THE LOGIC OF VULNERABILITY . AND CIVILIAN VICTIMIZATION Shifting Front Lines in Italy (1943–1945)

By STEFANO COSTALLI, FRANCESCO NICCOLÒ MORO, and ANDREA RUGGERI

https://www.cambridge.org/core/terms abstract What causes civilian victimization in conventional civil wars and in conventional wars that experience insurgencies? The authors argue that a key driver of civilian victimization is the vulnerability of the incumbent forces, specifically when the conflict’s front line is shifting. Vulnerability is a function of informational and logistical challenges: when the front line is moving, incumbents face increased informational uncertainty and unstable supply chains that augment their vulnerability. Thus, incumbents will increase the use of civilian victimization in response to a scarcity of high-quality information on the loca- tion and identity of insurgents, to limit possible information leaks, and to contain supply disruption and logistics support to adversaries. The authors support their argument using matched difference-in-differences analyses of original subnational data on Nazi-Fascist violence in World War II Italy (1943–1945) and qualitative evidence.

Introduction N early October 1943, German forces in southern Italy were rapidly , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Iwithdrawing toward the Gustav Line, a defensive position north of Naples. On the night of October 13th, near the town of Caiazzo, a small group of German soldiers noticed flares being set off from a country house close to their tactical command. Shortly after, a young German lieutenant ordered the troops to shoot twenty-two civilians, including

26 Sep 2021 at 00:01:48 at 2021 Sep 26 women and children, on the conjecture that they attempted to pro-

, on on , vide information about the location of the German command center to advancing English troops.1 This episode of civilian victimization is far from unique in World War II Italy. According to the most accurate historical research, Nazi-Fascist forces killed 9,977 civilians throughout 2

University of Athens of University Italy. Some of the most brutal massacres, such as those at Sant’Anna . .

1 Klinkhammer 1993, 43. 2 Data are from the website, Atlante delle Stragi Nazis e Fasciste in Italia, as of April 2016. Avail- able at www.straginazifasciste.it, accessed July 24, 2020. Nazi forces were mostly composed of Wehr- macht and Schutzstaffel (SS), which conducted the bulk of military operations and civilian victimiza- tion. Fascist forces comprised a more heterogenous set of groups, including official forces of the Re-

World Politics 72, no. 4 (October 2020) 679–718 Copyright © 2020 Trustees of Princeton University

doi: 10.1017/S004388712000012X

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 680 world politics di Stazzema and Marzabotto, were committed by German troops as they withdrew toward northern Italy after breaking the Gustav Line . and before the new Gothic Line was established across the Apennines between Florence and Bologna, starting in the late summer of 1944. In Poland, on the eastern front line of World War II, “the Germans would kill civilians after taking new territories. They would also kill civilians after losing ground. If they took casualties at all, they would blame whoever was at hand: men in the first instance, but also women, and children.”3 The Korean War, the second deadliest conflict since 1945,4 also featured high levels of civilian victimization in moments of

pressure on or movement of the front line: “As usual in most wars, when https://www.cambridge.org/core/terms the atmosphere at the front was relaxed, communist prisoners were per- fectly properly used and sent to the camps in the rear. But at periods of special stress or fear, especially in the first six months of the war, many UN soldiers shot down enemy prisoners—or even Korean civilians— with barely a moment’s scruple.”5 In the recent war against isis in Iraq, the dynamics of civilian victimization can also be linked to the retreat of the front line. During the Battle of Mosul, from October 2016 to July 2017, isis killed many civilians while they were retreating from the front line under pressure by Iraqi and US forces. According to some accounts, spikes in civilian victimization were linked to concerns that escaping ci- vilians could leak information to anti-isis alliance forces, facilitating the alliance’s advancement.6

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , What explains the spatial and temporal variation in civilian victim- ization by conventional armies? Violence against civilians in civil wars has been explained by territorial control,7 knowledge of previous polit- ical identities,8 and internal organization of forces,9 and also as a func- tion of resource scarcity10 or desperation.11 The use of indiscriminate

26 Sep 2021 at 00:01:48 at 2021 Sep 26 violence against civilians has also been an important issue in studies on

, on on , counterinsurgency.12 But limited attention has been given to the inter- action between macrotrends in hostilities, such as conventional armies’

pubblica Sociale Italiana (the puppet regime installed by the Nazis in September 1943) in the North, and local militias. Fascist forces were largely operating as agents of the Nazis. In this article, we use the

term Nazi-Fascist to describe the agency of both. University of Athens of University . . 3 Snyder 2011, 121. 4 Lacina and Gleditsch 2005, 154. 5 Hastings 2010, 329. 6 Kesling 2016; Prickett 2017. 7 Kalyvas 2006. 8 Herreros and Criado 2009; Balcells 2017. 9 Humphreys and Weinstein 2006; Mitchell 2004. 10 Wood 2014; Zhukov 2017. 11 Downes 2008.

12 Kocher, Pepinsky, and Kalyvas 2011; Lyall and Wilson 2009.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 681 frontline movements, and local patterns of civilian victimization. Fur- thermore, most theories of civilian victimization tend to be static. The . examples outlined above suggest that shifting front lines can influence the propensity of conventional armies to kill civilians when they are also facing a local insurgency. This article demonstrates how an incumbent force13 fighting a con- ventional war varies the use of violence against civilians as a tool of counterinsurgency according to the movement or stability of front lines. We argue that incumbent forces will respond to increased vulnerabil- ity due to changes in the overall pattern of conventional war by inten-

sifying violence against civilians. When armed forces are on the move, https://www.cambridge.org/core/terms the front line is under pressure. Moving forces face significant chal- lenges in protecting their communication and supply lines, thus in- creasing the incentive to deter and retaliate against a civilian population that may provide vital information to their enemies. Conversely, when stable conventional front lines are (re)established, the vulnerability of the incumbent troops to irregular warfare decreases, and we expect an incumbent to reduce its use of civilian victimization. This article in- troduces the logic of vulnerability, which is based on changes and disrup- tions to the flow of information and logistical supply at the front line, to explain the variation in civilian victimization by conventional forces. The scope of our argument applies to conventional wars that experi- ence the presence of an insurgency operating against at least one of the

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , contending parties. The examination of conflicts in different time periods and regions suggest that our argument is relevant for all conventional forces that, while fighting a conventional army, advance or retreat through areas inhabited by aggrieved populations. Because the logic of vulnerabil-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 ity is particularly salient for conventional forces that rely on long sup-

, on on , ply lines, such as armies operating abroad, the incumbents engaging in civilian victimization can be the regular forces of an invading state. Examples include several World War II theaters, such as the German invasion of the Soviet Union (Operation Barbarossa) and its invasion of

University of Athens of University the Balkans, during which German troops faced Yugoslav partisan re- . . sistance that also assumed a conventional nature.14 Furthermore, these patterns can be extended to other wars, such as the Korean War (1950– 1953),15 the Iran-Iraq War (1980–1988), and the more recent Eritrean-

13 For definition of incumbent, see Kalyvas 2006. 14 Shepherd 2012.

15 Hastings 2010.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 682 world politics Ethiopian War (1998–2000).16 Our theory also applies to conventional civil wars, which are marked by direct military confrontation, well- . defined front lines, and armed columns.17 In these conflicts, “there is a clear distinction between offensive and defensive actions.”18 Exam- ples include the American Civil War (1861–1865), the Spanish Civil War (1936–1939), and the war in Bosnia-Herzegovina (1992–1995). Noticeably, “high levels of external support or external intervention in favor of the rebel side may turn an irregular war into a conventional one,”19 leading to an overlap of conventional and internationalized civil war. Such was the case, for instance, during the late phase of the Viet-

nam War. https://www.cambridge.org/core/terms We test our hypotheses by studying the violence perpetrated by Ger- man and Fascist forces in Italy during World War II. Italy is a particu- larly appropriate case to evaluate our theory because the conventional warfare between German and Allied forces clearly alternated between moments of extreme mobility of the front line and periods of prolonged stability. Moreover, the Italian case provides exceptionally fine-grained data on timing, geography, and perpetrators and victims of violence, which we use to construct two original georeferenced data sets at dif- ferent levels of analysis. To begin, we test the links between conven- tional operations, incumbents’ vulnerability, and civilian victimization by performing matched difference-in-differences (dd) analyses with data from throughout Italy. Then, we replicate the analysis at a more

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , granular level for the Emilia-Romagna region. In this key area, we are able to complete a more in-depth analysis on the effects of frontline movement and evaluate the possible influence of specific, local-level factors. Last, we profit from detailed historical studies that allow us to integrate our quantitative findings by using historical material to qual-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 itatively identify our mechanisms and to evaluate possible alternative

, on on , logics of civilian victimization. Although popular historical accounts have interpreted Nazi-Fas- cist violence as a result of racism or political indoctrination,20 our find- ings show that Nazi-Fascist forces significantly changed their behavior

University of Athens of University toward the Italian civilian population in different phases of the con- . . ventional war, depending on their vulnerability. Beginning in the late spring of 1944, while withdrawing after the fall of the Gustav Line in southern Italy and before they consolidated their position on the 16 Abbink 2003. 17 Kalyvas and Balcells 2010. 18 Kalyvas and Balcells 2010, 419. 19 Kalyvas 2006, 92.

20 Bartov 2001; Wette 2007.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 683 Gothic Line in northern Italy, German troops notably increased their use of civilian victimization. German forces also changed their behavior . again—in the opposite direction—as they strengthened their position on the northern Gothic Line in the fall of that year, civilian victimiza- tion was reduced. This reduction was greatest in the areas closest to the front, with less reduction as distance from the front line increased. This article contributes to the literature on civilian victimization in war in three ways. First, it explains how the logic of vulnerability can drive the victimization of civilians by conventional forces, and specifies its constitutive elements: information and logistical challenges. While

most explanations of civilian victimization tend to be static, the logic https://www.cambridge.org/core/terms of vulnerability provides a dynamic theory that does not assume a de- terministic linearity in which violence is anticipated to increase as in- cumbents retreat and their defeat approaches. The logic of vulnerability predicts not only punctuated equilibria of violence, but also increases and decreases in civilian victimization, depending on the level of the forces’ vulnerability produced by movements of the front line. Second, the article focuses on the underappreciated links between the macro- dynamics of regular warfare and civilian victimization. Given the in- creasing number of internationalized civil wars in the global security environment in which insurgency is often combined with conventional warfare, it is important to shed light on these links.21 Third, this article is one of the first studies to use a mixed-methods approach to demon-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , strate the nexus between vulnerability and civilian victimization using high-quality disaggregated data from a historical case of major conflict.

Incumbents’ Logics of Civilian Victimization

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Violence against unarmed civilians, though reprehensible, can follow a

, on on , logic.22 In this section, we present the different logics of civilian victim- ization found in the extant literature. First, scholars attempt to explain civilian victimization as a tool used to directly diminish insurgents’ effectiveness (tactical logic). There is a

University of Athens of University lively debate on the tactical effectiveness of indiscriminate violence tar- . . geting the population. On the one hand, practicing widespread civilian victimization against those not engaged with insurgent forces can be counterproductive.23 On the other hand, some find that under specific conditions, indiscriminate violence can work as an effective counter-

21 Pettersson, Högbladh, and Öberg 2019. 22 Kalyvas 2006.

23 Kalyvas 2006.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 684 world politics insurgency tool. 24 Acts of barbarism, defined as the “systematic viola- tions of the laws of war in pursuit of a military or political objective,”25 . including targeting of the civilian population and other mass atroci- ties, when used by the strongest actors in asymmetric wars can lead to favorable outcomes. Furthermore, indiscriminate attacks against civil- ians can decrease insurgents’ activities in the short term, but in the long term, insurgents can recover and change their areas of activity.26 Second, other scholars argue that violence will occur in areas where information about political allegiances is available to parties in con- flict (logic of allegiance).27 In conventional ideological civil wars, vio-

lence against civilians takes place mainly in the rear guard and targets https://www.cambridge.org/core/terms those who supported the adverse political faction before the onset of the war.28 Laia Balcells states that in conventional civil wars, armed groups will resort to violence in their rear guard “if there has been mobilization from the enemy group in this territory.”29 Similarly, in conflicts in which groups are mobilized along ethnic lines, warring actors will engage in civilian victimization in areas where ethnic minorities are present,30 as well as in areas populated by the enemy’s ethnic group, to disrupt net- works of civilian support.31 Third, research finds that that exasperated armies resort to violence against civilians as a final attempt to win a conflict or, conversely, to minimize their losses (logic of desperation). For example, Alexander Downes argues that in long wars, actors are more likely to resort to ci-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , vilian victimization if they believe doing so will serve to coerce their adversary to give up or will minimize their own losses.32 Long deploy- ments can also have the effect of inducing combatants to increase vi- olence against civilians as a result of frustration or erosion of social and moral norms, especially when structures of command are dysfunctional.33

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Fourth, works also find that racism, rather than strategic advantage,

, on on , can lead to civilian victimization (racist logic). Historical research on Nazi military operations and occupation in eastern and southern Eu- rope finds that racism was deeply embedded in the education and train-

ing of Wehrmacht soldiers, as well as of the more politicized units of

University of Athens of University . .

24 Downes 2008; Lyall 2009. 25 Arreguín-Toft 2001, 101. 26 Souleimanov and Siroky 2016. 27 Belge 2016; Steele 2011; Steele 2017. 28 Balcells 2012. 29 Balcells 2017, 28. 30 Di Salvatore 2016. 31 Fjelde and Hultman 2014. 32 Downes 2008.

https://www.cambridge.org/core 33 Manekin 2013.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 685

the Schutzstaffel (ss).34 The apparent complete disrespect among Ger- man forces for the lives of civilians as demonstrated throughout Russia . and the Balkans is often linked to the indoctrination of German sol- diers to think of the enemies as Untermenschen (inferior persons). In Italy, according to Wolfram Wette, “the Germans now [1943] began to regard Italians as inferior too, and treated them in a manner hardly less degrading and inhumane than their treatment of Jews and Russian prisoners of war.”35 Fifth, studies find that organizational factors can also explain ci- vilian victimization (organizational logic). Armed groups that rely on

short-term gains to recruit personnel and that are not endowed with https://www.cambridge.org/core/terms internal enforcement structures appear to be more likely to kill civil- ians.36 Specific organizational cultures, peer pressure, and particularly brutal commanders seem to explain why some units engaged in civil- ian massacres whereas others did not. Historians note that particular German units were responsible for multiple and particularly brutal in- stances of civilian victimization. For example, the Sixteenth ss Panzer- grenadier Division Reichsführer-ss, which operated in Italy from 1943, was responsible for several massacres, including those at Sant’Anna di Stazzema (August 1944) and Marzabotto (September–October 1944), which killed more than one thousand civilians.37 Similarly, particular German units, such as the Reserve Police Battalion 101 in Jozefow and Lomazy, were responsible for the large-scale killing of Jews in World

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , War II Poland.38 Sixth, scholars also find that civilian victimization can be the out- come of spiraling retaliation between the groups involved (logic of re- taliation). In such cases, the action of one group leads to a reaction from the opposition, leading to a circular pattern of violence independent

26 Sep 2021 at 00:01:48 at 2021 Sep 26 from the prime motives of the conflict. Some scholarship on state re-

, on on , pression stresses the link between states’ action and their perception of the threat. For example, Christian Davenport argues, “authorities gen- erally employ some form of repressive action to counter or eliminate the behavioral threat.”39 Revenge against an enemy’s actions can be a pow-

University of Athens of University erful motive for engaging in violence and, under certain conditions, in . . civilian victimization.40 Two final logics are more directly connected with our logic of vul-

34 Bartov 2001. 35 Wette 2007, 137–38. 36 Humphreys and Weinstein 2008; Arjona 2016; Green 2018. 37 Gentile 2015. 38 Browning 1998. 39 Davenport 2007, 7.

https://www.cambridge.org/core 40 Balcells 2017.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 686 world politics nerability: the logics of control and of resources. According to Stathis Kalyvas’s theory of control, incumbents would use selective violence in . areas where they enjoy hegemonic but incomplete control and can rely on the collaboration of the population for gathering high-quality in- formation about insurgents. Separately, incumbents should cause high levels of civilian victimization through indiscriminate violence in ar- eas where insurgents enjoy full control.41 Incumbents may also alter their strategic use of civilian victimization when their resources are un- der threat. Yuri Zhukov argues that insurgent attacks on infrastructure compel incumbents to use their forces for defensive duties, limiting

their capacity to carry out reprisals against civilians.42 As war continues, https://www.cambridge.org/core/terms conditions on the ground change. Reed Wood argues that battlefield losses exacerbate resource scarcity and lead insurgents to increase ci- vilian victimization.43 Similarly, Lisa Hultman shows that rebel groups facing high battle losses engage in civilian victimization to maximize their impact with fewer resources.44 We thus know that incumbents re- spond to changing balances of forces vis-à-vis insurgents, but a key question emerges: How do the dynamics of conventional warfare influ- ence the strategies of incumbents against insurgency?

The Logic of Vulnerability and Civilian Victimization We argue that incumbents’ vulnerability affects civilian victimization.

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Vulnerability, in turn, varies as front lines shift. When incumbents’ forces are on the move, they will resort to higher levels of civilian vic- timization. Conversely, when the front line is static, incumbents commit lower levels of civilian victimization. Opposing forces in a conventional civil war or armies engaging in conventional war that simultaneously

26 Sep 2021 at 00:01:48 at 2021 Sep 26 fight a counterinsurgency campaign face difficult and competing stra-

, on on , tegic choices. Incumbents need to ensure that their forces are protected and adequately supplied to work effectively and to survive. Two key factors consistently affect the vulnerability of an incumbent’s military forces: information availability and logistical constraints.

University of Athens of University As a battlefront shifts, the availability of reliable information de- . . creases and logistical constraints intensify, therefore increasing an incumbent’s vulnerability and leading to an increase in civilian victim- ization. When on the move, incumbent forces operating in areas marked

41 Kalyvas 2006. 42 Zhukov 2017. 43 Wood 2014.

44 Hultman 2007.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 687 by insurgency face two key informational problems. First, incumbents have difficulties acquiring information about the location and identity . of their enemies. Second, especially if incumbent forces are retreating, there is a high risk that the civilian population will defect to the insur- gents45 or, more important, to the incumbent’s conventional enemies. Civilians living in conflict areas may be able to provide information to the enemy’s troops about the incumbent forces’ position and move- ments. Effective counterinsurgency using selective violence necessitates high-quality intelligence. Such information requires a functioning net- work of reliable informants that can be developed only over time in a

stable environment. When incumbent forces move, these networks are https://www.cambridge.org/core/terms not available and, as the situation on the ground evolves, the popula- tion’s preferences change over time. As such, armies have less time to acquire information and, once acquired, to use it appropriately. Thus, incumbent forces may have only a vague idea of who and where the in- surgents are located and civilian victimization may be the best available strategy to kill insurgents. In the German advance during the Russian campaign in 1941, a German officer in Ukraine noticed that villagers were seen moving from the territory acquired by Germans to provide intelligence to the opposing forces. “On reaching the Soviet positions,” the officer recalled, “they had provided detailed information to the Rus- sian troops on the east bank, outlining our positions and strength.”46 At the same time, Sonderkommandos—special groups of deportees who

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , collaborated with Nazi troops—were conducting extensive “pacifica- tion” through “countless atrocities” toward the civilian population.47 Facing diminished security together with limited information on the ground increases the incentives for incumbents to use violence against civilians and insurgents. The incumbent’s inability to identify insur-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 gents makes the civilian population its most likely target. In short,

, on on , when the front line is moving, incumbents have greater difficulty gath- ering information and face a higher risk of information leaks. Hence, they employ civilian victimization as a tactic to both eliminate the in- surgents and to limit information loss. This process is evident among

University of Athens of University the US forces during the Vietnam War. As US troops moved into vil- . . lages, they found it difficult to distinguish between insurgents and ci- vilians and resorted to civilian victimization to avoid further attacks.48 Although information availability also plays a key role in the logic of

45 Kalyvas 2006. 46 Bidermann 2000, 42. 47 Bidermann 2000, 43.

48 Greiner 2010.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 688 world politics vulnerability, our theory diverges from Kalyvas’s both in its scope and in its predictions. In scope, Kalyvas’s theory aims to explain static situ- . ations, whereas we argue that the dynamics of civilian victimization is fundamentally different when front lines are in flux. Substantively, ac- cording to Kalyvas, incumbents’ indiscriminate violence would typically occur in areas of full insurgent control and takes the form of raids and bombings,49 while “the most contested areas are predicted to be oases of peace in the midst of violence.”50 Our theory posits that in phases of frontline movement, these heavily contested areas are unlikely to be oa- ses of peace for civilians; rather, most civilian victimization will occur in

areas close to the front line, which is under the prevailing, though con- https://www.cambridge.org/core/terms tested, control of the incumbents. Organizing deep raids far from the incumbents’ bulk of the forces, when communication lines are insecure and a major military effort is occurring, would be too risky. The second element of incumbents’ vulnerability relates to logistics. Mobility limits the possibility of incumbents creating functioning war economies and providing logistical support to guarantee the provision of army resources. Previous research finds that rebels resort to civil- ian victimization more frequently when they are facing severe losses that constrain their access to resources.51 For incumbents and conven- tional forces, the problem is possibly even more pressing. Our argument here differs from analyses of civilian victimization perpetrated by reb- els because it applies to a different phenomenon (conventional armies)

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , that requires developing different, though partially overlapping, mech- anisms and empirical indicators of how resources affect incumbents’ strategies. For conventional armies, the establishment of stable com- munication networks with fortified positions guarantees the possibil- ity of a well-functioning military supply chain. As highlighted by T. E.

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Lawrence, insurgents can prosper by attacking poorly defended com-

, on on , munication networks, thereby creating major obstacles to the normal functioning of regular armed forces.52 During the Korean War, shifting front lines and disruptions of logistical supply were tactical obsessions for military officials: “As the supply system cracked, men grew desper-

University of Athens of University ate in their hunger. Morton saw two soldiers discover an abandoned, . . half-empty can of peas coated in days of dust. They simply scraped off the dust with a bayonet and wolfed the remains. The young lieutenant found this spectacle, of thousands of men on the margins of panic, very

49 Kalyvas 2006, 223. 50 Kalyvas 2006, 204. 51 Wood 2014; Hultman 2007. 52

https://www.cambridge.org/core Lawrence 1920.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 689 frightening.”53 The challenge of logistics is still relevant. During Oper- ation Iraqi Freedom (2003–2011), US planners considered guarantee- . ing the security of the supply chain as “combat operations.”54 As front lines move, supply chains and logistics are stretched. This extension increases an incumbent’s vulnerability by making the incum- bent’s supply chains more vulnerable to attacks and by constraining ready access to resources. Without an effective supply chain, the incum- bent’s troops will be more likely to live off the land and to engage with the civilian economy through forced seizures of fuel, food, raw materi- als, and shelter. Mobility intensifies the emergency, exacerbates scarcity,

and creates further incentives to punish noncompliant civilians who https://www.cambridge.org/core/terms compete for resources. To borrow from Mancur Olson, when moving, incumbents will act as “roving bandits,” maximizing short-term gains. Incumbents facing high uncertainty will shorten and refocus their time horizon on decision-making and opt for short-term benefits. They will not privilege medium- and long-term time horizons in which incen- tives for maximizing resource extraction capacity prevail and make in- cumbents “stationary bandits.”55 A final logistical problem is that armies on the move create displaced, internal refugees in the areas through which they pass. Managing the flow of civilians represents a high cost for conventional armies for sev- eral reasons. First, incumbents allocate scarce resources to deal with displaced civilians, such as building and managing camps. Second, dis-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , placed civilians represent potential defectors and insurgents. In both cases, the incentive for conventional forces is to engage in less costly and more rapid solutions, and large-scale civilian victimization is a po- tentially effective tool. In sum, we posit that conventional forces’ vulnerability, with particu-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 lar reference to their supply chain, creates incentives for them to engage

, on on , in civilian victimization. Incumbent forces need not experience harsh losses, as rebels do, according to Wood and Hultman; they might well be prevailing, yet vulnerability makes them perceive a potential risk to

their own safety.56 University of Athens of University

. . —H1. When front lines move, incumbents’ armed forces will increase civilian victimization in the areas under their control near the front line. When incumbent forces are not on the move, that is, when a static front line is established, vulnerability is much lower because informa-

53 Hastings 2010, 195. 54 Wright and Reese 2008, 506. 55 Olson 1993. 56

https://www.cambridge.org/core Wood 2014; Hultman 2007.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 690 world politics tion availability and logistical safety work in the opposite direction re- garding the use of violent means. The reestablishment of front lines . provides four opportunities for incumbents. First, they can recreate for- tified positions that reduce the possibility of insurgents launching effec- tive attacks.57 Second, in these less vulnerable periods, incumbents can reestablish defensible supply chains between home bases and forward- deployed units. Third, thanks to this resource stability, incumbents reduce their need to interact with civilians, diminishing the information avail- able to the population itself and consequently, its strategic value to the enemy. As incumbents enjoy less vulnerability, even if they do not pos-

sess sufficient information to clearly identify insurgents, they reduce the https://www.cambridge.org/core/terms need to kill civilians as a way to retaliate against or to deter insurgents. Last, as new front lines are established, incumbents can devote most of their resources to fighting the conventional enemy, who now rep- resents the major threat to their immediate survival. In these phases, counterinsurgency is more effectively pursued through the develop- ment of a functioning network of informants and, therefore, the capac- ity to apply selective violence against insurgents. Moreover, incentives exist to strike informal deals with insurgent forces, which have been recognized as an important, if often overlooked, local outcome in in- trastate conflict.58 In these phases of conventional war, we should expect incumbents to modify their strategies, following incentives to exert less violence against civilians than in the phases preceding the instauration

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , of the new front lines. —H2. When new front lines are established, the levels of incumbent civilian victimization should decrease in areas near the new front line. We do not claim that the logic of vulnerability necessarily excludes

26 Sep 2021 at 00:01:48 at 2021 Sep 26 all other logics presented above. Our research design allows us to con-

, on on , trol for some alternate logics and to gauge if these logics operate in par- allel with our proposed logic.

The Italian Case University of Athens of University . . Several specific features make Italy from 1943 through 1945 an in- teresting and suitable case to study how the macrodynamics of a con- ventional war affect the microlevel dynamics of counterinsurgency and civilian victimization. The conflict between the Italian resistance move- ment and the Nazi-Fascist forces was fought in the broader context

57 Collard-Wexler 2013.

58 Staniland 2012.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 691 of World War II. Within Italy, conventional military operations and asymmetric civil war occurred simultaneously for almost two years. At . the macrolevel, German forces attempted to prevent the northern ad- vance of Allied forces following the latter’s invasion of southern Italy. At the local level, Nazi and Italian Fascist forces also carried out coun- terinsurgency operations against the Italian resistance movement. The conventional international conflict within Italy alternated between mo- ments of extreme mobility of the front lines and periods of prolonged stability. The fluctuation between movement and stability allows us to demonstrate how these different phases had diverging effects on the

incumbents’ use of civilian victimization. Including military personnel https://www.cambridge.org/core/terms and civilians, the international conflict and insurgency within Italy re- sulted in more than two hundred thousand Italian casualties.59 The civil war in Italy began on September 8, 1943—the day that King Vittorio Emanuele III signed an armistice with the Allied Forces withdrawing Italy from World War II—and ended in April 1945, with the complete liberation of Italy. In the immediate aftermath of the ar- mistice, Benito Mussolini was made formal leader of the second and last of his Fascist governments, the Italian Social Republic, a Nazi pup- pet state that included the territories of central and northern Italy. With the signing of the armistice, German forces invaded Italy from the north to prevent the advancement of Allied forces from the south. With no formal orders, parts of the Italian army attempted to resist

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , the German invasion. In September 1943 alone, the resisting forces endured 18,965 casualties.60 In southern Italy, a new government was formed, supported by a coalition of political parties that reemerged af- ter twenty years of Fascist dictatorship. Italy was divided into two parts, and an armed resistance movement began to form in the regions occu-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 pied by Nazi forces.

, on on , Armed resistance formed to fight the foreign occupiers and the new Fascist state. In turn, Nazi forces and Italian Fascist militias, often op- erating under the command of the Germans, carried out counterinsur- gency. In the first months, the resistance movement was located mostly

University of Athens of University in northern and central Italy. Initially, it was fragmented, formed by . . small bands of fighters that had only limited contact with other bands and the local branches of the reemerging parties. The partisans had scarce resources, with light weapons provided by former soldiers and officers of the Italian army in the days of the disbandment and arms

59 ISTAT 1957.

60 Torsiello 1975, 643.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 692 world politics that had been seized from Fascist forces.61 But the resistance grew dur- ing the first months of 1944, and by spring 1945 it has become a sig- . nificant force. The Italian resistance benefited from the advancement of the Allied forces from the south, whose landing in Sicily in July 1943 was followed in September by a second series of landings on mainland Italy. These forces played a fundamental role in the liberation of the country and in the development of the civil war between the partisan bands and the Nazi-Fascist forces. The Allied forces needed to break through two ma- jor German defensive lines to reach northern Italy. The first, the Gustav

Line, below Rome, blocked Allied troops between the end of 1943 and https://www.cambridge.org/core/terms the spring of 1944. The second, the Gothic Line, ran along the Apen- nines north of Florence and held back Allied advancement from win- ter 1944 until spring 1945.

Research Design To test our hypotheses of the logic of vulnerability, we triangulate three different empirical analyses. First, we perform a statistical analysis us- ing spatially and temporally disaggregated data relative to all the Ital- ian territory involved in frontline movements. Second, we increase the spatial and temporal disaggregation of the data for a second statistical analysis of the Emilia-Romagna region. Third, we present qualitative

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , historical evidence to provide a richer understanding of the local dy- namics. In our research design, the movement and stability of front lines di- vide the temporal dimension into three distinct phases (Figure 1) and the geographical space into three geographic areas (Figure 2). Two cru-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 cial events divide the temporal dimension: the collapse of the Gus-

, on on , tav Line in May 1944 and the establishment of the Gothic Line in the Emilia-Romagna region in December 1944. The collapse of the Gus-

tav Line divides time into two phases: phase one (before the collapse, when the front line is stable) and phase two (after the collapse, when

University of Athens of University the front line is shifting). The establishment of the Gothic Line marks . . the end of phase two and the beginning of phase three, a period when the front line is again stable. The geographic space comprises three areas: A, B, C (Figure 2). Area A is delimited by the Gustav Line (in the south) and the Gothic Line (in the north). Area B is a seventy-five-kilometer-deep strip of territory that runs above the entire Gothic Line, from the Tyrrhenian Sea (on

https://www.cambridge.org/core 61 Pavone 2013.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logIc of vulnerabIlIty 693

Gustav Line Gustav Line Gothic Line

. Collapses

Phase 1 Phase 2 Phase 3 Warfront Stable Warfront Shifts Warfront Restabilizes

fIgure 1

frontlIne phases

https://www.cambridge.org/core/terms , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject ,

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Area C Area B , on on , Area A Not in sample

fIgure 2

Italy and research desIgn areas

University of Athens of University . . the west) to the Adriatic Sea (on the east). Last, area C is the territory between the northern border of Italy and area B. Hence, area A is con- trolled by the Germans and Italian Fascists (incumbents) in phase one, becomes an area of moving front line in phase two, and disappears from our research in phase three because it is under the control of the Allies.

In the empirical analysis of all three areas across the three phases, our

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 694 world politics units of analysis are cell months. Considering the nature of the anal- ysis, we do not rely on administrative units because they had no role . during the combat operations. Using cell months allows us to compare units with consistency. We opt for a grid of squared cells with sides of twenty-five kilome- ters. Previous research finds that cells of this size are the best balance between overaggregation and sample inflation.62 We have a time-series cross section with twenty-six months and a maximum of six hundred- fifty cells. The data include more than five thousand violent events. But because we opted for the cell month as our analytical unit, the tempo-

ral and spatial merging lead to a total of 1,419 cell months with vio- https://www.cambridge.org/core/terms lent events and 961 events involving civilian victimization. The sample shrinks because several events happened on different days but within the same month and same location. Cell months with events amount to 10 percent of the overall sample (N = 15,600).

Testing the Two Hypotheses on the Whole Territory of the Conflict In the first part of our national analysis, we test whether civilian vic- timization in area A changed due to the movement of the front line. We analyze the difference in the level of civilian victimization in area A before and after the collapse of the Gustav Line (phase one and two, respectively). We compare this difference with the levels of civilian vic-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , timization during same time span in area C, which was not affected by the movement of the front line. More precisely, we use a dd research design in which the grid cells in area A represent the treated sample and the grid cells in area C are the control group. We exclude area B from the untreated sample because keeping B would likely undermine

26 Sep 2021 at 00:01:48 at 2021 Sep 26 the stable unit treatment value assumption (sutva) and violate the as-

, on on , sumption of spatial independence resulting from possible spillovers of violence as an inertial by-product of the moving forces.63 In the supple- mentary material, we provide evidence that our data does not violate the parallel trend assumption.64

University of Athens of University In the second part of our national analysis, we investigate whether the . . cessation of frontline movement and the establishment of the Gothic Line had any effect on civilian victimization in area B, which was still under control of the incumbent and close to the front line. In this sec- ond dd analysis, the grid cells in area B are the treated sample, and the

62 Costalli and Ruggeri 2015. 63 We also run models with area B in the control group and our results hold (see Table 7A in the supplementary material; Costalli, Moro, and Ruggeri 2020b).

https://www.cambridge.org/core 64 Costalli, Moro, and Ruggeri 2020b.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 695

C .

B

A

Phase 1 Phase 2 Phase 3

Figure 3

Areas, Phases, and Expected Violence https://www.cambridge.org/core/terms

cells in area C are again the control group. We compare the two areas at

phase 2 (when the front line is moving) and at phase 3 (when the front line is again stable). Figure 3 depicts our expectations for the evolution of violence against civilians in the three areas over time. According to our theory, we should see low levels of violence in all three areas in phase one, but we should see a sudden increase of violence during phase two in area A, when the front line moves. During phase two, violence should also increase in area B, but with a delay because of the front line moving north and the regrouping of Nazi-Fascist forces in the area. During phase three,

though, violence in area B should decrease due to the stability of the , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , front line. In area C, we expect to see no change in the level of violence against civilians, which is why we use it as our control group. We match our units before performing the dd analysis to tackle pos- sible selection bias in the treatment and control groups and to make the parallel assumption more credible.65 We match cells with a similar

26 Sep 2021 at 00:01:48 at 2021 Sep 26 political history of voting radical left because variation in radical net- , on on , works may explain the variability in resistance to incumbents66 and be- cause incumbents may also use this information to target underground networks of the resistance.67 Moreover, we match areas with similar mountainous terrain because counterinsurgency could target areas with

University of Athens of University 68 . . rough terrain since they provide better protection for insurgents. Be- cause we need to handle cross-section time-series data,69 we opt for ker- nel matching and dd for repeated cross sections.70 65 Abadie 2005. 66 We use data from the 1921 elections, the last democratic elections before the Fascist regime. Costalli and Ruggeri 2015; Costalli and Ruggeri 2019; Corbetta and Piretti 2009. 67 Balcells 2017. 68 Fearon and Laitin 2003; Pavone 2013. 69 Blundell and Dias 2009.

https://www.cambridge.org/core 70 Villa 2016.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 696 world politics

Testing the Two Hypotheses in Emilia-Romagna Increasing the spatial and temporal disaggregation of our data, we repli- . cate our research design to conduct a within-region analysis focused on Emilia-Romagna. This region is the only one that includes territory in all three areas (A, B, C) and is one of the regions that endured signifi- cant confrontation between the Nazi-Fascist forces and the resistance. From summer 1943 through December 1944, the Nazi-Fascists were responsible for 2,153 killings in Emilia-Romagna, 780 in Piedmont, and 3,711 in Tuscany. By the end of the conflict, Emilia-Romagna had around forty thousand partisans, compared with twenty-two thousand https://www.cambridge.org/core/terms in Piedmont, and sixteen thousand in Tuscany. The main goal of our within-region analysis is to reduce the problem of omitted variable bias. To achieve this, we use squared cells with sides of ten kilometers and weeks as our temporal units of analysis. Moreover, important histori- cal works claim that Nazi-Fascists, aware of the extreme mobility of the resistance, directed their actions mainly in areas they presumed to hold partisan operational bases rather than in areas where the rebels carried out attacks.71 But it could be argued that the civilian victimization per- petrated by the incumbent was influenced by the actions of the parti- sans. To account for this possibility, we leverage the extremely detailed data on partisan activity that are available for Emilia-Romagna and use them in our analysis to explicitly control for the partisans’ actions. We

replicate the research strategy used in our statistical analysis on the en- , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , tire territory of the conflict, performing two matcheddd s for the two phases of the war (moving and stable front lines). Figure 4 shows the three areas into which we divided Emilia-Romagna. In this figure, area

B is twenty kilometers deep (instead of seventy-five), and phase one be-

gins when the front line is established in the region. 26 Sep 2021 at 00:01:48 at 2021 Sep 26

, on on , Data We evaluate our hypotheses using new, disaggregated data gathered through extensive archival research. First, a team of historians in Italy

collected information on every act of violence against civilians or un- University of Athens of University

. . armed rebels perpetrated in the country during the German occupation ( July 1943–May 1945). Following this research, we geocode more than five thousand events for which we know the actors of violence, the tar- get of violence (partisans or civilians), the number of victims, the date, and the location.72 We use the number of civilians killed by Nazi-Fascist

71 Pezzino 2008; Klinkhammer 1993.

72 See the website at http://www.straginazifasciste.it, accessed July 10, 2020.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded

logIc of vulnerabIlIty 697 .

https://www.cambridge.org/core/terms Area C Area B Area A

fIgure 4 emIlIa-romagna: cells and research desIgn areas

forces in a cell month as a dependent variable.73 On average, a cell month has 0.64 killings with a standard deviation of 8.2. The data refer to civilians killed without clear motivation based on group membership, excluding collaborators of the insurgents. We also control for the pres- ence of partisans after geocoding the data on partisan bases collected by the historians.74 In Italy overall, only 35 percent of the cells had parti- san bases between 1943 and 1945, while in our sample areas, 65 percent

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , of the cells had such bases. We also geocode data on partisan activities in Emilia-Romagna: Nazi-Fascists killed 3,768 civilians in 202 events, and partisans killed 1,814 Nazi-Fascist soldiers and committed 1,397

acts of sabotage.75 26 Sep 2021 at 00:01:48 at 2021 Sep 26

results , on on , Figure 5 presents the empirically observable evolution of civilian vic- timization across time in our three areas. Comparing Figure 5 with Fig- ure 3, which represents our theoretical expectations, it is evident that

our expectations closely refl ect this simple description of the data. As University of Athens of University . . expected, the level of violence against civilians in area A is relatively low in phase one, but increases signifi cantly at the beginning of phase two, when the Gustav Line collapses and the front line moves. Violence in 73 In the supplementary material, we also provide Tables 1A and 2A, in which the dependent vari- able is operationalized as the number of violent events. Our fi ndings do not change. Costalli, Moro, and Ruggeri 2020b. 74 Baldissara 2000.

75 Casali and Gagliani 2014.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 698 world polItIcs

1500 1000

. 500 0 Civilians Killed 1943m4 1943m10 1944m4 1944m10 1945m4 Area C

1500 1000 500 0 Civilians Killed

https://www.cambridge.org/core/terms 1943m4 1943m10 1944m4 1944m10 1945m4 Area B

1500 1000 500 0 Civilians Killed 1943m4 1943m10 1944m4 1944m10 1945m4 Area A

fIgure 5

three areas and three phases: actual levels of cIvIlIan vIctImIzatIon , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , this area drops in October 1944, when most of the Gothic Line is es- tablished and the front line is moving only on the Adriatic shore of Emilia-Romagna. In December, when the front line eventually stabi- lizes in that zone and phase three begins, the level of violence in area A

26 Sep 2021 at 00:01:48 at 2021 Sep 26 stabilizes to around zero. Area B exhibits a similar path, but with a de-

, on on , lay, as expected. Last, area C—the control group—shows no relevant change in the level of violence until April 1945, which marks the col- lapse of the Gothic Line, the end of phase three, the end of the war, and therefore also the end of the time span we consider.

But we cannot infer any causal relationship from Figure 5. To

University of Athens of University . . strengthen our argument, Table 1 reports the results of our dd analy- sis of civilian victimization (comparing area A with area C) for the fi rst phase, when the front line is moving. We use fi xed effects at the cell- month level and fi nd that when the front line is moving, each cell hosts on average almost three more civilian killings per month than when the front line is stable. This difference is statistically signifi cant (model 1).

Performing the same analysis on the sample of units with similar shares

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 699

Table 1

Effect of the Moving Front Line on Violence against Civiliansa . Model 1 Model 2 Model 3 DD Matched DD Matched DD and Controls Diff. A–C before moving front 0.304 0.407 0.623** (0.212) (0.297) (0.309) Diff. A–C moving front 2.958*** 2.781*** 3.009*** (0.300) (0.394) (0.404) DD 2.654*** 2.374*** 2.386*** (0.367) (0.493) (0.493) https://www.cambridge.org/core/terms Partisan presence 0.690** (0.276) R-squared 0.03 0.03 0.03 Observations 5940 4584 4584

***p < 0.01, **p < 0.05, *p < 0.1; standard errors in parentheses a In matched models, we matched units of analysis by share of mountainous terrain and 1921 vote share of left-wing parties.

of mountainous terrain and votes for the radical left in the 1921 elec- tions produced by kernel propensity score matching confirms the pre- vious findings (model 2) with only a small mitigation of the average treatment effect. If the logic of allegiance was the only logic at work, matching for historical local political preferences should dissolve the , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , effects of the moving front and challenge the logic of vulnerability. On the contrary, our findings suggest that the logic of vulnerability has a crucial role in explaining the level of civilian victimization. Addition- ally, we control in a matched dd model (model 3) for the presence of partisan bases in a given cell month. If civilian victimization was caused

26 Sep 2021 at 00:01:48 at 2021 Sep 26 only by a reaction to the presence of partisan bands, in a logic of retali- , on on , ation, our dd analysis should no longer be significant. However, the ef- fect of the logic of vulnerability is unchanged, but a logic of retaliation also appears to be present.

In Table 2, we use a different approach to study the effect of the University of Athens of University

. . moving front line. Instead of comparing areas that experienced the movement of the front line with a control group of units that did not, we disaggregate the frontline movement on a monthly basis following US military maps (see the supplementary material)76 and code the cell months accordingly. A grid cell is included in the staggered treatment if the Allies liberated it in the next month. We also modify the sample

76 Costalli, Moro, and Ruggeri 2020b.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 700 world politics

Table 2

Moving Front Line as Staggered Treatment . Model 4 Model 5 Model 6 Staggered treatment 7.186*** 7.186*** 4.956** (2.038) (2.100) (2.109) Lead-staggered treatment 1.334 (0.782) Lag-staggered treatment 2.100 (2.730) Partisans 2.105*** (0.000)

https://www.cambridge.org/core/terms Constant –0.035 –1.041*** –0.028 (0.185) (0.229) (0.197) Cell FE yes yes yes Month FE yes yes yes N 6.131 6.131 5.773

***p < 0.01, **p < 0.05, *p < 0.1; robust standard errors in parentheses

as the front line moves, dropping the cells that have been treated two months before because they can no longer be considered either part of the control group or part of the treated group, since the Nazi-Fascist forces are no longer there. We estimate the effect of our staggered treat- ment of the moving front line through two-way fixed-effects (months 77

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , and cells) models. Table 2, model 4, shows the average monthly effect of the moving front line. It increases the level of civilian victimization by Nazi-Fascist forces substantially. A twenty-five-kilometer-square cell subject to frontline movement experiences on average seven kill- ings more than untreated comparable cells. In model 5, we control for

26 Sep 2021 at 00:01:48 at 2021 Sep 26 whether the cells hosted bases of partisans, finding that the presence of

, on on , partisans increases civilian victimization but that the moving front line is still an important factor for Nazi-Fascist violence. In model 6, we add the lead and lag of the staggered treatment.78 We think of those two variables as placebos. If the increase of civilian victimization is mainly

due to the movement of the front line, the two placebos should not University of Athens of University . . reach standard statistical significance.79 Hence, our first hypothesis is supported using different model specifications and methods. We next analyze the effect of frontline stabilization on violence against civilians by looking at the difference in area B at phase three 77 Autor 2003; Angrist and Pischke 2008, 237–42. 78 Angrist and Pischke 2008, 237. 79 In the supplementary material, we provide further models with jackknife resampling and the

cubic polynomial of the overall temporal trend; Costalli, Moro, and Ruggeri 2020b.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 701 and phase two. As we note above, area B is a seventy-five-kilometer- deep strip of territory that runs above the Gothic Line; we study the ef- . fect of frontline stabilization on different treated areas, using strips of twenty-five, fifty, and seventy-five kilometers to gauge whether changes in treated areas affect the results. To further reduce the chances of non- random selection of units into the treatment group, we focus on the sample of grid cells produced by matching. The results in Table 3 are consistent and confirm our second hypothesis. The level of civilian vic- timization in the area under control of incumbents decreases dramati- cally when the front line is reestablished and stabilizes, but it is worth

noting that the reduction in civilian victimization is smaller as the dis- https://www.cambridge.org/core/terms tance from the front line increases. This finding is coherent with the logic of vulnerability: logistical and information needs are most press- ing near the front line. In other words, the vulnerability of the incum- bent’s forces increases near the front line and so do the incentives to kill civilians. As a result, when the front line stabilizes and the conven- tional forces’ security is somehow restored, the difference between the phase of movement and the phase of stability is largest close to the front line. Given that the presence of resistance groups increased over time in central and northern Italy, we also control for whether cells hosted a partisan base.80 Model 7 shows that even if stabilization considerably attenuates violence, in areas closer to the front line (twenty-five kilo- meters) civilian victimization is still triggered if there is partisan pres-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , ence, as expected by the logic of retaliation. In addition, as reported in Table 4A in the supplementary material, we control for a temporal spatial lag of Nazi civilian victimization based on an inverted distance W-matrix, lagged by one month. We find ev- idence of diffusion effects, but our findings on moving and stabiliza- 81 26 Sep 2021 at 00:01:48 at 2021 Sep 26 tion phases hold.

, on on , The idea that the movement of the front is strictly connected to the occurrence of massacres features prominently in work by Lutz Klink- hammer on the occupation of Italy. Klinkhammer finds:

[T]he zones of massacres moved rapidly, leaving the impression of a uniform University of Athens of University . . distribution over the entire territory. In fact, most of the massacres took place in the area of the front line or in the military zone immediately behind the front line. . . . In my view, the speed of German withdrawal considerably influenced the protection of logistical and withdrawal lines. . . . It was also the speed of withdrawal that influenced the behaviour of many military units.82

80 Costalli and Ruggeri 2015. 81 Costalli, Moro, and Ruggeri 2020b.

82 Klinkhammer 1997, 19–21.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 702 world politics

Table 3

Effect of Frontline Stabilization on Violence against Civilians a . Model 7 Model 8 Model 9 Matched DD Matched DD Matched DD and Controls and Controls and Controls 25 km 50 km 75 km Diff. B–C moving front 16.655*** 14.562*** 11.229*** (2.240) (2.000) (1.751) Diff. B–C stable front –0.770 –0.450 –0.167 (2.241) (1.957) (1.718) DD –17.425*** –15.011*** –11.396*** https://www.cambridge.org/core/terms (3.167) (2.800) (2.448) Partisan presence 5.138* 2.640 1.788 (2.942) (2.445) (2.188) R-squared 0.05 0.04 0.03 Observations 1848 1824 1812

***p < 0.01, **p < 0.05, *p < 0.1; standard errors in parentheses a Units of analysis matched by share of mountainous terrain and 1921 vote share of left-wing parties.

Similarly, Carlo Gentile argues:

[S]ince fighting the aggressors directly was very difficult, the soldiers victimized the population. In many cases, during the withdrawals, the situation caused a

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , real “psychosis of partisans.” . . . At the same time, the violence against civilians was not necessarily connected with the fight against the resistance bands: the regular warfare at the front line also caused an escalation of violence against ci- vilians. In these cases the violence came in the form of waves, strictly associated with the evolution of the conflict.83

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Mark Mazower suggests a relationship between front line shifts and , on on , the level of civilian victimization:

After months on the retreat, military necessity dictated securing the Apen- nines—the last line of defence before the Po valley—and in the summer of 1944

this was assured. It was scarcely a coincidence that the Marzabotto massacre University of Athens of University . . occurred at the very moment when the Germans defending the central section of the Gothic Line came under pressure from the advancing Allied forces.84

If we combine local accounts of violent episodes, it becomes evident that the geography of civilian victimization closely followed frontline movement. Large-scale civilian victimization along the Tuscan coast 83 Gentile 2015, 468–69.

84 Mazower 2009, 501. For more on the Marzabatto episode, see below.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 703 occurred as the German front line moved in the days that preceded the arrival of Allied troops. On June 13, 1944, German forces killed eighty- . three civilians in the small village of Niccioleta in southern Tuscany. Prior to these killings, Nazi-Fascist forces monitored the area, and de- spite partisans’ activism, there are no documents or testimonies indicat- ing a growing or altered concern for hostility in that particular village.85 The situation changed suddenly between June 10th and 15th. A band of insurgents stopped to rest in Niccioleta for a day. When they left, the local anti-Fascist inhabitants put together a small group of men armed with a few shotguns to protect the village from the Nazi-Fascist forces,

assuming that the partisans were now strong in that area and that the https://www.cambridge.org/core/terms incumbents would soon withdraw. The German troops were withdraw- ing northward after the fall of Rome, and the Allied forces were at their heels. On the morning of June 13th, a German battalion arrived in Niccioleta. This battalion was moving inland to conduct counterin- surgency designed to protect withdrawing German troops along the coast.86 When the German soldiers saw the few armed civilians guard- ing the town, they assumed that Niccioleta was now aligned with the insurgents. As postulated in our argument, the German units assumed the partisans now had a stronghold in Niccioleta and, pressed by the need to secure the rear of the moving front and lacking more precise information, immediately responded by killing nearly one hundred ci- vilians.87

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Two weeks later, a similar massacre of fifty-five civilians occurred in Guardistallo, about sixty kilometers north of Niccioleta. Both cases are significant examples of the many episodes that coincided with the movement of German forces north. American troops were quickly ad- vancing northward from Rome as Germans forces attempted to re-

26 Sep 2021 at 00:01:48 at 2021 Sep 26 group along the so-called Frieda Line, which ran from Lake Trasimeno

, on on , to the Tyrrhenian coast.88 The Frieda Line was located only a few miles from Niccioleta, making the area a logistic bulwark for German forces. On June 14th, American and French troops were only a few miles south of the line. Like Niccioleta, Guardistallo was located in the hills sur-

University of Athens of University rounding the coastal road, where Allied forces were advancing toward . . the front. The Guardistallo Massacre occurred only three days before the American Thirty-Fourth Infantry Division took control of the area. The massacre of Civitella della Chiana, in central Tuscany, on June 29th, also suggests a strong link between the advancement of Allied forces and 85 Pezzino 2001. 86 Pezzino 2001. 87 Pezzino 2001.

88 Fisher 1989, 242–47.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 704 world politics civilian victimization. The area had not been hosting insurgents’ bases, which were located in the surrounding mountains. Civitella was cer- . tainly central in the conventional operations of German forces, show- ing the importance of logistical concerns in the Germans’ most violent counterinsurgency operations. The massacre occurred when the Ger- man rear guard was regrouping in the face of the thrust from American, British, and French forces. In a detailed account of the massacre, Mi- chael Geyer explicitly mentions how Germans acted to minimize prob- lems in their vulnerable rear guard, with counterinsurgency becoming “a matter of immediate and systematic concern for front lines units be-

cause of their rapid, fighting retreat.”89 On August 12th, approximately https://www.cambridge.org/core/terms two hundred German soldiers of the Sixteenth ss Panzergrenadier Di- vision Reichsführer-ss executed 394 civilians, including women and children in Sant’Anna di Stazzema, in northern Tuscany. The massacre occurred just a few weeks before the completion and stabilization of the Gothic Line in the same area. In that period, the resistance was reach- ing its apex in Tuscany and, lacking high-quality information on the lo- cation and identity of the rebels, the Nazi-Fascist forces wanted to clear the southern side of the Gothic Line of insurgent threats.90 As argued by Klinkhammer, “The fact that partisan bands could control the com- munication routes behind the front line in some areas of the Apennine Mountains between Tuscany and Emilia-Romagna looked particularly dangerous for the German forces. The partisans could have blocked the

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , supplies and caused problems in case of a German withdrawal.”91 The resistance “represented a danger for the [German] fighting troops, for their supplies, and for the frequent patrolling missions that were re- quired in view of the establishment of the Gothic Line,”92 and “the fight against the bands could succeed only if reliable spies previously infil- 93 26 Sep 2021 at 00:01:48 at 2021 Sep 26 trated among the partisans and provided precise information.” “When

, on on , Fascist militias or German troops managed to completely crush a parti- san group, this was almost always due to the action of a spy or to under- ground investigative work,”94 but such activities take time and require relative stability.

University of Athens of University Most of the Gothic Line was already established by October 1944, . . but the front line was not completely stabilized until December. Dur- ing this period of stabilization, the incumbents were able to focus their

89 Geyer 1997, 189. 90 Pezzino 2008. 91 Klinkhammer 1993, 354. 92 BA-MA 1944c. 93 Klinkhammer 1993, 354.

94 Klinkhammer 1993, 338.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 705 counterinsurgency action on actual partisans.95 In those months, “the [German] troops managed to arrest many young men who, at the end . of ‘in-depth interrogatories’ and some cross-checking turned out to be partisans. . . . At the end of the [counterinsurgency] actions of Decem- ber 1944, a period of considerable quiet finally began.”96 The defensive actions of the Nazi-Fascist forces and the counterinsurgency strategies they carried out in this period of stability posed significant challenges to the resistance bands. Some of the latter even entered into explicit or implicit agreements with the incumbents.97 For instance, the so-called Banda Beretta was in control of several areas in the province of Parma,

a key area for the communications and supplies for the German Four- https://www.cambridge.org/core/terms teenth Army. The leaders of this partisan band negotiated with the German army and with the Fascist police to not attack the German forces and the communication lines in exchange for a similar restraint from the incumbents.98 Similarly, on November 5th, in the area of Pa- via, a partisan band entered into negotiations with representatives of the Fascist militias. Due to the nature of these negotiations and agree- ments, it is difficult to trace their outcomes,99 but “the situation for the population improved: the number of civilians killed remarkably de- creased compared to the previous summer.”100 These qualitative accounts demonstrate the importance of integrat- ing dynamic factors into the logics that aim to explain variation in ci- vilian victimization. Violence in World War II Italy cannot be reduced

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , (not exclusively, at least) to the actions of some units, because civilian victimization was widespread and involved several different divisions.101 It also cannot be linked exclusively to the desperation of the retreating German forces or to racism toward Italians because the killings were not occurring linearly through time and space but rather in response 102 26 Sep 2021 at 00:01:48 at 2021 Sep 26 to specific strategic incentives. To gauge more precisely whether the

, on on , logic of retaliation could have a crucial role and be an inferential threat for the logic of vulnerability, we provide further disaggregated empiri- cal analysis of the region of Emilia-Romagna.

95 Klinkhammer 1993, 365.

96 Gentile 2015, 186, 187. University of Athens of University . . 97 BA-MA 1944a. 98 US NARA 1945. 99 BA-MA 1944b. 100 Gentile 2015, 174. 101 Some of the literature, e.g., Gentile 2015, argues that the Sixteenth SS Panzergrenadier Division Reichsfuhrer-SS and the First Paratroop Panzer Division Hermann Goering perpetrated most of the violence against civilians in Italy. The new, complete data set that we use shows that these two divisions combined, although certainly violent, were responsible for only about 6 percent of the violent events. 102 But desperation might be a key factor in explaining violence in April 1945, the last phase of

war.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 706 world politics

The Logic of Vulnerability in Emilia-Romagna

. This analysis provides two major empirical gains. First, not only can we identify the presence of partisan bases, but we can also identify their actions against Nazi-Fascist forces. Second, we can increase the geo- graphic and temporal disaggregation of the data, moving from twenty- five-kilometer-square cells to ten-kilometer-square cells and from months to weeks. Both the data on partisan actions and the smaller units of analysis allow us to assess the relevance of the logic of retali- ation even more carefully. In Table 4, we provide the results of the dd analysis for the period of the front line’s movement (phase two). We https://www.cambridge.org/core/terms find that the level of civilian victimization increases compared to the premovement period (phase 1) and that the difference is statistically significant. Our results suggest that each treated cell-week experienced on average one civilian death at the hands of incumbent forces. When we consider all treated cells, the area affected by the moving front line in phase two suffered more than one thousand civilian casualties. The re- sults also hold when, in model 11, we match the grid cells by the share of votes to radical left in 1921 before performing the dd analysis and we control for partisan activity, expressed by the number of Nazi-Fascist soldiers killed in a given ten-kilometer-square cell in a given week. 103 In the supplementary material, we also consider different types of par- tisan activity, such as the number of attacks on the incumbents and the

number of sabotages conducted by the partisans.104 The results of inter- , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , est do not change in these additional models. In Table 5, we show the results of examining the period of frontline stabilization (phase three) in Emilia-Romagna; area B is twenty kilo- meters deep. Even at this level of analysis, and accounting for the num- ber of incumbents killed by partisans, we find a statistically significant

26 Sep 2021 at 00:01:48 at 2021 Sep 26 decline of civilian victimization. Matching grid cells by the share of , on on , votes to radical left in 1921 before testing the effect of frontline stabi- lization does not change the results. In this area too, qualitative historical evidence supports our theory

and quantitative findings. On September 29, 1944, immediately prior University of Athens of University . . to the stabilization of the central-western section of the front line, Ger- man troops started a campaign against civilians in the surroundings of Marzabotto, about fifteen to thirty kilometers north of the front line in the Apennines between Florence and Bologna, which lasted sev-

103 The territory covered by this analysis, especially the treated area, is much smaller and more ho- mogeneous than the one in the previous analysis, thus we do not match for the share of mountainous territory.

104 Costalli, Moro, and Ruggeri 2020b.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded Table 4 Effect of the Moving Front line on Civilian a . Victimization in Emilia-Romagna Model 10 Model 11 DD Matched DD and Controls Diff. A–C before moving front 0.026 –0.204 (0.139) (0.149) Diff. A–C moving front 1.049*** 0.837*** (0.280) (0.302) DD 1.022*** 1.041*** https://www.cambridge.org/core/terms (0.312) (0.336) Nazi-Fascists killed by partisans 4.914*** (0.122) R-squared 0.01 0.08 Observations 18084 14522

***p < 0.01, **p < 0.05, *p < 0.1; standard errors in parentheses a In matched models, we matched units of analysis by 1921 vote share of left-wing parties.

Table 5 Effect of Frontline Stabilization on Violence against Civilians in Emilia-Romagnaa

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Model 12 Model 13 DD Matched DD 20 km and Controls 20 km Diff. B–C moving front 2.190*** 1.664**

26 Sep 2021 at 00:01:48 at 2021 Sep 26 (0.706) (0.761)

, on on , Diff. B–C stable front 0.266 –0.095 (0.636) (0.686) DD –1.923** –1.760* (0.950) (1.024) Nazi-Fascists killed by partisans 5.986***

(0.249) University of Athens of University . . R-squared 0.01 0.10 Observations 5452 2444

***p < 0.01, **p < 0.05, *p < 0.1; standard errors in parentheses a In matched models, we matched units of analysis by 1921 vote share of

left-wing parties.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 708 world politics eral days. By October 5th, seven hundred-seventy people, the vast ma- jority of whom were civilians, had been killed.105 In the days preceding . the campaign, the Second Corps of the US Army was pushing north- ward, having previously reached and conquered the mountain passes. In the days following the massacre, the Sixty-Fifth, Thirty-Fourth, and Ninety-First divisions of the US Army had reached these areas, where the front then remained stable until the spring.106 Unrestricted civil- ian victimization unleashed in the area was linked to its strategic im- portance: for the Nazis, losing territory would have meant opening the door to the Allies’ penetration into a large plain where the latter would

have easily prevailed, thanks to their superior resources.107 A battal- https://www.cambridge.org/core/terms ion of the ss division, led by Major Walter Reder, was tasked with pre- venting partisans from disrupting this vulnerable and important line of communication. The core concern was the potential for an attack on German troops behind the front lines. In Reder’s words, the possibil- ity of an “arranged, simultaneous attack with the Americans” was per- ceived as a major danger, given the strategic value of the area, that had to be avoided at all costs.108 The decision to attack the civilian popu- lation rather than insurgents was due to both informational problems and the fact that units found civilian victimization convenient at a time when attrition from the conventional campaign was severely draining their human and material resources. In an ex post justification of what had happened in and around Marzabotto, a German officer claimed

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , that the behavior of German units had been “deplorable” yet “under- standable.” German soldiers faced a superior opposition and also had to defend themselves from attacks from behind front lines at a time when “supplies and ammunitions were becoming scarce and there was in- creasing worry about the military situation on other fronts.”109

26 Sep 2021 at 00:01:48 at 2021 Sep 26 As the central-western section of the front line stabilized between

, on on , mid-October and November 1944, civilian victimization in the area decreased. But along the Adriatic coast, the eastern section of the front line continued moving until December. The absence of major geo- graphical obstacles had allowed Allied forces to proceed further north

University of Athens of University in the autumn of 1944, creating a heightened sense of vulnerability for . . German forces. In response to this vulnerability, German forces re- sorted to sustained civilian victimization. The province of Ravenna was at the center of this violence. The largest massacre of the period oc- 105 Baldissara and Pezzino 2009. 106 Fisher 1989, 378–80. 107 Baldissara and Pezzino 2009, 68. 108 Baldissara and Pezzino 2009, 104.

109 Baldissara and Pezzino 2009, 324.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 709 curred on November 27th in the village of Madonna dell’Albero. The front was fragmented in this area, with partisans joining forces with . Canadian troops. The 721st Regiment of the 114th Jager (Alpine) Di- vision of the Wehrmacht was defending the outposts that guaranteed access to Ravenna, the provincial capital. After a firefight in the area of Via Nuova, German troops resorted to a scorched-earth strategy, killing fifty-six civilians who had taken refuge in nearby houses. The killing of civilians was not retaliatory but rather was directly linked to the need for German forces in the outpost to maintain the link with their rear guard. There was also the risk that the civilian population could pro-

vide support to Allied troops and partisans in further house-to-house https://www.cambridge.org/core/terms firefights. According to Enrica Cavina, “The massacre of the people of Via Nuova would have made the presence of other people more imme- diately visible and made the control of the outpost, and the possible re- treat, safer.”110 Fighting between German and Allied forces continued into the early days of December, with soldiers of the Canadian Fifth Armored Divi- sion and First Infantry Division forcing German troops to retreat, lib- erating a number of villages. To slow the Canadians’ advance, German forces began destroying all lines of communication, including bridges. Withdrawing forces also killed thirty-two civilians in their homes along the river. According to a former local partisan, when the retreating Ger- man forces arrived, they mined houses, aware that they were inhabited.

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , The mining of civilian houses was not retaliatory because there had been no partisan violence against the Germans in the area.111 Instead, the violence aimed to guarantee the safety of German forces operating in an area where their conventional enemies were already present. Ger- man forces sought to prevent civilians from providing information to

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Allied forces.

, on on , During the winter of 1944–1945, the front line remained stable with almost no direct clashes between the opposing conventional armies. On their side of the front line, the Nazi-Fascist forces organized counterin- surgency initiatives focused on partisan bands,112 but civilian victimiza-

University of Athens of University tion decreased dramatically. In this period, the partisans liberated areas . . of the territory under the official control of the incumbents near the front line, and some interesting forms of governance emerged. During this period, the so-called Repubblica di Montefiorino in the Apennines of Emilia-Romagna had a population of more than twenty-five thou-

110 Cavina 2016, 4. 111 Golinelli 1985.

112 Gorrieri 1966, 599.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 710 world politics sand inhabitants distributed among approximately thirty hamlets.113 Often, local representatives of pre-Fascist political parties reached . agreements with the Fascist militias so that these hamlets were on pa- per ruled by local Fascist authorities, but in reality were under civilian administration.114 The civilian population profited from this form of de facto democratic self-government while the incumbents benefitted from the fact that the inhabitants of these areas were not actively sup- porting or joining the armed partisan bands.

Violence and Advancing Front Lines: Qualitative Evidence https://www.cambridge.org/core/terms The nature of the conflict in World War II Italy does not allow for sys- tematic testing of how advancing troops engage in civilian victimization because a central condition of our theory is that such violence occurs in areas where regular forces face an aggrieved population with its related uncertainties. In Italy, the advancing Allied forces were largely wel- comed as liberators and the insurgents were coordinating their activities with the Allies. But as suggested above, recalling the German advance during the Russia campaign in 1941, our argument should also apply to cases of advancing troops facing insurgencies. Focusing on Italy, it could be argued that there was a short temporal window during which the local population was uncertain about the advancing Allied forces. In fact, the early phase of the Allies’ advance in Sicily provides some an-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , ecdotal evidence on how forces moving forward might face similar in- centives to use violence based on the logic of vulnerability. As noted by historians of the period and contrary to the rest of Italy, “the Sicilian case was characterized by the fluidity of the alignments and alliances, as it was close to July 25 [the fall of Fascist regime] and the armistice,

26 Sep 2021 at 00:01:48 at 2021 Sep 26 and discounted the unclear perception that the civilian population and

, on on , the Italian troops could have of who the enemy was and who the ally was.”115 In this context, we can observe how even the Allied troops could potentially face an aggrieved population. Coherent with our ar- gument, Sicily had been the theater of episodes of civilian victimiza-

University of Athens of University tion during the advancement of Allied forces. In the area of operations . . around Gela, soldiers of the US Forty-Fifth Division, which had lib- erated the area after fierce combat ( July 10–12, 1943), were responsi- ble for killing more than a dozen civilians in several episodes. Among those killed were the local Fascist leader and his relatives, signaling that

113 Gorrieri 1966, 553. 114 Gorrieri 1966, 548–49.

115 Mangiameli 2012, 142.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 711 the Allies were worried about potential defectors in the area who were yet to be fully controlled.116 . Other World War II theaters showed similar patterns of civilian vic- timization. For example, during Operation Barbarossa, the Wehrmacht, advancing toward Stalingrad, committed widespread atrocities—simi- lar to those undertaken during the retreat—in areas including Kiev and Kharkov. The Wehrmacht faced a growing partisan movement and “the well-organized partisan units would become an increasing menace to our rears. As the war continued, the people came to trust and support the partisans to a great extent, and they were able to find shelter and

protection everywhere.”117 Moreover, the Germans found it difficult to https://www.cambridge.org/core/terms organize camps and wanted to avoid creating a “potential nucleus of fu- ture partisans.”118 In the American Civil War, Sherman’s March to the Sea provides an- other example of an advancing army engaging in civilian victimization. As noted by James McPherson:

Sherman wrote that “we are not only fighting hostile armies, but a hostile peo- ple, and must make old and young, rich and poor, feel the hard hand of war.” Union armies must destroy the capacity of the Southern people to sustain the war. Their factories, railroads, farms—indeed their will to resist—must be dev- astated. “We cannot change the hearts of those people of the South, but we can make war so terrible . . . [and] make them so sick of war that generations would

pass away before they would again appeal to it.”119 , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , More recent conflicts also featured civilian victimization by regu- lar armies that were advancing. In the Eritrean-Ethiopian War, while moving forward the Eritrean army occupied cities on the front lines, such as Badme, where, according to Jon Abbink, “they neither were

26 Sep 2021 at 00:01:48 at 2021 Sep 26 hailed as liberators nor did they behave as such.” This was because “Er-

, on on , itrea had no administrative or other foothold in Badme and was not recognized by local people as legitimately having one.” In that context, Eritrean troops killed and abducted civilians and also destroyed schools and hospitals. Uncertain about the allegiance of the population and

constrained by the dire requirements of frontline fighting, they “lived University of Athens of University . . off the field” and took no risk.120

116 Mangiameli 2012. 117 Bidermann 2000, 44. 118 Shepherd 2009, 107. 119 McPherson 1988, 809; brackets in the original.

120 Abbink 2003, 223.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 712 world politics

Conclusion

. Scholars have attempted to explain the use of civilian victimization through several logics. This article both challenges and contributes to these logics by introducing the logic of vulnerability. We argue that the use of civilian victimization in counterinsurgency operations and its use in conventional warfare are strictly related. In conflicts where incum- bents have to fight both an insurgency and a conventional campaign, the logic of vulnerability explains temporal and geographic variation of civilian victimization as a function of frontline movements. We demon- strate the logic of vulnerability through an analysis of two key moments https://www.cambridge.org/core/terms in World War II Italy during the campaign of Allied forces against Nazi-Fascist forces as well as during the fighting of local armed resis- tance. In the phase of frontline movement between June and August 1944, Nazi-Fascist incumbents significantly increased their use of ci- vilian victimization in the rear guard of the front line. This case sup- ports the logic of vulnerability because the incumbents attempted to deal with the pressing needs of securing unstable supply lines against insurgents’ attacks and of struggling with scarce information on the lo- cation and identity of insurgents. The need for acquiring resources from the population and increased exposure to defection of the civilian pop- ulation to opposition forces can be a critical trigger of civilian victim- ization. After the stabilization of a new front line in fall 1944, German

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , forces notably reduced their violence against civilians. In line with the logic of vulnerability, civilian victimization was not a particularly use- ful tool once the incumbents’ vulnerability to insurgents’ actions dimin- ished. The incumbent forces’ need for resources could be managed with a restructured supply chain, and the need to avoid defection from the

26 Sep 2021 at 00:01:48 at 2021 Sep 26 civilian population decreased. In particular, a decline in civilian victim-

, on on , ization can be observed in the territories close to the front line. The ef- fect of the stabilization of the front line decreases as the distance from the front increases. We do not dismiss extant logics of civilian victimization, but in-

stead aim to demonstrate how some logics are more prevalent than University of Athens of University . . others. For example, even though racism and extreme ideological in- doctrination were notoriously pervasive amid the Nazi-Fascist forces, our research shows that the logic of vulnerability stands as a major ex- planation for civilian victimization in Italy and it poses the conditions under which other logics may drive violence. That our findings hold under multiple robustness checks suggests that the logic of vulnerability

should be even more relevant for incumbents with lower levels of racism

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 713 and extreme ideological indoctrination. Desperation might also play a role in the decision to victimize civilians. Records show that there were . frequent instances of isolated acts of violence against civilians by Ger- man soldiers, but the bulk of the violence perpetrated by Nazi-Fascist forces was in response to precise strategic incentives. Our study suggests that besides the logic of vulnerability, retaliation is a powerful mecha- nism to explain massacres by incumbents. Given the recent increase of internationalized domestic conflict, which often meets our scope conditions combining insurgency with conventional warfare, we suggest future studies should include three

specific analytical dimensions. First, further attention should be de- https://www.cambridge.org/core/terms voted to the different incentives that exist for domestic and foreign in- cumbents. The need—or aspiration—to govern the territory in the long term can change the structure of the incentives of domestic incumbents vis-à-vis foreign ones, leading to different strategies. Second, although civilian victimization has been carried out in history by both demo- cratic and nondemocratic armies, further research could engage with how democratic countries’ armed forces respond to the logic of vulner- ability. Evidence from accounts of the US war in Vietnam and, more recently, in Afghanistan and Iraq, could lead to more rigorous analysis of the conditions under which large-scale human rights violations and massacres of civilians have occurred. Third, future investigations should include theoretical insights on the different incentives on the part

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , of incumbents vis-à-vis the insurgents for using civilian victimization.

Supplementary Material Supplementary material for this article can be found at https://doi.org/10.1017

S004388712000012X. 26 Sep 2021 at 00:01:48 at 2021 Sep 26

, on on , Data Replication files for this article can be found at https://doi.org/10.7910/DVN/ ADYQWK.

References University of Athens of University . . Abadie, Alberto. 2005. “Semiparametric Difference-in-Differences Estimators.” Review of Economic Studies 72, no. 1: 1–19. doi: 10.1111/0034-6527.00321. Abbink, Jon. 2003. “Badme and the Ethio-Eritrean Border: The Challenge of Demarcation in the Post-War Period.” Africa: Rivista Trimestrale Di Studi e Documentazione Dell’Istituto Italiano per l’Africa e l’Oriente 58, no. 2: 219–31. At https://www.jstor.org/stable/40761693, accessed July 11, 2020. Angrist, Joshua D., and Jörn-Steffen Pischke. 2008. Mostly Harmless Econometrics:

An Empiricist’s Companion. Princeton, N.J.: Princeton University Press.

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 714 world politics Arjona, Ana. 2016. Rebelocracy. New York, N.Y.: Cambridge University Press. Arreguín-Toft, Ivan. 2001. “How the Weak Win Wars: A Theory of Asymmet-

. ric Conflict.”International Security 26, no. 1: 93–128. doi: 10.1162/01622880 1753212868. Autor, David H. 2003. “Outsourcing at Will: The Contribution of Unjust Dis- missal Doctrine to the Growth of Employment Outsourcing.” Journal of Labor Economics 21, no. 1: 1–42. doi: 10.1086/344122. Balcells, Laia. 2012. “The Consequences of Victimization on Political Identities: Evidence from Spain.” Politics & Society 40, no. 3: 311–47. doi: 10.1177/0032 329211424721. Balcells, Laia. 2017. Rivalry and Revenge: The Politics of Violence during Civil War. New York, N.Y.: Cambridge University Press.

https://www.cambridge.org/core/terms Baldissara, Luca. 2000. Atlante storico della Resistenza italiana. Milan, Italy: Bruno Mondadori. Baldissara, Luca, and Paolo Pezzino. 2009. Il massacro: guerra ai civili a Monte Sole. Bologna, Italy: Il Mulino. BA-MA. 1944a. RH 20-14, AOK 14, Vol. 44, Sheets 87–89. Freiburg, Germany: Bundesarchiv-Militararchiv. BA-MA. 1944b. RH 20-14/120, AOK 14, Bandenlagebericht. Freiburg, Ger- many: Bundesarchiv-Militararchiv. BA-MA. 1944c. RH 20-14, AOK 116. Freiburg, Germany: Bundesarchiv-Mili- tararchiv. Bartov, Omer. 2001. The Eastern Front, 1941–45: German Troops and the Barbari- sation of Warfare. New York, N.Y.: Palgrave McMillan. Belge, Ceren. 2016. “Civilian Victimization and the Politics of Information in the Kurdish Conflict in Turkey.”World Politics 68 no. 2 (April): 275–306. doi: 10.1017/S0043887115000398.

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Bidermann, Gottlob Herbert. 2000. In Deadly Combat: A German Soldier’s Memoir of the Eastern Front. Lawrence, Kans.: University Press of Kansas. Blundell, Richard, and Monica Costa Dias. 2009. “Alternative Approaches to Evaluation in Empirical Microeconomics.” Journal of Human Resources 44, no. 3: 565–640. doi: 10.3368/jhr.44.3.565. Browning, Christopher R. 1998. Ordinary Men: Reserve Police Battalion 101 and

26 Sep 2021 at 00:01:48 at 2021 Sep 26 the Final Solution in Poland. New York, N.Y.: Harper Perrenial.

, on on , Casali, Luciano, and Dianella Gagliani. 2014. “Guerra e Resistenza: azioni par- tigiane e stragi nazifasciste in Emilia Romagna.” Department of History, Cultures, and Civilization, University of Bologna. At https://disci.unibo.it/it /biblioteca/collezioni/basi-di-dati/azioni-partigiane-e-stragi-nazifasciste-in -emilia-romagna, accessed July 24, 2020.

Cavina, Enrica. 2016. “Episodio di Madonna dell’Albero, Ravenna, 27.11. 1944.” University of Athens of University . . At http://www.straginazifasciste.it/wp-content/uploads/schede/MadonnaDell Albero_Ravenna_27novembre44.pdf, accessed July 11, 2020. Collard-Wexler, Simon. 2013. “Understanding Resistance to Foreign Occupa- tion.” Ph.D. diss., Columbia University. At https://academiccommons.colum bia.edu/doi/10.7916/D8BZ6D8M, accessed August 9, 2020. Corbetta, Piergiorgio, and Maria Serena Piretti. 2009. Atlante storico-elettorale d’Italia: 1861–2008. Bologna, Italy: Zanichelli.

Costalli, Stefano, and Andrea Ruggeri. 2015. “Indignation, Ideologies, and

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 715 Armed Mobilization: Civil War in Italy, 1943–45.” International Security 40, no. 2: 119–57. doi: 10.1162/ISEC_a_00218.

. Costalli, Stefano, and Andrea Ruggeri. 2019. “The Long-Term Electoral Lega- cies of Civil War in Young Democracies: Italy, 1946–1968.” Comparative Po- litical Studies 52, no 6: 927–61. doi: 10.1177/0010414018784057. Costalli, Stefano, Francesco N. Moro, and Andrea Ruggeri. 2020a. “Replication data for: The Logic of Vulnerability and Civilian Victimization: Shifting Front Lines in Italy (1943–1945).” Harvard Dataverse, V1. doi: 10.7910/DVN/ADY QWK. Costalli, Stefano, Francesco N. Moro, and Andrea Ruggeri. 2020b. “Supplemen- tary material for: The Logic of Vulnerability and Civilian Victimization: Shift- ing Front Lines in Italy (1943–1945).” doi: 10.1017/S004388712000012X.

https://www.cambridge.org/core/terms Davenport, Christian. 2007. “State Repression and Political Order.” Annual Review of Political Science. 10: 1–23. doi: 10.1146/annurev.polisci.10.101405.143216. Di Salvatore, Jessica. 2016. “Inherently Vulnerable? Ethnic Geography and the Intensity of Violence in the Bosnian Civil War.” Political Geography 51: 1–14. doi: 10.1016/j.polgeo.2015.11.008. Downes, Alexander B. 2008. Targeting Civilians in War. Ithaca, N.Y.: Cornell University Press. Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97, no. 1: 75–90. doi: 10.1017/S000 3055403000534. Fisher, Ernest. 1989. United States Army in World War II: Mediterranean Theater of Operations; Cassino to the Alps. Washington D.C.: Center of Military History, United States Army. Fjelde, Hanne, and Lisa Hultman. 2014. “Weakening the Enemy: A Disaggre- gated Study of Violence against Civilians in Africa.” Journal of Conflict Resolu-

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , tion 58, no. 7: 1230–57. doi: 10.1177/0022002713492648. Gentile, Carlo. 2015. I crimini di guerra tedeschi in Italia: 1943–1945. Turin, Italy: Giulio Einaudi. Geyer, Michael. 1997. “Civitella in Val Di Chiana, 29 Giugno 1944: ricostruzione di un ‘intervento’ tedesco.” In Leonardo Paggi, ed., La memoria del Nazismo nell’Europa di oggi. Florence, Italy: La Nuova Italia.

26 Sep 2021 at 00:01:48 at 2021 Sep 26 Golinelli, Ercole. 1985. Interviewed by Gianluigi Meandri, Bagnacavallo, Italy,

, on on , August 20. Ravenna: Istituto Storico della Resistenza e dell’età Contempora- nea in Ravenna e Provincia. At https://istoricora.it/sites/default/images/down loads/interviste/Golinelli_Ercole.pdf, accessed July 11, 2020. Gorrieri, Ermanno. 1966. La Repubblica di Montefiorino: per una storia della Resi- stenza in Emilia. Bologna, Italy: Il Mulino.

Green, Amelia Hoover. 2018. The Commander’s Dilemma: Violence and Restraint in University of Athens of University . . Wartime. Ithaca, N.Y.: Cornell University Press. Greiner, Bernd. 2010. War without Fronts: The USA in Vietnam. London, UK: Vintage Books. Hastings, Max. 2010. The Korean War. New York, N.Y.: Pan Macmillan. Herreros, Francisco, and Henar Criado. 2009. “Pre-emptive or Arbitrary: Two Forms of Lethal Violence in a Civil War.” Journal of Conflict Resolution 53, no. 3: 419–45. doi: 10.1177/0022002709332208.

Hultman, Lisa. 2007. “Battle Losses and Rebel Violence: Raising the Costs for

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 716 world politics Fighting.” Terrorism and Political Violence 19, no. 2: 205–22. doi: 10.1080/0954 6550701246866.

. Humphreys, Macartan, and Jeremy M. Weinstein. 2006. “Handling and Man- handling Civilians in Civil War.” American Political Science Review 100, no. 3: 429–47. doi: 10.1017/S0003055406062289. Humphreys, Macartan, and Jeremy M. Weinstein. 2008. “Who Fights? The De- terminants of Participation in Civil War.” American Journal of Political Science 52, no. 2: 436–55. doi: 10.1111/j.1540-5907.2008.00322.x. ISTAT. 1957. Morti e dispersi per cause belliche negli anni 1940–45. Rome, Italy: Istituto Centrale di Statistica. At https://ebiblio.istat.it/digibib/Cause%20 di%20morte/IST3413mortiedispersipercausebellicheanni1940_45+OCRotti mizz.pdf, accessed July 11, 2020.

https://www.cambridge.org/core/terms Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge, UK: Cambridge University Press. Kalyvas, Stathis N., and Laia Balcells. 2010. “International System and Tech- nologies of Rebellion: How the End of the Cold War Shaped Internal Con- flict.” American Political Science Review 104, no. 3: 415–29. doi: 10.1017/S00 03055410000286. Kesling, Ben. 2016. “ISIS Herds Civilians to Mosul as Human Shields.” Wall Street Journal. October 28. At https://www.wsj.com/articles/isis-herds-civil ians-to-mosul-as-human-shields-1477675640, accessed March 30, 2019. Klinkhammer, Lutz. 1993. L’occupazione tedesca in Italia, 1943–1945. Turin, Italy: Bollati Boringhieri. Klinkhammer, Lutz. 1997. Stragi naziste in Italia: la guerra contro i civili, 1943–44. Rome, Italy: Donzelli. Kocher, Matthew Adam, Thomas B. Pepinsky, and Stathis N. Kalyvas. 2011. “Aer- ial Bombing and Counterinsurgency in the Vietnam War.” American Journal of

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Political Science 55, no. 2: 201–18. doi: 10.1111/j.1540-5907.2010.00498.x. Lacina, Bethany, and Nils Petter Gleditsch. 2005. “Monitoring Trends in Global Combat: A New Dataset of Battle Deaths.” European Journal of Population/ Revue Européenne de Démographie 21, no. 2–3: 145–66. doi: 10.1007/s10680 -005-6851-6. Lawrence, Thomas Edward. 1920. “The Evolution of a Revolt.” Army Quarterly

26 Sep 2021 at 00:01:48 at 2021 Sep 26 & Defence Journal 1, no. 1: 55–69. At https://permanent.fdlp.gov/lps68452

, on on , /lawrence.pdf, accessed July 24, 2020. Lyall, Jason. 2009. “Does Indiscriminate Violence Incite Insurgent Attacks? Evi- dence from Chechnya.” Journal of Conflict Resolution 53, no. 3: 331–62. doi: 10 .1177/0022002708330881. Lyall, Jason, and Isaiah Wilson. 2009. “Rage against the Machines: Explaining

Outcomes in Counterinsurgency Wars.” International Organization 63, no. 1: University of Athens of University . . 67–106. doi: 10.1017/S0020818309090031. Manekin, Devorah. 2013. “Violence against Civilians in the Second Intifada: The Moderating Effect of Armed Group Structure on Opportunistic Violence.” Comparative Political Studies 46, no. 10: 1273–300. doi: 10.1177/001041401 3489382. Mangiameli, Rosario. 2012. “Le stragi americane e tedesche in Sicilia nel 1943.” Polo Sud 2, no. 1: 141–78. Mazower, Mark. 2009. Hitler’s Empire: How the Nazis Ruled Europe. New York,

https://www.cambridge.org/core N.Y.: Penguin.

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded logic of vulnerabilit y 717 McPherson, James M. 1988. Battle Cry of Freedom: The Civil War Era. New York, N.Y.: Oxford University Press.

. Mitchell, Neil J. 2004 Agents of Atrocity: Leaders, Followers, and the Violation of Hu- man Rights in Civil War. London, UK: Palgrave Macmillan. Olson, Mancur. 1993. “Dictatorship, Democracy, and Development.” American Political Science Review 87, no. 3: 567–76. doi: 10.2307/2938736. Pavone, Claudio. 2013. A Civil War: A History of the Italian Resistance. New York, N.Y.: Verso Books. Pettersson, Therése, Stina Högbladh, and Magnus Öberg. 2019. “Organized Vio- lence, 1989–2018 and Peace Agreements.” Journal of Peace Research 56, no. 4: 589–603. doi: 10.1177/0022343319856046. Pezzino, Paolo. 2001. Storie di guerra civile: l’eccidio di Niccioleta. Bologna, Italy:

https://www.cambridge.org/core/terms Il Mulino. Pezzino, Paolo. 2008. Sant’Anna di Stazzema: storia di una strage. Bologna, Italy: Il Mulino. Prickett, Ivor. 2017 “In Mosul, Revealing the Last ISIS Stronghold.” New York Times. August 1. At https://www.nytimes.com/2017/08/01/world/middle east/mosul-isis-survivors-rights.html, accessed March 30, 2019. Shepherd, Ben. 2009. War in the Wild East: The German Army and Soviet Partisans. Cambridge, Mass.: Harvard University Press. Shepherd, Ben. 2012. Terror in the Balkans: German Armies and Partisan Warfare. Cambridge, Mass.: Harvard University Press. Snyder, Timothy. 2011. Bloodlands: Europe between Hitler and Stalin. New York, N.Y.: Vintage Books. Souleimanov, Emil Aslan, and David S. Siroky. 2016. “Random or Retributive? Indiscriminate Violence in the Chechen Wars.” World Politics 68, no. 4 (Octo- ber): 677–712. doi: 10.1017/S0043887116000101.

, subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Staniland, Paul. 2012. “States, Insurgents, and Wartime Political Orders.” Perspec- tives on Politics 10, no. 2: 243–64. doi: 10.1017/S1537592712000655. Steele, Abbey. 2011. “Electing Displacement: Political Cleansing in Apartadó, Co- lombia.” Journal of Conflict Resolution 55, no. 3: 423–45. doi: 10.1177/0022002 711400975. Steele, Abbey. 2017. Democracy and Displacement in Colombia’s Civil War. Ithaca,

26 Sep 2021 at 00:01:48 at 2021 Sep 26 N.Y.: Cornell University Press.

, on on , Torsiello, Mario, ed. 1975. Le operazioni delle unità italiane nel settembre–ottobre 1943. Rome, Italy: Ufficio Storico, Esercito Italiano. US NARA. 1945. “Sonderstab Ewringmann, Bandenfuhrer Beretta.” Miscella- neous German Records Collection, Part III, RG 242.16, T84E, Roll 249, Im- age 6-607298. United States National Archives and Records Administration

II: College Park, Md. University of Athens of University . . Villa, Juan M. 2016. “Diff: Simplifying the Estimation of Difference-in-Differ- ences Treatment Effects.” Stata Journal 16, no. 1: 52–71. doi: 10.1177/1536867 X1601600108. Wette, Wolfram. 2007. The Wehrmacht: History, Myth, and Reality. Cambridge, Mass.: Harvard University Press. Wood, Reed M. 2014. “From Loss to Looting? Battlefield Costs and Rebel Incen- tives for Violence.” International Organization 68, no. 4: 979–99. doi 10.1017/ S0020818314000204.

https://www.cambridge.org/core Wright, Donald P., and Timothy R. Reese. 2008. On Point II: Transition to the

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded 718 world politics New Campaign; The United States Army in Operation Iraqi Freedom, May 2003– January 2005. Fort Leavenworth, Kans.: Combat Studies Institute Press.

. Zhukov, Yuri M. 2017. “External Resources and Indiscriminate Violence: Evi- dence from German-Occupied Belarus.” World Politics 69, no. 1 ( January): 54–97. doi: 10.1017/S0043887116000137.

Authors Stefano Costalli is an associate professor of political science in the Department of Political and Social Sciences at the University of Florence, Italy. His research interests include civil wars, political violence, peacekeeping, democratization pro- cesses, political realism, and quantitative methods for political research. He can be

reached at [email protected]. https://www.cambridge.org/core/terms

Francesco N. Moro is an associate professor of political science in the De- partment of Political and Social Sciences at the University of Bologna, Italy, and an adjunct professor of International Relations at the Johns Hopkins University School of Advanced International Studies–Europe Campus. His research inter- ests include collective violence, organized crime, and defense and security policy. He can be reached at [email protected].

Andrea Ruggeri is a professor of political science and international relations and director of the Center for International Studies at the University of Oxford. He is Fellow in Politics at Brasenose College, Oxford. His research interests include civil wars, political violence, peacekeeping, and methods for political research. He is the author, with Vincenzo Bove and Chiara Ruffa, of Composing Peace (2020).

He can be reached at [email protected]. , subject to the Cambridge Core terms of use, available at at available use, of terms Core Cambridge the to subject , Acknowledgments We thank Bernd Beber, Alessandro Belmonte, Vincenzo Bove, Lars-Erik Ceder- man, Fabrizio Coticchia, Elias Dinas, Jessica Di Salvatore, Leandro Elia, Simon Hug, Stathis Kalyvas, Juan Masullo, Adam McCauley, Toni Rovatti, Julian Wuch- erpfennig, three anonymous reviewers of World Politics, and workshop participants

26 Sep 2021 at 00:01:48 at 2021 Sep 26 at the University of Amsterdam, University of Genoa, Hertie School, London

, on on , School of Economics and Political Science, University of Oxford, University of Warwick, and the 2019 Conference of the Italian Political Science Association (sisp) for helpful comments on previous versions of this article. We thank Jorien Van De Mortel and Silvia D’Amato for their research assistance.

University of Athens of University ey ords

. . K W

civil war, civilian victimization, counterinsurgency, Italy, World War II

https://www.cambridge.org/core

https://doi.org/10.1017/S004388712000012X Downloaded from from Downloaded