While the study of both genocide dynamics and civil war insurgencies has attracted a lot of scholarly interest in the past two decades (e.g., Balcells Reference Balcells2017; Balcells and Kalyvas Reference Balcells and Kalyvas2014; Braun Reference Braun2016; Charnysh and Finkel Reference Charnysh and Finkel2017; Fujii Reference Fujii2017; Humphreys and Weinstein Reference Humphreys and Weinstein2008; Kopstein and Wittenberg Reference Kopstein and Wittenberg2018; Lyall Reference Lyall2009; Staniland Reference Staniland2018; Straus Reference Straus2006; Valentino Reference Valentino2013; Wood Reference Wood2003), few social scientists have studied the relationship between genocide targets and insurgents.Footnote 1 This is surprising, because genocides and mass killings typically coincide with other conflicts such as civil wars in which insurgents are present (Eck and Hultman Reference Eck and Hultman2007; Kalyvas Reference Kalyvas2019; Schwartz and Straus Reference Schwartz and Straus2018; Shaw Reference Shaw2003). Indeed, insurgents were frequently involved in major events of mass categorical violence, such as those in the Ottoman Empire, Rwanda, Bosnia, Myanmar, or South Sudan. In the case at hand, anti-Nazi insurgents, a pan-European phenomenon (Petersen Reference Petersen2001; Wieviorka Reference Wieviorka2019), were Jews’ potential allies whose widespread presence across France and Europe in WWII could have affected Jews’ survival. The topic has been debated by historians studying the Holocaust: for example, viewing European partisan networks as dangerous (Ainsztein Reference Ainsztein1974; Bańkowska Reference Bańkowska2005), or at best inconsequential to Jewish survival (Poznanski Reference Poznanski2008), although cases of genuinely pro-Jewish insurgent groups are well documented (Kerenji Reference Kerenji2016; Kermish Reference Kermish2011; Ringelblum Reference Ringelblum1992; Tec Reference Tec2008). Is it possible to abstract from individual cases and find out the average influence of insurgent presence on the survival of the targets of genocide? I study this question using the example of non-Jewish insurgents during the Holocaust in France.
Existing studies of civil war and mass violence that study insurgents’ impact on civilians consider civilians as a uniform group facing similar constraints and focus only on the victimization during counterinsurgency and reprisal campaigns. They regard mass killings of civilians as a military strategy of the incumbent employed when insurgents pose a heightened threat (Balcells Reference Balcells2017; Kalyvas Reference Kalyvas2006; Schwartz and Straus Reference Schwartz and Straus2018; Shaw Reference Shaw2003; Straus Reference Straus2015; Sullivan Reference Sullivan2012; Valentino, Huth, and Balch-Lindsay Reference Valentino, Huth and Balch-Lindsay2004; Vargas, Anderton, and Brauer Reference Vargas, Anderton and Brauer2016). However, firstly, they do not consider the impact on genocide targets, a special group of civilians who face heightened persecution. Secondly, they do not look at the impact of insurgents on civilians outside the times of increased violence, an important consideration in longer-spanning conflicts. This narrow temporal focus makes them conclude that the insurgent threat will always increase state targeting leading to higher civilian casualties. Lastly, they do not specify the meaning of “insurgent threat,” which could refer to insurgent strength, presence, or nature (pertaining to a specific ethnic or religious group). My theoretical intervention complements the literature by studying the three-way relationship between the incumbent, insurgents, and genocide targets over the entire conflict span. It unpacks the meaning of “threat” by focusing on insurgent presence specifically.
Being relatively weaker, insurgents develop a clandestine toolkit—“hit and run” skills—to fight the incumbent (Finkel Reference Finkel2015; Kalyvas Reference Kalyvas2019). These include establishing clandestine evasion networks and hiding their real identities—skills and resources that are sought after by genocide targets as well (Finkel Reference Finkel2017, 126–58). Insurgents are willing to assist genocide targets out of sheer “moral” motivations, or because they hope to “materially” benefit from it.
To test the relationship between insurgent presence and genocide targets’ survival, I apply a mixed-methods design combining large-n analysis with qualitative case studies. My empirical approach consists of two steps. First, I instrument WWII insurgent presence (number of La Résistance members) with the number of soldier deaths in World War One (WWI)—assigning both to their respective places of origin—and study its effects on Jews’ victimization. What links the WWI deaths with WWII insurgent enlistment is the anti-German sentiment stemming from WWI. Second, to explain the involved causal mechanisms, I zoom in on six “typical” or “deviant” French provinces that differ in levels of insurgent density and type (left-wing/right-wing) and in levels of Jews’ victimization (high/low). I analyze mentions of insurgency in survivors’ and rescuers’ testimonies.
My analysis relies on twelve archival collections and includes the complete list of French Holocaust victims compiled from the “Memorial to the Jews Deported from France” (Klarsfeld Reference Klarsfeld2012), the French Ministry of Armed Forces’ extensive lists of French engagement in WWI and WWII, USC Shoah Foundation’s survivor testimonies, and Yad Vashem’s (YV) Righteous rescuers’ testimonies.
My instrumental variable (IV) analysis indicates that a higher concentration of WWII insurgents led to fewer Holocaust victims. The case studies reveal that what drove the negative relationship between WWII insurgent density and Holocaust victimization rates was the irregular character of the French occupation and civil wars. To help the Jews, insurgents used the resources and skills they had developed to evade and fight the German occupier and the collaborationist French state. They warned Jews against impending raids (shared information), connected them with other rescuers (shared networks), forged their documents and smuggled them across borders (used skills). Insurgents who provided sustained help pertained to left-wing groups and acted out of “moral” motivations. Those from right-wing groups helped out of “material” motivations in a more one-off fashion. Due to military concerns, help or cooperation however were not official policy of La Résistance leadership. Both were initiated at the discretion of individual insurgents.
These results delimit two scope conditions of the theory. We are more likely to observe the positive effects of insurgent presence in those genocides (1) where insurgents harbor inclusive ideologies, and (2) that take some time to unfold, so that the weak actors (insurgents and genocide targets) have time to adapt their response strategies to the realities of the persecution. With time, genocide targets become aware of the dangers they face and are better positioned to seek help from insurgents whose skills match their needs.
This article contributes to the scholarship on genocide and political violence in general and to that on the Holocaust in particular, both empirically and theoretically. It builds on the burgeoning field of the social science of the Holocaust (among other works see notably, Balcells and Solomon Reference Balcells and Solomon2020; Braun Reference Braun2016; Reference Braun2019; Charnysh Reference Charnysh2015; Charnysh and Finkel Reference Charnysh and Finkel2017; Finkel Reference Finkel2015; Reference Finkel2017; Kopstein and Wittenberg Reference Kopstein and Wittenberg2018; Kopstein, Subotić, and Welch Reference Kopstein, Subotić and Welch2022; Subotić Reference Subotić2019).
Empirically, the article advances our knowledge about how Jews survived in the Holocaust. While research in Jews’ survival abounds in the field of history, the disjunction between micro and macro-level dynamics of insurgent assistance makes the contrasting evidence difficult to reconcile. The comparative approach and social scientific methods I exploit enable the quantification of average outcomes. By drawing on extremely rich yet underutilized archival resources on the Holocaust (Lerner Reference Lerner2021), WWI and WWII, and applying a mixed-methods design, my approach provides first robust empirical evidence about surprising moderating effects of anti-Nazi insurgency on the severity of the Holocaust violence.
Theoretically, it qualifies the findings of existing studies on violence against civilians and complements those on the rescue of genocide targets. Firstly, I propose that if we take a longer view and consider the special needs of civilians who are ascriptively targeted, it transpires that insurgents benefit their survival. Thus, my findings qualify the consensus in the literature that insurgent threat will inevitably spur genocidal violence. Secondly, this article is, to the best of my knowledge, the first social scientific study to explore the relationship between insurgents and genocide targets. It does so from the perspective of the weaker actors and not the state, as has been done until now in studies of insurgent influence on civilian victimization. Thirdly, I move away from conceiving of rescue in genocide as synonymous to hiding by altruistic civilians. I incorporate other helping modes to the concept of rescue—warning, sharing networks, forging documents and smuggling—making the concept better suited to explain survival.
EXISTING EXPLANATIONS: CIVILIANS SUFFER FROM COUNTERINSURGENCY AND REPRISALS
Historians and social scientists who studied survival in genocide analyzed it through three lenses: the perpetrator’s strategy and motivations, the actions of civilian rescuers, and the agency of the victims. They concentrated on actor dyads, i.e., the behavior of one set of actors in relation to another.Footnote 2 The literature has not theorized the role of insurgents in relation to survival in genocide. This is surprising, because genocides typically occur in times of interstate or civil wars (Eck and Hultman Reference Eck and Hultman2007; Harff Reference Harff1987; Kalyvas Reference Kalyvas2019; Krain Reference Krain1997; Melson Reference Melson1992; Shaw Reference Shaw2003; Straus Reference Straus2006).
Scholars of mass violence and insurgency in civil war did study states’ anti-insurgent policies (counterinsurgency campaigns and reprisals) in relation to civilians. They found that irregular warfare correlates with the highest number of non-combatant deaths as compared to other types of civil wars (Balcells and Kalyvas Reference Balcells and Kalyvas2014). This is because irregular civil wars are longer and generate more indiscriminate (Kalyvas Reference Kalyvas2006) and group-selective (Fjelde and Hultman Reference Fjelde and Hultman2014; Steele Reference Steele2009) violent events —a calculated military strategy by incumbents to drain insurgent groups of manpower and resources, especially when perceived insurgent threat is high (Schwartz and Straus Reference Schwartz and Straus2018; Shaw Reference Shaw2003; Sullivan Reference Sullivan2012; Valentino Reference Valentino2013; Vargas, Anderton, and Brauer Reference Vargas, Anderton and Brauer2016).Footnote 3
The existing theories are nevertheless incomplete. Firstly, because they do not address the specific characteristics and interests of genocide targets—a special category of civilians. Secondly, because they do not study extended conflict duration.
THEORY: OVER TIME, GENOCIDE TARGETS BENEFIT FROM INSURGENTS’ IRREGULAR WAR SKILLS
A complementary view posits that the insurgents are uniquely positioned to help the genocide targets, because (1) they possess adequate resources and skills, and (2) they could be willing to use them in the genocide targets’ favor. The two actors’ physical proximity could therefore translate into a lower incidence of genocidal violence.
The (im)balance of power between belligerents determines the difficulties and shapes the strategies of each warring side (Kalyvas Reference Kalyvas2019). In irregular civil war, the relative weakness of insurgents vis-à-vis the incumbent forces them to develop an “insurgent toolkit” (Finkel Reference Finkel2015), i.e., clandestine tactics to evade and attack the stronger enemy. Such strategies include: gathering intelligence about the enemy and their collaborators; developing national and cross-border evasion networks to conduct sabotage and build the organization; forging documents to hinder identification by blending into the civilian population. These skills and resources match genocide targets’ needs.
Although both genocide targets and regular civilians ultimately seek to maximize their survival chances, the two actors will adopt different strategies to achieve this. The genocide targets, unless they comply with their persecutor or fight them, will try to escape and/or hide either by disappearing from public life or by passing for non-targets (Finkel Reference Finkel2017; Paulsson Reference Paulsson2002). Their strategies can evolve over time based on changes in persecutory strategy and the targets’ risk appreciation (Finkel Reference Finkel2017, 51–68). Regular civilians will instead strategically choose to associate themselves with the locally strongest warring actor (Baines and Paddon Reference Baines and Paddon2012; Kalyvas Reference Kalyvas2006, 12). These allegiances will fluctuate during conflict based on local conditions and beliefs about the likely outcome of the conflict (Kalyvas Reference Kalyvas2006, 95, 229). In sum, insurgents’ specific skills and resources might benefit the genocide targets, but will not be sought after by regular civilians.
Current studies concentrate on the effects of counterinsurgency campaigns and reprisals, i.e., violence inflicted over short periods. Following Finkel and Straus (Reference Finkel and Straus2012, 63), I take a longer-term approach that reflects changes occurring throughout the conflict duration. In the longer term, both insurgents and genocide targets might adapt to conflict conditions and come up with ways to seek and offer help. This would explain why in the testimonies that recount less volatile stages of WWII in France, survivors sought help in insurgent-filled areas.Footnote 4
Various historical and sociological theories shed light on why insurgents may choose to assist or forgo helping. In Table 1, I outline motivations likely present in the studied context, differentiating between “material” and “moral” incentives that relate to, respectively, presence or absence of cost/benefit-centered reasoning. These motivations may overlap, but I anticipate that, at the micro level, individual insurgents are inclined towards reasoning favoring pro-social behavior.Footnote 5
Table 1. Insurgents’ Motivations for or Against Helping
Among the “material” motivations, we have the “financial,” “compromising information,” “reputational,” and “capacity” theories. Firstly, insurgents may have assisted Jews for financial gain or as part of their network’s financing strategy (Brethour Reference Brethour2019; Grabowski Reference Grabowski2008; Tec Reference Tec2013). Secondly, Gambetta and Przepiorka (Reference Gambetta and Przepiorka2019) propose a mechanism explaining mutual trust between insurgents and genocide targets, rooted in the sharing of “compromising information” to demonstrate loyalty, fostering cooperation (see also Gambetta Reference Gambetta2011). Since both insurgents and genocide targets run the danger of denunciation, one could trust not to be betrayed by the other. Thirdly, “reputational” motivations may drive collectives to aid, as seen with Yugoslav Communist partisans who viewed aiding Jewish inmates as a means to showcase inclusivity and serve propaganda (Kerenji Reference Kerenji2016). Conversely, if the persecuted group is unpopular, insurgents might refrain from assistance to avoid losing public support. Lastly, “capacity” reasoning suggests insurgents may abstain from helping due to limited organizational resources.
Among the key “moral” motivations are the “empathy,” “structural balance,” and “antipathy” theories. Historical literature often depicts rescuers during the Holocaust as highly moral individuals driven by a sense of ethical duty (Fogelman Reference Fogelman2011; Monroe Reference Monroe2013; Oliner and Oliner Reference Oliner and Oliner1992; Paldiel Reference Paldiel1993). This perspective aligns with findings by Braun (Reference Braun2016; Reference Braun2019), who observed that rescue efforts within religious minority networks stemmed from “empathy” (Hoffman Reference Hoffman2001). Additionally, the theory of “structural balance” suggests that groups tend to form positive or negative sentiments based on shared friendships or enmities (“enemy of my enemy is my friend”), making cooperation likely between insurgents and genocide targets (Cartwright and Harary Reference Cartwright and Harary1956; Heider Reference Heider1946; Simmel Reference Simmel1922).Footnote 6 Conversely, “antipathy” towards genocide targets may deter insurgents from assisting or even lead to aggression against them (Bańkowska Reference Bańkowska2005; Fein Reference Fein1979, 68–71, 82–90).
Despite potential reasons for non-engagement, the average insurgent is likely motivated by positive inclinations towards helping. Insurgents possess personality traits predisposing them to aid others, akin to the community of rescuers. Suedfeld and de Best (Reference Suedfeld and de Best2008, 37) found that both rescuers and anti-Nazi insurgents are risk-takers who differ from the general population. They prioritize Benevolence and Security over Conformity, reflecting shared values emphasizing helpfulness, responsibility, and national security concerns (Schwartz Reference Schwartz and Zanna1992). These core life values, combined with insurgents’ urge to resist against occupiers, may lead them to view helping as another form of resistance.
In sum, we could expect to find a negative net effect of insurgent presence on genocidal violence, as insurgents engage in helping genocide targets out of material and/or moral motivations.
BACKGROUND
Insurgency in France
What is typically known as the French anti-Nazi resistance effort, La Résistance, was in fact a pool of many organizations that mushroomed across France and abroad after the French–German Armistice of June 1940 and until the Normandy landings in June 1944. They represented the entire ideological spectrum, from conservative to communist, and are typically divided into two categories, networks and movements.
La Résistance groups performed six types of activities: organization of the evasion of prisoners of war and allied soldiers, printing and distribution of anti-Nazi press, document forgery, intelligence-gathering, political assassinations, and armed actions (sabotage). While propaganda and document forgery were organized even in urban centers, other activities such as evasion and sabotage took place in rural and mountainous regions.
The first insurgent groups to form were right and centre-left. They formed in the summer of 1940, shocked by the defeat and believing that the armed effort needed to continue. Due to the lack of weaponry, their activities were at first limited to the distribution of underground press and the smuggling of Allied soldiers and refugees (Wieviorka Reference Wieviorka2016). Three major subsequent events drove widespread underground enlistments. First, Nazi Germany’s invasion of the Soviet Union in June 1941 led to the emergence of the Communist insurgency; second, the introduction of two compulsory work service legislations (STO) in September 1942 and in February 1943 propelled by far the largest recruitment waves into the guerilla French Forces of the Interior (FFI); third, the eventual imminent Allied victory signalled by the Normandy landings in June 1944 turned La Résistance into a popular movement (Wieviorka Reference Wieviorka2016). Overall, of the 323 insurgent groups that emerged in France during WWII, 116 (36%) did so before the first deportation of Jews, which took place on March 27th, 1942 (see Figure 1).
Figure 1. Emergence and Survival of La Résistance Groups with Chosen Deportation Convoys
Note: Timeline includes 279 La Résistance networks (“réseaux”) and 44 movements (“mouvements”) (Marcot Reference Marcot2006, 104–11; 114–43; 1082–8). The vertical lines refer to the dates of key deportations of Jews from France: Convoy 1 stands for the first train of Jewish deportees; Convoy 3 for the first train of Jewish deportees that included women and children; Convoy 79 was the last convoy to depart from France.
La Résistance leadership condemned the premeditated murder of the Jews by the Nazis and their collaborators, but did not adopt official pro-Jewish measures to counter the genocide that developed in front of their eyes (Bruttmann Reference Bruttmann2003, 195–202; Poznanski Reference Poznanski2008, 194–6; Poznanski Reference Poznanski1995, 136, 139; Yagil Reference Yagil2010, 239). This decision was due to concerns about military costs of such rescue operations, which is apparent in the underground press publications and statements by La Résistance leadership.Footnote 7
Jews in France
Around 330,000 Jews are believed to have lived in metropolitan France before the outbreak of WWII. Some two hundred thousand of them were French and 130,000 were foreign Jews. No official data exist on their exact distribution across the French territory, but from the distribution of the synagogues it is estimated that the biggest communities were concentrated in Paris, Alsace-Moselle, Marseille, Bordeaux, Marne, Nord, and Seine Maritime (Klarsfeld Reference Klarsfeld1983, 14–5).
While Paris was the site of the biggest and best well-known round-up of Jews in July 1942, most of the victims came from the territories administered by the Vichy government in the south (see Figure 2).Footnote 8
Figure 2. Holocaust Victims by Place of Arrest or Murder
Note: Units are counties (“cantons”) by their 1940 boundaries.
Several Jewish rescue organizations formed already at the beginning of the occupation, in 1940. Among them were Rue Amelot Committee, the Cimade, Jewish Guides and Scouts of France (EIF, la Sixième), Children’s Aid Society “OSE” and a military resistance network, the Jewish Army (Latour Reference Latour1970; Poznanski Reference Poznanski1995). They became best known for their efforts to hide and evacuate Jewish children from France. These organizations had some ties with La Résistance and were not opposed to by the latter, but, on the strategic macro-level, the two operated separately from one another.
QUANTITATIVE DATA: SOURCES AND MEASUREMENT
Outcome
Genocidal Violence
Similarly to Kalyvas (Reference Kalyvas2006, 19–20), I define a Holocaust victim as a Jewish person who is either executed or captured for deportation.Footnote 9 To construct the measure, I rely on the fine-grained “Memorial to the Jews Deported from France,” which contains records of all 79,253 victims of the Holocaust deported from France, 97% of whom I am able to geocode (see Figure 2). The database was compiled by Klarsfeld (Reference Klarsfeld2012) based on over 300 local historical sources: collections of 90 French departmental archives, ghetto and concentration camp records, YV Center’s archive, the publications of the Official Journal of the French Republic and legal documentation from proceedings against Nazi criminals and their collaborators. Additionally, a minority of Jews were murdered on the French territory. To include these victims I apply dictionary approach to biographies from the online “Dictionary of the Shot and Executed Civilians in France Between 1940–1944,” a database gathered and curated by the CNRS Institute and University Paris 1 (Pennetier et al. Reference Pennetier, Ponnavoy, Marquer, Pennetier and Jondeau2015). I code as a Jewish victim every civilian identified as Jewish.
To compute local extermination rates, I geocode the information about each victim’s place of arrest or murder and aggregate it at the county level using historical 1940 vector data by Gay (Reference Gay2021). The final variable summarizes information about 77,198 victims.Footnote 10 Lastly, to ensure that higher values of the outcome reflect higher victimization rates, I divide the count of the victims by the Jewish population statistics.
The measure of Jews’ wartime presence is based on data from 1940/1941 census of Jews, collected by the Nazi and Vichy police. I gather census province-level data from Klarsfeld (Reference Klarsfeld1983) (N=229,577 across 90 provinces) and employ an OLS prediction model to allocate people across the province-nested counties.Footnote 11 The police records relied on voluntary registrations and intelligence work, since not all Jews decided to register (for a micro-history on the subject see Mariot and Zalc Reference Mariot, Zalc, Zalc and Bruttmann2017, 85–98). The data therefore contain biases as they do not involve those Jews who decided to and succeeded in hiding their identity early during the war. However, the genocide initiated in Eastern Europe in the second half of 1941 and spread to France only in the summer of 1942. Without the benefit of contemporary hindsight, a large majority of Jews did follow the registration order at the beginning of the occupation (Becker and Wieviorka Reference Becker and Wieviorka1998, 200), therefore most historians consider the census to be—unfortunately—reliable (see, e.g., a study of the reliability of the census in the Dordogne province by Reviriego Reference Reviriego2003, 96–8).
Explanatory Variable
Insurgent Presence
To gauge insurgent presence across France, I rely on a collection of the French Ministry of Armed Forces on the members of La Résistance (Mémoire des hommes 2018). Compiled between 1941 and 1951 from wartime evidence and postwar certification requests, it consists of 597,632 individuals and includes names, surnames, place of birth and insurgent affiliation.Footnote 12 In constructing the measure of insurgent presence I consider potential alternative explanations that could invalidate my theory. The first relates to the causal mechanism, and the second to potential temporal mismatch between the outcome and the explanatory variable.
Firstly, since I aim to test my theory at the example of non-coethnic insurgents in order to isolate the effects of insurgency from the effects of coethnicity,Footnote 13 I discard from my measure all individuals who were foreigners and/or Jewish.Footnote 14 Furthermore, since the social networks of included individuals could still reflect the underlying networks even after the deletion, I limit my measure to insurgents who formed part of the French Forces of the Interior (FFI) and the French Fighting Forces (FFC). The former were the so-called maquis guerrillas, while the latter specialized in intelligence gathering, sabotage actions and evasion. I do so because FFI and FFC had the lowest proportions of Jews and foreigners among their ranks. Figure 3 presents the breakdown of French Résistance groups by nationality and ethnicity.
Figure 3. La Résistance Group Membership by Group Type, Nationality, and Ethnicity
Note: Left-hand-side bars in each group type denote foreigners and right-hand bars French insurgents. The final explanatory variable includes only the insurgents of the second and fourth columns from the left, FFI and FFC French, gentile insurgents (in red). FFI were maquis guerrillas and irregular volunteer soldiers. FFC were networks gathering intelligence, organising sabotage actions and evasion routes. Forces Fraçaises Libres (FFL) were members of De Gaulle’s regular army. Résistance Intérieure Française (RIF) were not insurgents, but people who due to some services for the resistance obtained military pension.
Secondly, the sole inclusion of the FFI and FFC insurgents addresses the second concern related to the potential post-treatment bias in my specifications. The measure of insurgent presence does not vary over time, e.g., by recruitment, and therefore could reflect the local group manpower in the post-deportation period. Fortunately, most of FFI guerillas were recruited after the establishment of the obligatory labour service (STO), between September 1942 and February 1943.Footnote 15 The same was true for the FFC groups which were constituted between 1942 and 1943 (Musée de la Résistance 2022). This precedes most of the deportations, alleviating concerns about the post-treatment bias in my findings.
Thus, the final measure consists of 255,658 insurgents (43% of the total).
Insurgent Command
To account for different incentives, resources and strategies for helping the Jews, I construct an indicator of each county’s military division of La Résistance (N = 26; AERI 2004).Footnote 16
Instrument
WWI Military Deaths
I derive the measure of WWI military deaths from Gay and Grosjean (Reference Gay and Grosjean2023), who utilize the extensive “Grand Mémorial” database of French soldiers honored with the title “Mort pour la France” (Mémoire des hommes 2014). This database covers 2.4 million individuals who served in the French army during WWI. They summarize statistics for those killed in action, extract birthplace information, and calculate WWI military death rates relative to the 1911 French census population figures. The final variable consists of 1,239,887 French combatant victims.Footnote 17
WWI Exposure
To address variations in soldiers’ exposure to different battles, which could impact their likelihood of death and affect the randomness of my instrument (explained further below), I document data on county belonging to military subdivisions from Gay (Reference Gay2021, N = 152).
Confounders
My quantitative empirical tests rely on the instrumental variable (IV) estimation method (see more detail in “Method” section below). While using causal inference tools normally alleviates concerns about the confounding problem, I gather a number of controls in order to satisfy the exclusion restriction assumption (see “Method: Exclusion Restriction” section below for more details). Controlling for the below-specified confounders serves to ensure that my instrument, WWI military casualties, affected Jews’ survival only through the insurgents’ presence and not through alternative backdoor channels relating to civilian propensity to betray or help insurgents and Jews alike.
Collaboration
Individual French collaborators were crucial in assisting German and Vichy states with conducting the antisemitic and anti-insurgent policies (Marrus and Paxton Reference Marrus and Paxton2019; Paxton Reference Paxton2001). Their presence affected the survival of the Jews and the ability of insurgents to form and operate. Higher WWI casualties decreased local propensity to collaborate, generating increased sympathy towards both insurgents and the Jews.Footnote 18 Therefore, controlling for collaboration captures the variation in Holocaust survival caused by WWI-stemming civilian attitudes towards the Nazi occupier and the resulting civilian actions concerning insurgents and Jews.
I create the measure of collaborators’ presence from Liste Paillole, a collection established by the former French Ministry of War, deposited at the Departmental Archive of Gironde in Bordeaux. The list comprises 96,492 presumed WWII collaborators compiled by the head of insurgent counter-intelligence Paul Paillole. Until the German occupation of the South of France in November 1942, Paillole worked for the Vichy government (Kitson Reference Kitson2019). Therefore, unlike traditionally used information from postwar trials, which contains information about higher-ranking officials only, these data are more representative of the actual collaboration phenomenon. The database includes information about each individual’s name, surname, place of birth or residence, and organizational belonging. To construct a measure of number of collaborators by county, I use information about place of birth or residenceFootnote 19 of collaborators engaged in organizations that were either overtly antisemitic (fascist political parties Francisme, Parti populaire français, Rassemblement national populaire and Mouvement social révolutionnaire) or were directly involved in tracking insurgents and Jews (the Milice, Milice’s precursor Service d’ordre legionnaire, Gestapo, Schutzstaffel, Sicherheitsdienst, and the German intelligence service). The final variable consists of 54,197 individuals.
Incumbent Control
German and Vichy territorial control enhanced incumbent’s ability to capture Jews and insurgents. Additionally, by projecting power, it steered local opinion in incumbent’s favor. I operationalize “incumbent control” (Kalyvas Reference Kalyvas2006) as the presence of offices and buildings of German and Vichy state institutions, capturing local civilians’ structural propensity to collaborate. Thus, the collaboration variable discussed above reflects only idiosyncratic pro-German sentiments.
I gather data from the “France under the occupation” collection at the Deutsches Historisches Institut in Paris—a register compiled from telephone directories, detailing 33,381 unique posts for each month-year throughout the war (Deutsches Historisches Institut 2021). I manually categorize entities into military, political, economic, religious, health, or cultural types, and cross-check them with the help of a research assistant.Footnote 20 I summarize information on political and military posts only. To mitigate post-treatment bias, main aggregate regressions incorporate the June 1942 post count, a month before the first deportations began (N = 8,939).
Ideology and Antisemitism
Local ideological leanings—inclusive attitudes and antisemitism alike—could have affected insurgent enlistment and civilian propensity for betraying or helping the Jews. To make sure that my main explanatory variable does not reflect the compound effect of local help by civilians and insurgents which would violate the assumption of exclusion restriction, I control in my specifications for prewar electoral results digitized by Cagé et al. (Reference Cagé, Dagorret, Grosjean and Jha2023; see also Cagé and Piketty Reference Cagé and Piketty2023). I follow their protocol of assignment of parties to left-right ideological spectrum.Footnote 21 I control separately for 1936 parliamentary turnout and vote shares. I include separately two variables meant to capture antisemitism specifically—i.e., votes for extreme right parties—the fascist Action Française in 1919 and Franciste party in 1936. Additionally, I collect open-source data on the distribution of Catholic churches in France.
Population Statistics and Administrative Conditions
To account for the mechanical relationship between population densities and victim statistics, all models include controls for total population and for the presence of synagogue (a dummy). I geocode data gathered for “Historic Synagogues of Europe” project (FJH 2010), which lists locations of 672 prewar synagogues in France. This measure captures places of prewar residence of French Jewish communities and thus reflects likely destinations of Central and Eastern European Jewish immigrants who came to France in the interwar period.
Administrative conditions in the first two years of the occupation determined incumbent and civilian strategies vis-à-vis insurgents and Jews, including civilian populations’ propensity for collaboration. They also shaped different fighting and survival strategies by insurgents and Jews. These could have persisted over the entire war, even after the occupation of Vichy in November 1942. To control for these time-invariant characteristics, all models include fixed effects of the five occupation zones: Belgian, German, Italian, Vichy, and the German-annexed Alsace-Lorraine.
METHOD
Given that La Résistance enlistment was not random, the local presence of insurgents was influenced by regional war dynamics and various social and geographical characteristics. The standard OLS approach may not reliably control for all possible confounders and crucial indicators are irretrievable after over 80 years (see “OLS Results” section below). Therefore, I employ the IV method to explore the relationship between insurgent presence and genocidal violence.
IV Instrument: WWI Deaths
I leverage WWI battlefield death statistics to capture random variation in WWII insurgent enlistment. I create county-level rates based on individuals’ origins, linking WWI-resulting anti-German and nationalist sentiment to descendants’ insurgent engagement. The logic of the connection is rooted in the psychological impact of indirect, community-level exposure to the war toll. Losses inflicted by an identifiable enemy contribute to later animosity and heightened nationalism (De Juan et al. Reference Juan, Alexander, Koos, Riaz and Tichelbaecker2024; Lupu and Peisakhin Reference Lupu and Peisakhin2017; Rozenas and Zhukov Reference Rozenas and Zhukov2019).
Given the brief direct hostilities between France and Germany in WWII, communities with anti-German and nationalist sentiments found an outlet for resistance in the underground. Assuming WWI soldier mortality was as-good-as-random and its resulting anti-German sentiment solely influenced the Holocaust through insurgent activity (more below), the community-level WWI death toll allows a random prediction of WWII insurgent enlistment. This approach facilitates an examination of how a heightened insurgent presence may have impacted Holocaust victimization.
Strong First Stage
To test the first assumption that the WWI death toll—the assignment—is a strong predictor of joining insurgency in WWII—the treatment—I estimate the following specification:
$$ \begin{array}{c}{\mathrm{Insurgents}}_{i(b,r)}={\gamma}_0+{\gamma}_1{\mathrm{WWIdeaths}}_{i(b,r)}\\ {}\hskip8em +{\gamma}^{\prime }{\mathrm{X}}_{i(b,r)}+{\tau}_{Z,Di}+{\upsilon}_{i(b,r)},& \end{array} $$
where my unit of analysis i are 1940 counties (canton; N = 2,912)Footnote 22 that belong (1) to military recruitment bureaus b listing soldiers for WWI efforts and (2) to Résistance military regions r leading La Résistance groups between 1940-1944. Counties are the second-smallest administrative units in censuses and the smallest units for which prewar GIS data are currently available.
$ {Insurgents}_{i(b,r)} $
is the outcome variable, the density of WWII insurgency, measured as the logarithm of the share of insurgents by their county place of birth i, normalized by 1936 population.
$ {WWIdeaths}_{i(b,r)} $
denotes the WWI soldier victimization rate, measured as the logarithm of the share of military personnel born in county i who died in WWI, normalized by 1911 population.
$ {\gamma}_1 $
is the main coefficient of interest. The vector
$ {X}_{i(b,r)} $
stands for a matrix of controls in line with the second-stage estimation (see below). To ensure spatial independence of observations and account for the impact of regional norms and political cultures on engagement in insurgency and helping, I include occupation zone and district (“arrondissement,” N = 283) fixed effects
$ {\tau}_{Z,Di} $
in each regression. I employ two-way clustering of errors at the level of the assignment (county military bureau b that determined level and character of exposure to WWI violence and strategy) and the treatment (county Résistance military region r that determined leadership strategy and the resulting incentives and resources for helping the Jews).Footnote
23
As evidenced by the result in Table 3 (columns 1 and 3), WWI military deaths are a strong predictor of later insurgent engagement. The p-values on the WWI death rates coefficient across all specifications are consistently below 0.01. The coefficients suggest that a 1% increase in the community-level WWI death rate increased WWII insurgent enlistment in that community by approximately 0.2%. The F-statistics range between 17 and 32. Wu-Hausman tests of endogeneity show p-values of below 0.05 in all models, which indicates that IV approach is preferable to OLS.
Ignorability
Secondly, the viability of my identification strategy hinges on the random assignment of WWI soldier deaths across the French territory. If variation in human losses followed a pattern, e.g., was related to the affected regiments’ deployment, the instrument would not allow me to claim orthogonality of the predicted variation in WWII insurgent recruitment. This in turn would preclude me from estimating the effect of each additional WWII insurgent on each additional Jewish death.
The engagement of the French male population in the WWI conflict was massive and its human cost unfathomable. The mobilization order from August 1, 1914 resulted in the immediate calling of 14 cohorts into their regiments, totalling approximately 3 million reservists. Following the costly Western Front battles, over the duration of the conflict France ended up conscripting men up to the age of 45. The French army suffered an estimated 6 million casualties, which constituted approximately 71% of the 8.8 million people called up to fight. As many as 1.3 million people, or 4.3% of the population, were killed during the conflict (Hart Reference Hart2013). The French army was mostly engaged on the fronts fighting Germany.
Due to the unparalleled conscription needs that the WWI conflict created, few eligible males were able to avoid active military service. Additionally, a regimental rotation system of troop deployment, aimed at ensuring re-strengthening and regular provision of equipment to the front and at tackling potential mutinies (Gilbert and Bernard Reference Gilbert and Bernard1959), meant that all regiments were at some point engaged in the main battles. This first conflict to mobilise all industrial powers’ resources against one another resulted in a long war of attrition, inflicting gigantic human losses among competing armies (Philpott Reference Philpott2015).
To assess the plausibility of the as-good-as-random military death occurrence, I conduct a series of covariate balance tests of the instrument on electoral, social and economic pre-WWI indicators.Footnote 24 Only 2 out of 23 covariates turn out to be statistically significant, a result that would be expected by statistical chance.
Exclusion Restriction
Thirdly, the exclusion assumption states that the instrument (French military deaths in WWI) had an effect on the outcome (murder of Jews in WWII) only through the variation it induced in the treatment (joining La Résistance). However, the very logic of the relationship between my instrument and my treatment, anti-German, nationalist sentiment induced by WWI deaths, calls this assumption into question.
My IV analysis relies on the logic that anti-German sentiments incentivized the populations in regions more affected by WWI deaths to become insurgents. That same sentiment however made other parts of those populations more supportive of insurgents. This introduces a challenge into the viability of my instrument, because it means that the instrument might have generated fewer Jewish deaths through that alternative channel, namely, the civilian population that was more supportive of the insurgents. In such a scenario, the coefficient could show a compound effect of insurgents and civilians either actively helping or simply not betraying Jews to the Germans. This would mean my exclusion restriction assumption is violated.
I deal with this challenge in a twofold way by controlling in my specifications (1) for the willingness of non-insurgent populations to betray the Jews (and insurgents) to the Germans, and (2) for their likelihood to help either group. Firstly, to capture civilian propensity for betrayal, I use a measure of the presence of collaborators who tracked down insurgents and Jews. Counties which were more impacted by WWI deaths and became, as a result, more anti-German and pro-insurgent would recruit less collaborators during WWII. I compile this measure based on a recently declassified wartime intelligence documentation, which includes anti-insurgent and antisemitic collaborators (see “Confounders: Collaboration” section above for a detailed description of the archival collection and the way I operationalize the measure). Secondly, since the collaboration phenomenon is a product not only of ideology, but also of survival maximization (Kalyvas Reference Kalyvas2006, 124–32), I control in my specifications for incumbent presence and different wartime occupation zones (see “Confounders: Incumbent Control” and “Confounders: Population Statistics and Administrative Conditions” sections above). Thirdly, to capture civilian propensity to help, in additional specifications I include controls of local ideological leanings (1936 electoral results), which crucially involve information about the distribution of communist and left-leaning attitudes. The electoral controls also capture antisemitic attitudes: vote shares for fascist Action Française (1919) and Francisme (1936) parties.Footnote 25 Fourthly, I gather data on the distribution of Catholic churches (OSM 2024). In sum, I collect 12 different measures to account for local pre-existing willingness to betray and help the Jews. Altogether, these ideological, religious and administrative controls account for the variation in the rates of Holocaust survival that were induced by WWI military deaths, but caused not directly by insurgents but the rest of population.
I then conduct tests to check whether WWI military deaths affected collaborator recruitment and vote choice in interwar elections (1919, 1924, 1932, and 1936).Footnote 26 I find that, in two elections, the WWI soldier deaths are positively correlated with voting for right-wing parties, and negatively correlated both with voting for left-wing parties, and with joining collaborator ranks in WWII. This test makes evident the importance of controlling for ideological leanings in my models.
In an additional placebo test, I analyze reduced-form specifications of Jewish victimization on WWI military death rates, focusing on counties where insurgency did not form (approximately 25% of all counties). If assistance operated through alternative channels, such as pro-Jewish civilians, WWI deaths should be linked to reduced victimization in those regions. The results reveal that in counties with no or low insurgency, the WWI coefficient is not statistically significant.Footnote 27 To address concerns about statistical power of this test, I randomly draw 25% of the counties 1,500 times and rerun the main specification.Footnote 28 Almost 90% of these randomly-drawn samples yield statistically significant results, indicating that the null reduced-form result in non-insurgent areas is not due to lack of statistical power. The results mean that those counties where insurgency did not form did not see decreased Holocaust violence severity even in the presence of high WWI victimization.
Second Stage
In this step, I estimate variations of the following second-stage specification:
$$ \begin{array}{rl}{Y}_{i(b,r)}={\beta}_0+{\beta}_1{\hat{\mathrm{Insurgents}}}_{i(b,r)}+{\beta}^{\prime }{\mathrm{X}}_{i(b,r)}+{\varphi}_{Z,Di}+{\varepsilon}_{i(b,r)},& \end{array} $$
where
$ {Y}_{i(b,r)} $
is the outcome variable, the Holocaust victimization, measured as the logarithm of the share of Holocaust victims by 1941 Jewish population in county i.
Footnote
29
$ {\hat{Insurgents}}_{i(b,r)} $
is the share of insurgents by county i predicted by the share of WWI military deaths in that county.
$ {\beta}_1 $
is the main coefficient of interest. The vector
$ {X}_{i(b,r)} $
stands for a matrix of pre-treatment and wartime covariates discussed in the “Confounders” section above. Just like in the first stage, I include occupation zone and district fixed effects
$ {\varphi}_{Z,Di} $
in each regression and cluster the errors at the level of the instrument (military bureaus b) and the treatment (Résistance regions r). I calculate and report the Moran statistic for every specification.
OLS RESULTS
Before approaching the results of the introduced IV approach, I briefly discuss the results of ordinary least squares (OLS) models (Table 2). I employ specifications that are equivalent to the second stage of the IV.
Table 2. OLS Results
Note: *
$ p<0.1 $
, **
$ p<0.05 $
, ***
$ p<0.01 $
. Robust standard errors in parentheses.
Table 3. Main Results. Dependent Variable: Proportion Holocaust Victims (Logged)
Note: *
$ p<0.1 $
, **
$ p<0.05 $
, ***
$ p<0.01 $
.
The OLS coefficients turn out statistically insignificant for two most likely reasons. Firstly, because of the omitted variable bias. Factors such as occupier strategy, economic conditions, preexisting social networks, local animosities, and private attitudes towards Nazi Germany and the Vichy government all played a role in observed insurgent enlistment and Shoah victimization. They are therefore likely to interfere in the estimation of the independent effect of insurgent presence. Another unobserved confounder is the spillover of the anti-insurgent violence onto the Jews in the times of counterinsurgency/reprisal campaigns, which may cancel out the “helpful” effects of insurgent presence. (Note that the coefficients oscillate around 0.) Secondly, unlike my IV strategy which is more likely to isolate the effects of the “morally”-motivated insurgents, the raw measure includes also those insurgents who joined La Résistance groups for non-ideological reasons (Kalyvas Reference Kalyvas2006, 44–6). This means that, unlike my instrumented measure of insurgent presence, the OLS measure, and specifically the “non-ideological insurgents” in it, might dampen the effect of the insurgents who have more latent inclination to help.
IV RESULTS
Across all second-stage models, I find a negative, statistically significant relationship between insurgent presence and Holocaust victimization (Table 3, columns 2 and 4; see also additional tables in Appendix section A). The magnitude of the estimates suggests that a 1% increase in the density of insurgents decreased the share of the victims by between 0.82 and 1.2%, which corresponds to a 0.2 standard deviation change. In a county with the mean number of 75 Jews, this would correspond to one more person being saved. On the national scale, this indicates that ultimately between 1,754 and 1,782 people less would fall victim to the Holocaust.Footnote 30 This result is statistically and substantively similar to an alternative specification, where the dependent variable is coded as a count of victims (Table A2 in the Dataverse Appendix). This estimate is plausible for three reasons. First, because insurgents were only a minority of the population (around 1–3%), so even a 1% increase in their number is considerable. Second, because for every saved person many more were helped. Third, because the substantive effect is comparable with the total number of 1,210 Jews who were rescued by insurgent French Righteous. Given the bias in favor of successful, altruistic rescue attempts in the Righteous data, the true number of assisted Jews must have been higher.
Importantly however, the victim statistics alone do not enable us to comment on the type and extent of interactions between insurgents and Jews. I tackle these questions in the following section.
HOW AND WHY DID LA RÉSISTANCE HELP THE JEWS? PROCESS TRACING OF TYPICAL AND DEVIANT PROVINCESFootnote 31
To uncover the causal mechanisms underpinning the established relationship, I employ process tracing of six cases, three “typical” and three “deviant” provinces, identified using my main results (Dunning Reference Dunning2012, 222–4; Seawright Reference Seawright2016). As per Table 4, I choose cases with different treatment types and contrasting treatment values to uncover potential differences in how the treatment operated (through different mechanisms) which, in turn, is meant to formulate scope conditions of my theory (Lyall Reference Lyall2015, 199–203). I zoom in on provinces with low insurgent density (Gironde and Somme), high left-wing insurgent density (Dordogne and Creuse) and high right-wing insurgent density (Vosges and Doubs). I analyze evidence from survivors’ and rescuers’ testimonies, as well as secondary sources on French insurgency. With regards to the primary sources, I draw on (1) the collections of survivor testimonies housed in the USC Shoah Foundation’s “Visual History Archive” (VHA) and on (2) the rescuer testimonies from Yad Vashem Center’s “Righteous Among the Nations Database” (YV).
Table 4. Overview of Selected Cases
The VHA project was founded in 1994 by Steven Spielberg with the revenues from Schindler’s List, a drama film about the rescue efforts of a German industrialist. The VHA library contains, i.a., over 50,000 interviews with Shoah survivors and witnesses from all over the world. The video testimonies span 116,000 hours, or the equivalent of 13 years of recordings.
I proceed in two steps. Firstly, using VHA’s geographic indexing, I select a sample of people who mention in their testimonies a locality in at least one of the six selected provinces. Secondly, I proceed to coding the interviews. I look for general context in which the province is mentioned and for any remarks about insurgency. If the survivor mentions insurgency, I write down the entire passages for further interpretation.
The YV database is an inventory of non-Jews who have been recognized as rescuers of Jews in the Holocaust (Gutman et al. Reference Gutman, Bender, Lazare, Michman, Flim, Krakowski and Mikhman2005). The “Righteous” titles have been awarded—also posthumously—since 1963 to altruistic individuals who did not expect material benefits in return for rescuing. The database consists of 15,233 rescuer households (46,400 people) and spans entire Europe. The rescue activities (“rescue modes”) that qualify for the award include hiding, arrangement of shelter, forgery of documents, and organization of evasion.
I construct two datasets: one for France (2,455 households), and one for the rest of Europe. I code key information about each household (rescue modes, number of rescuers in household, rescuers’ fate, and number of saved Jews). Moreover, I read and manually classify each individual household as either consisting of insurgent individuals (“resister”; N = 1,825), those who were in contact with insurgents, but were not insurgents themselves (“contact”; N = 2,028), and regular civilians (“civilian”; N = 11,447).Footnote 32 Lastly, from among the rescuers who are thus categorized as insurgents and “civilians in contact with insurgents,” for further analysis, I note down the passages which feature insurgency.
I contrast the VHA and YV results with secondary historical accounts and other primary resources, where available. One such additional source are the interviews with civilians and insurgents from Dordogne (Département de la Dordogne 2009).
I analyze those qualitative data to uncover (1) the types of interactions between the insurgents and genocide targets (the “how”: Do insurgents and Jews interact? If yes, what do they do?) and (2) the reasons for them to engage in these actions (the “why”: Do insurgents/Jews mention why they interact?). To answer this last question, I look for evidence supporting the motivations discussed in the theory section (see Table 1) (Bennett and Checkel Reference Bennett and Checkel2012, 17–31). In the absence of explicit motivation, I consider help performed against benefits as “material” (e.g., smuggling against payment) and help not accompanied by benefits as “moral” (e.g., smuggling “for free”).
The “How”
Three main help modes, which I call “information,” “network,” and “skill,” explain how insurgents contributed to Jews’ strive for survival in Creuse, Dordogne, and Doubs—three provinces with highest occurrence of help (see Table 5 for an overview of the results). Each help mode relates to rebellion technology the insurgents developed to fight the Nazis. Insurgents (1) warned the Jews of impending raids (information); (2) connected them with civilians who were ready to help (network); and (3) forged their documents and helped smuggle them across internal and external borders (skills).Footnote 33
Table 5. Contexts in Which Insurgency Is Mentioned in Survivors’ (USC) and Rescuers’ (YV) Testimonies
Note:
$ {}^{\mathrm{a}}\hskip0.3em $
+ up to 32 via the Circuit Garel network.
Information: Insurgents Warn Jews of Impending Raids
Warnings by insurgents announcing upcoming raids are one of the most often mentioned ways in which Jews were assisted. Twelve survivors (17%) in Creuse and ten in Dordogne (13%) explicitly mention receiving such repeated warnings. The warnings allowed the Jews either to hide for a short time spanning a few hours to a few days, or flee the given locality altogether. For example, sisters Annette Thau and Berthe Prasquer, who hid in a small town Fresselines (Creuse), mention being urged by the local insurgents to hide in the nearby forest. Denise Attia, hiding in Anzême (Creuse), evokes a similar memory,
The maquis would come and warn us and say, “We think the Germans are going to come and have a look around here.” (…) I didn’t know how they knew (…) they came to warn the Jewish families who were there in Anzême, so we had to hide. (Tape 2; 23:54)
In Dordogne, Esther Bendeck explains how the partisans would keep an eye on the Jewish refugees in Saint-Pierre-de-Chignac, “The maquis knew we were there. (…) It was the maquis who came to tell us, ‘You stay quiet, you don’t move, because if you go down, they’ll round you up straight away”’ (Tape 2; 14:42, 18:40). Similarly, David Ruzié hiding in Périgueux recounts, “I escaped a roundup [in February or March 1944] thanks to the help of a neighbour who was (…) involved in the Resistance” (Tape 1; 18:38). Similar accounts are found among the insurgent Righteous Among the Nations such as Emilie and Pierre Blanchet who hid two Jews in a small town of Felletin (Creuse): “Pierre Blanchet, an activist in the French underground, warned Odette, her mother, and other Jews he knew, of impending raids by French police in the area.”
Network: Insurgents Recruit Civilian Helpers
Each survival attempt in wartime France involved looking for people ready to offer help. This was especially true in urban centres, where helpers were only able to shelter people for short periods and an average Jew was helped by dozens of individuals (Düring Reference Düring2016). Such indispensable points of contact were provided by networked individuals.
Nine survivors in both Creuse and Dordogne samples (13% and 12%, respectively) as well as one survivor in each Doubs and Vosges mention how insurgents would ask civilians they knew and trusted whether they would help shelter Jews. For example, survivor Annette Thau explains how her family found their six consecutive hideouts: “Docteur Lapine and Delbecque from Guéret, both members of Resistance, found a variety of places for my sister and me to hide [around Creuse]” (Tape 1; 9:39). In Périgueux (Dordogne), Eric Sommer remembers how the maquis helped him, his sister and a friend of his to find hiding places among civilians and clergy.
Additionally, using the YV data, I conduct negative binomial regressions and ANOVA tests of the number of helped people on the rescuer status.Footnote 34 I find strong evidence that households of Jews’ rescuers that had links to the French underground (“contact”) managed to save on average more lives than other rescuer groups (“insurgent” and “regular civilian” rescuers). As evidenced in Table 6, rescuers in contact with insurgents helped on average one more person than other rescuer groups. This finding would lend further validity to the theory by Varese and Yaish (Reference Varese and Yaish2000) about the “rescue market” dilemma faced by the potential rescuers: while some people want to help, they do not always know how and whom. In such a market, the insurgents act as trusted mediators between the potential rescuers and the rescuees in need.
Table 6. Rescuers and Rescuees in Righteous Households
Note: One-way ANOVA test shows there is a significant difference between the groups (p-value: 0.008) and the Tukey test of multiple comparison of means between the three groups shows a statistically significant difference of 0.9 (p-value: 0.006) between the average number of Jews helped by the civilians who were connected with insurgents and those civilians who were acting on their own. The corresponding tests of the differences in means of average number of rescuers per household indicate that there were statistically significantly less rescuers per each insurgent versus civilian household (one-way ANOVA p-value: 0.006; Tukey test’s p-value: 0.004).
Lastly, I explore how—by performing what activities—each of the identified rescuer groups helped Jews. The results of linear probability models in Figure 4 indicate that insurgent rescuers were more likely than the other two rescuer groups to recruit other rescuers (“finding shelter”). Compared to a civilian rescuer, both insurgent rescuers and “contact” rescuers were more likely to do it (by 13% and 7%, respectively). Notably, insurgent rescuers were however 22% less likely to shelter Jews themselves, which is routinely explained in the testimonies by a heightened risk of discovery the two groups ran.
Figure 4. Linear Probability Estimates of Likelihood of Performing Specific Rescue Activities, by Rescuer Profile
Note: Civilians without links to insurgents are baseline for comparison. The coefficient lines are 90% and 95% confidence intervals with robust standard errors. Units of analysis are individual YV Righteous rescuers from France (N = 2,466). See Tables A9–A12 in the Dataverse repository for results in table format.
Skills: Insurgents Forge Documents and Organize Evasion
While the hiding of Jews was an integral part of the survival facilitation activities, other help modes such as document forgery and organization of evasion were at least equally important to ensure Jews’ survival. Those latter activities were performed most often by specialized Jewish and non-Jewish networks.
Three survivors from Creuse (4%), nine from Dordogne (12%) and two in Vosges (7%) mention having had their documents falsified by members of the French insurgency. Jacqueline Birn, Manuela Bornstein and their parents were smuggled in August 1942 from Paris to Le Got/Mazeyrolles (Dordogne) by two resisters. The town’s mayor, also a resister, warmly received them, provided fake IDs for their parents, and secured a farming job for their father. They lived there until 1945. Among the rescue accounts in YV database, ten insurgent rescuers forged papers for their rescuees (six in Dordogne, two in Gironde, one in Creuse, and one in Vosges). For example, Marie-Thérèse Goumy, a town clerk during the war in Crocq (Creuse), “[t]hrough her connections with the Resistance, (…) provided false papers for a number of the Jews in danger [in Creuse], as well as for their families remaining in Paris.”
Mentions of cross-border evasion networks are also extremely common in the testimonies. Thirty-two people in Creuse (45%), nine in Dordogne (12%), thirteen in Doubs (25%) and one in Vosges (3%) were smuggled either through the demarcation line or the Spanish or Swiss border. For example, Colette Kleinberg recounts how, after both of her parents got arrested and deported to Auschwitz in September 1942, the local partisans smuggled her and her sister from Saint-Antoine-Cumond (Dordogne) to Indre province to rejoin their grandfather. Among the YV accounts, five insurgent rescuers organized their rescuees’ evasion (two in Creuse, two in Dordogne, one in Vosges).
One of the most noteworthy escape rescue mission was organized by a strictly Jewish organization from Creuse, Oeuvre de Secours aux Enfants (OSE), in collaboration with La Résistance insurgents and their contacts. In the second half of 1942, OSE contacted Georges Garel, a Jewish member of the Communist Combat insurgency and charged him with the mission of dispersing all children under their auspices away from Creuse (Serge Klarsfeld, Tape 2, 06:00; Georges Loinger, Tape 5, 10:00). The OSE directors had learned about the extermination camps and worried that big groups of Jewish children would soon attract the state’s attention. In a matter of months, the “Circuit Garel” managed to rehouse across France or evacuate through the Swiss border at least 1,600 children (Poznanski Reference Poznanski1995, 143).Footnote 35 Circuit Garel relied on non-Jewish resistance groups for contact with external, non-Jewish institutions and people (Garel Reference Garel2012, 94, 128, 155) and sometimes for the forgery of documents (Garel Reference Garel2012, 92). La Résistance also facilitated the hiding in Creuse (Garel Reference Garel2012, 118). Thus, Righteous Jean-Baptiste Robert—a corporal in the 1914-1918 War and the director of La Souterraine school (Creuse)—hid Jewish children via the Garel network.Footnote 36
In line with these qualitative findings, the above-mentioned linear probability analysis of all French Righteous rescuers (Figure 4) shows that the activities of the two civilian rescuer groups were different from those of the insurgent rescuers. As compared to civilian rescuers, insurgent rescuers were 20% more likely to help the Jews forge documents and 15% more likely to organize their evasion.
The “Why”
The in-depth study of the six cases reveals that survivors in provinces with majority left-leaning insurgent groups were more aware of and reliant on insurgent help (see Table 5). In those provinces, 49-62% of survivors mention La Résistance (vs. 28-32% in provinces with right-leaning groups and 10-17% in provinces with low insurgent density). Left-leaning provinces and the “typical” right-leaning province show the highest incidence of helping. This higher incidence of helping correlates with more “moral” motivations. “Material” motivations are less common in provinces with left-leaning groups, and so are the “moral” motivations in provinces with right-leaning groups. In the following sub-sections, I quote the most typical motivations. In the Dataverse Appendix Section H I report full results and describe each province.
“Moral” Motivations Among Left-Leaning Groups
Among the survivors who mention being helped by insurgents in Creuse and Dordogne (102 people), 37 explicitly ascribe to them “moral” motivations (see Table 5). However, the other 65 are helped for free and in ways that equally point towards the “moral” interpretation.Footnote 37 Moreover, relatively many Righteous rescuers belong to insurgent groups (4 rescuers or 11% in Creuse and 12 or 19% in Dordogne).Footnote 38
For example, three survivors in Dordogne and one in Creuse mention the antipathy towards the Germans as a reason for engaging in helping activities (Table 1, “structural balance”). Survivor Pierre Chanover, hidden by underground members Madame Rosier and her son in Sainte-Foy-la-Grande (Dordogne), speculates that the sentiment was provoked by the experience of WWI, “Madame Rosier’s husband died in WWI—was killed in the war. And she had no love for the Germans. And the actions that she took [had] always something to do against the Germans” (Tape 3). A further 14 people in Dordogne and 37 in Creuse are helped in ways which clearly do not involve material benefits. For example, at the arrival of Gisele Israel (maiden name “Dahan”) and her family in Creuse, a police officer refuses to register them as Jewish:
We went to register [on arriving from Paris] (…) [, but] we never got a “Jewish” stamp, because the police officer in Guéret who was in the Resistance said, “Your name doesn’t tell me anything, I won’t stamp you.” It was Monsieur Saintoire whose family helped us a lot, they were very, very good people. (Tape 3; 16:30)
“Material” Motivations Among Right-leaning Groups
Among the survivors who mentioned being helped by insurgents in Doubs and Vosges (18 people), nine identified “material” motivations. Moreover, further five were helped in ways which point to “material” incentives due to the character of provided help. No one in the sample from either province was hidden over a longer period or warned of a raid. Only one Righteous rescuer in each sample belonged to an insurgent group.Footnote 39 On the whole, purely material motivations play an important role in assistance, but seem less reliable in providing longer-term support (see also Brethour Reference Brethour2019; Tec Reference Tec2013, 106–7).
For example, in Doubs, all Jews who were helped were either smuggled (13 or 25% of the sample) or contacted with smugglers (1 person or 2%). Four specified paying for the service. Frederique Epstein from Besançon recounts: “We left via Dole, the zone that was still occupied. There were SNCF smugglers working with the maquis” (Tape 1; 23:08). Her sister, Suzanne Gibersztajn, specified that their mother paid 15,000 French francs for the journey.Footnote 40 In Vosges, comparatively less people were helped: two (7%) received false documents, one was smuggled and one was connected with a trustworthy civilian helper.
Scope Conditions
The process tracing results suggest two scope conditions of the theory. Firstly, insurgent assistance to the persecuted is most likely to be driven by ideologically inclusive groups.Footnote 41 This interpretation could explain why provinces considered “deviant,” with high insurgent density and high victimization rates, exhibited more extreme but directionally similar behavior to their ideologically “typical” counterparts that had lower victimization rates. The insurgents in the “deviant” left-wing province, Dordogne, assisted the Jews even more than insurgents in the “typical” left-wing Creuse. Conversely, the insurgents in the “deviant” right-wing province, Vosges, assisted the Jews less than insurgents in the “typical” Doubs. Secondly, strategic targeting by the state might lessen the effects of insurgent help in the short term, which makes the theory more applicable to longer-unfolding genocides. The studied “deviant” provinces were exposed to higher levels of persecution of partisans and Jews, and to indiscriminate violence against civilians during the volatile beginning and end of the occupation (1940 and 1944). I discuss these scope conditions in detail and contrast them with evidence from Poland, Rwanda and the Ottoman Empire in Section I of the Dataverse Appendix.
CONCLUSION
This article re-examined the relationship between insurgency and civilian victimization in genocide, challenging the common assumption that insurgents uniformly endanger all civilian populations. Although vulnerable at moments of increased state targeting, civilian genocide targets benefit from insurgents’ support at other times. By shining the spotlight on the overlooked dynamics of co-habitation and alliances, the study shifts focus from the state’s perspective to highlight the strategies and decisions of genocide targets and insurgents. Rather than viewing résistants and Jews solely as objects of state strategy, the article demonstrates how insurgents develop a toolkit that can be transformed into an assistance repertoire in support of genocide targets.
The main finding of this article is likely to travel to other countries of occupied Europe and other genocides where resistance groups harbored inclusive ideologies, or where they were coethnics of the targeted group. In an additional analysis of YV testimonies from the whole of Europe (N = 15,232) I find that, similarly to France, insurgent rescuers in most countries tended to use insurgent skills to help the Jews (they were less involved in hiding, but more involved in document forgery and organization of evasion than civilian helpers). Moreover, they tended to help on average more people than civilian rescuers.Footnote 42
Yet, how do we reconcile the above with the widespread accusation that anti-Nazi resistance movements across Europe did not do enough to counteract the genocide they witnessed? Two reasons might contribute to this view. One is related to the constraints insurgent leadership faced and the other to macro-level bias in historical studies of anti-Nazi insurgencies.
Firstly, strategic military considerations among insurgent leadership can limit the strength of the theorized inclination for helping. Unless they have a “reputational” motivation to help, insurgent command prioritizes military goals and seeks popular support, often avoiding open assistance that could jeopardize their mission by attracting unwanted state attention, or by antagonising parts of the population who hold negative outgroup attitudes. A case example I came across involved a high-ranking insurgent, Daniel Cordier—the private secretary to President of the National Council of La Résistance Jean Moulin—who did not want to “endanger his functions” and the entire organization by sheltering a Jew. He agreed to do it for only one night after a colleague urged him by saying that “the Jews are even harder to place than the airmen” (Cordier Reference Cordier2009, 440–1). This illustrates that observed effects reflect individual insurgents’ decision-making. The limited macro-level involvement both restricts potential assistance and may discourage individual members from helping. This also explains the small size of the estimated effects (see Section “IV Results”). The total IV estimate of 1,944 lives corresponds to only about 2% of all Jews who perished in the Holocaust on the French soil. In other words, insurgents’ potential for saving lives out of their own initiative seems limited without their leadership’s or international backing. Understandably, Jews and concerned witnesses during and after the war have therefore both expected help from insurgents and deplored its actual, limited extent.
Secondly, the micro-level character of assistance obfuscates the true extent of the helping phenomenon. One, because reliable data about clandestine acts of help in war contexts are notoriously difficult to gather. Two, because historical studies do not use methodological tools that enable the summary of large amounts of micro-level data and conclusions about average effects. In any case, historians are not concerned with average effects or generalizability, but with establishing facts and their causes in their specific context (Sewell Reference Sewell2005). Given that macro and micro-level conflict dynamics are inherently different (Kalyvas Reference Kalyvas2006, 10–1, 364–76), it is therefore difficult for them to integrate and reconcile these opposing dynamics. This might explain why previous history scholarship on La Résistance concluded that the help to the Jews, if any, was only minimal (Poznanski Reference Poznanski2008; Reference Poznanski2018) even though positive accounts about La Résistance abound in individual survivors’ testimonies.
This article sheds light on the role of insurgency within the studies of genocide by pointing to an important source of variation in the observed survival rates. It helps us formulate expectations about which armed groups are likely to help. It also contributes to the study of genocidal violence within studies of (counter)insurgency. In contrast to these studies’ focus on counterinsurgency campaigns, it demonstrates that in longer-spanning conflicts civilians who need to hide and evade will benefit from the proximity with insurgents. Lastly, it invites future studies on war alliances to assess whether in conflicts with a shared, stronger incumbent enemy, the relatively weaker actors devise collaborative strategies based on “moral” motivations and whether such alliances are more durable than the strictly “materially” motivated ones.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424001345.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/UBKN7E. Limitations on data availability are discussed in the appendix and the readme file.
Acknowledgments
I would like to thank the four reviewers, Robert Braun, Elias Dinas, Stathis Kalyvas, and Scott Straus for multiple rounds of exceptional feedback and encouragement. Other people who contributed to this work and to whom I am deeply grateful include Lars-Erik Cederman, Volha Charnysh, Jeffrey Checkel, Philippe Douroux, Thad Dunning, Vicky Fouka, Nerea Gándara Guerra, Miriam Golden, Anna Gromada, Alba Huidobro, Mark Kayser, Joe Kendall, Otto Kienitz, Jeffrey Kopstein, Krzysztof Krakowski, Sebastián Lavezzolo, Sandra León, Sergi Martínez, Kasia Mikołajczak, Joanna Nalewajko-Kulikov, Marta Ratajczyk, David Su, Vicente Valentim, Jason Wittenberg, Maciej Wzorek, the participants of the Center on the Politics of Development at UC, Berkeley, and of the Political Behaviour Colloquium and the Sparks Seminar at the European University Institute.
FUNDING STATEMENT
This research was funded by the Ph.D. grant provided by the Polish Ministry of National Education and the European University Institute.
CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The author affirms this research did not involve human participants.
Comments
No Comments have been published for this article.