Democracies die when autocrats convince enough citizens to vote for them. Across countries, there is substantial variation in the willingness of voters to support leaders who agitate for autocratic institutions. Cross-sectional studies have shown that the character of local historical political institutions is strongly correlated with subsequent democratic institutions and attitudes toward democracy (Bentzen, Hariri, and Robinson Reference Bentzen, Hariri and Robinson2017; Giuliano and Nunn Reference Giuliano and Nunn2013). This is also true within Europe (Downing Reference Downing1989; Ertman Reference Ertman1997). Supporting this, countries with additional historical experience with inclusive urban institutions exhibit higher levels of democracy today (see Fig. A1 in the Online Appendix). There are, however, two unanswered questions. First, whether this reflects the persistence of inclusive institutions, their impact on state-making, or their influence on the development of democracy-supporting attitudes is unknown. Second, the lack of exogenous variation in institutions in prior analyses combined with the extensive number of causes of institutional change makes it hard to gauge whether the relationship can be interpreted as causal.

In most cases, autonomous city republics were destroyed following the advent of the modern state in nineteenth- and twentieth-century Europe. Yet, self-governing cities often played an important role in the state-making and regime change processes that led to modern democracy (Ertman Reference Ertman1997). To reduce the risk of these mechanisms being at play, this letter examines the relationship between historical self-government and later support for democracy in a setting where historical institutions were destroyed and had no impact on state-building efforts (Bentzen, Hariri, and Robinson Reference Bentzen, Hariri and Robinson2017). A possible link between early institutions and democracy in such cases is by teaching citizens the mores of democracy (Bentzen, Hariri, and Robinson Reference Bentzen, Hariri and Robinson2017; de Tocqueville Reference de Tocqueville1945), thus making citizens less likely to support autocrats once democracy is introduced. A second possible link is local informal institutions. To provide an initial examination of these mechanisms, this letter examines whether historical experience with local self-government in the County of Hauenstein (located in the German state Baden-Württemberg) made citizens of the former county less likely to vote for the Nazi party.

The lack of exogenous variation in institutions in prior studies makes it hard to claim that any observed relationships can be given a causal interpretation as many (potentially unmeasured) factors, such as geographical conditions ( Ahmed and Stasavage Reference Ahmed and Stasavage2020), determine the character of institutions. I mitigate this issue by using a geographic regression discontinuity (GRD) design that compares the voting patterns of towns and villages within the county to adjacent towns and villages just outside. Using newly coded data on the voting behaviour and political history of 177 towns, I find that areas within the county were substantially less inclined to vote for the Nazi party in July 1932.

I argue that this result reflects the county's history of extensive self-governing institutions that left its former citizens more appreciative of inclusive institutions. I consider several implications of this argument. First, I provide evidence that the adjacent towns and villages that had experience of local self-government had levels of support for the Nazi party that were similar to the ones exhibited by areas within the county. Second, I show that areas within the county were also less inclined to vote for the DNVP (which sought to re-introduction the monarchy) in 1919. Third, I find that the difference in political behaviour persisted as Hauenstein areas had a higher turnout and less support for the DNP in 1969. Fourth, I show that the difference was not visible before the emergence of self-government in Hauenstein, as pre-county political institutions were similar on both sides of the border. In addition, I provide evidence that this relationship cannot be explained by the religious affiliation of the towns and villages (Spenkuch and Tillmann Reference Spenkuch and Tillmann2018); their economic development; their geographical, agricultural, and climatic endowments (Ahmed and Stasavage Reference Ahmed and Stasavage2020); their inheritance laws (Hager and Hilbig Reference Hager and Hilbig2019); their economic conditions during the depression; and their exposure to warfare, prior antisemitism (Voigtländer and Voth Reference Voigtländer and Voth2012), pre-Nazi Jewish presence, or migration.

Empirical Strategy

To mitigate possible confounding from pre-existing differences between Hauenstein and its surroundings, I use a geographic regression discontinuity design (GRD) that limits comparisons to areas close to the county's boundaries. This design identifies the effect of self-government under the assumption that support for autocratic parties would vary continuously around the boundary in the absence of self-government. To ascertain the plausibility of this assumption, I first present my data on electoral support and the county's boundary. Next, I examine the behaviour of several predictors of self-government (and autocratic support). If these vary continuously around the boundary, it raises confidence in the assumption.

Data

My sample consists of all historical towns and villages located within 10 km of the border of the County of Hauenstein. A total of 177 towns were identified based on the Historischer Atlas von Baden-Württemberg (Schröder Reference Schröder1988). The county's boundary is drawn based on a map presented by Bader (Reference Bader2020). In addition, the ownership of each town is checked against information found in the atlas. Based on this, 72 out of 177 towns were part of the county between 1371 and 1745. Figure 1 depicts the location of the county and the towns in the dataset. Only three towns were misclassified based on the map (1.7%), indicating that it captured the county's territory well.

Figure 1. The County of Hauenstein.

Note: The upper graph depicts the County of Hauenstein within contemporary Germany. The lower graph shows the towns in my sample, including whether they were part of the county.

The takeover by the NSDAP is one of the most prominent examples of a democratic breakdown instigated by a party with popular support. Thus, I measure the vote share of the NSDAP party in the 1932 July election as my outcome (coded based on the Historischer Atlas von Baden-Württemberg Schröder Reference Schröder1988).Footnote ¹ The average share in the sample is 0.35, while the standard deviation of the measure is 0.21. Figure A2 in the Online Appendix plots the vote share of each town together with the borders of the county. I later present analyses that use support for other autocratic parties as the outcome.

Specification and controls

To ascertain whether a legacy of self-government impacts later support for autocrats, I use a GRD design and limit comparisons to towns around the county's boundary. Thus, in expectation, making the assignment of territorial ownership closer to as-if random. I estimate variants of the following two models:

(1)$$Y_i = \beta {\rm Count}{\rm y}_i + \gamma ( {\rm geographic}-{\rm loc}.) _i + \delta _d{\rm dist}.-{\rm borde}{\rm r}_i + \epsilon _i$$

(2)$$\eqalign{Y_i = & \beta {\rm Count}{\rm y}_i + \gamma \; ( {\rm geographic}-{\rm loc}.) _i + \delta {\rm dist}.-{\rm borde}{\rm r}_i + \alpha ( {\rm Count}{\rm y}_i \times {\rm dist}.-{\rm borde}{\rm r}_i) + \epsilon _i} $$

where Y _i is the NSDAP vote share in town i, and County_i is an indicator equal to 1 if town i belonged to the County of Hauenstein, and 0 otherwise. γ denotes a vector of coefficients for controls measuring the geographic location of town i: latitude, longitude, and four geographic segment fixed effects based on splitting the observations along their latitude and longitude using percentiles (see Fig. A3 in the Online Appendix for a visualization of the segments). δ _d are fixed effects for the Euclidian distance to the county border in Equation 1. The distance segments, d, are in 1 km bins. δ is the coefficient for the Euclidian distance to the border of the county in Equation 2, while α is the coefficient for the interaction between the distance (the running variable) and belonging to the county. All regressions are estimated with ordinary least squares (OLS).

Limiting comparisons to towns near the border should mitigate problems of omitted variable bias. Yet, one might worry that other factors sharply vary around the border. I check this using the following geographic and climatic controls: altitude, slope, suitability for agriculture, variation in suitability, annual mean temperature, annual mean rainfall, and river access. I also examined social and political characteristics: Alemannic settlements, market presence in 1500, population size in 1500, and years since an agglomeration was first mentioned. Sources are presented in the ‘Data’ section of the Online Appendix, along with the spatial distribution of all variables.

The identifying assumption of my design implies that there is no jump in characteristics before the adoption of self-government when comparing towns close to the county's borders. Figures 2 and A9 show little evidence of a discontinuity in pre-adoption covariates. Furthermore, in Figs A10 and A11, I provide evidence that the same is true for post-adoption covariates. Thus, there is support for this key assumption.

Figure 2. Do observables vary smoothly at the border?.

Note: The x-axis depicts p-values from regressions that estimate whether each covariate is discontinuous at the border. Specifications are based on equation (2) (p-value for β). The dashed vertical line marks the 0.05 level of significance.

Results

The average vote share of the NSDAP in Hauenstein was 27. This is lower than the national average (37) and the average in Baden (37) (results from Statistischen-Reichsamt 1932). I also examine support for the NSDAP (graphically) around the borders of the former county in Fig. 3. There is a negative jump in electoral support when moving from areas outside the county (dashed grey lines) to the county (black lines). Towns within the county have a vote share for the NSDAP of around 13 percentage points less (0.62σ).

Figure 3. Self-government and support for the NSDAP.

Note: The area to the right of the vertical line contains towns within the county, represented by black lines. The dashed grey lines are towns outside the county. Dots correspond to means for 2.5 km bins, while the lines are based on a linear fit for each side of the border. The surrounding lines are 95 per cent confidence intervals.

In the Online Appendix, I present regression results based on Equation (1) in Table A2 and Equation (2) in Tables A3 and A4. I find a similar negative association. The correlations are robust in accounting for geographic location, distance to the border, and segment fixed effects, and they remained when I introduced the above-mentioned controls. Finally, this persists when I consider a number of alternative explanations that occurred after the adoption of self-government: Protestantism, rules of inheritance, industrialization, economic growth (and crisis), pre-1932 migration, warfare, historic antisemitism, and pre-Nazi Jewish presence. Sources and spatial distribution for these variables are presented in the ‘Data’ section of the Online Appendix. Accounting for these explanations does not change my findings. Based on the most restrictive model (Model 4 in Table A4), towns within the county have, on average, a 14.4 percentage point lower vote share for the NSDAP (equal to a 0.68σ change in vote share and relative to a mean of 35). Thus, the estimated impact is substantial.

In the Online Appendix, I present additional checks. Table A5 shows that the results are not dependent on bandwidth choice. Table A6 recovers similar estimates using non-parametric models (using both sharp and fuzzy estimations). Estimating a ‘Donut’ RDD excluding towns located at the border does not alter the findings (see Table A7). Table A8 examines whether the findings are driven by a particular area. The results hold when excluding different geographical parts of the sample. Table A9 shows that the relationship is unaffected by including contemporary election-district fixed effects. Figure A12 finds a similar discontinuity when using a quadratic fit and a local polynomial smooth fit rather than a linear fit. Figure A13 presents evidence from placebo tests where the border is moved west, east, north, and south. I find little evidence of similar discontinuities at the placebo borders. Table A12 shows that it is not proximity to Switzerland that explains the findings.

Empirical Implications

Areas within the County of Hauenstein exhibit lower levels of support for the NSDAP. However, it is not clear whether this reflects the county's historical self-government or whether an alternative county characteristic explains it.

If a legacy of self-government drives this difference, I would expect neighbouring areas with self-government to hold similar lower levels of electoral support for autocrats. To test this, I coded data on the ownership of all adjacent towns and the political regime in place in each polity. I provide information on this coding in the ‘Information on the surrounding polities’ section of the Online Appendix.

Using this data, I plot the predicted level of NSDAP support across historical experience with self-government in Fig. 4. I find that non-Hauenstein towns without self-government have the highest level of support for the NSDAP. The vote share of the NSDAP drops around ten percentage points in towns with self-government experience, and it is lowest in Hauenstein towns. This suggests that the difference in support for the NSDAP is related to the institutional history of each town. In Table A15 in the Online Appendix, I test whether my overall results depend on the inclusion of a specific polity in the comparison group but find little evidence that this is the case.

Figure 4. Support for the NSDAP by historical experience with self-government.

Note: Estimated using OLS. Based on Models 1 and 3 from Table A2. Lines are 95 per cent confidence intervals.

Furthermore, I assess additional implications. First, historical self-government should affect support for autocratic parties generally and not just the NSDAP. Therefore, I test whether support for the DNVP, an extremist right-wing party that sought to introduce monarchy (Jones Reference Jones2020, 35–41) in the 1919 election, is lower in the county. This also serves to rule out any impact of the Great Depression. Second, it is argued that historical self-government creates politically engaged citizens who oppose undemocratic leaders (Bentzen, Hariri, and Robinson Reference Bentzen, Hariri and Robinson2017). I evaluate this by examining whether former county areas were more inclined to vote in the first election with data on turnout (data on vote share for the DNVP, mean = 0.03 & SD = 0.08, and turnout, mean = 0.79 & SD = 0.07, in the 1969 election is coded from Schröder Reference Schröder1988).Footnote ² Third, if variation in support for autocratic parties can be traced to the county's self-government, there should be no differences in political institutions before 1371, when autonomy was first introduced. I distinguish between towns that were controlled by the Imperial family (the Hohenstaufens) prior to the collapse of the dynasty and other towns. The Imperial family had significant holdings in the area and was the strongest political unit before their demise. Thus, they are likely to cause pre-county differences in political attitudes. Consequently, I test whether Hauenstein towns were more likely to be under Imperial political rule in 1250 (coded based on Vollmer Reference Vollmer and Schröder1988; mean = 0.12 & SD = 0.32). The upper left graph in Fig. 5 shows a negative jump in support for the DNVP when moving from areas outside Hauenstein (grey dashed lines) to areas inside the county (black lines). A similar but positive jump is visible for voter turnout in 1969, which can be seen in the upper right graph. The lower left graph shows that there was no discontinuity in political institutions prior to the introduction of self-government in Hauenstein. The lower right graph confirms the above findings using regressions based on Model 3 from Table A2 with the three alternative outcomes.

Figure 5. Test of implications

Note: The area to the right of the vertical line contains towns within the county, represented by black lines. The dashed grey lines are towns outside the county. Dots correspond to means for 2.5 km bins, while the lines are based on a linear fit for each side of the border. The surrounding lines are 95 per cent confidence intervals. Coefficients for belonging to the county are based on Model 3 from Table A2 with the three different outcomes. Vote share and turnout are re-scaled to 0–1 for comparability.

In the ‘Additional implications’ section of the Online Appendix, I interrogate additional implications. First, I provide suggestive evidence that this difference was present in the first election in Imperial Germany. Second, I show that former county areas also exhibited lower support for autocratic parties after 1932. Third, I explore whether a legacy of self-government is related to support for the Communist Party: it is not. Thus, I find that a legacy of self-government seems to depress support for (right-wing) autocratic parties.