Military Repression and Restraint in Algeria

The Algerian military’s response to the 2019–2020 Hirak protests was relatively peaceful. In contrast to its violent repression of protests in 1988, and subsequent coup and civil war in the 1990s, the military showed considerable restraint toward the Hirak. Leveraging a survey of 2,235 self-reported military personnel, I show that the military’s restraint emanated from protesters’ use of nonviolence and fraternization, as well as from a recognition that the military’s more repressive approach in the 1990s was a mistake. At the same time, a priming experiment suggests that the military’s willingness to repress increases when protesters threaten the military’s corporate interests, and when Russia, Algeria’s primary arms supplier, reiterates its support for the regime. Overall, the results show how protester tactics, international reactions, and political learning can condition the military’s repression or restraint during times of unrest.


INTRODUCTION
O ver the past 5 years, protesters have taken to the streets from Hong Kong to Khartoum and Minsk to Managua in an effort to overthrow their dictators or reform their political systems.Whether these mass movements succeed, however, often turns on the behavior of the military.When militaries refuse to repress the protests, dictators are almost inevitably forced from power.But when militaries instead agree to repress, protests are typically crushed or escalate into civil war.Chenoweth and Stephan (2011, 58) thus find that major security sector defections make mass uprisings 60% more likely to succeed.
Scholars have accordingly sought to understand the conditions under which militaries "defend or defect" from the dictator in the face of a mass uprising (i.e., Barany 2016;Bellin 2012;Lee 2014).The most recent wave of scholarship, after the Arab Spring, has convincingly shown that a dictator's "coup-proofing" strategy shapes the military's response to mass uprisings (Albrecht, Croissant, and Lawson 2016;Bou Nassif 2021;Brooks 2017).Militaries that are stacked with the dictator's ethnic or sectarian group, like in Bahrain or Syria, are more likely to defend the regime from a mass uprising emanating from the out-group (Allen 2019;Bellin 2012;Harkness 2018;Hassan 2020;Makara 2013;McLauchlin 2010;Nepstad 2013;Roessler 2016).By contrast, militaries that are neglected and counterbalanced, like Tunisia's, are more likely to defect (Bou Nassif 2015;Brooks 2013;De Bruin 2020;Grewal 2019a;Lutscher 2016;Lutterbeck 2012).
While ethnic stacking and counterbalancing produce relatively clear effects on repression and defection, more theoretically indeterminate are a dictator's efforts to buy the military's loyalty through a share of material wealth and political power, a strategy I call powersharing.These militaries, like in Egypt or Algeria, enjoy high salaries, large budgets, and political influence, giving the generals incentives to stick with the regime during mass uprisings (Barany 2016;Croissant, Kuehn, and Eschenauer 2018;Koehler 2017;Pion-Berlin, Esparza, and Grisham 2014).Yet, as Bou Nassif (2021) astutely observes, these benefits accrue disproportionately to the top ranks, raising the specter of vertical fragmentation, with the lower ranks more likely to side with the people.Depending on how the generals weigh these competing considerations, these cases can either produce restraint, as in Egypt in 2011 or Algeria in 2019, or repression, as in Egypt in 2013 or Algeria in 1988.In other words, militaries that have been coupproofed through power-sharing can go either way during times of mass uprising.
Understanding this variation is important for several reasons.Buying off the military through high salaries and political influence is one of the most common strategies for coup-proofing the military (Powell 2012;Quinlivan 1999;Svolik 2013).Indeed, many of the regimes that have yet to fully democratize feature these powerful, politicized militaries, including in Egypt, Sudan, Pakistan, Thailand, Myanmar, and North Korea (Albertus and Menaldo 2012;Bellin 2012).Understanding what additional variables can tip the scales and lead them to defect is therefore crucial for generating lessons on how to democratize in these hard cases.
I argue that in cases of power-sharing, where the regime's strategy vis-à-vis the military does not predetermine the military's behavior, other variables gain greater explanatory power.The first are protester tactics.Scholars have long argued that nonviolence and fraternization can induce security force defections (Binnejdik and Marovic 2006;Chenoweth and Stephan 2011;Ketchley 2014;Morency-Laflamme 2018;Nepstad 2011).However, such tactics are often less relevant in countries with ethnic stacking, where out-group protesters are perceived as violent and threatening regardless of their actual tactics (Edwards and Arnon 2021;Manekin and Mitts 2022), and in countries with counterbalancing, where neglected militaries are likely to defect either way.In countries with power-sharing, protester tactics should become relatively more important.Nonviolent protesters who are fraternizing with the soldiers increase the likelihood of vertical fragmentation, in turn producing a begrudging defection from the top.Meanwhile, violent protests, and those threatening to curtail the military's budget and influence, should induce greater support for repression.
Second, how the international community responds during an uprising may also shape these militaries' calculations.Several scholars and particularly policymakers have emphasized the role played by the United States in encouraging the Egyptian military not to repress protesters in 2011, for fear of losing U.S. aid (Gates 2015;Obama 2020;Taylor 2014).More generally, it is possible that the great power most involved in supplying and training a military may have outsized influence over its behavior.The reaction of that great power, whether in support of repression or defection, is likely to shape the military's calculations as well.
A third and relatively novel factor is political learning.I argue that militaries learn from their responses to previous uprisings.If repression went poorly last time, by, for instance, triggering a civil war, then the generals are more likely to exercise restraint next time around.Meanwhile, if defection last time led to a new leader who curtailed the military's interests, then they are more likely to repress next time around.Historical experience should thus factor into their calculations as well when responding to an uprising.In short, in these cases where regime-military relations do not pre-determine the military's behavior, these three additional factors-protester tactics, international reactions, and political learning-should help tip the scales toward repression or restraint.
To test this theory, this article leverages the case of Algeria, where dictators have historically coup-proofed the military through power-sharing, and not also through ethnic stacking or counterbalancing.Algeria thus features variation in how its military has responded to mass uprisings.It brutally repressed riots in 1988 and then staged a coup and fought a bloody civil war in the 1990s.But in 2019, when a mass uprising emerged against President Abdelaziz Bouteflika, the military abandoned Bouteflika, forcing his resignation.As the Hirak protests continued on for another year, demanding deeper reforms, the military likewise showed surprising restraint, not once deploying troops to fire on protesters.Why did the military exercise restraint in 2019-20, in contrast to its repression one generation earlier?
I shed light on the Algerian military's decisionmaking calculus through an innovative survey of 2,235 active and former military personnel conducted in 2019-20, during the Hirak protests.The survey data show how each of the factors outlined above shaped the military's calculations.The military's restraint in 2019-20 appeared to stem primarily from (1) the protesters' tactics, particularly their use of nonviolence and fraternization, and (2) the military's political learning, realizing their earlier repression in the 1990s was a mistake.On the other hand, the survey also uncovers evidence through a priming experiment that the protesters' calls for civilian control over the military, and Russia's statements in support of the regime, increased the military's willingness to repress.The results thus show how in these theoretically indeterminate cases, where militaries could either repress or defect, these three factors can tilt the scales in either direction.
Beyond these substantive contributions, this article also offers a new methodology for surveying hard-toreach populations like the military, in difficult times like during a mass uprising.To do so, it builds off of pioneering work by Potzschke and Braun (2017) and Jäger (2017) using targeted advertisements on Facebook, to target them in this case to military personnel.While the survey samples it generates may not be representative, this method allows us to recruit uniquely large samples of military personnel for surveys.By doing so during a mass uprising, we can gauge soldiers' willingness to repress protesters, and the factors that influence this decision, even before they are asked to do so.

SHOOT OR SHIRK
"There is no doubt that the fate of every revolution at a certain point is decided by a break in the disposition of the army," wrote Trotsky (1932, 88).A growing literature in political science today echoes Trotsky's observation, noting that while the police regularly agree to repress protests, the military's response is both more lethal and more variable, making it the pivotal actor that makes or breaks a revolution (Barany 2016;Lee 2014;Nepstad 2013).
The dominant paradigm regarding why some militaries defend their dictators and crush protesters stresses regime-military relations (Brooks 2017), particularly the dictator's "coup-proofing" strategy (Quinlivan 1999).Dictators who stack their militaries with their own in-group create a "protection pact" (Slater 2010) with a shared aversion toward the outgroup (Harkness 2018;McLauchlin 2010).In the Arab Spring, Syria's Bashar al-Assad could thus rely on his Alawi-stacked military to repress a largely Sunni uprising, while Bahrain's monarchy could likewise deploy its Sunni military against a largely Shia uprising.
Other coup-proofing strategies, however, leave dictators vulnerable during times of mass uprising.Dictators who neglect and counterbalance their militaries by building up militias or paramilitary forces may be able to rely on the latter for repression, but are likely to be abandoned by a military that resents them (Bou Nassif 2015;Brooks 2013).If ethnic stacking produces loyalty in times of unrest, counterbalancing is thought to produce defection (Allen 2019;Barany 2016;Bellin 2012;Lutscher 2016;Morency-Laflamme 2018).
The literature is conflicted, however, over a third major coup-proofing strategy: power-sharing.Here, the dictator attempts to win the military's loyalty through a share of material resources and political power.On the one hand, these militaries have vested interests in the dictator, and fear that any change in leadership might result in a decline in their institutional privileges (Croissant, Kuehn, and Eschenauer 2018;Pion-Berlin, Esparza, and Grisham 2014).On the other hand, as Bou Nassif (2021) points out, these benefits largely accrue to the top ranks, creating the possibility of vertical fragmentation, with the lower ranks preferring defection.Depending on how the generals assess the loyalty of their subordinates, these militaries could either defect like in Egypt in 2011, or repress like in Egypt in 2013.
In this article, I examine what might tip the scales in these theoretically indeterminate cases.I outline three primary sets of variables concerning protester tactics, international reactions, and political learning.While these factors may be inconsequential in countries with ethnic stacking or counterbalancing-where the military is likely to repress or defect regardless-I argue that they may be critical to shaping military behavior in countries with power-sharing.

Protester Tactics
The first set of variables concerns the protesters' tactics.Protesters who are nonviolent and who fraternize with the military facilitate defection, whereas those who are violent and threaten the military's interests engender repression.
Violence is a blessing to senior officers who are trying to justify repression to their subordinates.Violence allows them to paint the protesters as criminals and terrorists, in turn legitimizing repression (Binnejdik and Marovic 2006;Chenoweth and Stephan 2011;Nepstad 2011).While violence should encourage repression among all levels in the military, I contend that this effect will be particularly strong for senior officers in powersharing regimes.For the top ranks seeking to preserve their privileges, they are likely to seize the opportunity presented by the use of violence to push for repression.While the lower ranks may view the use of violence as an isolated incident, not representative of the protesters, the senior officers have interests in using it to paint all protesters as criminals deserving of repression.
A second protester tactic is fraternization.To fraternize-literally, to turn people into brothers-refers to tactics protesters use to get the soldiers on their side by seeing their common humanity.Getting physically close to soldiers, handing them roses or water bottles, reminding them of their duty to the people and not just the regime, can each raise the moral costs of repression and make defection more likely.In Egypt, for instance, Ketchley (2014) documents how protesters hugged and kissed soldiers and chanted that "the army and the people are one hand" in order to encourage defection.
These tactics of fraternization should be particularly effective on conscripts, who are most literally the people's brothers.Conscripts are drawn randomly from the population, including from the same families who are protesting, while careerists more often tend to be selfselected from those willing to join the regime (Barany 2016;Brooks 2017;Lutterbeck 2012).Moreover, given their short time in service, conscripts tend to maintain their pre-existing friendships and social networks, while for careerists these ties to civilians tend to weaken over time (Cebul and Grewal 2022).In short, I argue that careerists should on average have greater social distance from the protesters, making fraternization more difficult, while conscripts should be more readily susceptible to these appeals.
Finally, protesters can also encourage or discourage repression if they spell out what their political agenda might mean for the military's corporate interests.Morency-Laflamme (2018, 10), for instance, argues that the opposition in Benin in 1989 was able to convince the military through a national dialogue that their interests would be enhanced under democracy, including by dismantling the paramilitary and redistributing its equipment into the regular army.By contrast, protesters who threaten the military's interests are more likely to be met with repression.For militaries that had been enjoying considerable material and political power under the current dictator, even simple calls for civilian control over the military might represent a decline in the military's influence.These signals shape the military's expectations for what might happen if they defect, pushing the ledger toward repression.Finally, these interests should be most salient for the officers who benefit or stand to benefit from power-sharing, rather than the low-ranking soldiers (Bou Nassif 2021).
Hypothesis 3: Threatening the military's interests encourages repression, particularly among officers.

International Reactions
A second set of variables that might tip the scales in cases of power-sharing are international reactions to the protests.Chenoweth and Stephan (2011, 52-4) highlight how nonviolent protests can succeed in part by getting foreign powers to apply diplomatic and economic pressure on the dictator.
It is plausible that such reactions also shape the military's decision to repress.After all, international reactions shape the military's calculations in other arenas, like coups and post-coup trajectories (Thyne et al. 2018;Thyne and Hitch 2020).Though like with coups, the reactions that likely matter the most are those from the great power that is training and supplying the military (Grewal and Kureshi 2019), typically the United States, France, or Russia.Indeed, many of the regimes that engage in power-sharing today do so with the support of a foreign power whose assistance helps win the military's loyalty (Bellin 2004).
I argue that the reaction of this foreign power, moreso than others, can tip the scales in times of unrest.If the military's arms supplier pressures it not to repress, as the United States did in Egypt in 2011 (Gates 2015;Obama 2020;Taylor 2014), the military is more likely to comply, for fear of losing its foreign assistance and training.Where these powers instead encourage repression, a crackdown becomes more likely.
Hypothesis 4: The reaction of the foreign power supplying the military will help shape its behavior, moreso than the international community writ large.

Political Learning
A final factor is political learning.I contend that militaries learn from their responses to previous uprisings.If the military repressed last time around, and it succeeded-in the sense of crushing the protesters, preserving power-sharing, and avoiding international sanction-such as in Bahrain in 2011-then they are more likely to do so again next time around.If, by contrast, repression led the military to fragment, or plunged the country into civil war-as in Algeria in the 1990s-then the military may be more hesitant next time around.Similarly, if the military defected last time, and that went poorly-in the sense of bringing to power a new leader who curtailed the military's interests-then the military may be more likely to choose to repress next time around.These historical experiences should thus factor into the military's calculations as well when responding to an uprising.Moreover, these historical lessons should be especially salient for those generations who experienced them directly, rather than younger generations born after the fact.
Hypothesis 5: Political learning from previous uprisings will shape decisions to repress, particularly among those who experienced them directly.

THE CASE OF ALGERIA
To test these hypotheses, I explore the case of Algeria.Algeria features a military coup-proofed through power-sharing, and not also through ethnic stacking or counterbalancing, allowing us to isolate what factors matter in these cases.I argue that protester tactics, international reactions, and political learning indeed tipped the scales in Algeria, leading the military toward repression in the 1980s and 1990s but toward restraint during the 2019-20 Hirak protests.

Background
The Algerian military has dominated the country's politics ever since winning its war of independence from France (Addi 1998).It staged its first coup just 3 years later, ousting President Ahmed Ben Bella in 1965 after his threats to counterbalance the military with popular militias (McDougall 2017, 250).The coup leader, Colonel Houari Boumédiène, led the country for the next 13 years, followed in turn by Colonel Chadli Bendjedid (1979-92).
In 1988, a severe economic crisis prompted widespread protests and riots against the regime (Zoubir 1993).The military agreed to repress on Bendjedid's behalf, crushing the protests and killing hundreds of Algerians (Allouche 2016;Delany 1988).Nonetheless, Bendjedid initiated a brief liberalization of the regime, permitting relatively free elections for municipalities in 1990 and parliament in 1991-92.The Islamic Salvation Front (FIS) was on the verge of winning these elections when the military stepped in once more, concerned the FIS might purge the senior ranks or even replace it outright with an Islamist militia (Willis 1997;Quandt 1998, 63).The military cancelled the elections and banned the FIS.When the Islamists rose up in protest, the military moved to repress them, sparking a brutal, decade-long civil war that would claim over two hundred thousand lives (Martinez 2000).
The military, eventually victorious, then sought to retreat from the limelight, installing President Abdelaziz Bouteflika (1999-2019) as a civilian façade.Despite appearances, the military remained the dominant voice behind the scenes, "ruling but not governing" (Cook 2007).Beyond determining security policy, the officer corps profited from vast networks of oil rents and other, often illicit, business interests (Cunningham 2013;Ouali 2018).
Seared by the memory of the civil war, Algerians largely avoided the 2011 Arab Spring protests, which saw only a few thousand Algerians take to the streets, easily repressed by the police (Entelis 2011;Pearlman 2013).However, grievances toward the regime grew in the mid-2010s.Global oil prices collapsed in 2014, making it difficult to sustain price subsidies and public sector employment (Ghanem 2019a;Zoubir 2016).Moreover, although a stroke in 2013 left Bouteflika almost fully paralyzed, the regime continued to re-field him for president in 2014 and 2019, reinforcing perceptions of a corrupt "pouvoir" (power) running the country and exploiting its wealth from the shadows (Meddi 2019).Bouteflika's nomination for a fifth term in 2019 lit this growing tinderbox of economic and political frustrations into the mass uprising of the Hirak.
The regime's initial response was to try to scare the protesters, threatening them with repression.On February 28, Prime Minister Ahmed Ouyahia tried to evoke the trauma of the 1990s civil war (Zeraoulia 2020), warning protesters: "I remember 1991, it was like today.I read today that there is a call to strike, I remember the strike of 1991" (EFE 2019).The regime also evoked regional examples.Commenting on protesters handing out roses to the police, Ouyahia noted that "In Syria, it started with the roses, too" (Souames 2019).
But the threat of military repression never materialized.Instead, as protests continued to grow week after week, the military abandoned Bouteflika.On March 26, the army chief of staff, General Ahmed Gaid Salah, called for the constitutional removal of Bouteflika, invoking article 102 based on Bouteflika's medical inability to govern.Gaid Salah reiterated his call on April 2, this time calling for Bouteflika's immediate removal from power.Bouteflika complied 2 hours later, handing in his resignation.
Protests, however, continued for the next year, pushing for a complete change in the system."Yetnahaw ga3 [they all should go!]," they demanded, not just Bouteflika.By fall 2019, protesters understood that the military was the true power within the regime (Ghanem 2019b).They thus began to target it directly, adopting new chants of "civilian, not military state" and "Generals to the dustbin [of history]" (Allahoum 2019).
Despite these (vocal) attacks, the military continued to proceed with caution.The regular army was never deployed to deal with protesters, and the gendarmerie (part of the military) was deployed in September 2019 only to restrict movement from the countryside into the capital.Not once did soldiers-or for that matter, the police-fire upon the protesters.Only one protester was killed over the course of an entire year of mass protests, highlighting the restraint on both sides.
Yet the military's restraint should not be interpreted as a defection to the protesters.On the contrary, the regime continued to ignore the protesters' demands.It pressed forward with its own road map of presidential elections in 2019, a new constitution in 2020, and parliamentary elections in 2021, all without the participation of the Hirak (Volpi 2020).
Although the military exhibited restraint, the police did engage in repression.The police arrested hundreds of protesters over the course of the uprising, though typically out of the limelight, not during protests.Likewise, when protests briefly reemerged in spring 2021 after the initial COVID-19 lockdown, the police repressed the movement, but once again without any bloodshed in the street, nor any participation from the military (Rachidi 2021).

Explaining the Algerian Military's Behavior
The military's cautious and restrained approach during the Hirak stands in stark contrast to its repression of protesters in 1988 and coup and civil war in the 1990s.Its restraint is also puzzling theoretically, since a military that has ruled Algeria for decades, pre-selecting every president (Benchikh 2016, 370) and profiting from vast networks of corruption, is one that the literature tells us has structural incentives to repress.
I argue that the military's restraint in 2019-20 stemmed from two major factors.The first is the protesters' tactics.The protests were entirely peaceful, eschewing the violence that helped justify the military's earlier crackdown on riots in 1988.Moreover, the protesters also explicitly fraternized with the soldiers (Grewal 2019b).From the start, protesters were chanting "the army and the people are brothers, brothers" (jaysh wa sha'ab, khawa, khawa).1 Given that Algeria's army is conscripted, such that the soldiers may well have been the protesters' brothers, these attempts at fraternization likely resonated.Algerian political scientist Dalia Ghanem (2019b) observed that "many junior officers and enlisted men showed their support to the people because they identify with the average citizen.The Algerian army's ranks are filled by ordinary citizens, mostly from the lower and working classes, as conscription for 12 months is mandatory for all men from the age of 19." Algerian sociologist Nacer Djabi accordingly concluded that the regime "can't be certain of the instruments of repression themselves."2In short, the protesters' tactics, in tandem with the military being conscripted, may thus be one explanation for the Algerian military's restraint.
While nonviolence and fraternization came to characterize the protests, it is also important to note that other, less salient, protester tactics might have encouraged military repression.As noted above, the protesters later switched gears to openly criticizing the military, calling for a civilian state and threatening to throw the top generals in the dustbin.Even less known is that at the start of the protests, several opposition figures, including Mustapha Bouchachi, Karim Tabbou, and Zoubida Assoul, had united into a National Coordination for Change and put forward a platform calling for "democratic control over the armed forces."3Given the military's historic role in politics, that call may have been viewed as a threat to its corporate interests, and thus encouraged repression.If protester tactics truly shape the military's calculations, we should see both of these effects occur, even if the former was far more salient and thus came to dominate overall.
A second explanation for the military's restraint is that it has learned from the 1990s not to repress protesters or get involved in politics.Some observers claim that the Algerian military views its coup of 1992 as a mistake, as having brought the country to the "Black Decade" of civil war.The Algerian journalist Adlène Meddi (2018), for instance, argued prior to the protests that "an entire generation of army leaders was profoundly marked by the 'parentheses' of the 1990s, when the army put down the Islamist uprising.Senior officers found themselves involved in politics 'against their will'," quoting Khaled Nezzar, who led the 1992 coup, as saying that "we were drawn into politics in spite of ourselves."Meddi likewise quotes army chief Gaid Salah as saying "I have previously stressed and made clear that ANP is an army that knows its limits and its constitutional missions.Therefore, it can never get involved in the entanglements of parties and politicians."Accordingly, scholars argue that the Algerian military has become increasingly professional and less willing to enter politics or repress protesters (Mortimer 2004).
The least relevant factor in Algeria was the international community's reactions to the protests.Early on, United Nations (UN) Secretary-General António Guterres declared his support for the Hirak, stating on March 31, 2019 that he "welcomes the efforts toward a peaceful and democratic transition in Algeria."4By contrast, the Algerian military's primary arms supplier, Russia, declared its support for the regime.Foreign Minister Sergei Lavrov on March 19 met with the Algerian deputy prime minister and noted that "Russia was concerned by protests in Algeria and saw attempts underway to destabilize the situation."5Other than these statements, however, the international community was largely silent (Grewal 2021).Moreover, Algeria has historically been resistant to international pressure, even from its allies (Benantar 2016). 6As a result, despite Russian support for repression, international factors should be relatively weak in the Algerian military's calculus.
In short, I argue that the military's restraint in 2019-20 stemmed from the protesters' use of nonviolence and fraternization, along with the military's political learning.While some protesters had called for civilian control, and Russia had declared its support for the regime, these factors were less salient and thus outweighed, producing restraint.To test these hypotheses, I turn to a survey of the Algerian military.

SURVEYING THE ALGERIAN MILITARY
Algeria is a challenging environment for research on the military, due to both extensive state repression and animosity toward researchers.For instance, the local partner of the Arab Barometer was placed under house arrest for more than a year for conducting a survey prior to the protests.Even a French MP, who met with protesters in October 2019, was arrested and deported (Latrous 2019).These concerns made an in-person survey, particularly of the military, simply infeasible.
To avoid putting any enumerators in danger, I instead recruited Algerian military personnel directly through advertisements on Facebook.Facebook ads have become increasingly common for recruiting survey samples in both the United States and the developing world (i.e., Cassese et al. 2013;Guiler 2021;Samuels and Zucco 2014).About 45% of Algerians, or 19 million, were active monthly users of Facebook in 2019, and they clicked on average four advertisements per month.7I therefore purchased advertisements that invited Algerians to take "an academic survey about Algerian politics."The advertisements were displayed for almost the entire protest period-from April 1, 2019 to February 21, 2020-recruiting respondents on a rolling basis to take the survey throughout this time.Clicking on an advertisement took users out of Facebook and into Qualtrics, a survey platform, where they filled out a consent form and then answered the survey. 8o reach military personnel, I pursued two approaches.The first was to simply leverage the natural rate of military personnel in the population, and advertise the survey to all 19-million Algerians on Facebook.Over the course of the year, 13,847 of them clicked on this advertisement and completed the survey.Of these, 17.7% (2,453) answered in the demographics section that they have military experience, including 6.8% who said they were active-duty and 10.9% who said they had completed their service.The second approach was to attempt to oversample military personnel in the first place by targeting the advertisement just to the 1.8-million Algerians that Facebook algorithms have classified as being interested in the military (see Appendix A of the Supplementary Material).Of these, 4,897 completed the survey, 31.4% of whom (1,540) said they had military experience, including 13.6% active-duty and 17.8% former personnel.The targeted advertisement thus succeeded in nearly doubling the rate of military personnel, though the general advertisement, by reaching far greater numbers, in fact reached more military personnel.
In total, across the two advertisements, 18,744 Algerians clicked on the ad and completed our survey.Of these, 3,993 reported in the survey that they had military experience.There are several reasons to believe that respondents would not lie about their military experience (see Appendix B of the Supplementary Material), but I administer a number of tests and filters to be sure.First, I ask respondents a military knowledge question: "which rank is higher: sergeant or corporal?"Over 72% of the self-reported military personnel correctly answered sergeant, whereas only 37% of the selfreported civilians did so, a statistically significant difference (p ¼ 0:02).These results suggest that military personnel might have honestly reported their military service.I also remove from the sample those who answered this knowledge question incorrectly.
Second, I filter the data for inconsistencies in answers.I removed the respondents who (1) were too young for the rank they claimed to be, (2) claimed to be conscripted but also an officer, (3) said they were active-duty but unemployed, (4) claimed to have been commissioned as an officer without a college degree, and (5) said they were a general.For a more conservative test, I also remove the 716 respondents who did not provide their rank at all and for whom we cannot apply these filters.Overall, these filters leave 2,235 military personnel.
Of these 2,235, the vast majority were soldiers (680) or non-commissioned officers (NCOs, 1,073), with 412 junior officers and 70 senior officers (Majors, Lt. Colonels, and Colonels).About 33% said they were active-duty, and the rest former military personnel.The survey sample largely mirrors the actual Algerian military in terms of branch, with about 65% in the land army and another 20% in the national gendarmerie (see Table A.2 in Appendix C of the Supplementary Material).
While branch is the only publicly available comparison point, the remaining survey demographics are internally consistent (see Table A.3 in Appendix C of the Supplementary Material).The higher ranks in our sample are older, wealthier, and better educated than the lower ranks.For instance, about 26% of junior officers and 51% of senior officers say they have graduate degrees, compared with just 4% of soldiers and 8% of NCOs.Likewise, about 26% of junior officers and 60% of senior officers have monthly household incomes over one hundred thousand Algerian dinars, compared with just 14% of soldiers and 15% of NCOs.Finally, 77% of senior officers say they have received foreign training, compared with only 36% for junior officers and 20% for soldiers and NCOs.
Although the data are internally consistent, they are unlikely to be representative of the Algerian military.Given the sampling strategy, they by necessity exclude those without internet access, and thus likely skew younger, wealthier, and more educated than their offline counterparts.Yet none of the results vary by age, income, or education (see Table A.4 in Appendix D of the Supplementary Material), suggesting that a more representative sample on these demographics might not have changed the results.Moreover, most of the hypotheses are tested through survey experiments, which generally see consistent results whether conducted on online, convenience samples or representative ones (Coppock, Leeper, and Mullinix 2018).

Attitudes toward Repression
Our outcome of interest is the military's willingness to repress the protests.The survey asked: "Suppose, hypothetically, that military personnel are ordered to repress the protesters.Do you think the military would agree or refuse to repress the protesters?"Respondents answered on a 5-point scale from "very likely to refuse" to "very likely to agree." Notably, this question asked what respondents expected "the military" to do, not how they personally would have responded.That distinction was intentional, and motivated by two concerns.First, a respondent's attitude toward repression is a sensitive topic, either revealing a willingness to disobey orders or to hurt protesters.It is not something that military personnel may want to reveal openly.Asking instead about what "the military" would do removes individual culpability even if their answers are still colored by their personal preferences of what they want the military to do.
Second, scholarship tells us that this is in reality how military personnel respond: based in part on their expectations of how others in the military will respond.Geddes (1999) and Singh (2014), for instance, show that militaries highly prioritize cohesion, thus creating a "coordination game" when put in such situations like repressing protesters or staging a coup.Accordingly, asking how "the military" would respond is likely a more accurate way of capturing the actual likelihood of repression.
Figure 1 (left) presents the results.Overall, expectations of repression were low: 73.5% said the military would refuse to repress, whereas only 10.5% said it would agree (the remaining 16% answered "neutral").Figure 1 (right) shows that expectations slightly increased over time, consistent with our qualitative reading of the protesters' chants as increasingly threatening the military. 9However, expectations remained relatively low throughout the course of uprising, consistent with the military's lack of repression.
It is worth underscoring that the protests were ongoing during this time.The survey question thus captures the military's willingness to repress in the moment, not in hindsight or in some abstract future scenario.These are real, pre-formed expectations of their willingness to repress, and thus likely carry considerable external validity.Moreover, given that these attitudes are likely real and pre-formed, it should be difficult to influence them.As such, evidence that our informational primes (discussed below) shaped these expectations of repression would constitute particularly strong support for the theory.

Independent Variables
To test the five hypotheses, I leverage both observational and experimental variables.First, to test Hypothesis 1 about violence, the survey asked respondents observationally whether they perceived the ongoing Hirak protests as violent, mostly violent, mostly nonviolent, or nonviolent. 10In reality, the protests were exclusively nonviolent, showing remarkable discipline given the millions amassed every Friday.Indeed, 85% of military personnel said the protests were completely nonviolent, with another 14.6% saying mostly nonviolent (see Figure 2).I therefore dichotomize this variable into those who believe the protests were fully nonviolent or had some violence, but results are substantively similar when using the full scale (Table A.6 in Appendix E of the Supplementary Material).I anticipate that those who perceive the protests as completely nonviolent should be less willing to repress, and that 9 There were only five military respondents surveyed in December 2019.However, that was also the month of the regime-imposed presidential elections which might also explain the low expectation of repression. 10I chose to examine this hypothesis observationally, since a prime saying that the protests were violent simply would not be believed.
this should be particularly pronounced for senior officers.
I test Hypotheses 2-4 experimentally.Prior to gauging respondents' expectations of repression, the survey contained a priming experiment attempting to shape these attitudes. 11Respondents were randomly assigned to either a control group or one of four treatment groups, with about 330 in each group.The control group received no text.Each of the treatment groups received a factual, informational prime highlighting either protester tactics or international reactions (Table 1).I anticipated that respondents were already aware of the information provided, but that reminding them would briefly heighten the salience of these factors as they answered the subsequent questions.
The first treatment group, Fraternization, primed respondents to think about how the protesters were praising the military and highlighting their commonalities through their well-known chant "the army and the people are brothers."Indeed, 99% of respondents said they had heard this chant, underscoring its salience.If Hypothesis 2 is correct, fraternization should breed support for restraint, particularly among conscripts.
The second treatment group, Civilian Control, highlighted the opposition's potential threat to the military, noting that the opposition recently united and called for "democratic control over the armed forces."Given the military's historic role in Algerian politics, such a call would represent a direct threat to its corporate interests, and should thus encourage repression if Hypothesis 3 is correct.Only 20% of respondents had heard of this platform, underscoring that these threats were not very salient in reality.
The two remaining treatments focused on international reactions to the protests, one in favor and one against.The Russia condition primed respondents to think of Russian support for regime repression, noting that Russia is a major supplier for the Algerian army and that it had recently expressed concern about the protests.About 57% reported knowing about Russia's position.Finally, the United Nations condition primed respondents to think of the UN's support for the protests, which likewise the majority of respondents (68%) already knew.If Hypothesis 4 is correct, then respondents should become more supportive of repression when primed of Russia's support, but should be unmoved by the UN prime, given that it is not Algeria's arms supplier.
Finally, to examine political learning, the survey asked, observationally, whether respondents oppose the coup of 1992.Overall, about 45% of military respondents said they now oppose or strongly oppose the coup, with only 12% saying they support it (see Figure 3).Results do not appear to be influenced by social desirability bias; a list experiment similarly uncovers 51% opposition to the coup (see Appendix G of the Supplementary Material).I accordingly use The experiment was only included in the survey between April 1 and August 13, 2019, during which 1,659 of the 2,235 respondents completed the survey.To maintain the full sample, I, therefore, create an additional, "post-experiment" bin, to house the remaining respondents.
the direct question as a measure of regret over the 1992 coup.If Hypothesis 5 is correct, opposition to the coup should correlate with less support for repression today, particularly among the older generations that experienced the civil war.
To demonstrate that opposition to the coup represents political learning, I also control for two alternative possibilities.First, it could reflect support for the Islamists who were ousted in the coup.I accordingly control for whether respondents self-describe as Islamist or Salafist (N ¼ 317, or 14%), and whether they believe laws should be in accordance with shari'a (66% agreed or strongly agreed).Second, it could reflect that they were forced to fight in the 1990s, not that they necessarily regret the coup.I therefore control for whether respondents actually fought in the 1990s civil war (N ¼ 435, or 20%).
I also control for a number of other factors that might shape respondents' views of repression.I control for whether respondents received foreign training in the West (3%), in Russia (11%), or in China (1%), to account for arguments that Western training decreases support for repression (Taylor 2014), while Russian training increases loyalty (Casey 2020).I control for conscription (43% of soldiers and NCOs), which may be correlated with restraint regardless of the fraternization prime.I control for their rank, whether they are active-duty (33%), and whether they serve in branches that would be used in repression (i.e., the land army and gendarmerie).I control for whether they themselves have protested, whether they want the Hirak to continue, whether they perceive the government as corrupt and the economy as poor, and their level of support for democracy and for opposition parties.Given spatial variation in the protests (Kilavuz, Grewal, and Kubinec 2023), I also include fixed effects for governorate.Finally, I include their age, gender, education, whether they live in urban areas, whether they are employed or a student, whether they speak Arabic at home (rather than Tamazight or French), and the month they took the survey, since expectations of repression increased over time.All covariates are re-scaled between 0 and 1 to compare effect sizes.

Results
Table 2 presents the results.Model 1 shows the base model, whereas model 2 adds in interactions to examine where effects are concentrated.For ease of interpretation, I dichotomize the dependent variable, but Note: Final column does not include DK/refuse (N = 1, 1, 4, and 3, respectively).

FIGURE 3. Algerian Military's Attitudes toward 1992 Coup
Military Repression and Restraint in Algeria results are similar using the full 5-point scale (Table A.6 in Appendix E of the Supplementary Material).Figure 4 plots the marginal effects of the core hypotheses. 12ypothesis 1 enjoys strong support.Respondents who perceived the protests as nonviolent were significantly (p < 0:001) more likely to say the military would show restraint.Substantively, nonviolence appears to move respondents about 10 percentage points more supportive of restraint.Moreover, an interaction (model 2) suggests that these effects are particularly pronounced for senior officers.Figure 5 shows that the impact of nonviolence was almost four times as large for senior officers, increasing support for restraint by about 35 percentage points. 13In short, nonviolence matters, and particularly for the top ranks.Hypothesis 2 also sees support in the data.While the fraternization prime did not shape attitudes overall, it did for a subset of respondents: conscripts.In line with Hypothesis 2, conscripts became about 10 percentage points more supportive of restraint when primed with fraternization (Figure 6).Notably, this is an experimental finding, allowing us to conclude that fraternization has a causal effect on conscripts.Moreover, it suggests that fraternization through chants alone-and not also through mingling with soldiers or presenting them with roses-can induce military restraint.
At the same time, fraternization appeared to slightly backfire among volunteers, who became about 8 points less supportive of restraint.It is possible that the chant led conscripts-randomly chosen from the population -to realize they are indeed similar to the population.But it may have led volunteers-a self-selected and thus more distinct group-to instead think about how they are in fact different from the population.The    2.
results thus suggest that fraternization may be less effective on volunteer armies than conscripted ones.
This interaction between the fraternization prime and conscription is robust to two alternative dependent variables, as well.Beyond their expectations of the military repressing in general, the survey also asked about who in the military they thought refuse to repress.The first asked for their level of agreement with the statement, "Soldiers would not repress the protesters because they are brothers" (84% agreed or strongly agreed).The latter asked for their agreement with the statement "While officers may wish to preserve the system, soldiers will not fire on their countrymen" (71% agreed or strongly agreed).For both, fraternization led conscripts to become significantly more likely to agree that the soldiers (as opposed to the officers) would not repress (Table A.7 in Appendix E of the Supplementary Material), but had no effect or even a negative effect on volunteers.
These first two hypotheses-nonviolence and fraternization-go a long way in explaining why the Algerian military did not repress protesters between 2019 and 2020.Protesters were remarkably disciplined in their nonviolence, such that the vast majority of military personnel surveyed (84%) believed that the protests were fully nonviolent.Similarly, the protesters' attempts at fraternization, particularly slogans that the army and the people are brothers, were so widespread that 99% of those surveyed had heard of them.With nonviolence shifting the attitudes of the senior ranks and fraternization shifting those of the conscripts, the Algerian military had little choice but to exercise restraint.
Still, other protester tactics made repression slightly more likely, though not enough to tip the scales.The protesters' calls for democratic control over the armed forces, which would curtail the military's political influence, indeed fueled support for repression.In line with Hypothesis 3, the civilian control prime significantly reduced expectations of restraint.Notably, this effect is entirely concentrated among the officers, whose interests are actually threatened, rather than the soldiers, who benefit little from power-sharing (Figure 7).These results suggest that had these calls been more salient in Algeria, they may have tipped the scales toward repression.
Similarly, international reactions also bred support for repression.Reminding the military that Russia supports the regime and has expressed concern over the protests decreased expectations of restraint by about 7 percentage points.Had Russia more actively supported repression, and had Algeria in general been more responsive to foreign pressure, these reactions may likewise have tipped the scales.Conversely, the UN prime had no effect (Figure 8), underscoring how it is the reaction of the arms supplier, and not the international community writ large, that shapes the military's attitudes toward repression.
Finally, there is also some support for the notion that the Algerian military learned from its earlier experience with repression in the 1990s.Military personnel who said they now oppose the 1992 coup were about 5 percentage points less supportive of repressing the Hirak.Given that a plurality of military respondents surveyed, 45%, said they opposed the 1992 coup, and only 12% supported it, this explanation likewise helps to explain the military's overall lack of repression.Moreover, in line with Hypothesis 5, this effect emerges only for military personnel old enough to experience the civil war firsthand (see Figure 9).For the 18-24 year-olds who did not, their attitudes toward the coup were not salient enough to shape their calculations visà-vis today's protests.
In sum, the survey reveals support for each of the hypotheses.Protester tactics, international reactions, and political learning all shaped the military's decisionmaking calculus in deciding whether to repress the Hirak.The protesters' nonviolence and fraternization, combined with the military's political learning, pushed the military toward restraint.Meanwhile, the  protesters' calls for civilian control and Russia's support for the regime pushed the military toward repression.At least during the 2019-20 period, the former effects were more salient and outweighed the latter, contributing overall to the military's restraint.
Beyond these core hypotheses, two control variables are also worth First, conscription appeared to have no effect on its own without the interaction with fraternization.This null effect is somewhat surprising, given that conscription's effect on defection is a seemingly well-established finding in the literature both globally and particularly in Algeria (i.e., Barany 2016;Cebul and Grewal 2022;Lutscher 2016).However, this null effect may be specific to our survey, and a product of the socioeconomic biases in the survey sample.The sample excludes the poorest conscripts who do not have Internet access, who are likely the least satisfied with the status quo and most frustrated with inequality, and thus likely to desert even without fraternization (Lyall 2020).
A final unexpected result was Western training.While there are not many Algerian military personnel in the sample with Western training (N ¼ 72), these individuals appear to be significantly (p < 0:01) more supportive of repression.This does not appear to be driven by collinearity; the positive correlation remains even when dropping all other variables.While additional research is needed, these results run counter to assumptions that Western training socializes democratic norms (Atkinson 2014;Ruby and Gibler 2010), corroborating instead recent literature that socialization might be mixed (Grewal 2022;Joyce 2022).Russian and Chinese training, meanwhile, seemingly had no effect on these attitudes.

CONCLUSION
The Algerian military's relatively peaceful response to the 2019-20 Hirak protests came as a surprise, as structurally it was a politicized military profiting from the regime, and had repressed on multiple occasions in the past.To explain the military's restraint, this article privileges the agency of individuals to overcome these structures.Protesters were remarkably disciplined and nonviolent, eschewing the justification senior officers used in previous cases of repression.Likewise, protesters intentionally fraternized with the conscript soldiers, highlighting their commonalities.For its part, the military also appears to have learned lessons from the 1990s, when its coup and repression of Islamists sparked a bloody civil war.Wary of doing so again, the military tread cautiously, not once deploying troops to repress the Hirak.
On the other hand, the results of our priming experiment also highlight the power of these agency explanations to shape the military's calculus in the opposite direction.When protesters threaten a powerful military like Algeria's with civilian control, they risk inviting military repression.Likewise, when Russia reiterates its support for the regime, it appears to have significant effects emboldening the military to repress.However, at least during the 2019-20 period, the former effects appeared to outweigh these latter ones, contributing overall to the military's restraint.
The results in the Algerian case suggest that these three factors-protester tactics, international reactions, and political learning-may help tilt the scales toward either repression or restraint in these theoretically indeterminate cases.While ethnically stacked militaries are likely to repress, and counterbalanced ones likely to defect, militaries coup-proofed through power-sharing are more variable, with their calculations open to be shaped by these ancillary factors.Where the structures are indeterminate, the agency of militaries, protesters, and the international community shines through.
A quick look beyond Algeria suggests these findings may generalize to other cases of power-sharing.In Egypt, nonviolence, fraternization, and international reactions shaped the military's restraint in 2011 (Bellin 2012;Nepstad 2013), while political learning and threats to the military's interests shaped its repression in 2013 (Bou Nassif 2017; Grewal 2023a).In China, in 1989, the protesters' initial nonviolence and fraternization produced restraint, but their escalation to violence led to the Tiananmen Square massacre (Nepstad 2011).In Myanmar post-2021 coup, fraternization has secured some defections, but threats to the military's interests have led the majority of the force to repress (Barany 2021;Dunant 2022).And, in Sudan, threats to the military's interests and international support from the Gulf similarly facilitated the 2019 massacre (Gallopin 2020;Hassan and Kodouda 2019).
Moreover, these factors seem less relevant in countries where the coup-proofing strategy pre-determines the military's behavior.In Bahrain, a Sunni-stacked military repressed the largely Shia protesters, despite their strong adherence to nonviolence.In Syria, although fraternization led some conscripts to defect, the core of the Alawi-stacked military remained loyal (Droz-Vincent 2016;Koehler, Ohl, and Albrecht 2016).In Tunisia, the counterbalanced military refused to repress despite protesters' use of violence against the police and ruling party.Where the structures point  2.
clearly to repression or defection, these ancillary factors are unlikely to tip the scales.
Beyond these substantive contributions, this article also hopes to advance the field methodologically, by offering a creative new method of surveying military personnel on a large scale.In this article, this approach permitted a survey two thousand military personnel during a mass uprising, allowing for the systematic examination of multiple hypotheses of military repression.Beyond repression, this approach to surveying military personnel could be useful to the literatures on civil-military relations, coups, and wars.Even more generally, the method outlined herein, of using targeted Facebook advertisements, could be adopted in any arena in which scholars need to generate a specific rather than nationally representative survey sample.

FIGURE 4 .
FIGURE 4. Marginal Effects on Military Restraint

FIGURE 5 .
FIGURE 5. Nonviolence Matters More for Senior Officers

FIGURE 6 .
FIGURE 6. Fraternization Works on Conscripts, but not Volunteers

FIGURE 7 .
FIGURE 7. Calling for Civilian Control Threatens Officers, but Not Soldiers

FIGURE 8 .
FIGURE 8.Only the Reaction of the Arms Supplier Matters

FIGURE 9 .
FIGURE 9. Political Learning among the Older Generations

TABLE 1 .
Priming Experiment (N ¼ 1, 659) United Nations "At the Arab League summit on March 31, United Nations Secretary-General Antonio Guterres said that he 'welcomes the efforts toward a peaceful and democratic transition in Algeria.' Did you know about this UN statement?"