Education and Revolutions: Why do Revolutionary Uprisings Take Violent or Nonviolent Forms?

Is there a relationship between education and the type of revolutionary action – violent or nonviolent? Past studies found a positive relationship between the education and nonviolence, but the influence that education produces on the form that revolution takes has not yet been explored. We show several possible mechanisms that push the educated population to choose nonviolent tactic: (1) education changes people’s preferences toward peaceful solutions and increases support for civil liberties; (2) it enhances human capital that makes it feasible to use unarmed tactics successfully and (3) it increases the relative costs of engaging in armed action. Thus, it is reasonable to assume that the higher education in a country, the higher the probability that revolution will be nonviolent. This paper examines it at a cross-national level with an analysis of 470 NAVCO ‘maximalist campaigns’ and 265 revolutionary events recorded between 1950 and 2020. Overall, we find robust evidence that the higher the level of education in a country, the lower chance that the revolution there would take a violent/armed form.


Introduction
In 2001 Jack Goldstone made the following observation: 'Until very recently, revolutions have invariably failed to produce democracy. The need to consolidate a new regime in the face of struggles with domestic and foreign foes has instead produced authoritarian regimes, often in the guise of populist dictatorships such as those of Napoleon, Castro, and Mao, or of one-party states such as the PRI state in Mexico or the Communist Party-led states of the Soviet Union and Eastern Europe. Indeed, the struggle required to take and hold power in revolutions generally leaves its mark in the militarized and coercive character of new revolutionary regimes. It is therefore striking that in several recent revolutionsin the Philippines in 1986, in South Africa in 1990, in Eastern European nations in 1989-1991 the sudden collapse of the old regime has led directly to new democracies, often against strong expectations of reversion to dictatorship' (Goldstone, 2001, p. 168).
Note that most of these authors prefer to denote revolutions as "maximalist campaigns". Following Ackerman and Kruegler (1994, p. 10-11), Chenoweth and Stephan (2011, p. 14) define "campaign" as "a series of observable, continual, purposive mass tactics in pursuit of a political objective." What is more, the abovementioned studies consider campaigns "with goals that are perceived as maximalist (fundamentally altering the political order); …we deliberately choose campaigns with goals commonly perceived to be maximalist in nature: regime change, antioccupation, and secession" (Chenoweth & Stephan, 2011, p. 68). Thus, the abovementioned works study "series of observable, continual, purposive mass tactics in pursuit of fundamentally altering the political order: regime change, antioccupation, and secession". We ourselves rely on such definitions of revolution as "a collective mobilization that attempts to quickly and forcibly overthrow an existing regime in order to transform political, economic, and symbolic relations" (Lawson, 2019, p. 5); "anti-government (very often illegal) mass actions (mass mobilization) with the following aims: (1) to overthrow or replace the existing government within a certain period of time; (2) to seize power or to provide conditions for coming to power; (3) to make significant changes in the regime, social or political institutions" (Goldstone et al., 2022b, pp. 50-51), or "an effort to transform the political institutions and the justifications for political authority in a society, accompanied by formal or informal mass mobilization and noninstitutionalized actions that undermine existing authorities" (Goldstone, 2001, p. 142). Thus, we find that "maximalist campaigns" are just nothing else but revolutions (including national liberation ones); hence, the abovementioned works actually study revolutions (rather oddly denoted as "campaigns"). This point is further supported by the fact that Chenoweth's database of Nonviolent and Violent Campaigns and Outcomes (NAVCO) designates as campaigns all the indisputable revolutions since 1900including Russian revolutions of 1905-1907and 1917, Constitutional Revolution in Iran, Xinhai Revolution in China, Mexican Revolution of 1910-1917 and so on (Chenoweth & Shay, 2020a). Thus, the results of the abovementioned studies on the outcomes of "maximalist campaigns" turn out to be perfectly relevant for our understanding of the outcomes of revolutions.
It is highly remarkable that the main finding of the abovementioned studies (supported by a number of rather rigorous tests on the basis of a very wide range of worldwide empirical data) is that violent revolutions ("campaigns") are very unlikely to lead to the formation of stable democratic regimes, whereas this is much more probable as a result of nonviolent revolutions (Ackerman & Karatnycky, 2005;Stephan & Chenoweth, 2008;Chenoweth & Stephan, 2011;Butcher & Svensson, 2016;Kim & Kroeger, 2019;Rasler et al., 2022). However, there is a further need to study the structural conditions that explain when a country is more likely to experience one form of revolution instead of the other.
Against this background, it is a bit surprising that there seem to be only a few systematic quantitative global cross-national studies of factors of violent versus non-violent revolutions (Butcher & Svensson, 2016;Dahlum, 2019). students were not the only ones who participated in the demonstrations. Indirectly, one can look at the social composition of the protesters: most of the petitions and resolutions during the revolution were signed by workers who also shared the ideals of nonviolence and democratic reforms (Wheaton & Kaban, 2018, p. 69, 95). At the same time, the revolution was also supported by judges and university rectors who silently approved the nonviolent way of fighting the regime (Wheaton & Kaban, 2018, p. 62).
The Houthi Revolution in Yemen (or Al-Houthi Rebellion) began very differently in Yemen in 2004. Initially, it was a protest against the corruption of President Ali Abdullah Saleh and his close ties with the United States, led by opposition politician Hussein al-Houthi (the group of rebels, the Houthis, is named after him). When the uprising began, it was based on rural poorly educated youth from the organization "Believing Youth" (Brandt, 2017;Palik, 2017). The President initially tried to negotiate with the leader of the uprising, Hussein al-Houthi, but was refusedthe opposition did not want to negotiate. Then the conflict escalated into violent clashes: government troops tried to arrest the protesters and their leader, but they took up arms. In the very near future, the leader of the uprising was killed, but the conflict did not end. It escalated into the so-called "Six Saada Wars" (after the region where the uprising began and developed) and eventually led to the Houthis takeover of power in Sanaa in September 2014 (Brandt, 2017;Korotayev & Issaev, 2021).
One can see how the events in Yemen differ from the events in Czechoslovakia in 1989. If the Velvet Revolution was bloodless, the protesters' slogans were "we do not want violence", and the protesters actively contacted the authorities, then the events in Yemen were completely opposite and escalated into a civil war with thousands of death toll. One explanation for such a clear difference may be the level of education in these countries. If the highly educated population of Czechoslovakia 1 chose unarmed methods of resistance (marches, petitions and strikes) and demanded negotiations with the government (one of the slogans also was "Dialogue cannot be conducted on the street" (Wheaton & Kaban, 2018, p. 42)), then the poorly educated protesters in Yemen 2 took up arms almost immediately and did not want to come to an agreement with the government, although in both cases the government somehow began with the suppression of protests. Such association between education level of population and the type of tactics during the uprising can be found for many other cases: the armed 2003 "Seleka Rebellion" in Central African Republic (3.1 mean years of schooling) and the bloodless 2004 "Orange Revolution: in Ukraine (10.7 mean years of schooling); the armed Anti-Mutharika movement in 2011 in Malawi (4.3 mean years of schooling) and the unarmed "Jordan spring" in the same year (9.9 mean years of schooling). In other words, these cases suggest a clear connection between the level of the population education and the type of revolution that took placearmed or unarmed.
In part, this pattern has already been considered by some authors. Of special importance is that Butcher and Svensson (2016, p. 324-325) show that the likelihood of onset of nonviolent revolutions (but not violent ones) increases significantly with the increase of the level of education (operationalized through average years of schooling) of the population of a respective country. Of course, in conjunction with the abovementioned finding that the nonviolent revolutions are more likely to lead to a stable democracy than the violent ones are, this suggests that in a country with a very highly educated population a revolution is quite likely to be non-violent and to lead to the establishment of a stable democracy; but, on the other hand, this implies that in a country with a very poorly educated population a revolution is much more likely to be violent and, thus, is very unlikely to produce a democratic outcome.
Similar conclusions were arrived at by Dahlum (2019): campaigns with a large number of students and graduated participants are more likely to be nonviolent and more likely to be successful (see also Grinin et al., 2017). Note that the main independent variable used by Dahlum, the Combined Education Index, takes into account the education level of the campaign participants only. 3 Essentially, Dahlum created a fundamental base for our research and provided robust results that education promotes nonviolent strategies within protests campaigns. However, these findings must be expanded and tested on a more focused sample, including exclusively revolutionary events, not just the large protest campaigns analyzed by Dahlum. 4 Moreover, her combined education index, the main explanatory variable, was created by herself and cannot be updated constantly. In other words, if one wants to replicate revolutionary events with this index, one will be hard-pressed to find an enormous amount of information for incomplete data for years since 2006. Moreover, the main problem with it is that "the majority of criteria are based on the sources' descriptions of who the movement consisted of" (Dahlum, 2019, p. 288). Thus, it can only be collected just post facto and has no predictive power for an understandable reason. Of course, this index may have more explanatory power, but it is not the ultimate cause, only the immediate cause determining violent versus non-violent form of the revolutionary action.
Thus, the novelty of the present article is not only the analysis of the influence of education on the form of revolutionary events but also the identification of a viable independent variable that can not only describe the form of revolutions post factum but also can predict the form of forthcoming revolutions.
Turning to the main point, the studies we have reviewed postulate a connection between the level of education and the choice of tactics for nonviolent resistance by the protesters, but this hypothesis has not been tested. Previous studies have found a consistent relationship between the likelihood of a peaceful protest and education (Machado et al., 2011;Brancati, 2014;Butcher & Svensson, 2016;Korotayev, Bilyuga, & Shishkina, 2017Kostelka & Rovny, 2019;Korotayev et al., 2020, but they have not investigated how education influences the choice of tactics of the revolutionaries. In fact, there are several reasons for educated people to choose the tactics of non-violence. They can be summarized as follows: (1) education changes ideological preferences in favor of democracy and commitment to civil liberties, fostering a culture of peaceful discussion and tolerance (Dahlum, 2019;Dee, 2004;Galston, 2001;Inglehart et al., 2015;Lipset, 1960); (2) as a result of getting education, individuals increase the level of human capital, which leads to a rise in the relative costs of participating in violent revolutions/ campaigns that require taking more risks and abandoning the usual life (Hall et al., 1986;Thyne, 2006;Barakat & Urdal, 2009;Dahlum & Wig, 2019;Østby et al., 2019;Dahl et al., 2021); (3) education lowers the cost of participating in a peaceful revolution/protest by facilitating cooperation and understanding political processes, which makes it possible to achieve success through nonviolent revolutions/campaigns recognized as more successful and less costly (Dahlum, 2019;Dee, 2004;Galston, 2001;Glaeser et al., 2007;Grinin & Korotayev, 2022;Abd Rabou, 2016;Stephan & Chenoweth, 2008).
Considering the first reason, it is worth noting that Lipset (1960Lipset ( , 1968) argued that education expands access to information and promotes democratic-liberal values (such as freedom of speech or recognition and respect for the rights of another person) with an emphasis on the development of civic knowledge fostering tolerance in individuals and allowing them to see the world from the perspective of other people who are not like themselves (Dahlum, 2019;Dee, 2004;Inglehart et al., 2015). In other words, education fosters interpersonal interaction and empathy, generating aversion to violence (Pinker, 2011), while its absence, on the contrary, makes people more militarized and inclined to accept the possibility of resolving disputes through violence due to a simplistic view of politics and an inability to understand the meaning behind tolerance and compromise with people you disagree with (Shayo, 2007). Thus, less educated individuals are more prone to violence, while educated people tend to be disgusted with violence due to their preference effect (Dahlum, 2019).
The second reason relates to human capital costs. It is theorized that education reduces the propensity to violence because of the high costs associated with the opportunity cost of their labor and investment in themselves (Hall et al., 1986;Dahlum & Wig, 2019). More specifically, education provides people with qualifications that increase the value of their labor in the market and improve their well-being, which makes violent protest tactics incredibly expensive (Thyne, 2006;Barakat & Urdal, 2009;Østby et al., 2019). This argument assumes that people with low opportunity costs are more likely to engage in collective violence. An armed conflict presupposes a rejection of the usual way of life, that is, of the familiar earnings, in favor of long and extremely dangerous actions associated with the risk of depriving the accumulated investment in human capitaldeath (Hegghammer, 2013). At the same time, peaceful protests allow one to quickly switch between ordinary life and protest, which naturally reduces the costs of missed opportunities (Dahl et al., 2021). In other words, the larger number of prospects for improving welfare that education offers will be lost, making armed tactics unattractive (Inglehart et al., 2015). Consequently, recruiting rebels becomes prohibitively expensive with respect to educated people, which makes it less likely that the opposition will choose the path of revolutionary armed uprising (Barakat & Urdal, 2009;Collier, 2004).
The third reason is that education not only increases the costs of participating in violent conflict but also lowers the costs of participating in nonviolent conflict. Thus, educated people have the necessary potential for peaceful protest and, therefore, are more likely to choose nonviolent tactics (Dahlum, 2019), because education provides them with the necessary resources to organize and succeed in peaceful protests, which are considered more effective but also more difficult from an organizational point of view (Beissinger, 2022;Dahlum, 2019;Stephan & Chenoweth, 2008). For example, educated people are more likely to use the media (Dee, 2004), which increases the speed of news dissemination necessary for mass mobilization and success. This factor has become especially important only recently due to technological progress and the emergence of various social networks that stimulate political expression and participation from below, giving everybody tribune and the opportunity to participate in the political process independently (Akaev et al., 2017;Enikolopov et al., 2020). In addition, education increases communication skills and teamwork (Green et al., 2001), as well as facilitates understanding of policy and reduces the cognitive costs of decisionmaking bypassing various bureaucratic and technological barriers to civic participation (Dee, 2004). In other words, more knowledge helps to better recognize public policy and more effectively promote one's point of view in the political sphere (Galston, 2001), which creates the necessary individual resource base for participation in peaceful protests (Chenoweth & Ulfelder, 2017;Dahlum, 2019;Stephan & Chenoweth, 2008). Consequently, the gains from choosing nonviolent tactics increase. Thus, education not only provides people with political knowledge but also facilitates cooperation between them, which expands the possibilities for choosing tactics of revolutionary protest. Simply put, education lowers the material and cognitive costs of political participation (Abd Rabou, 2016), making the possible benefits of participation greater (Glaeser et al., 2007), which Dahlum (2019) calls 'capacity-enhancing effect'.
Thus, education has a pacifying effect, because it increases the level of human capital, reduces the relative costs of organizing protests leading to an increase in the likelihood of peaceful revolutionary protests (Brancati, 2014;, and makes violence unacceptable on the personal level, instilling in people a tendency to tolerance (Jenkins & Wallace, 1996). In general, it is confirmed by empirical studies: researchers find that the average number of years of schooling is positively and significantly associated with the level of peaceful protests (Brancati, 2014;Butcher & Svensson, 2016;Korotayev, Bilyuga, & Shishkina, 2018;Korotayev et al., 2020Kostelka & Rovny, 2019;Machado et al., 2011;Romanov et al., 2021;. But at the same time, it is negatively associated with the likelihood of a civil war, which appears as an extreme form of violent revolutionary conflict (Barakat & Urdal, 2009;Collier, 2004).
Finally, our main hypothesis can be formulated as follows: H: revolutionary uprisings in countries with higher mean years of schooling 5 will be more likely to take a nonviolent 6 rather than a violent form compared with countries with lower mean years of schooling.

Methodology and Empirical Strategy
In order to test our hypothesis proposing a link between the mean education level in a country (the independent variable) and the degree to which revolutionary events are violent or nonviolent (the dependent variable), we focus on cross-national data of 622 Nonviolent versus Violent (NAVCO) 'maximalist campaigns' supplemented by our own data on 377 revolutionary events recorded between 1900 and 2020. Using logistic regression because the dependent variable is dichotomous, we also introduce control variables related to modernization (modernization group)-log per capita GDP, urbanization, and median age/share of youth (youth bulge). We also consider controls for the country's area, population density, population size, and index of electoral democracy. Note, the unit of analysis is a particular revolutionary event that could take violent or nonviolent forms. It is also important to emphasize that the logistic model, as we employ here, is supposed to describe a probability, which is always some number between 0 and 1. Thus, for the logistic model, we can never get a risk estimate either above one or below 0 (Kleinbaum et al., 2002). The sample and variables are described in more detail below.
Thus, the aim of the paper is to understand why revolutions are violent or nonviolent, but here we do not try to predict the occurrence of any type of revolution. In other words, this research is aimed at finding the answer to the following question: if revolution occurs, what type it will take?

Dependent Variable
We rely on data provided by The Nonviolent and Violent Campaigns and Outcomes (NAVCO) 1.3 data project (Chenoweth & Shay, 2020a), which identifies 622 maximalist/revolutionary campaigns that occurred between 1900 to 2019. These data combine numerous instances of violent and nonviolent maximalist/revolutionary campaigns with the goals of expelling foreign occupation, regime change or separatism, and in some cases other major types of social change (such as campaigns against apartheid). In addition, Chenoweth and Shay identify whether the campaign was successful, achieved its goals or failed, and some other characteristics. However, we are interested in another variable provided by Chenoweth and Shaywhether the campaign was violent/unarmed or not. 7 It is this variable that will be the dependent variable in this paper. Note that it is a binary variable, where '1' is a nonviolent/unarmed revolutionary uprising/campaign and '0' is a violent one. Also, worth noting, the absence of any type of event is recorded in NAVCO as a missing value.
Note that NAVCO also contains a considerable number of so-called quasi revolutionary episodes, which Korotayev (following Beissinger, 2017, 2022) define as episodes with "mass mobilization, but there are no demands to overthrow the government, or sufficient efforts are not made to overthrow the government and seizing power (there is no evidence that there have been serious attempts to overthrow the government and seize power)" (Goldstone et al., 2022b, p. 7). As an example, we can recall such cases recorded in the NAVCO as "Denim revolution" (or the so-called "Jeans Revolution") in Belarus (in 2006) or similar "Dissenter's March" in Russia (2007)(2008) which cannot be called revolutions/failed revolutions, or even revolutionary episodes. "Denim revolution" (post-election anti-Lukashenko protests) took place just a few days in March 2006, most of which the number of participants was only a few hundred (see, e.g., Korosteleva, 2009), and no serious attempts to take power were undertaken. 8 Similarly, during the "Dissenters' Marches" in Russia (2007Russia ( -2008, or "Kefaya" movement events in Egypt (2000Egypt ( -2005 no serious attempts to take power were undertaken (Clarke, 2011), and, together with Mark R. Beissinger (2017;2022), we prefer to denote such events as "quasi revolutionary". 9 Thus, in order to check if the results obtained for NAVCO turn out to be valid for the revolutionary events proper, we implement the dataset for 1950-2021 that describes only revolutionary events (revolutions, failed revolutions/ revolutionary episodes, revolutionary guerrilla warfare, analogues of revolutions and failed revolutions) without any quasi-revolutionary episodes. It contains events from the database of the 21 st century revolutions (reproduced in the online supplement material, Table S8) augmented for the 1950-1999 period with the events from the dataset of 20 th century revolutions developed by L.  as well as those revolutionary events of the 20 th century that are present simultaneously both in Franziska Keller's dataset of revolutions (Keller, 2012) and in NAVCO (Chenoweth & Shay, 2020a). In this way, we have a comprehensive list of revolutionary events between 1950 and 2021. This dataset will allow us to test whether conclusions drawn from the NAVCO work for revolutionary events proper (excluding quasirevolutionary episodes). Moreover, this dataset is necessary for testing the ability of Dahlum's index to explain the violent/nonviolent format of not only campaigns, but also revolutions.

Independent Variable
Our main explanatory variable is the mean years of schooling sourced from the United Nations Development Programme reports database (UNDP, 2020) and the Barro and Lee database (Barro & Lee, 2010). The Barro and Lee data only range from 1950 to 2010, therefore their figures were combined with the UNDP ones to cover the period between 2011 and 2019. The method for the calculation of mean years of schooling in the UNDP report is similar to the method used by Lee (1993, 2010), thus making data transformation possible. The United Nations Development Programme defines this variable as the "average number of years of education received by people ages 25 and older, converted from education attainment levels using official durations of each level" (Jahan, 2016, p. 213). For each age group, the proportion that attained a given level of education is multiplied by the official duration of that level in a given country. The sum of the resulting values yields the mean years of schooling for the population in that specific country.
As one can see, this is very different from the independent variable, the combined education index, proposed by Dahlum (2019), as discussed above. Our variable, mean years of schooling, is easy to assemble, covers a long array of country-years, and it allows us to predict the type of revolutionary events rather than simply describing them, as is the case with Dahlum's combined educational index. Of course, the mean years of schooling is likely to have less explanatory power, but this is because it is presumably the ultimate cause, while Dahlum's variable, which deals with the education of the campaign members, is more proximate. Presumably, the number of students and graduated members in campaigns is a consequence of the general education of the population, which is fully reflected by mean years of schooling.
Thus, we maintain that using the mean years of schooling is a qualitatively new step in the study of the possible influence of education on the revolutionary event format. It is easy to use, it covers a large time series, and, most importantly, can both help explain past revolutions and forecast the future format of forthcoming revolutions, which positively distinguishes it from the index proposed by Dahlum. A comparison of descriptive statistics for both variables can be found in Table S1 online (especially note the difference in the number of observations).
In addition, we introduce an education index that distinguishes countries by the level of formal education proliferation. It is the variable 'mean years of schooling' (which is discussed below) that we divide into six equal parts from the global samplesextiles. Thus, we obtain: Sextile 1. Very low level of education (up to 2.08 mean years of schooling). Sextile 2. Low level of education (from 2.08 to 3.81 mean years of schooling). Sextile 3. Lower-middle level of education (from 3.81 to 5.58 mean years of schooling). Sextile 4. Upper-middle level of education (from 5.58 to 7.45 mean years of schooling). Sextile 5. High level of education (from 7.45 to 9.4 mean years of schooling). Sextile 6. Very high level of education (9.4 mean years of schooling and more).
With this index, we analyze cross-tabulation with revolutionary protest and revolution formats to determine the joint distribution and relationship between the level of education and frequency of types of revolutionary uprisings.

Control Variables
This section discusses previous research and our reasoning for employing the following control variables: Youth bulges, GDP per capita, urbanization, democracy level, discrimination, area, population density and population. Cincotta and Weber (2021) found that violent revolutions are significantly more likely in countries with a very high proportion of the youth in the total adult population of this societythe so-called 'youth bulge'. This finding is very congruent with other research on demographic structural factors of revolutions. 10 This relationship is associated with the fact that young people are easier to engage in violent revolutionary actions because, as "most young people have fewer responsibilities for families and careers, they are relatively easily mobilized for social or political conflicts. Youth have played a prominent role in political violence throughout recorded history, and the existence of a 'youth bulge' (an unusually high proportion of youths 15 to 29 relative to the total adult population) has historically been associated with times of political crisis" (Goldstone, 2002, p. 11-12; see also, e.g., Weber, 2019). Thus, the higher the proportion of young people in the population, the higher the likelihood of violent destabilization and the lower stability of the regime (Cincotta & Doces, 2012;Cincotta & Weber, 2021;Farzanegan & Witthuhn, 2017;Korotayev et al., 2011. We operationalize 'youth bulge' through two approaches found in these studies: (1) through the proportion of people in the population between the ages of 15 and 29 in the total adult (15+) population and (2) through the median age of the population. These variables are provided by the United Nations Population Division (UNPD) World Population Prospects database (UNPD, 2022a), or have been calculated on its basis.
GDP per capita may favor nonviolent forms of revolutionary action (e.g., Grinin & Korotayev, 2016;Inglehart & Welzel, 2005;Korotayev, Bilyuga, & Shishkina, 2018). For instance, Inglehart and Welzel (2005) claim that the explosive growth of wealth (using proxy through GDP per capita) generates a growing need for self-expression including political participation; and the expansion of markets and trade has always been a crucial factor in reducing violence due to the demand for nonviolent communication (Inglehart et al., 2015). So, higher well-being may be associated with higher nonviolent protest activity because economic development and the natural expansion of the middle class have led to a greater public interest in expanding political and civil liberties (Chenoweth & Ulfelder, 2017;Massoud et al., 2019). Researchers find robust evidence that GDP per capita is positively associated with nonviolent protests and negatively with violent destabilization (Dahl et al., 2021;Gleditsch & Rivera, 2017;Korotayev, Bilyuga, & Shishkina, 2018;Korotayev, Vaskin, & Bilyuga, 2017) and civil wars (Hegre & Sambanis, 2006). This relationship presumably exists because high well-being dramatically increases opportunity costs for protesters: people have bigger accumulated investments, and the risk of losing everything overrides all possible benefits. So, if the pre-conflict state equilibrium provides people with a small level of utility, the marginal utility of each increase in goods will be higher, which pushes people to risk giving up their usual life (Besançon, 2005;Sambanis, 2001). Moreover, the elites of rich countries can actively use various redistributive policies or co-opt the opposition elite to mitigate general discontent, which is possible due to soft resource constraints (Wimmer et al., 2009). Thus, we add the GDP per capita logarithm from the V-Dem dataset (Coppedge et al., 2022b) to control our model for the level of well-being.
The level of urbanization is also relevant. For example, mass mobilization is more likely in the most urbanized and complex societies with dispersed social power (Gleditsch & Rivera, 2017), where a high concentration of the population and human capital helps disaffected groups find a larger audience (Beissinger, 2022;Butcher & Svensson, 2016;Chenoweth & Ulfelder, 2017;Dahl et al., 2021). It is also important to note that in urbanized areas there is a high likelihood of peaceful protests, whereas at the periphery radical groups choose another method of disagreementviolent actions (Buhaug & Lujala, 2005;Dahl et al., 2021;Korotayev, Sawyer, Gladyshev, et al., 2021). For our urbanization variable, we take the share of the population that lives in urban areas. These data are from the United Nations Population Division (UNPD) World Urbanization Prospects database (UNPD, 2022b).
Moreover, a large number of researchers notice that the political regime type also matters. Relatively long ago, Karl Popper said that he calls 'the type of government that can be eliminated without violence "democracy", and the other "tyranny"' (Popper, 1949, p. 90). Modern researchers have similar findings: there is a greater likelihood of peaceful protest mobilization in democracies than in autocratic regimes (Caren et al., 2016;Chenoweth & Ulfelder, 2017;Dahl et al., 2021;Korotayev et al., 2016;Ustyuzhanin & Korotayev, 2022). This association is postulated to be because in a democracy: (1) it is easier for dissatisfied citizens to present their demands to the government or to mobilize in a democratic country, where the structure of institutions presupposes the inclusion of broad masses in governance (Nam, 2007); (2) a relatively higher level of freedoms and less power of the repressive apparatus leads to an increase in the likelihood of a nonviolent revolutionary protest (Massoud et al., 2019). In other words, a higher level of political repression may increase the likelihood of violent uprising (Regan & Norton, 2005). Thus, the reasoning is that democracy does not in itself lead to an increase in discontent but opens the way for its expression through peaceful mass mobilization in polling stations and streets (Dahl et al., 2021). On the other hand, such revolutions as the Ukrainian (Euromaidan) Revolution of 2013-2014, or the Armenian Velvet Revolution of 2018 demonstrate that revolutions may well topple democratically elected presidents, but within an even partly democratic system a revolution is much more likely to take an unarmed form than in a full autocracy. So, type of regime matters, and we introduce index of Electoral Democracy from Varieties of Democracy (V-Dem [Coppedge et al., 2022b]), that 'is formed by taking the average of, on the one hand, the weighted average of the indices measuring freedom of association, clean elections, freedom of expression, elected officials, and suffrage and, on the other, the five-way multiplicative interaction between those indices' (Coppedge et al., 2022a, p. 43), and it scales from 0 to 1.
Moreover, it is also necessary to control our model by introducing a variable that can explain ethnic discrimination, because this factor is found to be quite important. The majority of papers on civil war find that the likelihood of violent uprising onset is positively associated with ethnic discrimination (Gurr, 2000;Besançon, 2005;Buhaug & Lujala, 2005;Regan & Norton, 2005;Wimmer et al., 2009). The suggestion is that the discriminated group is likely to choose violent tactics rather than nonviolent ones. Firstly, such groups usually do not have enough opportunities for a successful nonviolent uprising, because dominant ethnic groups own most of the resources and use the state to restrict the access of minorities to various goods as, for example, education (Besançon, 2005). Secondly, costs for collective violent action for discriminated ethnic groups are smaller, because: (1) there are stable social ties and trust; (2) opportunity costs are not big, because the well-being of the discriminated group is usually low and they tend not to have big amounts of accumulated investments to human capital, and (3) consequently, their possible gain from success is large (Sambanis, 2001). Thus, we introduce a control for the share of the discriminated population from the Ethnic Power Relations (EPR) Dataset (Cederman et al., 2022) that gives the following description: 'group members are subjected to active, intentional, and targeted discrimination by the state, with the intent of excluding them from political power. Such active discrimination can be either formal or informal, but always refers to the domain of public politics (excluding discrimination in the socio-economic sphere)' (Vogt & Rüegger, 2021, p. 6).
In addition, we include several geographical variables that may explain the tactic of protestors, which was noted by several pieces of research (Sambanis, 2001;Fearon & Laitin, 2003;Wimmer et al., 2009). These are the logarithm of the area (in sq. miles) and population density that are provided by the CNTS Database (Banks & Wilson, 2021). Of course, it is also necessary to control for the total population that is also positively associated with violent uprisings (Hegre & Sambanis, 2006). This variable is presented in thousands by the United Nations Population Division World Population Prospects database (UNPD, 2022a). Finally, we include a decade variable to control for time trends that is necessary while one uses panel data.
In sum, a total of 377 maximalist (revolutionary and quasi-revolutionary) campaigns and 227 revolutions/revolutionary episodes were taken from 14,669 country-year samples with all controls with few missing values from 1950 to 2020 (see Table S1 online).

Multicollinearity Problem
As one can see from Table S2 (online), there is a strong multicollinearity between the variables from the 'modernization' group (log per capita GDP, urbanization, and median age/share of youth). All correlations between them are significant and mostly greater than 0.5. For example, the Pearson correlation between per capita GDP and urbanization is 0.839 (p < .001), while the correlation between per capita GDP and median age is 0.755 (p < .001) Therefore, to minimize multicollinearity we will separately model each of these highly correlated variables. To check that we have indeed eliminated the problem, all combinations of variables, presented in the models, are checked for multicollinearity through VIF (variance inflation factor) analysis. It can be seen that the VIF scores are not higher than 5, which is regarded as generally acceptable (all VIF tables can be found in the supplementary online material). For example, Craney and Surles (2002) distinguish a decisive boundary at 20 that is much higher than our scores.

Results
In this section, we first present the results of the distribution of campaigns and revolutions across six groups of countries, distinguished by six levels of education, to examine how educational attainment relates to the distribution of violence and nonviolence. We then present logistic models with different controls to test our main hypothesis. Following that, predicted probabilities of campaigns and revolutions on the level of education are shown. Finally, a comparison of effects between mean years of schooling and Dahlum's combined education index is presented. Overall, we find robust evidence that the higher the level of education in a country, the lower chance that the revolution there would take a violent/armed form.  to nonviolent revolutionary methods is also different between these groups. One can see that while in the first and second groups most cases are violent, in the last two groups more than 85% of revolutionary events are non-violent. Countries with high and very high levels of education have almost no violent campaigns. In countries with lower-and upper-middle levels of education, the proportion of violent revolutionary campaigns is higher than in the countries with high and very high values of the mean years of schooling, but lower than in the states with low and very low levels of education. Thus, there is a clear relationship: as education increases, the proportion of violent maximalist (revolutionary & quasi-revolutionary) campaigns decreases.

Level of Education and the Format of Revolutionary Events
Similar results are presented in Figure 2, which compares the frequency of revolutions by six groups of countries on the basis of our own dataset of revolutionary events described above and in the online supplement to this article. Obviously, there is the same trend as in the first figure (Figure 1): when the education level increases, violence drops remarkably. In the first group there is the highest proportion of violent/armed revolutions, in the last group this proportion is the smallest. However, the total number of revolutionary events per group changes not much for countries with low levels of education (sextiles from one to 2). Following that, there is a decrease in the number of revolutionary events from the second to the third group. Meanwhile, that transition from the high level to a very high level does not affect the number of  Table S8). total cases. However, there is a difference in the amount of violence. Thus, in the case of revolutions, there is also a rather clear relationship: as education increases, the share of violent/armed revolutions decreases.
Taken together, the results suggest that there is a clear trend towards a decrease in the amount of revolutionary violence with the growth of education level. However, while the number of episodes also falls in the case of the campaigns, there is no such thing with revolutions. Interestingly, the most remarkable decrease in the level of violence is seen with the transition from the very low level to the lower-middle level of education. For both NAVCO's campaigns and revolutionary events, the difference between the first and third levels is more than 50%, while the distinction between high level and very high level is negligible.

Mean Years of Schooling, Revolutionary Protests, and revolutions
The results of the logistic regressions of mean years of schooling on NAVCO's revolutionary campaigns with the set of control variables are presented in Table 1. As might be expected based on our theoretical research, education is indeed positively and significantly associated with the nonviolent type of campaigns in all models with the introduction of all kinds of controls.
Thus, even in M1, which presents a pairwise regression, it is seen that mean years of schooling significantly (at the level p < 0.001) affect the protesters' choice of nonviolent tactic. After controlling for the logarithms of population, area, and population density within M2 and M3, education is still significantly related to nonviolence. Note that none of these controls has a significant effect on the protesters' choice of tactics. In M4 and M5, we introduce the index of electoral democracy and the share of the discriminated population. The most surprising is that democracy itself is not a significant predictor (for a possible explanation of this result see Ustyuzhanin & Korotayev, 2022;Ustyuzhanin & Korotayev, 2023), while the share of the discriminated population is appreciably significant (p < 0.1) and associated with a greater likelihood of violence in all subsequent models.
In the further models (M6-9), the variables from the "modernization" group are added in turn to the already ruminated controls. Thus, the GDP per capita logarithm appears in M6, and is significantly (p < 0.1) positively associated with nonviolent tactics, and at the same time reduces the significance level of education (from p < 0.01 to p < 0.1) and its effect (from 0.281 to 0.144). Note, this model has the smallest prediction error and, therefore, the highest statistical quality (the Akaike's Information Criteria (AIC) is 379.709) 11 among all the models. In M7 urbanization is introduced. It is a significant (p < 0.1) and positive predictor of nonviolence. However, in contrast to the logarithm of GDP per capita, it does not diminish the significance of education and does not affect its effect as much. M8 and M9 add the share of youth and the median age. These controls are not significant and do not affect the relationship between  education and nonviolence. Interestingly, these two variables, which operationalize the 'youth bulge', are not significantly different from each other in our analysis, suggesting that it is possible to operationalize it through either of them.
To evaluate the effect of education on nonviolent revolutionary campaigns the plot in Figure 3 is introduced. It shows the probability of a peaceful case depending on mean years of schooling with a 95% confidence interval. Note, this graph is based on M6 from Table 2 (with logarithm of GDP per capita as a 'modernization control' and the smallest AIC) with mean values of all controls. From the plot, one can see that the highest probability of revolutionary nonviolence is in countries with the highest level of education. For instance, if the mean years of schooling are 12, the predicted probability of revolutionary protest/uprising being unarmed/nonviolent is almost at 85%. At the same time, in countries with the lowest level of education probability of revolutionary uprising being unarmed/nonviolent is rather low. If the mean years of schooling are close to zero, 12 the probability of peaceful revolutionary campaign is just about 40%. Overall, the same tendency can be seen as in Figure 1: when the population becomes more and more educated, the nonviolent/unarmed tactic becomes preferred by the revolutionaries more and more.
Consider now the results of the second series of regression tests (see Table 2): The second table (Table 2) is similar to the first (Table 1), but the data from NAVCO is replaced with our dataset of revolutionary events (thus, excluding quasi revolutionary episodes, as shown in Table S5 in the supporting online materials). At the same time, the set of variables in the models and the dependent variable remain unchanged. As expected, mean years of schooling is   Note. ***p < .01; **p < .05; *p < .1; decade fixed effects in M6-M9 models are excluded because after their introduction all variables become totally insignificant. This is due to small number of observations in relation to number of predictors with dummy variables for each decade.
a consistently significant and positive predictor of nonviolent revolutions, as in the case of previous tests with maximalist (revolutionary & quasirevolutionary) campaigns. However, in contrast to previous tests GDP per capita and urbanization are losing their significance. On the other hand, controls from the 'modernization group' reduce the significance of the main independent variable. Overall, this test on revolutions further confirms our hypothesis: the higher the mean years of schooling in a country, the more likely that revolutionary events there will be nonviolent rather than violent. Figure 4 shows a plot rather similar to the one found in Figure 3, but it is based on our dataset of revolutionary events. Broadly put, the conclusions are consistent with those that we have arrived at above with NAVCO. However, the curve slope is higher, suggesting that education is an even stronger inhibitor of the violence of the revolutions than in the case of NAVCO's maximalist campaigns (remember that, in addition to revolutionary events, NAVCO dataset includes some quasi-revolutionary episodes).
To compare effects of mean years of schooling and other variables on the probability that revolution or revolutionary protest will take a peaceful form, the standardized average marginal effects (AME) within models with NAVCO (Table 1) and revolutionary (Table 2) datasets with 95% confidence intervals are presented in Figure 5. Standardized AME illustrates the effect of a one standard deviation increase of each predictor on the probability that revolutionary protest or revolution will take an unarmed/nonviolent form. In other words, AME shows comparable effects of variables on different scales. One can see that the change of mean years of schooling is significant and a strong factor, which increases the probability of nonviolence during revolution by about 15% and during NAVCO's campaign by a little less than 10%. GDP per capita has a comparable effect, but only during revolutionary protests, but during revolutions its effect is close to zero, which is associated with low significance. Effects of the other variables are much weaker than the mean years of schooling or totally insignificant.
Taken together, results of our analysis of the NAVCO dataset and our dataset of revolutions support the hypothesis that there is a significant positive relationship between the mean years of schooling and the probability of revolutionary events taking a nonviolent/unarmed form. At the same time, in the case of NAVCO, a high level of well-being also encourages protesters to choose peaceful tactics, while a large proportion of the discriminated population predicts that the rebellion will be armed. In contrast, in the case of pure revolutions these factors are insignificant, while the index of electoral democracy is an important and positive predictor of nonviolence. Moreover, the predictive curves and AME confirm the significance and effect of mean years of schooling. Nevertheless, the effect of education is greater during pure revolutions than within revolutionary protests.

Comparison of Explanatory Variables: Mean Years of Schooling Versus Combined Education Index
The results of standardized logistic regression from the Table 3 show the comparable effect of mean years of schooling and Dahlum's combined education index, 13 on the one hand, and NAVCO as the dataset of revolutionary and quasi-revolutionary events, on the other.  As one can see from M1 and M2 with pair regressions from the NAVCO data, the significance of mean years of schooling and combined education index are the same, at the p < 0.01 level. M3 and M4 are similar and confirm previous results in the case of our dataset. Interestingly, Dahlum's variable can also be successfully applied to revolutionary events proper, not just maximalist campaigns. The subsequent M5 and M6 show regressions with mean years of schooling and combined education index together. Variables in both models do not conflict with each other and have equally strong significance, but mean years of schooling has greater effect on response. In M7, with the full set of controls, mean years of schooling loses some effect size, while the combined education index does not, but in M8 mean years of schooling returns its effect strength. Interestingly, in the last models none of the controls are significant, unlike previous models (neither for the NAVCO's campaigns nor for the revolutionary events) where the first and second variables were not introduced together.
To evaluate the collaborative effects of mean years of schooling and combined education index on nonviolence, the plot in Figure 6 is introduced. It shows the AME of standardized models from Table 3 -M5 (which we call "restricted" because control variables are not added) and M7 (which we call "full")with a 95% confidence interval. Overall, the value of each variable falls within the range of the other, indicating approximately the same strength of the effect. Meanwhile, mean years of schooling has a slightly higher effect in a restricted model, while Dahlum's index is a little stronger within the full model. Thus, Dahlum's Combined Education Index is an important and strong factor, but far from exhaustive. First, when this factor is added together with mean years of schooling into one regression model, both variables are significant, suggesting they are not identical. Second, as can be seen from the standardized AME, the strengths of the two variables are roughly comparable. Third, the inclusion of both education variables into one model improves the quality of the respective models in a substantial way (as indicated by the AIC measure). Thus, our proposed explanatory variable does not replace Dahlum's one, but it is a significant, powerful, and qualitatively different predictor that shows that education of whole population has a strong effect on nonviolence during revolutions and revolutionary protests.

Discussion and Conclusion
In this article, we have attempted to analyze and close a gap in contemporary cross-national research on how education affects the form that a revolution takeswhether it will be violent/armed or nonviolent/unarmed or, in other words, how education affects revolutionaries' choice of violent or nonviolent tactics. Overall, our hypothesis is supported: using several logistic models with important controls, a higher level of education in a country does significantly predict that revolutionary events will take an unarmed/nonviolent form. This is consistent with the theory that mean years of schooling is a crucial factor pushing towards a nonviolent format of revolutions.
At the same time, it has been found that on average countries with higher mean education have fewer violent events as a proportion of maximalist (revolutionary and quasi-revolutionary) campaigns/revolutions than countries with lower mean years of schooling. More specifically, the greatest difference between violent and non-violent maximalist revolutionary events is between countries with very low and lower-middle level of education. This suggests that education has the highest pacifying effect in the early stages of modernization, when the average years of schooling moves to approximately 4 years. It appears that this effect gradually decreases: overcoming the mark from upper-middle to high level has a markedly less pronounced pacifying effect. 14 Moreover, in already developed societies the difference between high and very high level no longer plays a very important role both in the decreasing of violence and the total number of revolutionary events. Also, our analysis suggests, contra Dahlum, that it is not only the presence among the campaign participants of students and university graduates that mattersthe primary and secondary education of the populace matters too. We saw that the greatest effect of the education proliferation on the movement from the violent to nonviolent revolutionary tactics is observed between countries with very low and lower-middle education level, which is achieved first of all through the proliferation of primary and secondary, but not tertiary education. A natural possible continuation of the present work is to analyze this suggestion and test it, using different variables that characterize primary, secondary and tertiary education. Note that there are several articles on this topic, but they examine the onset of civil wars, but not the choice of violent versus nonviolent tactics by the rebels (see, e.g., Barakat & Urdal, 2009;Collier, 2004;Thyne, 2006). Moreover, it is also important to further analyze revolutions dynamically in order to see how the choice of tactics occurs during mobilization. Most likely, such research will be only possible through qualitative case studies. Another good point is to analyze, how education affects the likelihood of onset of armed or unarmed revolutionary situation. For example, Butcher and Swenson (2016) found negative relationship between education and armed revolutions, while Dahlum and Wig (2019) described a link between education and instability in African countries. However, the Chenoweth and Ulfelder's study (2017) shows the insignificance of education as a predictor of the onset of unarmed maximalist campaigns. Thus, the conclusions of the existing quantitative cross-national studies are contradictory, and the topic remains clearly under-researched. In the supplementary online material, we did a few tests to analyze the effect of mean years of schooling on revolutionary onset. Using bias reduction logistic regression for rare events data, we found that education is significant and positive predictor of unarmed revolutionary onset in most models, and is significant and negative predictor of armed revolutionary situation (Tables 6-S7). However, this topic should be studied in the further research more carefully.
If one insists on thinking about tactics as something that is chosen during mobilization then it would almost make more sense to look at changes once something has startedare we more likely to see shifts to violent or nonviolent strategies depending on level of education of the participants, for example? Also, it is important that the proposed explanatory variable, mean years of schooling, is a significant, positive, and strong predictor of nonviolence (both for NAVCO's campaigns and pure revolutions) that does not just copy Dahlum's index, but allows a qualitative depth for the research of the problem. Of special importance is that, although our factor is not stronger (but still also not weaker), it has real predictive power, unlike Dahlum's Combined Education Index, which can only be a descriptive variable, explaining the violent versus nonviolent tactics of rebels in retrospect. In other words, Dahlum's Combined Education Index cannot be used to forecast what kind of revolutionary episodes are more likely to be expected in the given country, as this index can only be calculated after the revolutionary events are over. In contrast to it, the mean years of schooling can be applied for forecasting purposes. Moreover, these are two qualitatively different and independent factors, because it is not only the tertiary education level of the participants that matters but also the general level of education of the country, which affects not only a much wider circle of the rebels (there are all grounds to expect that rebels with complete secondary education would behave significantly differently in comparison with illiterate rebels), but also the quality of institutions and people on the other side of the barricades. Thus, mean years of schooling can be seen as more applied and scientifically meaningful, allowing for the further study of violent versus nonviolent revolutionary action.
Turning to other results, it should be emphasized that although education is an important factor in protesters' choice of tactics, it does not the only one. Certainly, there must be other reasons why some revolutions take on a nonviolent character and others a violent one. Our analysis suggests that some economic and political factors might be also vital here. For instance, higher well-being (proxied through logarithm of GDP per capita) and urbanization seem to be important modernization factors making nonviolent/unarmed revolutionary tactics more likely. Though Huntington (1968) claimed that modernization leads to revolutionary conflict as the forces of participation exceed the state's ability to meet the new demands for representation (and Figure 2 above suggests that the revolutionary events at the intermediate phases of modernization might indeed be somehow more likely than at its initial and final stages), we add that it can also contribute to making revolutionary movements less violent. Although the other variables which characterize modernization were not significant in our analyses, this does not mean that they do not matter. So, a greater focus on these factors could produce interesting findings that have not been researched here.
Finally, our analysis demonstrates that the NAVCO dataset can well be used for the quantitative study of revolutions, as the presence of a number of quasi-revolutionary episodes in this dataset does not appear to lead to any significant distortionsthe analyses of the NAVCO dataset and our dataset of revolutionary events have produced qualitatively rather similar results.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This article is an output of a research project implemented as part of the Basic Research Program at the HSE University in 2023.

Supplemental Material
Supplemental material for this article is available online. Notes 1. In 1990, the mean years of schooling was 8.4 in the Czech Republic and 10.8 in Slovakia (a detailed explanation of this indicator is given in materials and methods section). 2. The mean years of schooling in Yemen in 2004 was just 1.4 years. 3. What is more, Dahlum only takes into account tertiary education; for her there is now difference between the situation when the majority of rebels are illiterate or have complete secondary education (if both rebellions have equal proportion of students and graduates). We do not find such an assumption really convincing. 4. The campaign data in her research were taken from the NAVCO database, which will be discussed below. 5. The discussion of this indicator can be found in the next section. 6. Or, to be more exact, "unarmed" form. In fact, Kadivar and Ketchley (2018) quite convincingly show that the participants in the majority of the so-called "nonviolent maximalist campaigns" resorted to violence on a fairly serious scale (here one can recall, for example, the Egyptian revolution of 2011 or the Ukrainian revolution ["Euromaidan"] of 2013-2014, which Chenoweth and Shay (2020a) quite confidently qualify as "nonviolent maximalist campaigns"), in connection with which they, with good reason, believe that it is wrong to call such revolutionary events "non-violent", suggesting rather to designate them as "unarmed". Yet, in order to facilitate the comparison with earlier research on this subject, we continue to denote them as "non-violent". 7. Chenoweth and Shay themselves emphasize that "campaigns are primarily nonviolent when the vast majority of participants are unarmed, and when they use mostly nonviolent practices to confound, impede, and challenge the regime and its supporters. Campaigns are primarily violent when most participants use force, especially armed force, to target regimes and their supporters" (Chenoweth & Shay, 2020b, p. 6). 8. This stands in a sharp contrast with the Belarusian Revolution of 2020-2021, when the protests continued for a few months, involved hundreds of thousands participants, with the opposition forming a specific body claiming its rights to take power in the country and undertaking specific steps to achieve this goal (e.g., Moshes & Nizhnikau, 2021). We are ready to denote such events as a (failed/ unsuccessful [at least, by the moment]) revolution/revolutionary episode, but we are not ready to denote as such the post-election anti-Lukashenko protests of March 2006. The point that some members of the Belarusian opposition opted to call the March 2006 events "Denim Revolution" or "Jeans Revolution" reflects first of all the unrealized hopes of some of them that those events would develop in a true revolution; the same is relevant, e.g., for the so-called "Snow Revolution" that took place in Russia in winter 2011/2012. 9. It still appears important to stress that the quasi-revolutionary events constitute a minority of all the episodes described in NAVCO, and, as we will see below, the NAVCO dataset can well be used for the quantitative study of revolutions.
10. See, e.g., Goldstone, 1991Goldstone, , 2002Goldstone, , 2017Korotayev et al., 2011;Korotayev & Zinkina, 2022. 11. In other words, the smaller size of the AIC, the higher quality of the model. 12. This can be the case for developing countries. For example, mean years of schooling in Yemen in 1994 was just 0.6, in Niger in 1991 -0.7, in Mali in 1995 -0.9. Nevertheless, the minimum value of education in the plot is "0", which is regression extrapolation helping to understand effect of variable within entire potential range, but not a real case. 13. We would remind the readers that Dahlum only takes into account the tertiary education level of rebels, ignoring the proliferation of education among the general population, and does not pay any attention to the proliferation of primary and secondary education even among the rebels (let alone the general public). 14. Note the important point, that we discuss average effects of different types of education levels, but not real transition from one level to another, because, as Lange shows (2011) the transition itself between education levels in the short term might be accompanied by an increase in violence.