WWI and the End of the German Empire

The German Empire was proclaimed in 1871 following Prussia’s victory over France. It was a constitutional monarchy under Prussia’s leadership, featuring the first publicly elected national parliament on German territory after an initial failed attempt in 1848. Under Emperor Wilhelm II, the German Empire started a period of unpredictable and provocative foreign policy, which isolated it from most of its former European allies, most notably Russia and the United Kingdom. As a result, it took only a spark in the form of the assassination of Archduke Franz Ferdinand of Austria to start WWI on the 28th of July 1914 (Berman Reference Berman2001; Hewitson Reference Hewitson and James2008). Among the German population, the initial excitement about the war faded very quickly with an increasing death toll, unpredictable duration of the war, frequent food shortages, and generally gloomy outlook. This resulted in several strikes but also the secession of the anti-war, farleft fringe (USPD) from the Socialist Party (SPD) and a new, informal coalition between the main leftwing and centrist parties to hold the Supreme Army Command (OHL) accountable (Mommsen Reference Mommsen1996).

By September 1918, the situation of the German Army had deteriorated to such an extent that the OHL admitted defeat to the Emperor. A new grand government was formed, and a few days later, Germany officially asked for an armistice. When the OHL rejected the conditions set by the Allied Forces in late October 1918, Chancellor von Baden sacked the military leadership and issued political reforms, which turned Germany into a parliamentary monarchy. The war, however, continued until the end of October, when a mutiny by the German Navy in the city of Kiel sparked a rebellion and the formation of workers’ and soldiers’ councils that took over civilian and military power.

This non-violent rebellion eventually led to the proclamation of the German Republic on the 9th of November (Büttner Reference Büttner2008). Two days later, WWI was officially ended by an armistice after four years of fighting and more than 10 m soldiers killed (Prost Reference Prost, Daniel, Gatrell, Janz, Jones, Keene, Kramer and Nasson2014).

A key reason for Weimar democracy’s failure 15 years later was that the German Army was still not technically defeated when the armistice was signed. This soon gave rise to the stab-in-the-back myth, a highly popular conspiracy theory according to which Germany had not lost WWI but was betrayed by an unreliable home front allegedly wound up by foreign Bolshevik activists and Jews with the support of the domestic political left and center. Organizations and parties on the right heavily used this myth as a propaganda tool to discredit the new democratic state and used it to connect Anti-Semitism with existing Red Menace fears of an imminent violent coup by the radical left (Barth Reference Barth2003).

Red Menace fears within the German population were fostered by the rebellions of November 1918 and the fact that the interim government was led by the two leftwing parties SPD and USPD. Especially problematic was that the USPD’s revolutionary part, which later became the Communist KPD, openly sought to create a Soviet Germany also by violent means (Brustein and Berntson Reference Brustein and Berntson1999). In light of this, the SPD started to collaborate with centrist parties but also increasingly relied on deeply reactionary paramilitary units, like the Freikorps, to maintain its power and territorial integrity.

Veterans during the Weimar Republic

Despite comprising at most 400,000 members, many of whom had never even served in WWI, the Freikorps significantly shaped the stereotypical image of Germany’s 11.2 m WWI veterans as brutalized and unable to fit into post-war society (Mosse Reference Mosse1990; Diehl 1993). Only recently has historical work begun to question this brutalization hypothesis (Schumann Reference Schumann2003; Ziemann Reference Ziemann2003). In fact, it has been noted that most veterans were presumably looking for stability after a period of war and revolution and returned to an unspectacular, civilian life. Also, the stylized image of the crippled, impoverished, and embittered former WWI soldier seems only true for a small fraction (Bessel Reference Bessel1993).

While most research has focused on significant groups or individual WWI participants, little is known about the political impact of the average veteran. This may be because it included almost all men born between 1869 and 1900 but a large part of those claiming to represent the so-called front generation was also not even part of it. Some scholars have noted that not all veterans supported the anti-democratic right, but that only the majority of those becoming politically active and claiming to represent the front generation did so (Diehl Reference Diehl and Stephen1975; Bessel Reference Bessel and Roseman1995). An actual investigation into veterans’ party orientation or electoral impact has, however, thus far not been conducted.

Panel Data of Reichstag Elections

To track changes in Germany’s voting behavior over time, I compiled a panel dataset covering all parliamentary elections held since the foundation of the German Empire until 1933. The panel is based on two existing datasets on elections in Imperial and Weimar Germany. ^{Footnote 4} I complement this with election results for the National Assembly in January 1919 that took place only two months after WWI ended and thus provide an invaluable measure of post-war political preferences unaffected by other significant events like the Versailles Treaty. In my analysis, I focus on elections from 1893 to 1933, covering 20 years before and after the onset of WWI in 1914. ^{Footnote 5} Comparing voting results over almost 40 years and different political regimes, however, raises several important questions.

One important issue is the comparability of parties across time. I tackle this by assigning each party to one of seven party groups based on classifications of Imperial and Weimar Germany’s party systems: Communist, Socialist, Left-Liberal, Catholic-Minority, Right-Liberal, Conservative, and Antisemite. ^{Footnote 6} To capture significant changes in political orientation and reduce complexity, I collapse party groups further into three political camps for most of my analysis: Leftwing (Socialist, Communist), Center (Left-Liberal, Catholic-Minority, Right-Liberal), and Rightwing (Conservative, Antisemite). These camps also roughly correspond to supporters and enemies of democratization since the left was the main driving force against the monarchy before WWI while the right was the fiercest opponent of Weimar democracy. ^{Footnote 7} Figure 1 shows aggregate vote shares in the area of post-war Germany for each camp in the sample period 1893 to 1933. One can see clearly that soon after WWI, the left’s rise and the right’s decline stopped and reverted while the downward trend of the center continued. The main boost in rightwing votes during the rise of the Nazi Party, however, did not happen until the 1930s.

FIGURE 1 NATIONAL ELECTION RESULTS IN THE GERMAN EMPIRE AND THE WEIMAR REPUBLIC

Notes: The dark grey-shaded area marks the WWI period. The vertical lines denote the last pre-WWI and the first post-WWI election. Data covers only the area contained in Weimar Germany’s borders and therefore may differ from official national aggregates for elections prior to 1922.

Source: Author’s calculations.

The second issue is the comparability of the electorate and the electoral system. The parliamentary elections held in the German Empire were relatively free and fair given the time period (Berman Reference Berman2001). Yet, suffrage was only granted to men aged 25 or older. The political reforms of 1918 introduced female suffrage and reduced the minimum voting age to 20, which effectively doubled the electorate. Since post-war data does not allow identifying these new voters, shifts in electoral behavior after WWI could also stem from simultaneous compositional changes in the electorate. Work on female voters in the Weimar Republic, however, has revealed that women voted along similar dimensions as men and predominantly stuck with their political camp (Sneeringer Reference Sneeringer2002). Crucially to this analysis, my empirical results also hold when controlling for the population share of women above 19 and men aged 20 to 25. ^{Footnote 8}

Finally, voting results during the German Empire were almost exclusively published for constituencies (Wahlkreise). Each single-member constituency typically consisted of 2–4 counties (Kreise) with occasional overlaps, and only followed political boundaries for small states and administrative districts. ^{Footnote 9} Conveniently, the initial 397 constituencies were never adjusted in response to population growth for political reasons and thus remained stable until WWI (Jesse Reference Jesse and Niedermayer2013). My units of analysis are the 362 constituencies contained within Weimar Germany’s borders. After WWI, Germany was divided up into a few large constituencies, but results were now published at the county level. This allows re-constructing post-war vote counts as well as other variables for each pre-war constituency. Constituency-level summary statistics are provided in Online Appendix Table B.1. ^{Footnote 10}

A Disaggregated Estimate of Germany’s WWI Veterans

Germany’s war veterans were never explicitly counted for statistical purposes. However, since the Prussian Army Archive was destroyed in WWII, official statistical publications remain virtually the sole source of information on WWI participation across constituencies. ^{Footnote 11} The key ingredient for estimating the number of surviving WWI soldiers is the wartime census of December 1917, which provides county-level figures of the female and civilian male population. Since the German army consisted exclusively of men, war participation led to a notable gender gap among the civilian population in the 1917 census. The male-female gap closed down again in the first post-war census conducted only 22 months later, in October 1919. ^{Footnote 12} The abnormal difference in civilian male and female population change, $$\Delta Me{n_{17 - 19}} - \Delta Wome{n_{17 - 19,}}$$ during this short period thus provides a close proxy for the number of war participants in 1917 who survived the war.

Equation (1) shows how the exact number of all WWI veterans could be recovered if one actually knew the amount of surviving soldiers in 1917. This would essentially require also accounting for soldiers who had not yet joined the army as well as those who had left for non-fatal reasons such as disability or work-related exemptions.

(1) \begin{align}Veterans = (Soldier{s_{17}} &- SoldiersDea{d_{18}}) + SoldiersJoi{n_{18}}\\ &+ SoldiersQui{t_{14 - 17}}\end{align}

The veteran estimate used in this paper is stated in the top line of Equation (2). It approximates the ideal measure in Equation (1) by summing up the difference in male and female population growth from 1917–1919 mentioned previously and the number of men born in 1900, which were the only cohorts drafted after 1917 (Nash Reference Nash1977). While this estimate gets reasonably close to the actual amount of WWI veterans, it also introduces several sources of measurement error. In short, my veteran estimate assumes no gender-specific differences in births, civilian deaths and migration between 1917 and 1919 (ε = 0), the draft of the entire 1900 cohort (v = 0), and no man quitting the army up to 1917 for other reasons than death (ξ = 0): ^{Footnote 13}

(2) \begin{align}& Veterans \approx (\Delta Me{n_{17 - 19 }} - \Delta Wome{n_{17 - 19}}) + MenDraftabl{e_{18}}\\ &\quad = \underbrace {(\Delta Me{n_{17 - 19}} - \Delta Wome{n_{17 - 19}} - \varepsilon )}_{(Soldier{s_{17}} - SoldiersDea{d_{18}} )} + \underbrace {(MenDraftabl{e_{18}} - V)}_{SoldiersJoi{n_{18}}} + \underbrace \xi _{SQui{t_{14 - 17}}}\end{align}

The aggregate number of surviving German soldiers for the German Empire, according to my estimate is about 9 m, which is below the 11 m stated in official sources. This gap can be explained by soldiers called back to work in war-related industries during the 1917 census (64.7 percent), pre-census dropouts due to non-fatal reasons (21.8 percent), and those still held as prisoners of war during the 1919 census (13.3 percent). ^{Footnote 14} This suggests that, at least in terms of size, ξ is by far the most important error term. The potentially significant mismeasurement due to gender-specific internal migration after WWI, however, cannot be assessed with countrylevel aggregates.

Bearing those caveats in mind, I calculate my main treatment variable, Veterans p.c., as a constituency-level version of the estimate noted previously divided by the last pre-war census population of 1910. This veteran share ranges from 0.06 to 0.20 and has a mean and median of roughly 0.14. Figure 2 shows the spatial distribution by deciles across the 362 constituencies discussed in the subsection on panel data of Reichstag elections. ^{Footnote 15} While veterans as a share of the population are quite dispersed across Weimar Germany, they are also visibly underrepresented in highly industrialized West Germany and over-represented in the more rural North and East. This is consistent with the aforementioned undercount of soldiers on temporary leave from the army to work in war industries during the 1917 census.

FIGURE 2 VETERANS PER CAPITA ACROSS WEIMAR GERMANY BY CONSTITUENCIES

Source: Author’s calculations.

Control Variables

In my empirical analysis, I control for several potential confounders. ^{Footnote 16} These include, for instance, the size of the draftable population (all men born between 1869 and 1900), pre-war military personnel, as well as male students, which were particularly likely to volunteer. I also add the differential growth between men and women in 1910–19 to approximate mismeasurement coming from internal gender-specific migration in 1917–19. A crucial driver of war participation were exemptions for employees in war-related industries. Lower draft rates in these occupations, however, were compensated by higher drafts in non-war industries. Due to the close alignment between blue-collar employment and support for leftwing parties, I control not only for men working in war-related industries but also blue-collar jobs in general. Lastly, I also adjust for the share of Catholics, which is negatively correlated with veterans and was a primary driver of political orientation at the time.

In Online Appendix Section A.1, I regress Veterans p.c. on several sets of variables. The controls mentioned earlier jointly explain almost half of the variation (R ² ≍ 0.44). The other half comes from two sources: one of these is differences in exemption rates within war-related industries. Accounting additionally for the share of occupations where labor input by young adult men was particularly difficult to replace, increases the R ² to about 0.49. In the instrumental variable estimates subsection, I use the prevalence of these low draft occupations as an instrument for war participation. The second source of variation is idiosyncrasies in recruitment and survival. I show this by adding fixed effects (FEs) for three tiers of prewar military territorial organization to the regressions. FEs for corps and brigade areas, which correspond to actual army units, increase the R ² only moderately to 0.67. The addition of recruitment area FEs, on the other hand, raises it to about 0.93. This suggests that most of the variation unexplained by the baseline controls comes from idiosyncrasies in draft stringency rather than different levels of battle exposure and survival probability.

DiD Estimates

The baseline DiD results for the three-party camps are reported in Table 1. The first specification in each column triplet includes only the interaction term Veterans pc _i × Post _t and FEs for constituencies and elections, and yields highly significant coefficients on all three outcomes. A 1 pp higher veteran share is thus associated with a 1.61 pp decrease in leftwing votes after WWI and a 0.68 and 0.93 pp increase in center

TABLE 1 DID ESTIMATES (BASELINE RESULTS)

Notes: Observations are at the constituency-level. The sample period is 1893 up to 1933 and includes 14 Reichstag elections. Standard errors clustered at the constituency-level in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ΔGrowth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics. All controls are interacted with Election FE.

Source: Author’s calculations.

and rightwing votes, respectively. The second column for each party camp is the preferred baseline specification stated in Equation (3), which also adjusts for covariates. This diminishes the effect on leftwing votes to –1.31 but does not affect significance. The effect on centrist votes, however, almost falls to zero and is no longer significant, while the one on rightwing votes climbs to 1.23. Adding constituency-specific linear trends in the third column of each triplet, on the other hand, only slightly changes the estimates and does not suggest that these are driven by linear pre-trends.

In terms of magnitude, these are large effects. Raising the veteran share by 0.021, the equivalent of one SD, implies an increase in the rightwing vote share by 0.025 or 12.8 percent of an SD, in the post-war period. The impact of WWI participants on rightwing votes is thus not far from that of other structural factors like medieval pogroms and association density on Nazi votes found in earlier research. ^{Footnote 19} Assuming the entire absence or no political effect of WWI veterans would have resulted in an average reduction of the rightwing share by 0.17 and a simultaneous increase in leftwing votes of 0.18. This hypothetical change in the parliamentary power balance would have prevented a rightwing majority at the end of the Weimar Republic and may have avoided the breakdown of democracy. In light of these implications as well as the potential threats to identification outlined previously, these results should be treated with caution at this stage and will be revisited later in the subsection on counterfactual.

Timing and Alternative Channels

Figure 3 uses an event study that disentangles the veteran effect for every election with respect to 1912. This allows for the inspection of non-linear pre-trends but also offers a perspective on the baseline results’ timing. Overall, the event study supports the assumption of no pre-trends. None of the pre-war coefficients for the right and center are significant, and the correlation with the veteran share in 1912 is very close to that in 1893, almost 20 years earlier. The left sees a significant small drop in 1903, which, however, quickly reverts and is mirrored by small gains for both the right and the center. No party camp exhibits clear anticipatory patterns. The post-war coefficients also suggest again that veterans predominantly affected left- and rightwing parties, whereas the center’s gains are tiny and never significant at conventional levels. ^{Footnote 20}

FIGURE 3 EVENT STUDY GRAPH FOR THE THREE PARTY CAMPS

Notes: Coefficients and 95 percent confidence intervals for the effect of veterans per capita on election results for each election between 1893 and 1933, normalized to the effect in 1912. The right-hand side of the regression specification is equivalent to the preferred baseline: Column (2) of Table 1. The light grey-shaded area marks the WWI period.

Source: Author’s calculations.

There are several additional important aspects to these results: firstly, the electoral shift from left to right already materialized in the January 1919 election, only two months after WWI officially ended. The Paris Peace Conference that would draft the Versailles Treaty had just started by then, and other polarizing events potentially related to the presence of veterans, such as the Kapp coup (1920), the Ruhr Uprising (1920), or the French Ruhr Occupation (1923), were still in the future. The Stahlhelm, which became the most active rightwing ex-servicemen club in the 1930s, had just been founded a month earlier (Berghahn Reference Berghahn1966). All in all, this leaves only a rather small time window of other post-war events, most notably the Spartacist uprising by the radical left, to confound the DiD results. Secondly, the veteran effect pre-dates the left-right swing in aggregate voting patterns after 1920 (see Figure 1). This implies that without veterans’ impact, pre-war electoral trends would have seen a sharp break in 1919, benefiting the left at the expense of the political right. Thirdly, the post-war coefficients for all camps are very stable over time, especially for the left, which hints at a genuine change in political preferences related to veterans rather than an issue-related short-term vote swing. The early and persistent anti-left effect also speaks against the widespread assumption that rightwing parties’ success among the German working class started only after the Great Depression in the 1930s (Brustein Reference Brustein1996). Finally, since the veteran measure was calculated based on the October 1919 census, the treatment onset in January 1919 provides evidence against politically selective migration in the following years.

The timing of the effect rules out several competing explanations but leaves a few alternatives open: some of these arise from the strong relationship between veterans per capita and the share of deceased WWI soldiers. For instance, casualties may have led to political radicalization among those bereaved by the war or reflected different levels of social cohesion in units’ main recruitment areas. Another possibility is that veterans could be linked to the strength of far-left groups. This seems plausible since the German revolution was sparked by a mutiny in the German Navy, and soldiers formed many of the revolutionary councils that replaced the pre-war administration during the interim period. If there was a correlation between veterans and the far-left, the baseline results could mask a simple political backlash after the Spartacist Uprising in Berlin only a few days before the 1919 election (Mommsen Reference Mommsen1996). Anti-Communism would thus constitute a confounder to the veteran effect rather than a mechanism through which it operated.

Table 2 provides evidence against both competing explanations. The first column for each party camp adds an interaction between WWI casualties per capita and Post . ^{Footnote 21} Doing so has virtually no effect on the veteran coefficient for any party camp. In the second column of each camp, I restrict the sample to Bavaria and a few other constituencies for which I can include an interaction with officially deceased WWI soldiers per capita. The left-right shift is more pronounced in this subset and shows an insignificant reduction in centrist votes, but does not qualitatively change my findings. In the third specification of each party group, I add an interaction based on the vote share of the far-left USPD in 1919 as an indicator of the far-left’s strength. The estimates show that Red Menace fears led to a marginally significant shift in votes from left to center but do not invalidate the veteran coefficients that provide evidence against this alternative explanation. ^{Footnote 22}

TABLE 2 ALTERNATIVE CHANNELS

Notes: Observations are at the constituency-level. The sample period is 1893 up to 1933 and includes 14 Reichstag elections. Standard errors clustered at the constituency-level in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ΔGrowth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics. All controls are interacted with Election FE.

Source: Author’s calculations.

Instrumental Variable Estimates

I tackle remaining endogeneity concerns by using war-related exemptions as an alternative source of variation. More precisely, I instrument veterans per capita with the share of male workers in low-draft occupations using a high share of draftable men (>45 percent) among their employees during wartime: iron and coal mining, extraction of graphite, asphalt, oil and amber, quarrymen, shipbuilding, manufacturing of non-rifle firearms, and pharmacists.

Instrument validity requires the following exclusion restriction to be satisfied: conditional on covariates, the share of low-draft occupations affects voting only through its negative effect on war participation. Several factors support this assumption: firstly, about half the low-draft employment is typically in mining or shipbuilding, where location is pre-determined by geography and thus less prone to endogeneity concerns. ^{Footnote 23} This is also shown in Figure 4, which depicts the spatial distribution of the instrument’s identifying variation after conditioning on baseline controls. Secondly, as the map shows, low-draft occupations are displaying considerable geographic dispersion and, crucially, do not coincide with the industrial heartlands and leftwing strongholds of West and Central Germany. Finally, the conditional correlations between low-draft occupations and pre-war vote shares are small and never go in the direction predicted by the IV. ^{Footnote 24}

FIGURE 4 MEN IN WAR-RELATED LOW-DRAFT OCCUPATIONS ACROSS WEIMAR GERMANY BY CONSTITUENCIES (RESIDUAL AFTER BASELINE CONTROLS)

Source: Author’s calculations.

Column (1) in Table 3 shows a strong, highly significant first-stage coefficient indicating that a lpp increase in low-draft male employment translated into a 0.72 pp decrease in the population share of veterans. The F-statistic on the excluded instrument of 34.0 renders a bias due to weak instruments unlikely. The first specification for each party camp repeats the baseline DiD results, while the second and third columns show the raw reduced-form effect of the instrument and the 2SLS estimates.

TABLE 3 INSTRUMENTAL VARIABLE ESTIMATES

Notes: Observations are at the constituency-level. The sample period is 1893 up to 1933 and includes 14 Reichstag elections. Standard errors clustered at the constituency-level in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ∆Growth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics. All controls are interacted with Election FE.

Source: Author’s calculations.

Qualitatively, the IV estimates confirm the OLS results, but in terms of magnitude, they are about twice as high for both left- and rightwing. Yet, one has to remember that the instrument only estimates a local average treatment effect on the compilers, which are workers who do not become veterans because they work in low-draft occupations or participate in WWI because they do not hold such a job. Since I control for the overall prevalence of all war-related occupations, the effect is identified in highly industrialized areas that may yield notably higher effects for several reasons: one is that the target population to persuade into voting rightwing was particularly high in those areas since they were the political heart-land of the Socialist party. In addition, research has pointed out that the tight social fabric of Germany’s working-class milieu may have provided a particularly fertile ground for the spread of rightwing extremist views after WWI (Szejnmann Reference Szejnmann and Fischer1996). While this does not invalidate the IV findings, it renders them less suitable for making general inferences, and they should thus be mainly regarded as additional evidence in support of my main results. The remainder of this paper will therefore continue to refer to the more conservative DiD baseline estimates.

The Role of Anti-Communism

As a first step to uncover the mechanisms behind my findings, I zoom into the party camps in Table 4 and estimate the baseline specification using individual party groups and turnout as outcomes. The first column

TABLE 4 DID ESTIMATES FOR TURNOUT AND PARTY GROUPS

Notes: Observations are at the constituency-level. The sample period is 1893 up to 1933 and includes 14 Reichstag elections. Standard errors clustered at the constituency-level in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ∆Growth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics. All controls are interacted with Election FE. Turnout and Antisemite votes are not available for 1919.

Source: Author’s calculations.

shows no impact on turnout. For the leftwing, we see that using only support for the Socialist SPD after WWI without Communist votes yields similar results as Table 1. ^{Footnote 25} This suggests that veterans mainly affected the left’s moderate and clearly pro-democratic parts rather than its radical fringes. Within the center, veterans had almost no effect on left-liberals, while yielding insignificant increases and reductions in Catholic-minority and right-liberal votes, respectively. Looking at the rightwing in Columns

(5) and (6) shows the conservatives as the clear winner and no significant impact on the Antisemite vote share. This contrasts with the widespread perception that veterans as a social group were predominantly supporting the Nazi Party. ^{Footnote 26}

Table 4 provides some first insights about the potential mechanisms behind my findings: the absence of a clear impact on centrist parties and the far-left favor a direct voter transition from moderate left to rightwing rather than a gradual shift across the entire political spectrum. This is already surprising from a rational viewpoint since soldiers’ support for the left, if anything, should have increased as WWI created more need for redistribution to make up for war-induced disabilities. Crucially, parties’ politico-economic orientations did not significantly change after WWI. A persistent voter transition without major changes in party platforms strongly hints at a shift in individual political preferences and suggests that moderate left veterans turned toward nationalism—the main selling point of rightwing parties. A general reorientation away from redistribution, on the other hand, seems unlikely since Communist votes were not affected by this shift.

Why would moderate left soldiers turn toward nationalism and rightwing parties after WWI? A plausible explanation for this could be anti-communist sentiments spread via the stab-in-the-back myth. While this would be broadly in line with the Red Menace theory in the sense that rightwing parties gained from Anti-Communism after WWI, it raises the question why such sentiments would also affect moderate left veterans as opposed to the more rational fears of the propertied middle class. A straightforward explanation for this could be that soldiers particularly longed for an easy and seemingly coherent rationalization of the traumatic defeat in 1918. Such demand for reason bears a strong resemblance to religious coping after natural disasters, whose historical relevance has also recently been discussed by economists (Belloc, Drago, and Galbiati Reference Belloc, Drago and Galbiati2016; Bentzen Reference Bentzen2019).

If veterans were indeed susceptible to the stab-in-the-back myth and Anti-Communism, neither the losses of the moderate left SPD nor the gains of the conservatives are surprising. ^{Footnote 27} While the far-left actually favored a Soviet Germany and the Catholic-bourgeois parties of the center could hardly be associated with Communism, the allegation of secretly collaborating with foreign Bolsheviks was mainly problematic for the moderate left. The conservative DNVP, on the other hand, had started to spread fears of a leftwing revolution and the stab-in-the-back myth since its creation in late 1918 (Thimme Reference Thimme1969). In spite of being founded by key figures of Imperial Germany’s nationalist and monarchic elites, the DNVP re-branded itself as an inclusive, classless people’s party under the unifying theme of nationalism and successfully wooed blue-collar workers. ^{Footnote 28} This appears to have been a key factor in attracting veterans disenchanted with the SPD since the right-liberal DVP, which used very similar anti-communist propaganda but remained skeptical of the working class, did not benefit at all (Liebe Reference Liebe1956; Frye Reference Frye1963). ^{Footnote 29}

The final three columns in Table 4 provide further support for the Anti-Communism channel. More precisely, they show a notable shift in veterans’ party preference during the last three elections in 1932 and 1933 that moved the effect on the Nazi Party entirely toward the DNVP. ^{Footnote 30} This seems puzzling at first since the Nazi vote share increased from 18 to 37 percent between 1930 and 1932, and voting for the far less successful DNVP was likely counterproductive in aiding a rightwing takeover of power. If veterans were susceptible to Anti-Communism, however, this shift is not overly surprising. Since 1930, the Nazis have upped their anti-capitalist rhetoric, fiercely attacked the DNVP for its middle-class links, and even founded their own trade union, which became involved in several strikes (Beck Reference Beck2016). This new leftwing-nationalist course was also actively denounced in the conservatives’ campaigns (Jones 2015). Hence, if veterans’ support for nationalist parties was driven by Anti-Communism, the change in the Nazi Party’s tactics should have precisely created the patterns observed in the data.

Alternative Mechanisms

There are two other potential explanations: first, veterans’ attitudes could have been altered by any other effect of war beyond the feeling of defeat. Second, rather than a change in veterans’ political preferences and political demand, my findings could be driven by a shift in political supply. Table 5 assesses the plausibility of these channels using a triple-diff design, which allows the baseline OLS results for left- and rightwing parties to vary with an additional factor. More precisely, I include interactions between Veterans _i × Post _t and a dummy to determine whether a variable linked to a particular mechanism is above the sample median. This new interaction term quantifies how the veteran effect in the half of the sample above the median differs from the one below. An interaction between the respective median dummy and Post _t is also included in each regression but not reported.

TABLE 5 ROLE OF DIRECT WAR EFFECTS AND POLITICAL SUPPLY

Notes: Observations are at the constituency-level. The sample period is 1893 up to 1933 and includes 14 Reichstag elections. Standard errors clustered at the constituency-level in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ∆Growth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics. All controls are interacted with Election FE. All constituent terms of interaction effects are included in the estimation but only relevant ones are reported in the table. Source: Author’s calculations.

I start by exploring the impact of direct economic and psychological effects of war: exposure to violence, disability, and political socialization. As discussed earlier in the subsection on veterans during the Weimar Republic, the historical literature on Weimar Germany has extensively debated whether the traumatic experiences of WWI soldiers paved the way for a violent post-war society and the rise of the Nazi Party. Columns (1) and (2) in Table 5 indicate a significantly different effect in the high-casualty part of the sample for the leftwing, which, however, goes in the opposite of the expected direction. Veterans’ effects on the rightwing, in turn, do not seem to depend on WWI casualties and thus speak against the brutalization hypothesis.

Closely related to this are the economic effects of war-induced disability. While the new government did its best to reintegrate the 2.7 m war-disabled veterans, their preferential treatment through protected job types, for example, also alienated them from the rest of the population (Bessel Reference Bessel1993; Cohen Reference Cohen and David2012). This alienation or general dissatisfaction with the benefit system could have led those disabled by the war into political extremism. In specifications 3 and 4, I allow the baseline effect to differ depending on the population share of male individuals eligible for WWI-related benefits from the 1924 census of the war-disabled. These new interactions are small

and insignificant at conventional levels, which rules against the important role of WWI-induced economic shocks and disability.

The last war-related mechanism concerns the importance of experiences from ages 18 to 25 for the development of political attitudes (Krosnick and Alwin Reference Krosnick and Alwin1989; Giuliano and Spilimbergo Reference Giuliano and Spilimbergo2014). Since many WWI draftees were in this age group, the war may have influenced elections through identity formation among its youngest participants. The literature has noted that the so-called front generation was still accustomed to the stability and predictability of the pre-war period but saw this world turned upside down through military defeat and the leftwing-led revolution in 1918 (Bessel Reference Bessel and Roseman1995; Verhey Reference Verhey2000). I test this possible explanation in Columns (5) and (6) by using the share of men aged at most 25 during wartime among the draftable population. For the leftwing, the interaction estimate goes against the expected direction. The effect on the right is statistically indistinguishable from zero. ^{Footnote 31}

Another competing explanation could be that veterans’ change in voting was triggered by a shift in political supply after WWI. For instance, veterans may have become rationally anti-communist through actual exposure to far-left activity, which was likely higher in leftwing strongholds. Alternatively, moderately left veterans could have been actively targeted by rightwing parties or groups after 1919. The remaining specifications in Table 5 test the plausibility of these mechanisms. Interactions with the 1919 USPD vote share in Columns (7) and (8) indicate that veterans benefited the right less in far-left strongholds. This suggests that Communist activists may have successfully obstructed the DNVP’s 1919 campaign rather than driving voters into the arms of the right. In specifications 9 and 10, I look at the impact of DNVP propaganda capacity in 1919. This is approximated by the 1914 chapter intensity of the German National Association of Commercial Employees (DHV), which, according to historians, was a crucial vehicle for the DNVP’s initial efforts in attracting blue-collar workers (Liebe Reference Liebe1956; Ohnezeit Reference Ohnezeit2011). The coefficients suggest that this mechanism also cannot explain the veteran effect. Online Appendix Table B.7 provides further evidence against a shift in political supply. ^{Footnote 32}

Insights from Referenda and Presidential Elections

So far, the analysis has two important drawbacks: first, one cannot distinguish political preferences from party loyalty and rule out the possibility that veterans supported the DNVP for other reasons than nationalism and Anti-Communism. Second, parliamentary votes do not allow inferring an order of preference over parties and policies. This forbids answering the historically relevant question of whether veterans’ strong support for the DNVP should be interpreted as a closer alignment of preferences or as a rejection of the Nazi Party and Hitler’s seizure of power. I tackle both issues by exploring veterans’ effects on two referenda and two presidential elections. To account for constant, yet unobserved, preferences for particular parties and policies, I add the (non-interacted) baseline controls and seven party strength variables. These are identical to the seven sets of estimated constituency fixed effects for each party group in Column (2) of Table 1 and Columns (2) to (7) in Table 4. ^{Footnote 33}

The first two columns in Table 6 present veterans’ impact on referenda turnout. ^{Footnote 34} The 1926 referendum was on expropriating Germany’s former aristocracy. This was a classic leftwing, redistributional topic, and the right made heavy use of anti-communist propaganda to discourage voters from supporting it (Ohnezeit Reference Ohnezeit2011). Column (1) shows that veterans reduced turnout in this ballot. The 1929 ballot was about rejecting the so-called Young plan, which finalized Germany’s WWI reparations. This was a classic rightwing issue and framed as a ballot on Germany’s national sovereignty and supposed “enslavement.” As Column (2) shows, former soldiers increased turnout in this referendum. Taken together, both results suggest that it was indeed the DNVP’s stance on nationalism and Anti-Communism that won them veterans’ political support.

TABLE 6 VETERAN EFFECT ON REFERENDA AND PRESIDENTIAL ELECTION RESULTS

Notes: Observations are at the constituency-level. Robust standard errors in parentheses: *p < 0.1; **p < 0.05; ***p < 0.01. Included control variables are % Male WWI-eligible, % ∆Growth M-F 1910–19, % Male blue-collar (all), % Male blue-collar (war), % Military, % Male students, % Catholics.

Source: Author’s calculations.

Specifications 3 to 6 look at the presidential election of 1925, where candidates are roughly classified by their endorsing party camps. The effects on the first round show a strong shift from left to right arising from war participation and thus strongly resemble the baseline party results. Importantly, the vast majority of former soldiers’ votes went to Karl Jarres who was a right-liberal endorsed by the DNVP. The Nazi Party candidate Ludendorff gained to a much lower degree. In the second round, rightwing parties agreed to nominate the independent former WWI General Paul Hindenburg, and, as shown in Column (6), the veteran effect moved almost entirely to this new rightwing candidate. As shown in Online Appendix Table B.8, this came at the expense of Wilhelm Marx of the Centre Party, who was endorsed by the Socialist SPD rather than the far-left Thälmann, and thus reaffirms that war participants only hurt the moderate part of the leftwing. ^{Footnote 35}

In 1932, Hindenburg became the candidate of the center and moderate left to prevent Adolf Hitler from becoming president (Mommsen Reference Mommsen1996). In this election, war participants completely withdrew their support for Hindenburg and voted for the even more rightwing candidates Hitler and Duesterberg, who was endorsed by the DNVP and leader of the paramilitary Stahlhelm group. According to Columns (8) and (9), the majority of veterans opted for Duesterberg. The effect on Hitler’s votes is smaller and less precisely estimated, with a p-value of about 0.18. After Duesterberg dropped out, the DNVP leadership advised its voters to abstain and endorsed neither of the remaining candidates (Jones 2015). In spite of this recommendation and the availability of another nationalist option in the form of Hindenburg, Column (10) shows that the majority of former soldiers apparently supported Hitler in the second round. These results indicate that, even though veterans on most occasions went with the DNVP and its candidates, they were not opposed to the Nazi Party and in 1932 supported Hitler’s (unsuccessful) bid for becoming president.

Counterfactual

My analysis finishes by revisiting the thought experiment in the DiD estimates subsection and, having established the baseline estimates’ robustness, providing a more accurate counterfactual of Weimar Germany’s political landscape without the political influence of WWI veterans. In order to do this, I take the event-study estimates from Figure 3 at face value and calculate counter-factual vote shares for the leftwing and rightwing in all parliamentary elections after WWI under the assumption of no veteran effect. ^{Footnote 36} This does not consider the actual seat allocation in parliament, but, given that the Weimar Republic used proportional representation, seat shares should be closely linked to popular support. Figure 5 shows the results from my calculations. ^{Footnote 37}

FIGURE 5 COUNTERFACTUAL EVOLUTION OF PARLIAMENTARY ELECTION RESULTS IN THE GERMAN EMPIRE AND THE WEIMAR REPUBLIC ASSUMING NO VETERAN EFFECT AFTER WWI

Notes: The dark grey-shaded area marks the WWI period. The vertical lines denote the last pre-WWI and the first post-WWI election. Data covers only the area contained in Weimar Germany’s borders and therefore may differ from official national aggregates for elections prior to 1922. For better visibility, the distance between elections in 1932 and 1933 has been artificially stretched.

Source: Author’s calculations.

The first thing to note is that the broad electoral trends between 1912 and 1920 would have been very different had it not been for WWI veterans. The rightwing would have suffered heavy electoral losses. Veterans’ shift in political preferences seems to have prevented this scenario as well as a much clearer victory for the left. According to my calculations, leftwing parties would have seen considerably higher vote shares throughout the Weimar period. Even during the last Weimar election in 1933, they would have received about 48 percent of the vote, compared to only 37 percent for all rightwing parties combined. While one can naturally only speculate what would have happened under such a scenario, it is certain that non-Nazi coalitions would have been much easier to achieve throughout the 1930s. As the graph clearly shows, the revival of the right in the wake of the Great Depression would have happened also without veterans’ influence. ^{Footnote 38}