Camps and Cognitive Dissonance

The prominence of these camps as symbols of out-group hatred helped legitimize such hatred and make it socially acceptable. We argue that, at the individual level, this occurred through the process of cognitive dissonance. Festinger (Reference Festinger1957) defines cognitive dissonance as the process by which an individual rationalizes new information that is inconsistent with his or her prior beliefs to reduce psychological discomfort stemming from the inconsistency.

The effects of cognitive dissonance on attitude change have been well documented in the literature on social psychology. Engaging in or witnessing violence against an individual or a group can breed negativity toward that individual or group (e.g., Davis and Jones Reference Davis and Jones1960). Similarly, acknowledging that one’s in-group has victimized an out-group can increase prejudice against that out-group (Imhoff and Banse Reference Imhoff and Banse2009). Cognitive dissonance, thus, serves as a likely mechanism for why ethnic or racial divisions and prejudice can be socially constructed from exposure to violence (Acharya, Blackwell, and Sen Reference Acharya, Blackwell and Sen2018; Fearon and Laitin Reference Fearon and Laitin2000; Hadzic, Carlson, and Tavits Reference Hadzic, Carlson and Tavits2017).

Applying cognitive dissonance theory to exposure to concentration camps suggests that individuals with relatively tolerant views of out-groups are likely to be confronted with psychological discomfort when living close to a camp. This occurs because the camp provides new information, discordant with the individuals’ prior beliefs, that out-group members are subhumans and can be mistreated. Although eliminating the camp is not an option for the individuals, one way in which they can reduce the unpleasant feeling of dissonance in this situation is to change their beliefs about out-groups. Local Germans had to rationalize the discrimination, enslavement, violence, and other inhumane treatment of people held in these camps, even if some of them had previously been their neighbors. A way to do so was to change individuals’ beliefs about the prisoners in these camps, to accept their status as out-groups, subhumans, and not worthy of the same rights. Some of this rationalization was necessary no matter where in the Third Reich an individual lived (Voigtländer and Voth Reference Voigtländer and Voth2015). However, as we argued above, the key difference that we capture in our study is that Germans who resided near concentration camps had to rationalize a more extreme example of intolerance than other Germans.^{Footnote 5} This enhanced rationalization effort led those living near the camps to maintain higher levels of out-group intolerance.^{Footnote 6}

The Persistence of Political Attitudes

Various studies have demonstrated that attitudes can persist historically through an institutional channel, continued communal interaction with out-groups, or cultural transmission (see, for example, Acharya, Blackwell, and Sen Reference Acharya, Blackwell and Sen2016a; Mazumder Reference Mazumder2018; Nunn and Wantchekon Reference Nunn and Wantchekon2011; Voigtländer and Voth Reference Voigtländer and Voth2012). Studies of legacies of American slavery, for example, show that Southern institutions such as Jim Crow helped sustain racism even after slavery was abolished (Acharya, Blackwell, and Sen Reference Acharya, Blackwell and Sen2016a). In our case, the institution (i.e., concentration camp) was removed together with the Nazi regime and was not replaced with any alternative institution reinforcing out-group subjugation. This rules out the institutional channel. Other studies show that continued (economic) competition between locals and out-groups (e.g., Jews and racial minorities) generates a communal bond against out-groups and helps sustain attitudes over time (e.g., Acharya, Blackwell, and Sen Reference Acharya, Blackwell and Sen2016a; Charnysh Reference Charnysh2015; Grosfeld, Rodnyansky, and Zhuravskaya Reference Grosfeld, Rodnyansky and Zhuravskaya2013). This presumes that after the removal of the institution out-groups remain in the community. In our case, the prison population was removed from the local area when the institution was erased, and there was no competition between residents who lived closer to camps and the prisoners of these camps. This makes transmission due to continued interaction and competition also an unlikely mechanism.

Rather, we argue that individuals transmitted their attitudes via family ties and social interactions, which led to the differences in out-group intolerance measurable even today. A meta-analysis of 60 years of research confirms that children’s and parents’ out-group attitudes overlap to a significant degree (Degner and Dalege Reference Degner and Dalege2013), and at least some scholars attribute these similarities to intergenerational transmission of ideology rather than to other shared characteristics among family members (e.g., Duriez and Soenens Reference Duriez and Soenens2009).^{Footnote 7}

Children acquire preferences from parents (and local peers) through altered social memory and by adaptation and imitation (Bisin and Verdier Reference Bisin and Verdier2000). Not necessarily aware that they are doing so, individuals pass on their belief systems to others in various contexts but especially through emotional and practical ties and relationships among generations. Furthermore, individuals tend to rely largely on the messages passed intergenerationally instead of history’s footprint (Anderlini, Gerardi, and Lagunoff Reference Anderlini, Gerardi and Lagunoff2010). This explains why anti–out-group sentiments can persist despite the fact that the truth about the atrocities committed at concentration camps was made public. Observers of post-war Germany often point out that for most people “education about the war begins at home” (Cowell Reference Cowell1995) because almost every German family has its own “complicated personal war history,” which is transmitted to subsequent generations (Sontheimer Reference Sontheimer2005). Parents’ views and expressed opinions about out-groups become part of that education. As a German educator put it, when children use the word “Jew” to give someone a bad name (despite having learned about the Holocaust at school), it often reflects the insensitivities they have obtained from their parents.^{Footnote 8}

In sum, we argue that through cognitive dissonance individuals rationalized the imprisonment and inhumane treatment of non-ethnic Germans. This process inflated out-group hatred closer to the camps. The out-group animosity was later transmitted through family and communal ties from one generation to the next, leading to the persistence of out-group prejudice long after the racist institution was erased. The observable implication of this argument is that the closer an individual lives to the site of a Nazi-era concentration camp today, the higher his or her expressed level of out-group intolerance. We test this hypothesis in the following sections.

Preexisting Attitudes and the Location of Camps in Germany

Our theoretical argument and empirical approach rely on the assumption that the decisions regarding the location of concentration camps in Germany were unrelated to pre-existing mass political attitudes. The historiography of Nazi Germany reviewed above is in line with this argument. In this section, we complement the historical accounts with systematic empirical tests of whether preexisting political attitudes explain camp locations using the pre-war Census and electoral data.

We measure preexisting political attitudes for each of the 946 German electoral districts in 1933 in two ways. First, we use early support for the Nazi party to capture anti-Jewish sentiment and political intolerance more broadly. Second, and in line with the literature on group conflict, we use data on the relative size of Jewish communities at the time to measure geographic variation in the levels of out-group threat. Electoral and Census data come from Falter, Lindenberger, and Schumann (Reference Falter, Lindenberger and Schumann2009) and Hänisch (Reference Hänisch1989); both variables are measured in proportions and range from 0 to 1.

Our outcome of interest here is the location of the concentration camps. Since the unit of analysis is the electoral district, we operationalize camp location in two different ways: (1) as the Euclidean distance between the centroid of a given electoral district and the closest Nazi camp, and (2) as a binary measure that takes the value of 1 if a camp was established in the district, and 0 otherwise.^{Footnote 14}

Based on these measures, Table 1 presents the results of four models that assess whether preexisting political attitudes explain concentration camp location (see Table SI2.1 for the full models). Columns 1 and 2 report estimates of OLS models with the distance to the closest camp (in 10 km) as the outcome variable, whereas Columns 3 and 4 present the results of logistic regressions with the binary measure of camp presence as the outcome variable.^{Footnote 15} Moreover, Models 2 and 4 extend the baseline specification by including additional socioeconomic covariates. Based on recent scholarship on the counter-extremist efforts of the Catholic Church during the rise of the Nazi party (Spenkuch and Tillmann Reference Spenkuch and Tillmann2018), we account for the religious composition of the districts. In addition, unemployment rates and district size capture the regional economic environment.

TABLE 1.

Interwar Political Attitudes and Camp Location in Germany

Note: Entries are coefficient estimates for the regression of Distance to Closest Camp (Columns 1 and 2), and Pr(Camp = 1) (Columns 3 and 4) on support for the Nazi party, Jewish presence in the district, and additional controls (standard errors in parentheses). Full models in Table SI2.1. * p < 0.05; ** p < 0.01.

Overall, the models show that district-level anti-Jewish sentiment is not a meaningful predictor of camp location. As an example, the coefficient for Nazi party support in Model 2 is indistinguishable from zero (−0.79; p-value = 0.56) and the point estimate suggests that a hypothetical increase from the lowest to the highest vote share obtained by the Nazi party in 1933 (from 13.3 to 83.0%) is associated with a decrease of Distance to Camp by only 5.5 km (≈3.4 miles).^{Footnote 16} In Model 1, which excludes the socioeconomic control variables, the coefficient flips sign (while remaining indistinguishable from zero at conventional levels). For the models predicting whether a given district has a camp, the coefficients for Nazi party support are also not statistically significant. Finally, the size of the Jewish community in a given district does not seem to predict the existence of a camp in any given district, but is positively associated with distance to the closest concentration camp. This result is in the opposite direction of what we would expect if the camps had been strategically placed close to Jewish communities.

We performed two further robustness tests. First, we used a dataset compiled by Voigtländer and Voth (Reference Voigtländer and Voth2012) with city-level information on the number of pogroms during the 1920s and in 1349, along with letters to the editor of the Nazi newspaper Der Stürmer. Previous research used these data to capture manifestations of antisemitism (e.g., Spenkuch and Tillmann Reference Spenkuch and Tillmann2018). Table SI2.3 replicates the models just described with this new set of covariates. Again, we find no systematic patterns that would explain proximity to the camps. Second, we looked at covariate balance between (a) electoral districts with and without a camp, and (b) between districts with camps and their neighboring districts. As the results in SI 2 show, the covariates are balanced across districts with and without camps in both types of comparisons.

Each of these results alone does not provide definitive proof that the location of concentration camps in Germany is completely unrelated to pre-existing beliefs. However, the consistency across the different empirical analyses and the fact that they line up with the historical accounts of the process through which German camps evolved gives us confidence that before the development of the Nazi camps, communities in the surrounding areas were no more prone to political intolerance, and xenophobia than those farther away.

The Consequences of Exposure to Camps

To test our expectations about the long-term consequences of exposure to Nazi camps in Germany, we rely on two nationally representative surveys of German adults: the German segment of the 2008 European Values Survey (EVS 2016), and the 2016 wave of the German General Social Survey (ALLBUS; GESIS 2017, 2018). These surveys were selected based on two criteria: (1) the inclusion of items on political tolerance and voting behavior, and (2) the provision of fine-grained regional identifiers.^{Footnote 17} To generate our quantities of interest, the surveys were combined with Census data from the interwar period, election results from the Weimar Republic, and the geolocation of the camps. Since we are ultimately interested in explaining current-day political attitudes, the unit of analysis is the individual survey respondent.

The key variable of interest in these analyses is Distance to Camp: the Euclidean distance between the location of a survey respondent and the closest Nazi camp.^{Footnote 18} This proxy for exposure to Nazi camps has two limitations that are worth noting. First, it does not account for historical roads and natural obstacles, since historical maps do not include a comprehensive description of the road system in place at the time. However, we have no reasons to believe that the measurement error generated by this decision is systematically related with current day political attitudes (see also Charnysh and Finkel Reference Charnysh and Finkel2017). Second, we identify the geolocation of respondents based on the centroid of the smallest regional identifier available in each survey: Kreis (N = 429) in the case of EVS, and Gemeinde (N = 11,084) in the case of ALLBUS.^{Footnote 19} Substantively, this means that all subjects within a given regional unit are assigned the same distance. This simplification may bias the results if subjects who live closer to the geographical center of an administrative unit are systematically different from those who live farther away, and if these differences are in turn systematically correlated with our outcome of interest. We do not have reasons to believe that this is the case. However, to assuage this possible concern we relied on different regional identifiers across our different survey and electoral analyses. The results produced across the different regional identifiers were substantively similar.^{Footnote 20}

We analyze three outcome variables in each survey, capturing different facets of political tolerance, xenophobia, and political behavior. To facilitate the interpretation of the different variables in each survey, we describe these variables in the respective subsections below. For each outcome of interest, we estimated two sets of models: a linear regression with interwar covariates (before the establishment of the camps), and a two-stage regression estimator—the sequential g-estimator (Acharya, Blackwell, and Sen Reference Acharya, Blackwell and Sen2016b; Vansteelandt Reference Vansteelandt2009)—that considers different contemporary mediators. This approach allows us to account for contemporary variables that may explain people’s attitudes without inducing posttreatment bias in our model estimates (Montgomery, Nyhan, and Torres Reference Montgomery, Nyhan and Torres2018). The method starts by estimating a model with both pretreatment and posttreatment covariates (first stage). Next, it recalculates the outcome variable by removing from it the effects of the mediating variables of interest. Finally, it estimates the effect of the treatment on this “demediated” outcome (second stage). The sets of pretreatment covariates and posttreatment mediators included in the models are as follows.

EVS Results

With data from the 2008 EVS, we generated three outcome variables measured as follows (the full question wording for all items is presented in SI 3.1).^{Footnote 26}

Support for Far-Right Parties

The EVS includes two items on party support: (1) “Which political party would you vote for,” and (2) for those answering ‘Don’t know,’ “If you don’t know, which party appeals to you the most?” Based on these two variables, we created a binary measure that takes the value of 1 if respondents mentioned an extreme right-wing party (National Democratic Party, NPD; German People’s Union, DVU; or the The Republicans, REP), and 0 otherwise.

Table 2 reports the estimates of the effect of camp distance on the three outcomes. For each outcome variable, a first specification includes only interwar covariates,^{Footnote 29} while the second model employs the sequential g-estimator to also account for contemporary mediators (Table SI4.3 reports the full set of coefficients and the first stage results). Overall, the different analyses provide support for our main prediction: exposure to concentration camps during the Nazi period is associated with significant differences in attitudes toward out-groups and support for far-right parties in present-day Germany.

TABLE 2.

Effects of Camp Proximity on Out-group Intolerance, Immigrant Resentment, and Support for Far-Right Parties (EVS)

Note: Entries are estimates of the effect of distance to closest camp on the different outcomes, described in column headers. Models 1, 3, and 5 account exclusively for interwar covariates (standard errors in parentheses). Models 2, 4, and 6 are the second stage of the sequential g-estimator to also account for contemporary predictors (bootstrapped standard errors in parentheses). Full model results in Table SI4.3. *p < 0.05; ** p < 0.01.

For example, Model 2 provides the controlled direct effect of camp distance on out-group intolerance using the sequential g-estimator. The coefficient of −0.02 is reliable and substantively meaningful. According to the model, a 50 km increase in distance to the closest camp is associated with a decrease of 0.10 points in out-group intolerance, representing 11.2% of the variable’s interquartile range (IQR).^{Footnote 30} To put this result into an appropriate context, we can compare it with the coefficient for conservatism, a common predictor of out-group intolerance. According to the model, a one unit increase in the 10-point scale of conservatism is associated with a 0.05 increase in out-group intolerance, equivalent to 6.0% of the outcome’s IQR. In other words, a 50 km increase in distance is estimated to have an effect similar to a two point shift on the 10-point conservatism scale.

Similarly, the estimated direct effect of camp distance in the model predicting immigrant resentment is −0.11 (Model 4). With the remaining predictors held constant, the model estimates that the difference in immigrant resentment between someone living next to a camp and someone living 50 km from a camp is equivalent to 10.0% of the outcome variable’s IQR. Finally, we also find that proximity to a former camp is associated with voting behavior.^{Footnote 31} According to Model 6, a 50 km increase in camp distance is associated with a two percentage point decrease in the probability of supporting an extreme right-wing party.^{Footnote 32} Considering that less than 3% of the 2008 EVS respondents reported supporting one of the far-right parties active at the time, this effect is meaningful. Relative to the mean value of the outcome variable, it represents a 66.7% change in the probability of supporting a far-right party.^{Footnote 33}

ALLBUS Results

We supplement the main analysis with an entirely new sample for several reasons. First, survey items in ALLBUS allow for a more detailed examination of the effects of camp proximity on attitudes toward different out-groups, including Jews and Muslims. Second, this survey includes lower level geographical identifiers than EVS [Gemeinde (N = 11,084) vs. Kreis (N = 429)] allowing for a more precise estimation of the respondents’ location. Third, it allows us to test the robustness of the EVS findings with a different sample from a different time period (2016 for ALLBUS vs. 2008 for EVS). If we find support for our theory in both datasets, the specific sampling pool adopted and the timing or political context of each survey is less likely to be driving our results.

Three outcome variables capture different facets of out-group intolerance. The full question wording and scaling for all questions is presented in SI 3.2.^{Footnote 34}

Intolerance Toward Muslims

Finally, given that Muslims have become an especially salient out-group in Germany (e.g., Jäckle and König Reference Jäckle and König2017), we would also expect to find our theorized effects when analyzing present-day attitudes toward Muslims. The ALLBUS data allows us to use another battery of six questions to capture these attitudes. Examples of these questions include “Islam fits into German society,” “Islamic associations/groups should be under state surveillance,” and “practicing the Islamic faith in Germany should be limited.” We again conducted a principal component analysis to find that the first dimension explains around 58% of the variation across the six items (Cronbach’s α = 0.84). This first dimension is extracted to produce our third outcome measure.

Table 3 presents the results of six models estimating the effect of camp proximity on present-day out-group intolerance. As before, we run two models for each outcome variable. The control variables are the same as in the EVS analysis (cf. Table 2, above). The full model and first-stage results are presented in Table SI4.4.

TABLE 3.

Effects of Camp Proximity on Intolerance Toward Foreigners, Jews, and Muslims (ALLBUS)

Note: Entries are estimates of the effect of distance to closest camp on the different outcomes, listed in column headings. Models 1, 3, and 5 account exclusively for interwar covariates (standard errors in parentheses). Models 2, 4, and 6 are the second stage of the sequential g-estimator to also account for contemporary predictors (bootstrapped standard errors in parentheses). Full-model results in Table SI4.4. * p < 0.05; ** p < 0.01.

The analyses provide further support for our theoretical argument: as a respondent’s distance to a former concentration camp increases, intolerance toward foreigners, Jews, and Muslims decreases. We find the largest effect sizes for Intolerance Toward Foreigners and Intolerance Toward Muslims. However, it is important to keep in mind that all three outcome variables are factor scores with a slightly different scaling. For example, the IQR of Intolerance Toward Foreigners is 6: 50% of all observations lie within −3 and +3. With that in mind, we can say that a distance increase of 50 km for an otherwise unchanged respondent should on average lead to a 0.24 point decrease in our measure of intolerance toward foreigners, representing 4% of its IQR. For Intolerance Toward Jews, the IQR is 4.4, so if a respondent’s distance to the closest camp increases by 50 km, their predicted factor score would decrease by 0.15 points, around 4% of the IQR. Finally, Model 6 estimates that a 50 km increase in distance is associated with a decrease of Muslim intolerance by 0.21 points, which would reflect a change of 4% of the outcome’s IQR.^{Footnote 35}

To summarize, we find consistent support for our expectations in both surveys. Evidence from EVS and ALLBUS shows that respondents who live closer to former Nazi concentration camps today are more likely to hold negative views of different types of out-groups in present-day Germany. They are also more likely to indicate support for right-wing parties. These patterns are in line with our argument that concentration camps have long lasting effects that still play a role in explaining political attitudes today.

Mechanisms: Cognitive Dissonance and Belief Transmission

We argued that the patterns observed above likely result from processes of cognitive dissonance triggered by proximity to the camps during the Third Reich. This argument implies that the pre-existing attitudes among individuals living close to the camps were not systematically different from those living elsewhere, and that it was the mental discomfort produced by this new social environment that led them to update their beliefs. While plausible, this mechanism is hard to demonstrate because a direct test of it would require a careful micro-level analysis of individuals who witnessed this period. Such data are not available, which is why prior work on historical legacies has generally not tested the mechanism at all. We aspire to do more, and offer an alternative strategy to assess the plausibility of this segment of our argument.

If cognitive dissonance is responsible for the relationship uncovered, individuals living in areas that were more supportive of the Nazi party before the creation of the camps should experience less cognitive dissonance. For these individuals, there should be less of a need to reconcile their pre-existing beliefs with the new social environment. Consequently, the effects of proximity to camps should be smaller. We test this implication by interacting our key predictor in the main analysis (distance to camp) with support for the Nazi party in 1933. The plots in Figure 2 present the marginal effects of distance to camp on out-group intolerance (left panel), immigrant resentment (central panel), and support for far-right parties (right panel), conditional on the vote share obtained by the Nazi party in the last contested election of the Weimar Republic.^{Footnote 36} In line with our expectations, the camp proximity effects are large and statistically significant in regions that previously had low levels of Nazi support. However, for all three outcome variables this effect decreases as Nazi support increases, and is no longer distinguishable from zero in districts with 50% or more support for the Nazi party. These results are consistent with the argument that cognitive dissonance links Nazi camps to contemporary intolerance, although they do not represent direct evidence for the mechanism proposed. Therefore, the findings should be interpreted as suggestive.^{Footnote 37}

FIGURE 2.

Marginal Effects of Camp Proximity on Contemporary Attitudes, Conditional on Support for the Nazi Party in 1933 (EVS)

Note: Plots depict the marginal effects of distance to camps on out-group intolerance (left panel), immigrant resentment (middle panel), and support for far-right parties (right panel), conditional on support for the Nazi party in 1933. Shaded regions represent 99% confidence intervals. The histograms at the base of each figure describe the distribution of support for the Nazi party. The full model results are reported in Table SI6.1.

We can also provide evidence consistent with the second part of our theoretical mechanism: the intergenerational transmission of beliefs. We argued that out-group intolerance was transmitted across generations through parental and social influence. This mechanism has been explored in Germany (Voigtländer and Voth Reference Voigtländer and Voth2012) and elsewhere (e.g., Lupu and Peisakhin Reference Lupu and Peisakhin2017; Nunn and Wantchekon Reference Nunn and Wantchekon2011). Still, to provide more direct evidence for this mechanism in the context of our study, we rely on two items from EVS.

First, respondents were asked in which Kreis they lived at the age of 14. Although communal transmission does not require that a given respondent, or their parents, were born in a specific district, we should expect clearer effects of camp proximity among respondents who lived in the same district all their lives, both due to longer cultural exposure and family ties in the area. Figure 3 provides evidence in line with this expectation. The marginal effects of camp proximity on out-group intolerance (left panel), immigrant resentment (middle panel), and support for far-right parties (right panel) are only distinguishable from zero for respondents who are still living in the Kreis of their childhood.^{Footnote 38} Second, EVS respondents were asked whether they discussed politics with their parents when they were young (“around 14 years old”). The argument about intergenerational transmission suggests that camp proximity should be a stronger predictor of contemporary attitudes among respondents who regularly discussed politics with their parents. Figure SI6.5 provides evidence in line with this expectation. The effects of camp proximity on out-group intolerance and immigrant resentment are only reliable among respondents who experienced this type of political socialization.^{Footnote 39}

FIGURE 3.

Marginal Effects of Camp Proximity on Contemporary Attitudes, Conditional on Place of Residence at Age 14 (EVS)

Note: Points are marginal effects of distance for EVS respondents who lived in the same Kreis at the age of 14 years and respondents who lived elsewhere. Outcome variables described at the top of each panel. At the time of the survey, 56.9% of respondents lived in the same Kreis they lived at the age of 14 years. Full model results in Table SI6.3.

Alternative Mechanisms

In this section, we explore three potential alternative explanations for our findings: economic conditions, geographic sorting, and contemporary cognitive dissonance. Recall that we also accounted for a potential economic explanation as part of our main analysis. Unless noted otherwise, detailed results of these additional tests are presented in SI 7.

Contemporary Cognitive Dissonance

Another alternative explanation for the relationship we uncovered is that the camps keep producing cognitive dissonance today. To varying degrees, the sites of most German camps are currently used as a memorial, documentation center, or museum. Hence, it is conceivable that these symbols of the Nazi era are still generating cognitive dissonance among those living in the surrounding areas. Research in psychology suggests that reminders of ingroup wrongdoing may prompt defensive reactions (Imhoff and Banse Reference Imhoff and Banse2009; Rotella and Richeson Reference Rotella and Richeson2013; but see Rees, Allpress, and Brown Reference Rees, Allpress and Brown2013). This mechanism would generate the same patterns uncovered here without requiring the transmission of attitudes across generations.

To assess the plausibility of this mechanism, we leverage the variation in current-day use of the German camps. For example, by the end of the 1940s, hardly anything remained to be seen of the former camp at Mittelbau-Dora. The grounds had rapidly been reclaimed by nature. A permanent exhibition on the camp history opened in a newly erected building only in 2006.^{Footnote 41} This stands in stark contrast to other former camps such as Dachau, where many of the original structures (e.g., the Jourhaus, the shunt room, prisoner baths, the bunker, the barracks, and the crematorium) have been preserved and their original utilizations displayed.^{Footnote 42} The memorial site gives visitors a very graphic and vivid experience about life in the camp, so realistic that an article described the experience as “A Day in Hell.”^{Footnote 43} If contemporary cognitive dissonance explains the patterns we observe, the effects should be most visible around camps that currently have a more noticeable presence.

We documented the current use of all camp locations and identified the existence of original physical structures as the most meaningful distinction between camps.^{Footnote 44} Finally, we reestimated the main models from EVS and ALLBUS interacting distance with a binary variable that identifies the current use of the camp.^{Footnote 45}

Figure 4 presents the main findings from both surveys, and reveals significant differences between camps with and without physical structures. However, the results are the opposite of what the contemporary cognitive dissonance argument would predict. The overall effects of proximity are mostly driven by respondents living near camps with a less noticeable presence. On the other hand, in the surroundings of those camps where original structures have been preserved as a museum, the effects of distance are either significantly smaller or indistinguishable from zero across all six model specifications.^{Footnote 46}

FIGURE 4.

Marginal Effects of Camp Proximity on Contemporary Attitudes, Conditional on Current-Day Use of the Camp (EVS and ALLBUS)

Note: Plots depict point estimates and 95% confidence intervals for the standardized effects of distance to camps on contemporary attitudes (described on the y-axis), conditional on current-day use of the camp. Gray (black) points/bars correspond to respondents whose closest camp does (not) include physical structures. The full model results are presented in Tables SI7.3 and SI7.4.

One possible interpretation of this pattern is that an educational effect of museums housed in the preserved physical structures of former camps is counteracting the legacy effects. Being able to see the actual structures of the institutions where atrocities were committed is likely to leave a strong impression on individuals visiting the camp locations. It is therefore possible that the Holocaust-related education promoted by these camps may offer a path to mitigate the effects of cognitive dissonance produced during the Nazi era. This interpretation is in line with recent policy debates in Germany: in spring of 2018, public discussion erupted in response to an anti-Semitic attack about whether school visits to concentration camps should be made compulsory because they can be transformative in promoting tolerance (Pastor Reference Pastor2018). In line with this, a museum director cited an example of how a camp visit eliminated a group of students’ prior fascination with neo-Nazi graffiti (Bennhold Reference Bennhold2018). As a further test of potential educational effects, we collected information on the number of years since a permanent exhibition about the Holocaust opened in the site of a former camp. Beyond the presence of original structures, any educational effect should be more prominent the longer an exhibit has been in place. This is exactly what we find as the results in Table SI7.7 show.^{Footnote 47}

Admittedly, our test here is imperfect, and we interpret it with caution. However, the potential implications of this mitigating effect can be far-reaching. So far, the literature on historical legacies has focused primarily on firmly establishing the legacy effects. The determinism of historical legacies, however, makes chances of progress appear rather bleak. That is why the finding here is important: it takes us a step closer to identifying ways to break the seemingly deterministic persistence of the effect of past institutions and affirms that efforts to re-educate the population may not be in vain.

Sensitivity Analyses

We also performed various robustness tests of our main analysis. These include (1) an analysis of far-right voting using recent electoral data from Germany, (2) an analysis of additional outcome variables in the ALLBUS survey, (3) a series of placebo tests, (4) a cross-national analysis with data from other Nazi camps in Europe, and (5) an analysis of closeness to the Treblinka concentration camp in Poland. We discuss each of these tests in turn (for details, see SI 8).

To analyze whether our results hold outside of survey data, we compiled a dataset of recent election results from the 2017 Bundestag election at the Wahlbezirk level (electoral district, N = 88,511), which we then combine with geographic and socio-demographic information at the Gemeinde level. Whereas the survey data we use above gives us a rich set of outcome variables, the electoral data provides behavioral measures. Additionally, it allows us to focus on specific geographic units. By restricting the analysis to small radii around each camp, these restricted samples are better matched in terms of potential confounders while still providing variation in closeness to a camp (see Charnysh and Finkel (Reference Charnysh and Finkel2017) for a similar approach). We use vote share for the radical right-wing parties—Alternative for Germany (AfD) and National Democratic Party (NPD)—as the outcome variable. Table 4 provides the key findings, which suggest stronger support for AfD and NPD in areas closer to former camps. The results hold both nationally and within 70 km of each camp. A variety of alternative modeling strategies reported in SI 8 render similar conclusions.

TABLE 4.

The Controlled Direct Effect of Camp Proximity on Support for Radical Right Parties in 2017

Note: Entries are estimates of the controlled direct effect of distance to closest camp on support for the AfD (Columns 1, 2) and AfD + NPD (Column 2, 3) in 2017. Full-model results and additional specifications are presented in Tables SI8.3–SI8.7; ** p < 0.01.

Given the comprehensive set of attitudinal questions asked in the ALLBUS survey, we tested the robustness of our findings on a complementary set of out-group intolerance measures. More specifically, we created factor scores for Perception of foreigners, Discrimination of foreigners, and a variable for respondents that indicated support of a far-right party. The perceptions items asked mostly about potential advantages and disadvantages of having foreigners in Germany, whereas the discrimination items asked about potential discrimination against or in favor of foreigners in different stages of life (education, job market, etc.). The Support for extreme parties variable takes the value of 1 if respondents indicated that they would vote for the NPD or AFD if there was an election next Sunday. The full question wording for these variables can be found in SI 3.2. We ran the exact same model specifications as in the main ALLBUS analysis above. The results are reported in Table SI8.8 and provide further support for our argument. Respondents living closer to concentration camps perceive foreigners more negatively, are less inclined to think that foreigners are discriminated against, and are more likely to support far-right parties.

Tables SI8.9 and SI8.10, in turn, report results from a series of placebo tests where we replaced the outcome variables with different contemporary attitudes that should not be meaningfully associated with either camp proximity or out-group intolerance. As expected, beliefs such as willingness to turn out, job and life satisfaction, or importance of leisure time are not explained by camp proximity. These results, although inevitably suggestive, indicate that the long-term effects of living close to Nazi camps are targeted and relate to attitudes toward out-groups.

Whereas our main analysis focuses exclusively on German survey respondents, the EVS includes respondents from other countries in Europe where the Nazi regime had also set up concentration camps. We use these data to test the generalizability of our findings. More specifically, we ran a cross-national analysis that includes EVS respondents from Austria, Belgium, Croatia, the Czech Republic, Estonia, France, Germany, Latvia, Lithuania, the Netherlands, and Poland (EVS 2016). For each respondent, we coded their distance to the closest former concentration camp. We used the same outcome variables as in Table 2: out-group intolerance, immigrant resentment, and support for far-right parties. The results of these model specifications can be found in Tables SI8.11 and SI8.12 and closely mirror our main results above. Across all model specifications, we find that respondents who live closer to concentration camps today are less tolerant of out-groups, exhibit stronger immigrant resentment, and are more likely to support far-right parties. The effect sizes tend to be smaller than in our main analysis, but are otherwise in line with our main findings and are equally statistically reliable.

Finally, we follow Charnysh and Finkel (Reference Charnysh and Finkel2017) for another empirical test of the implications of our theory with data from the area surrounding the Treblinka concentration camp in Poland. These authors argue that the location of this camp was “exogenous to the behavior and the views of the local population” (p. 802). We use their data and modeling strategy to analyze support for the PiS party in the 2015 national elections. We focus on PiS, because the party ran on a clearly anti-immigrant and anti-refugee platform. Our theory predicts that the party should have performed particularly strongly in communities that are closer to Treblinka. The results in Table SI8.13 confirm this expectation: across all different model specifications we find that the closer a community is to Treblinka, the higher its vote share for PiS.