy identification during the 2014–22 Donbas war, making the Russia Aggressor measure undiagnostic. In fact, our polling agency reported the question commonly elicited anger among respondents at the implied suggestion that an entity other than Russia could have attacked Ukraine. In this context, we use War Year (coded as 2022 = “1” and 2021 = “0”) as a proxy for geosocietal mobilization while controlling for microlevel war exposure effects with Loss and Stress. Footnote ³⁰

We use a linear mixed-effects model in SPSS with respondents’ ID as subjects, War Year as a repeated measure, fixed effects for all variables, and random intercepts estimated for subjects with default covariance type (variance components). Model 1 features only War Year and model 2 adds the same controls as in the 2017–18 tests: time-invariant factors and covariates based on 2021 data (Ukrainian language use, region, age aggregated into three groups, gender, and education level) plus time-variant covariates (income level, loss, and stress).Footnote ³¹ Using the opportunity to analyze panel data, we estimate war effects on a broader range of attitudes related to democracy support than in 2017–18. First, we examine trust in key democratic institutions (media, parliament) and the leadership (presidency). Second, we assess support for Ukraine’s international (geopolitical) identity as a democracy with support for EU and NATO membership: while the former could be valued primarily as an economic union and the latter as a military alliance, democratic government is a core requirement for new members in both. Seeking membership in such organizations is a natural way for a democratizing state to “reinforce the commitment to democracy and provide an external anchor against retrogression to authoritarianism” (Huntington Reference Huntington1991, 87). We use only scale measures. Unlike binary measures they capture longitudinal shifts across response categories that otherwise would fall within either “0” or “1”—particularly from “rather important” to “very important” on Democracy Importance (55% shift) and Free Speech Importance (60% shift).

We find unequivocal support for H5. The coefficients (table 1) show War Year giving a statistically significant, strong, and robust boost to support for democracy: 12.5% on democracy and free speech importance, over 22% on trust in the media and parliament, 50% on trust in the president, and 16% and 20% on support for EU and NATO membership. The effects are nearly identical with or without control variables (cf., models 1 and 2) and all are highly significant (p < 0.001). Notably, respondents reporting personal war losses are significantly more, not less, likely to support democracy. It means rallying for democracy in Ukraine has been resilient in the face of extensive, devastating, and persistent hardship: close to 70% of respondents in June–July 2022 reported at least one form of war loss (losing or fleeing their homes, getting wounded, losing friends, or having family or friends displaced), a more than threefold increase from December 2021 among the same individuals. Stress has a statistically significant negative effect only on one democracy support indicator.

We see the war galvanizing democracy support both among previously less supportive groups (older respondents) and more supportive groups (the better educated, the economically better-off, residents of Ukraine’s west and center in 2021). We also do not see regional polarization (no significant negative estimates for Ukraine’s east). Robustness checks are not indicative of social desirability bias. Exclusive reliance on Ukrainian national news media—a proxy for less-than-sincere rallying (Hale Reference Hale2021)—had inconsistent and weaker effects on democracy support indicators than War Year. Footnote ³² Other empirical data (Onuch and Hale Reference Onuch and Hale2022, chap. 7) characterizes the surge in support for democracy as predominantly endogenous and not top-down, reflecting Ukraine as “a country of 44 million Zelenskys,” a civic nation with spirited resilience and surging determination to join global democratic coalitions.