Abstract
In this article, we introduce an updated version of the Nonviolent and Violent Campaigns and Outcomes dataset (NAVCO 2.1), which compiles annual data on 389 nonviolent and violent mass movements for regime change, anti-occupation, and secession from 1945 to 2013. This version of the dataset corrects known coding errors in NAVCO 2.0, adds news cases (including the Arab uprisings), and codes attributes for each campaign year (such as participation size and diversity, the behavior of regime elites, repression and its effects, support from external actors, and campaign outcomes). In addition, NAVCO 2.1 adds several new attributes to each campaign-year, including more precise participation figures, more nuanced data about the scope, intensity, and degree of violent flank behavior and state repression, and further information about the parallel or alternative institutions developed by the campaign. The data reveal four key findings: (1) that the success rate of nonviolent resistance campaigns has declined since 2001; (2) that far more people have participated in nonviolent than violent campaigns in the postwar period; (3) that nonviolent campaigns suffer far fewer per-capita fatalities than armed campaigns; and (4) that incidental violence by dissidents has become a more common feature of contemporary nonviolent campaigns compared with earlier cases. The article concludes with suggestions for further research.
Keywords
Introduction to NAVCO 2.1
The Nonviolent and Violent Campaigns and Outcomes (NAVCO) data project tracks the onset and outcomes of maximalist political campaigns. Version 1.0 of the NAVCO dataset catalogued 323 campaigns of resistance from 1900 to 2006 (Chenoweth & Stephan, 2011). Version 2.0 changed the unit of analysis to the campaign-year, adding a number of new variables and observations to the 250 campaigns from 1946 to 2006 (Chenoweth & Lewis, 2013). Version 3.0 of the NAVCO data identified nonviolent and violent events in 26 countries between 1990 and 2011 (Chenoweth, Pinckney & Lewis, 2018).
Comparison of existing datasets
NAVCO 2.1 constitutes a significant improvement over its predecessor (NAVCO 2.0). It expands temporal coverage to 1945–2013, corrects known errors in variable coding, adds additional historical cases discovered through additional research, and adds novel variables to all campaign-years. This article showcases the dataset by testing several prevailing wisdoms about nonviolent resistance during and after the Arab uprisings, the prevalence of participation and fatality rates in nonviolent and violent campaigns, and the scope of violent flanks in otherwise nonviolent campaigns.
NAVCO’s contribution
NAVCO 2.1 differs from existing data projects on several important dimensions, summarized in Table I. Importantly, most up-to-date protest databases (i.e. Banks & Wilson, 2021; Clark & Regan, 2016; Weidmann & Rød, 2018; Salehyan et al., 2012; Raleigh et al., 2010) feature events-level data on protest. All of these datasets, which are useful and informative about day-to-day trends, implicitly treat events as independent from one another. However, the most meaningful impacts of contentious mobilization result from continuous campaigns rather than one-off events (Chenoweth & Lewis, 2013). Second, many existing databases have limited geographic scope (Asal, Johnson & Wilkenfeld, 2008; Chenoweth, Pinckney & Lewis, 2018; Salehyan et al., 2012; Raleigh et al., 2010), are limited by regime type (Weidmann & Rød, 2017) or campaign-type (Cunningham, Dahl & Fruge, 2020), or provide coverage only through 2006 (Chenoweth, 2011; Chenoweth & Lewis, 2013). Campaign-level databases with global coverage (Chenoweth, 2011; Chenoweth, 2021) present data at a highly aggregated unit of analysis, obscuring important variation on key attributes within multiyear campaigns.
In addition to expanding the temporal scope of data available at the campaign-year unit of analysis, NAVCO 2.1 also offers several novel variables not measured by existing datasets, including: (1) aggregate measures of annual participation (as opposed to peak participation); (2) the number of observed civilian fatalities caused by state violence per year; and (3) the scope, degree, and intensity of violent flank activity by primarily nonviolent campaigns. These new variables offer insights regarding the dynamics of nonviolent resistance, including their overall popular support, the lethality of campaigns across time and space, and the degree to which violent flanks help or hurt a campaign’s chances of success.
Data collection and methodology 1
Inclusion criteria
To be included in NAVCO, campaigns must pursue goals that challenge a ruling regime’s claims to sovereignty. More specifically, maximalist campaigns seek to overthrow or secede from an internationally recognized government, or to throw out an occupying state (usually a colonial ruler). Campaigns must attract at least 1,000 observed participants during at least one calendar-year of their existence, and this participation must be coherent and sustained over the course of at least two weeks. Campaigns may be either primarily violent or nonviolent, as the NAVCO project aims to assess the relative efficacy of violent and nonviolent methods.
Case selection
To create NAVCO 2.1, we used two methods to search for new campaigns (both within and outside of NAVCO 2.0’s temporal range). To search for primarily nonviolent campaigns, we compiled a list of 109 movements that were plausible candidates for inclusion. We developed this list by drawing on newly available data, such as MEC (Chenoweth & Ulfelder, 2017), SCAD (Salehyan et al., 2012), ACLED (Raleigh et al., 2010), and the Swarthmore Nonviolent Action Database (Lakey, 2011); consulting many social movement experts who suggested candidate cases; and integrating feedback from NAVCO 2.0 users, who had noted several absent campaigns. The codebook contains more description of how we adjudicated each case. To find primarily violent cases, we employed the UCDP/PRIO Armed Conflict Dataset (version 17.1), which documents all armed conflicts known to have produced at least 25 battle-deaths in at least one calendar-year during disputes over government or territory. The Armed Conflict Dataset (ACD) provided a list of 286 armed conflicts (2,384 conflict-years) between 1946 and 2018, 2 each of which was evaluated by research assistants (RAs) for NAVCO definitional criteria. ACD conflicts were only incorporated as NAVCO campaigns when RAs confirmed both the participation size requirements and the maximalist goals requirements (ACD conflicts fought over greater territorial autonomy, rather than territorial independence, were excluded from NAVCO). In some cases, RAs discovered new violent and nonviolent campaigns while researching the known candidate cases. These cases were coded as campaigns when the RA, project manager, and principal investigator all agreed that the case met definitional requirements.
All told, NAVCO 2.1 significantly expands on the number of observations captured by predecessors. Whereas NAVCO 2.0 provides data on 250 campaigns disaggregated to 1,726 campaign-years, NAVCO 2.1 includes 389 campaigns disaggregated to 2,733 campaign-years. This constitutes a 55% net increase in both the number of campaigns and the number of observations (campaign-years). Overall, the expansion of NAVCO 2.1’s temporal coverage from the years 1945–2006 to 1945–2013 accounts for most of the increase in observations by adding 53 new campaigns that began after 2006. Notably, this means that over 14% of all campaigns started in the final seven years (10%) of the data series.
As Figures 1 and 2 make clear, nonviolent campaign onsets have become more common than violent onsets during the past two decades.
Outcomes over time
From 1945 to 2006, over 62% of nonviolent campaigns succeeded, alongside about 30% of armed campaigns. From 2007 to 2013, the success rates declined to about 38% for nonviolent campaigns, and about 8% for violent campaigns. This apparent decline in aggregate success rates has prompted some to conclude that nonviolent resistance is declining in its efficacy (e.g. Chenoweth, 2021).
A closer examination of campaign outcomes attenuates that observation somewhat. Rather than focusing on
Onsets of nonviolent and violent maximalist campaigns Active nonviolent and violent campaigns Campaign success rates over time
It is clear from Figure 3 that both nonviolent and violent campaigns experienced several concurrent waves of success: during the anticolonial period in the late 1950s; during the mid-1970s; and during the late 1980s to early 1990s with the Soviet Union’s collapse. There was also a protracted period of unprecedented success for nonviolent campaigns in the late 1990s and early 2000s during the first wave of Color Revolutions. During that same period, violent campaigns continued to decline in their effectiveness. Even with the recent decline in success rates, nonviolent campaigns have had higher annual success rates than armed campaigns since at least 1969. However, nonviolent resistance campaigns have succeeded at continuously lower rates since 2001, the all-time peak of the five-year moving average. What explains this decline?
One possibility is regression to the mean – that after a cycle of extremes we should expect observations that accord with the baseline. It is clear from Figure 3 that after each wave of above-average annual successes (with peaks in 1974, 1991, and 2001), there were subsequent declines in annual success rates for unarmed campaigns. That the same pattern holds for violent campaigns lends even more support to this possibility.
A second explanation owes to the increased prevalence of new nonviolent campaigns after 2001 (Figure 2). With an increase in the number of campaigns each year, it is not surprising that annual success rates have declined. Indeed, because so many nonviolent campaigns have been active each year, it is actually surprising that nonviolent campaigns have maintained above-average success rates in most years post-2001. It is notable, for instance, that the greatest annual rates of success for armed campaigns occurred during the beginning of the time series, when the smallest number of armed campaigns was active.
Third, a growing global awareness of nonviolent tactics and the spread of social media may have reduced barriers to organizing campaigns without making it easier to oust regimes. In other words, more activists may have engaged in long-shot campaigns, many of which were defeated by governments who were increasingly savvy to such challenges after witnessing some contemporaneous successes. We explore this possibility by comparing the degree of institutional capacity among nonviolent campaigns pre- and post-2001. Using difference of means tests and non-parametric rank-sum tests, we find that post-2001 nonviolent campaigns tend to rely on fewer parallel institutions than pre-2001 campaigns, although post-2001 campaigns were not significantly less hierarchical. 3 This is a notable change, because centralized campaigns tend to be associated with a greater chance of success.
Fourth, it could be that nonviolent campaigns after 2001 have confronted more authoritarian opponents. To evaluate this explanation, we combine the NAVCO 2.1 data with the Varieties of Democracy dataset’s (V-Dem) Polyarchy indicator. Using difference of means tests and non-parametric rank-sum tests, we find that the opposite is true: nonviolent campaigns tended to arise in regimes with statistically significantly higher polyarchy scores from 2002 to 2013.
Finally, we explore the possibility that recent regimes have displayed greater brutality toward nonviolent campaigns than during prior periods. We first examine whether more recent campaigns have faced opponents with poorer human rights records. We combine the NAVCO 2.1 data with the V-Dem indicator of human rights abuses – the Freedom from Political Killings (FFPK) score – as a proxy for all regimes’ use of lethal repression at the country-year level (including repression not directed at campaigns). We then compare FFPK scores in onset years across nonviolent campaigns. We find, consistent with the previous observation, that nonviolent campaigns have arisen in countries with somewhat higher FFPK scores, while violent campaigns have not.
Next, we evaluate whether more recent campaigns have suffered more direct fatalities than their predecessors. We do this by looking at the average number of fatalities over time in nonviolent campaign-years, a measure derived from NAVCO 2.1’s fatalities estimates. As before, we employ rank-sum tests owing to a non-normal distribution of fatalities. This exercise reveals no statistically significant differences in the number of fatalities before and after 2001, suggesting that, if campaigns began targeting moderately less repressive regimes after 2001, those regimes responded with similar levels of direct violence.
Thus, the apparent decline in nonviolent efficacy may be due to changes in the composition of targeted regimes, but not for the reasons offered by conventional wisdom. We also see evidence that campaigns, too, have changed since 2001. We now turn to a discussion of the new variables included in NAVCO 2.1, which shows additional ways in which recent campaigns have diverged from their predecessors.
New variables
NAVCO 2.0 introduced some 40 variables identifying different attributes of maximalist campaigns, such as participation size and diversity, the behavior of regime elites, repression and its effects on the campaign, support or the lack thereof from external actors, and progress toward the campaign outcomes (Chenoweth & Lewis, 2013). NAVCO 2.1 includes these as well as 12 new variables, each of which either disaggregates an existing variable from NAVCO 2.0 (greatly increasing precision) or operationalizes a new campaign attribute of interest. Here we focus on several new variables.
Total participation
The total participation estimate provides a precise integer value for the total number of people observed to participate in a campaign during each calendar-year of its existence. This constitutes a significant increase in precision over NAVCO 2.0’s participation variable, which was ordinal rather than integer. For every campaign-year, RAs searched for exact indicators of participation, which were often available in primary sources. For example, the Zambian Anti-Single Party campaign’s 1990 year was derived from a campaign leader’s memoirs in which he recorded the number of cards printed and distributed to dues-paying members of the Movement for Multi-Party Democracy (yielding an estimate of 441,900). In cases where we could not find such precise first-hand estimates, RAs often found creditable unofficial estimates from news reports, journal articles, historical narratives, and policy briefs. Through triangulation, RAs were often able to infer total participation estimates from various independent sources.
NAVCO 2.1’s total participation estimate is unique in its scope and precision. For instance, the UCDP Armed Conflict Encyclopedia provides a yearly troop strength estimate for armed non-state groups, but its coverage is inconsistent and applies only to violent groups. The Social Conflicts in Africa Database (Salehyan et al., 2012) provides an ordinal value of participation size for individual events (violent and nonviolent) in Africa and Latin America, but does not aggregate these values to the movement level. The Manpower and Counter-Insurgency Database provides estimates of (violent) insurgent groups’ troop strength but does not disaggregate by time period (Friedman, 2011). Only NAVCO 2.1 reports an annual participation estimate for both nonviolent and violent maximalist campaigns.
Fatality estimates
Existing scholarship theorizes the relationship between repression and campaign mobilization to be complex yet crucial in determining outcomes (McAdam, Tarrow & Tilly, 2001; Schock, 2005; Chenoweth & Stephan, 2011; Sutton, Butcher & Svensson, 2014). Because no existing datasets have yet documented fatalities on a year-to-year basis, researchers have often had to settle for crude categorical measures of repression.
Precise fatality estimates enable more fine-grained analysis of this dynamic relationship. NAVCO 2.1 reports low and high annual fatality estimates for the number of people killed by the regime in the attempt to suppress campaigns. Here we are interested in the aggregate number of observed protestors, dissidents, or combatants as well as supporters, bystanders, or sympathizers killed directly by the state and its proxies in relation to campaign suppression. We therefore exclude deaths caused by campaign participants in this measure. As with the peak participation estimate, RAs also turned to news reports, academic books and journals, Amnesty International and Human Rights Watch reports, and policy briefs to create fatality estimates for campaign-years not covered by the UCDP’s Geolocated Events Dataset (GED) (Sundberg & Melander, 2013). Because reporting on such indicators can be uneven and imprecise, we also provide an ordinal fatalities range variable.
In terms of scope and precision, the NAVCO 2.1 fatalities variable constitutes a unique resource. Other datasets sometimes include fatality estimates for repressive episodes but do not provide global comparisons across nonviolent and violent resistance campaigns. For instance, the UCDP GED provides battle-death estimates for armed conflicts (1989–2021), but it rarely captures fatalities inflicted by states on nonviolent social movements. Sutton, Butcher & Svensson (2014) use UCDP’s One-Sided Violence Database (derived from the GED) to code the use of mass violence against protest in cases in which the regime kills at least 25 unarmed civilians during a calendar year. But this method does not capture the use of violence falling below UCDP’s 25 death threshold.
Violent flanks
Violent flanks refer to the tendency of some primarily nonviolent campaigns to develop or tolerate armed wings. This is significant in that nonviolent discipline may influence domestic and international audiences’ perceptions of a movement’s legitimacy (Chenoweth & Stephan, 2011; Pinckney, 2016; Chenoweth & Schock, 2015). Others argue that violent flanks are normal among nonviolent campaigns (Pressman, 2017) and potentially influential in allowing them to succeed (Kadivar & Ketchley, 2018).
NAVCO 2.1 augments variables identifying violent flank activity (among primarily nonviolent campaigns only) by adding information about the intensity and scope of violent flank behavior, as well as the campaign’s response to emergent violent flanks. Violent flank intensity is an ordinal variable that reports the degree of violence used by campaign participants: property damage, non-lethal violence, or lethal violence. The violent flank scope variable reports the extent of that severity with categorical values: incidental/isolated, routinized/institutionalized, or ambiguous (Day, Pinckney & Chenoweth, 2015). The violent flank response variable indicates whether campaign leaders tolerated the violent flank, urged participants to remain nonviolent, or signaled internal disagreement about the violent flank. RAs coded these variables using academic journals and books, news reports, and policy reports. Notably, the violent flank response variable is missing in 19% of campaign-years, as RAs often had difficulty discovering instances in which campaign leaders expressed explicit attitudes toward violent flanks.
Nonetheless, the violent flank variables improve on prior measures by reporting on the characteristics of violent flanks emerging within otherwise nonviolent campaigns. In conjunction with fatality estimates, violent flank variables provide the means to explore dynamic escalation processes between campaigns and regimes.
Describing seven decades of maximalist campaigns
Participation
Despite the preponderance of primarily violent campaigns in the post-World War II era, far more people participated in nonviolent campaigns than violent campaigns. In fact, nonviolent campaigns recruited more than twice as many participants (160,000,000) as their violent counterparts (74,100,000) over the entire series. As Figure 4 shows, the average participation size of nonviolent campaigns dwarfs that of violent campaigns in nearly all years.
As Figure 5 shows, nonviolent participation peaked in 1983 (21,300,000), but several other spikes are visible. Over 11,000,000 people joined nonviolent campaigns in
Annual participation Aggregate participation Major MENA uprisings (2007–13)
Fatalities
NAVCO 2.1 conservatively estimates that governments killed over 5,500,000 people while suppressing maximalist resistance campaigns from 1945 to 2013. 4 Crucially, nonviolent campaigns suffered just 1% of all fatalities despite attracting many more participants than violent campaigns. On average, states killed 25,353 people per violent campaign, as compared with 290 people per nonviolent campaign. Disaggregating by campaign-year, violent campaigns suffered an average of 2,851 fatalities per year as compared with 105 fatalities suffered per year by nonviolent campaigns. Since the fatality data (for both violent and nonviolent campaign-years) exhibit signs of a power-law distribution (Clauset, Shalizi & Newman, 2009), we employ a (non-parametric) Wilcoxon rank-sum test to show that random chance is very unlikely to account (p < 0.001) for the difference in fatalities.
Campaign-years associated with the Eastern European Revolutions (1989) had substantially lower fatality estimates on average than the global annual mean (six fatalities compared with 107), and a Wilcoxon rank-sum test shows that the difference was statistically significant (p < 0.05). Campaign-years associated with the 2011 Arab Spring uprisings (all nonviolent campaign years that occurred in MENA countries) suffered more fatalities on average (184 fatalities compared with 102), and a Wilcoxon rank-sum test indicates that this difference is statistically significant (p < 0.05).
Does this trend suggest that nonviolent campaigns have become more dangerous over time? We generate a fatality rate which provides the aggregate annual number of deaths per campaign participant (Figure 6). Once we account for the large number of participants in the East European Revolutions and the Arab uprisings, the overall trend in nonviolent fatalities rates has remained flat; the average nonviolent campaign has not become deadlier for the average participant. A t-test and Wilcoxon rank-sum test each confirm that the mean fatality rate of violent campaigns is greater than the mean rate of nonviolent campaigns (p < 0.01 and p < 0.001, respectively). Across time, participating in a nonviolent campaign seems to be far less lethal for the average participant than participating in a violent campaign.
Figure 6 brings to light an additional finding: violent campaigns pose much greater dangers to nonparticipant populations than do nonviolent campaigns. Recall that NAVCO 2.1 fatality estimates aggregate all deaths inflicted by the state while suppressing a campaign, including nonparticipants killed in indiscriminate violence or killed owing to misidentification by regime agents. Hence, fatality rates can exceed 1. 5
Figure 6 shows that the mean annual nonviolent fatality rate exceeded the violent mean annual fatality rate during only three years, owing to three campaigns: the Cameroon independence movement (1955), the
Annual fatality rates
In sum, the average nonviolent campaign is comparatively benign from the perspective of both participants and nonparticipants: those who choose not to participate in nonviolent campaigns rarely bear the brunt of regimes’ lethal retaliations against challengers. This is not so with violent campaigns, during which governments routinely crack down on broader swaths of society to root out dissent.
Nonviolent campaigns and violent flanks
Violent flanks and nonviolent campaign outcomes
Several patterns emerge when we disaggregate violent flank data to the campaign-year level. Although over 78% of nonviolent campaigns had violent flanks in at least one year of their existence, only 58.81% of nonviolent campaign-years possessed violent flanks (see Tables III and IV). This indicates variation in the use of violence over time within individual campaigns.
Violent flank scope across nonviolent campaigns
Number of nonviolent campaigns and violent flanks
More generally, Figure 7 shows that the prevalence of violent flanks over time is quite consistent except between 1989 and 1991 (corresponding to the Eastern European Revolutions), during which the number of nonviolent campaigns increased drastically and disproportionately compared with the number of violent flanks. This observation invites comparison with a more recent spike in active campaigns beginning in the post-2000 period. Unlike the Eastern European wave, the later increase in nonviolent campaigns was accompanied by a proportionate rise in violent flank activity. This serves to emphasize the uniquely peaceful character of the 1989–91 wave.
The data also show that violent flanks, when they do emerge, tend to be relatively severe: of the 291 nonviolent campaign-years with violent flanks, 165 (56.7%) involved lethal levels of violence. In comparison, only 82 campaign-years (28.2%) featured violent flank activity limited to non-lethal violence, and a small minority of 44 campaign-years (15.1%) limited this activity to property damage only. One possible interpretation of this finding is that violence, once it emerges, may tend to escalate quickly, or perhaps the maximalist inclusion criterion biases the sample toward higher-stakes campaigns, where even relatively mild forms of violence are perceived as politically transgressive and are therefore unlikely to occur absent other forms of more direct violence against persons.
NAVCO 2.1 also provides data on the scope or prevalence of violent flank activity within each campaign (Table IV). The modal scope of violent flank activity is its absence, which occurred in 208 (43.2%) nonviolent campaign-years. Of those nonviolent campaign-years where violent flank activity was present, 174 (59%) used
Scope of violent flanks in nonviolent campaigns
Finally, the scope of violent flanks has changed over time. Figure 8 reveals that most violent flanks in the post-2000 era were incidental in scope. Thus, the increase in violent flank activity after 2000 (evident in Figure 7) comprises primarily sporadic, rather than institutionalized, violence. It seems that nonviolent campaigns have failed to control (or refrained from controlling) violent flanks in recent years.
Discussion and conclusion
NAVCO 2.1 constitutes an improvement over its predecessors in terms of substantive and temporal coverage. Our investigation shows that the success rates of maximalist campaigns have been in decline since 2001. This finding is true for both nonviolent and violent campaigns. However, this shift does not appear to be a function of greater autocracy or brutality among the regimes targeted. Instead, it may be a function of the increasing prevalence of nonviolent campaigns, as well as the way such campaigns self-organize.
Second, we find that annual participation is generally higher for nonviolent campaigns than violent ones – both in terms of average campaign size on an annual basis and in terms of aggregate participation worldwide. Clearly, individuals worldwide have consistently preferred to engage in nonviolent action over militarized violence – even in similarly motivated campaigns against authoritarianism or foreign or colonial domination. This finding underscores the relative importance of nonviolent action as a political phenomenon and subject of study.
Third, the data show that participating in maximalist nonviolent campaigns is much less deadly for the average person than violent campaigns. This observation holds true even for campaigns associated with the Arab uprisings, which did not experience significantly greater per-capita danger than other nonviolent campaigns in the postwar era. These same data also illustrate that violent campaigns commonly trigger extremely broad repressive measures on the part of regimes, consistent with the coercive response logic prominent in the state repression literature (Davenport, 2007). We have shown that the breadth of repression against violent campaigns poses a much greater danger to nonparticipants than the repression usually brought to bear against nonviolent campaigns. Thus, when dissidents turn to primarily violent tactics, they may expose their communities to tremendous risks.
Fourth, we find that most nonviolent campaigns adopt or tolerate the limited use of violence against a regime at some point. Campaigns that have arisen since 2000 have become more likely to experience incidental violence – occurring on the fringes of otherwise nonviolent campaigns – compared with routinized or deliberate violence. It is unclear whether this stems from deliberate decisions to cultivate these types of flanks (over institutionalized flanks) or an inability to enforce nonviolent discipline. Regardless, violent flanks do not appear to be correlated with campaign outcomes.
These descriptive findings have helped us discover and map new patterns in global resistance struggles. In turn, they prompt us to pursue both new and old questions about civil resistance. For instance, why have maximalist campaigns become less successful over time, even as targeted regimes have become more democratic and less repressive? Which tactics and strategies aid campaigns in attracting greater participation? What explains variation in violent flank intensity and scope, and how does this variation affect mobilization dynamics and outcomes? How do nonviolent campaigns prevent and react to high fatality rates? NAVCO 2.1 can help researchers answer these questions with greater rigor and precision than was previously possible.
Footnotes
Replication data
Acknowledgements
Equal authorship is implied. The authors thank their research assistants for their excellent work.
Funding
The authors gratefully acknowledge financial support from the International Center on Nonviolent Conflict, the Social Science Foundation, the Compton Foundation, the University of Denver, and Harvard Kennedy School.