Evidence from the Code Napoleon in Germany∗

Law and Social Capital: Evidence from the Code Napoleon in Germany∗

Johannes C. Buggle†

Abstract

I test whether legal institutions crowd-in social cooperation in the long-run, using the introduction of the Code Napoleon in parts of 19th century Germany as a historical experiment. I find that the application of the Code Napoleon is associated with higher levels of trust and cooperation today. This finding is robust to an identification strategy that uses only individuals located around a discontinuity in the number of years the Code Civil was used. Results from a falsification test that moves this discontinuity artificially, as well as the comparison of pre-treatment characteristics support the interpretation of a causal effect. In addition, regions around the discontinuity are similar in post-treatment economic development and inequality. On the contrary, the positive social consequences of the Code Civil manifest themselves in less political fraud in elections from 1871 to 1900, and in more “bridging” social capital in the 1920’s.

Keywords: Institutions, Long-Term Persistence, Trust, Social Capital.

JEL-Classiﬁcation: N43, O10, P48, Z10

∗I am grateful to Yann Algan for his support and to Francesco Amodio, Sascha Becker, Davide Cantoni, Francesco D’Acunto, Barry Eichengreen, Pauline Grosjean, Emeric Henry, Marc Sangnier, Andrei Shleifer, Guido Tabellini, and Joachim Voth for helpful comments. Seminar participants at the FRESH Meeting London, the UPF Reading Group on Persistence, the Spring Meeting for Young Economists 2013, Sciences Po, and the Warsaw-Penn Workshop on “Institutions, Culture and Long- Terms Effects” provided useful suggestions. I thank Davide Cantoni and Joachim Voth for kindly sharing data, and the DIW Berlin for help with the SOEP data access. Valuable research assistance was provided by Ferdinand Lutz and Malte Syman. All errors remain my own. †Department of Economics, University of Lausanne, [email protected]. 1 Introduction

“If the laws are good, morality is good. If the laws are bad, morality‘s bad”- Diderot1

At least since the work by Banfield (1967), Coleman (1988) and Putnam (2001), social capital has been associated with many beneficial economic and social outcomes. Trust, as one dimension of social capital, for example eases cooperation between individuals and collective action.2 Why people trust each other and cooperate in some societies, but not in others, is therefore a pertinent question. In this paper, I research whether legal institutions that govern social interactions promote a culture of cooperation and trust. While there exists a strong correlation between the quality of legal institutions and trust across countries that seems to support Diderot’s notion (see above 1), this association does not necessarily reflect a causal impact of the law.3

[Figure 1 about here]

The identification of a causal link is made difficult by the endogeneity of legal institutions, which are themselves a function of cultural attitudes and preferences. To overcome this common identification problem, I make use of a historical experiment that is characterized by the imposition of a legal system on a society from outside. This historical experiment is the introduction of the Code Napoleon in parts of 19th century Germany, a dramatic positive shock to the quality of existing law. The Code Napoleon was the most modern legal code at that time, created to spread the ideas of the French Revolution and to modernize the pre-existing social order of European societies. Its most revolutionary concept was to treat all individuals as equals. This unprecedented degree of legal equality removed existing barriers to inter- and intra-class cooperation, and “encouraged the liberation of the individual from corporative bonds and the establishment of a civil society" (Fehrenbach, 2008). By showing that the application of the Napoleonic Civil Code throughout the 19th century goes along with higher levels of social trust today,

I provide novel evidence on the relationship between the law and social cooperation from estimating on the micro-level within a country.4

Formal theories have modeled a positive relationship between legal enforcement and norms of trust (e.g.

1From the “Continuation of the Dialogue between A and B” (Diderot, 1992). The extended quote is: “A: What do you mean by morality? - B: I mean a general obedience to laws, either good or bad, and such conduct as follows from that obedience. If the laws are good, morality is good. If the laws are bad, morality’s bad.” 2See Algan and Cahuc (2013) for a review of the literature on the impact of trust on economic outcomes. 3An association between historical political institutions and cooperation has recently been shown by Putnam, 1994; Tabellini, 2008 and Tabellini, 2010; as well as Guiso, Sapienza, and Zingales, 2008a. 4The long-run consequences of the law for economic outcomes has been intensively described by scholars researching different legal origins. See for example López-de-Silanes et al. (1998) and La Porta, López-de-Silanes, and Shleifer (2008).

2 Tabellini, 2008; Guiso, Sapienza, and Zingales, 2008b). Theoretically, a shift from a state of the world with weak (or partial) law enforcement and high incentives for exploitative behavior, to an environment where legal enforcement is strong (or impartial) and individuals are incentivized to cooperate, can lead to a permanent increase in social trust in the society. Rather than limiting their actions to a small circle of persons, individuals will apply good conduct towards everyone in a society if law enforcement is strong and impartial (“limited" versus “generalized" morality in Tabellini, 2008). A large, positive shocks to the quality of institutions can manifest themselves in higher trust even several generations later, as beliefs transmitted inter-generationally from parents to children incorporate past experiences from different institutional environments. To test whether the rule of law crowds-in norms of cooperation over time, one would ideally randomly select and expose parts of a society for several generations to different legal systems, and compare levels of social cooperation afterwards. The spread of the Code Napoleon to some areas of 19th century Germany approximates such an experiment. The Code was an institution that had not been created by the individuals to which it applied, but it had been imposed from outside on some regions of historical Germany during the Napoleonic Wars. It was the centerpiece of a modern legal system that established universal access and equality, that enlarged commercial freedom and that restricted elite’s privileges. Importantly, the Code lasted in several regions for a sufﬁciently long time period in order to affect beliefs long-lastingly, i.e. for more than three generations (Guiso, Sapienza, and Zingales,

2008b).

My empirical analysis of the long-term effects of the Code Civil starts by comparing social trust and cooperation of regions that have used the Code Civil in the 19th century to regions that did not apply it, defining treatment as the number of years the Code Civil was applied before 1900.5 I estimate a robust positive association between the duration of application of the Code Civil and contemporaneous levels of trust. However, the Napoleonic occupation did not only alter the law. In addition, serfdom and guilds were abolished, and land was freed. The treatment variable might proxy for the legacy of those reforms. Testing alternative institutional reform definitions against the Code Civil, I find that the Code

Civil has clearly the strongest inﬂuence on social capital levels today. However, the regional analysis is is susceptible to unobserved heterogeneity across regions. For example, differences in initial conditions across regions that applied the Code Civil could drive the observed results.6 In this case, rather than identifying a treatment effect, the estimation would just pick up path dependency of pre-existing differ-

5A uniﬁed civil code was installed in all German provinces in 1900, the Bürgerliches Gesetzbuch. 6Indeed, my identiﬁcation strategy requires the adoption of the Code Civil to be uncorrelated with initial social capital levels.

3 ences. To address concerns about the endogeneity of the Code Civil, I use a second empirical strategy that compares individuals living in neighboring districts separated only by the duration of application of the French legal system. While in some districts the Code Civil was used for more than 90 years, their neighbors either never used it or only for a short duration. I argue that close-by districts are very similar along many dimensions, and that treatment status along this artificial “border" discontinuity results from a number of historical accidents and district idiosyncrasies, but does not reflect any underlying systematic difference. I show that border districts are indeed very similar across a number of geographic and socio- economic pre-treatment characteristics. Empirical estimates are consistent with those obtained from the cross-regional comparison: treated localities have significantly higher levels of trust. This result appears robustly in local linear regressions, a spatial regression discontinuity design, and even when fixed effects for neighbor-pairs are controlled for. However, in the absence of any treatment one would not expect to find sizable differences in norms across nearby individuals. Indeed, a falsification test confirms this presumption. Coefficients from placebo treatments that move the discontinuity artificially outwards to the North-East or inwards to the South-West are small and insignificant.

I then investigate the mechanisms that explain the uncovered relationship. A ﬁrst likely candidate is economic development, as Acemoglu et al. (2011) document a positive effect of the Napoleonic institutions on urbanization. Contemporaneous trust might be a result of differences in economic development, and only indirectly be ascribed to the legal code. I investigate extensively whether there is evidence for different paths of post-treatment economic development. My results suggest that observable economic differences are not very likely to be the main driver of the observed relationship. I ﬁnd only limited evidence for persisting differences in long-run economic development when considering the cross-section of all sampled regions. More importantly, districts along the discontinuity are balanced regarding a broad range of early and late 19th century development outcomes, including growth in city population, and per capita income today.7

Regarding alternative channels, I analyze potential effects of the Code Napoleon on inequalities in the distribution of land. Land was, according to my estimations, not more equally distributed at the end of the 19th century on either side of the border. In addition, in the spirit of Fisman and Miguel (2007),

Guiso, Sapienza, and Zingales (2012), and Nannicini et al. (2013), I investigate whether the Code Civil is

7Results from a difference-in-difference estimation using city population over the period 1700 to 1910 and considering the entire sample of regions confirm the positive effects found in Acemoglu et al. (2011). However, the same estimation does not show any significant differences across the border. In addition, examining the evolution of urbanization rates as in Acemoglu et al. (2011) in a restricted sample consisting of only territories at the border, it appears that while “treated” regions grow faster in the first half of the 19th century, the gap in urbanization is closed by the turn of the 19th century. Therefore, differences in income cannot account for the differences in trust observed across the Code Civil discontinuity.

4 associated with stronger rule compliance and less “cheating”. My results show that elites were less likely to commit electoral fraud in national elections to the German Reichstag from 1871-1900 in the treated areas of the border. Although, elites are not representative for the entire society, I interpret this result as evidence for the spread of norms of generalized morality, potentially facilitated by the emergence of a new political class in the regions that used the Code Napoleon. Moreover, I document the existence of differences in historical cooperation in 1920’s Germany using the composition of social associations as indicators for social trust and generalized morality. Consistent with the results on contemporary trust,

I ﬁnd that the application of a legal system that emphasizes equality goes along with a higher share of associations that foster cooperation between members of different societal groups (“bridging” social capital Putnam, 2001). Supplementary evidence obtained from tracing the evolution of one particular type of club over time (shooting clubs, or Schützenvereine), shows an increase in the number of clubs founded in the second half of the 19th century in the treated province Rhineland compared to its neighbor

Westphalia.

This paper adds to the growing literature documenting the importance of historical events and institutions for explaining differences in current beliefs.8 In contrast to existing contributions researching the effect of a past institution on trust and social capital, in particular Guiso, Sapienza, and Zingales (2008a), and Tabellini (2010), I focus speciﬁcally on the effect of legal institutions that regulate interpersonal interactions. Apart from institutions, other historical shocks have been found to alter norms and beliefs of societies long-lastingly (see e.g. Africa’s slave trade in Nunn and Wantchekon, 2011, or the Stasi in East

Germany in Jacob and Tyrell, 2010) and Voigtländer and Voth (2012) document persistence of cultural norms over centuries.9 In addition, several papers modeled the persistence of cultural norms, following the seminal model in Bisin and Verdier (2001), and their interaction with formal institutions. Tabellini

(2008) studies the determinants of cooperation and the evolution of norms over time. In his model, an exogenous shock to the quality of the enforcement institution increases the scope of cooperation. He further argues that when individuals can choose institutions by participating in the political process, good institutions will be reinforced. Similarly, the model in Guiso, Sapienza, and Zingales (2008b) gives a rationale as to why differences in social capital originating from historical shocks are still observable today.10

A very interesting and related paper is Becker et al. (2015), who research the persistent impact of the

8See Nunn (2012) for a comprehensive summary of this literature. 9Aghion et al. (2010) provide evidence for a potentially detrimental role of the state for trust in the case of extensive regulation. 10Empirical evidence of the intergenerational transmission of values can be found in Dohmen et al. (2012).

5 Habsburg Empire, a kingdom characterized by its efﬁcient state bureaucracy, on trust in state actors.

Focusing on individuals living not more than 200 kilometers around the Empire’s former border, the authors ﬁnd that those living in the former Habsburg territory have more trust in institutions today, but are similar with respect to trust in other people. While methodologically the papers share many similarities, the novelty in my paper is the empirical investigation of a legal enforcement institution that regulates primarily citizen-citizen interactions and its effect on cooperation between citizens, rather than the interaction between citizens and the state. Moreover, the historical experiment under study gives additional insights about the role of imposed institutions and relates to the literature on the effects of legal transplants around the world (e.g. La Porta, López-de-Silanes, and Shleifer, 2008). Furthermore, I address some of the potential mechanisms that link institutions and trust in the long-run empirically.

While my research design is targeted at identifying long-term implications of institutions on social cooperation, laboratory experiments are generally better suited to investigate short-term behavioral changes associated with punishment and contract enforcement (see for example Fehr and Gächter, 2000 for experimental evidence; Fisman and Miguel, 2007 for evidence from a ﬁeld experiment). Most recently

Cassar, d’Adda, and Grosjean (2013) gathered interesting evidence for a causal link between institutions and norms of cooperation from experiments conducted in Italy and Kosovo. Finally, my ﬁndings add to the analysis of the socio-economic consequences attributed to institutions in general (see for example

Smith, 1776; Milgrom, North, and Weingast, 1990; De Long and Shleifer, 1993; Acemoglu, Johnson, and Robinson, 2001; La Porta, López-de-Silanes, and Shleifer, 2008; Cantoni and Yuchtman, 2014)11 and the Napoleonic institutions, as described in Acemoglu et al. (2011), in particular. Relative to Ace- moglu et al. (2011), my study differs in the outcomes studied, the long-term perspective, as well as the identiﬁcation strategy that exploits a geographic discontinuity on the micro level.

The paper is structured as follows. Section 2 gives a description of the historical context. Section 3 assesses the effects of the Napoleonic institutions on current social capital outcomes, both across historical regions and across districts along the Code Civil discontinuity. Section 4 tests for the importance of several potential channels. Section 5 concludes.

11Another strand of the literature disputes the fundamental role of institutions for economic growth and development. See for example Glaeser et al. (2004) for the debate on the importance of human capital versus institutions, or Gennaioli et al. (2014).

6 2 Historical Background: The Code Civil in 19th Century Germany

2.1 Introduction and Reception of the Code Civil

At the time of the French Revolution in 1789, Germany was fragmented into more than 300 states with a social organization characterized by feudal institutions privileging the nobility and the clergy (Grab,

2003). Germany’s legal system consisted of a plurality of legal norms that differed often within the same political entity from one town to another.12 Equality before the law did not exist, and peasants were subject to different courts and laws than the nobility (Acemoglu et al., 2011).

The introduction of the Code Napoleon parallels the territorial expansion of the French Empire and the strategic occupations of the German lands during the Napoleonic Wars. The French inﬂuence in Germany started in 1798 with the occupation and annexation of the Rhineland, the entire German part on the left bank of the Rhine. The occupation was quickly followed by the the introduction of French revolutionary institutions, chief among which was the Code Civil in 1802. Napoleon subsequently invaded large parts of the current German territory, primarily to defend the French Empire against Prussia.13 Between 1802 and 1814 Napoleon reorganized the German map and integrated many of the small Imperial Estates into larger states. The satellite states that Napoleon created, as well as other territories that came under direct French control, also adopted the French legal system. These states include Westphalia (Code

Civil from 1810-1815), Brunswick (1808-1814), Province of Saxony (1808-1815), Hanover (1808-1813) and Hesse-Kassel (1808-1814). Moreover, in the indirectly controlled states that became part of the

Confederation of the Rhine, Napoleon pushed for the adoption of his legal code (Arvind and Stirton,

2010). Some states, motivated by the fear of a French invasion, considered or announced its introduction, such as Bavaria, others, such as Baden in 1810, actually adopted the Code Civil.14

The reception of the Code Civil was not uniform. A number of German areas adopted the Code Civil

12These legal norms were derived from roman and canon law, the law of the church, and together constituted the so called common law (Gemeines Recht), not to be confused with the British law system. It was a mixture of customary law and traditions and was executed by the local elite, the church or judges appointed by them. 13The motives for occupying a particular region have been twofold, as described in Acemoglu et al. (2011). First, in order to defend France against Prussia and Austria, the territories in-between the Empires were strategically important. Second, the French revolutionaries wanted to enlarge the territory of the Empire to its natural borders: the Rhine, the Atlantic Ocean and the Alps, (Doyle, 1990, p. 199). The revolutionary leader Danton proclaimed on January, 31st 1793: “The limits of France are marked out by nature. We shall reach them at their four points: the Ocean, at the Rhine, at the Alps, at the Pyrenees.” (quoted in Doyle (1990, p. 200). 14The Code Civil offered the possibility to the newly formed Grand Duchy of Baden to unify the eight different legal systems used in its traditional and freshly gained territories (Sturm, 2011). However, while Baden was put under pressure by Napoleon to adopt the French legal system, it did not get the Code Civil through direct Napoleonic control. In the empirical section I show that results are robust to excluding Baden. In addition, Baden used a modiﬁed version of the Code with around 500 additions. The modiﬁcations concerned primarily the persistence of some peasant obligations towards landowners, although, serfdom had already been abolished in 1783 by the liberal Grand Duke Charles Frederick, and the inheritance legislation that forbade the splitting of large estates (Sturm, 2011).

7 and used it until 1900, the year when the ﬁrst civil code that applied to the entire uniﬁed German Reich, the Bürgerliches Gesetzbuch, was introduced. Other regions used it for a much shorter time period.15

For the persistence of the Napoleonic institutions the allocation of regions at the Congress of Vienna in

1815 proved to be crucial. As a result of the Congress, several of the old rulers regained their power.

They rolled back the Napoleonic institutions in regions such as Brunswick, Saxony, Hanover and Hesse-

Kassel. Several other territories that used the Code Civil were allocated to new rulers. The Code prevailed in those areas where the new rulers could not offer an equivalent civil code, as in the case of the Bavarian

Palatinate (Stein, 2004). In all former Prussian territories, as well as in those that it obtained after the

Congress, Prussia sought to install its own civil code, the Allgemeines Landrecht (ALR) of 1794. The introduction of the ALR in the Kingdom of Westphalia was therefore not a matter of discussion, as about half of the population belonged to Prussia before the French occupation. Only in the newly gained parts west of the Rhine which prior to 1815 were part of the French Empire, there was a debate about whether the Code Civil was to be replaced by the ALR.16 The Rhineland was in a unique situation as it had been applying the Code Civil particularly long before 1815. Acknowledging that the French reforms had already changed the Rheinish society considerably, the Prussian administration under the lead of the liberal Prime Minister Hardenberg decided that it was incompatible with the old-Prussian privileges codiﬁed in the ALR (Heymann, 1914). Therefore, Frederick William III of Prussia decided in

1818 to ﬁrst modernize the ALR and to introduce it afterwards in the entire Prussian Empire (Siebels,

2011). The revision of the Prussian ALR did, however, never take place. That the Code Civil was retained in the Rhine province can on the one hand be attributed to “reform-minded Prussian ofﬁcials" like Hardenberg that “played a crucial role in ensuring that [it] was not immediately swept aside", (Rowe,

2003, p. 281). On the other hand, its continuation was also favored by the local population, in particular the legal profession and the emerging commercial class of the Rhineland that, unsurprisingly, preferred the business friendly Code Civil.17,18

Overall, the length of application of the Code Civil can not be attributed to any singular factor, but it was the result of a complex historical process. Empirical evidence in Arvind and Stirton (2010) support this interpretation. They ﬁnd a diverse set of factors to be important, but none to be sufﬁcient, to explain

15The treatment in the empirical analysis is based on the number of years the Code Civil was used in a territory before 1900. 16Among the formerly Prussian parts of the Rhine province, the introduction of the ALR was only executed in districts on the right bank of the Rhine, such as the districts Duisburg, Essen, and Rees. 17The preference for retention is not surprising as the Code Civil had already changed the Rheinish society drastically. Feedback effects between a change in values and the choice of institutions are therefore likely, as in Tabellini, 2008. 18For the population of the Rhine area the Code Civil also provided a protection against the interference of the new unknown rulers from the East and even “[...] served as a a substitute constitution" (Rowe, 2003, p. 259).

8 the reception of the Code. Among the most important factors are direct French control, the number of territories a region gained in 1815, the presence of a liberal ruler, popular anti-french sentiment as a result of French occupation, and the presence of a proto-industrial economy before 1800.19 Results from regressing the duration of adoption on a set of pre-treatment variables conﬁrms that French rule and proximity to France are two important factors that determined the reception of French law, see Appendix

Table B1.20

As Figure 2 illustrates, the Code Civil existed throughout the 19th century in the Rhineland, consisting of the areas West to the Rhine river and small parts on the right bank of the Rhine, formerly belonging to the Duchy of Berg, as well as the Bavarian Palatinate (given to Bavaria), as well as Baden. In addition,

I follow the coding in Acemoglu et al. (2011) and regard the civil code of Saxony, the Sächsisches

Bürgerliches Gesetzbuch, as an equivalent legal Code guaranteeing similar rights as the Napoleonic

Civil Code. The areas that used the Napoleonic Code Civil in the Western parts of Germany bordered territories that either used common law or the Prussian ALR.

[Figure 2 about here]

2.2 The Code Civil versus alternative Legal Systems in 19th Century Germany

The Code Civil incorporated the ideas of the enlightenment and put into practice the principles of the

French Revolution (Schlosser, 1995). It established a new set of civil liberties that were exceptional for

19th century Germany (Sperber, 1992). The revolutionary characteristic that differentiated it the most from all other existing legal systems was universal equality before the law. Legal equality was granted to all men, removed any form of privileged treatment for the nobility and guaranteed the application of the same rules to everyone independent of income and status (Lyons, 1994). Legal equality was in stark

19In parts of the empirical analysis, I compare the regions of Rhineland that kept the Code to Westphalia that did not keep it. According to Arvind and Stirton (2010), the main dimension along which the Rhineland differed from its neighbor Westphalia was its more advanced economy prior to 1800. In the empirical specification, I take into account pre-treatment differences in economic development. 20Treated regions were occupied significantly longer by the Napoleonic Army, which allowed to install the institutions properly. Secondly, treated regions are located closer to Paris, equally important for whether a region was actually occupied or not. In contrast, one could think that, because of the geographical closeness to France, people in the Western territory were much more in favor of the Revolutionary ideas when it broke out in 1789, and thus much more involved in political activities, such as being member of Jacobin clubs. This was not the case. In the Rhineland for example, “there were hardly any local Jacobins or self-styled patriots sympathetic to the Revolution” (Doyle, 1990, p. 352) compared to other parts of Germany, and even if, these groups were small and consistent only of “scattered individual activists” (Sperber, 1992, p. 489). Contrary to Arvind and Stirton (2010), I find that that the treated regions were less urbanized before 1800. In addition, the literature on the determinants of social capital has shown that Protestantism is favorable for social attitudes and cooperation (Alesina and La Ferrara, 2002; Tabellini, 2008). The share of Protestants enters the regression with a negative coefficient, however. Finally, the number of territories from which the regions were created goes along with a longer adoption of the Code. Changing the institutions and reverting back to former institutions after the Napoleonic occupation was therefore much more costly in the light of the integration of those many small states. That a unified legal code made the integration of additionally gained territories easier was also emphasized by several scholars (see e.g. Becker, 1985; Arvind and Stirton, 2010).

9 contrast to the traditionally used common law, but also to the Code’s main competitor, the Prussian civil code, that was build on the idea of class differences.21 The ALR perpetuated these differences and the privileges of the nobility (Kleinheyer, 1979).22 Contrary to the French law that was constructed for a bourgeois civil society, the Prussian ALR aimed at cementing the feudal society in all aspects of the law (Kleinheyer, 1979). This difference in legal equality is, for instance, visible in the economic sphere.

The Code Civil guaranteed property rights and commercial freedom to all individuals (Lyons, 1994).

Starting a business, for example, only required a business license from the local city administration

(Becker, 1985). The ALR did not allow peasants to engage in commerce and put restrictions on property ownership, the bourgeoisie could not own feudal property, and commercial activities of the nobility were equally restricted (Schmon, 2007). Alexis de Tocqueville (2015) who studied the ALR in the mid-

19th century summarized the antiquity of the Prussian law: “Most of the landowners privileges are consecrated by the Code. [...] In this same Code the bourgois remains carefully separate from the peasants. [...] It is thus the case that this law code [...] put into practice after the outbreak of the French

Revolution, is the most authentic and recent legislative document that gives a legal basis to the very feudal inequalities that the French Revolution was going to abolish throughout Europe."

Furthermore, the Napoleonic Code increased the predictability of the law by unifying and replacing en- tirely traditional and local law. In contrast, the ALR and common law were only subsidiary to local law.

Where local legal customs differed from the written law they were given preference. Therefore, in areas where the Code Civil was not applied, nobles continued to administer the law themselves (Patrimonial- gerichtsbarkeit). The Code Civil also established a separation of powers between the judiciary and the administration and increased legal transparency through its clarity, as well as through the publicity of trials and hearings. In the Rhineland it went along with an extensive reformation of the entire judicial procedures and the installation of new courts of appeal and commercial courts.23

21Legal equality also differentiated the Code Civil from Bavarian civil code, the Bavarian Codex Maximilianeus created in 1756. 22Already the titles of the chapters in part II of the ALR (“The peasant class", “The bourgeoisie", and “The rights and duties of the nobility") make a clear reference to the desired social order. 23In these territories, ordinary people were appointed as Justices of Peace (JP), being in charge of pre-trial settlements of minor cases. They were, in particular, appealed in labor conﬂicts in an increasingly industrialized society (Haferkamp), settling disputes in the market about the quality of work between masters and manufacturers, as well as those between masters and journeymen or apprentices, who often breached contracts and left their masters on their own (Sperber, 1992, p. 55). “Courts would provide a remedy by ensuring the exact fulﬁllment of [...] contracts”, writes Sperber (1992, p. 56) and refers to the manufacturing tribunals, which provided a form of “industrial mediation” between manufacturers, workers and journeymen (Sperber, 1992, p. 55).

10 2.3 Social Consequences

The French institutions aimed at modernizing a social system that had been predominant for hundreds of years, with view to a more equal society, the abolition of privilege and the removal of social barriers

(Lyons, 1994). Its legal code guaranteed equal rights and encouraged cooperation between different parts of the population and the creation of a civil society. This transformation is most well documented for the Rhineland, a region that prior to 1800 was characterized by a “particularly old-fashioned, pre- absolutistic, feudal system"24 (Nipperdey, 1983, p. 78). However, "the French law and its idea of a society made up of free and equal property owners have penetrated the civil movement"25 (Nipperdey,

1983, p. 79) and led to a society that was “well advanced socially compared to the rest of Germany"26,

(Nipperdey, 1983, p. 78). Comparing the Rhineland to other provinces of 19th century Prussia, Sperber describes the latter as far less progressive, characterized by social structures and institutions “too different from those [...] along the Rhine” (Sperber, 1992, p. 38). While in the Rhenish territories, “the tone was set by the bourgeois elite of bankers, merchants, manufactures, rentiers, lawyers and notaries” (Sperber,

1992, p. 38), i.e. a large and important bourgeoisie and middle class, the nobility remained highly inﬂuential in Prussia.

3 Estimating the Effect on Contemporaneous Social Capital

This section estimates the impact of the Napoleonic Code Civil on norms of trust and cooperation in current day Germany. I start by analyzing the association between the duration of application of the

Code Civil and social capital in a sample of 16 historical territories that cover almost the entire current

Federal Republic. I then restrict the sample to observations around a border separating regions that applied the Code Civil for on average over 90 years from regions that applied it for only about 5 years.

To validate the border design, I test for pre-existing differences between the treated and untreated districts along the Code Civil discontinuity. In addition, I create a placebo treatment by moving the border as an additional test for the existence of a “true" treatment effect.

3.1 Data

Measures for contemporaneous social capital are obtained using individual survey data from the German

Socio-Economic Panel (SOEP) (Wagner, Frick, and Schupp, 2007), wave 2003. The outcome variable

24Author’s translation. 25Author’s translation. 26Author’s translation.

11 I am primarily interested in is interpersonal trust. In the survey, respondents were asked to evaluate

(“totally agree”, “slightly agree”, “disagree slightly”, “totally disagree”) the following three statements related to interpersonal trust: “On the whole one can trust people”, “Nowadays one cannot rely on anyone”, and “If one is dealing with strangers, it is better to be careful before one can trust them”. I recode the variables on a scale from 1 to 4, such that the highest value is assigned to the highest level of trust. The variable Trust is then computed as the average out of the individual scores for each question.27

Trust, while widely used, might be insufﬁcient as a stand-alone measure of social cooperation. I therefore construct two additional variables measuring norms of cooperation. The SOEP asks respondents about the perceived behavior of others with respect to fairness: “Do you agree that most people... a) exploit you if they had the opportunity or b) would attempt to be fair towards you?”, as well as with respect to helpfulness: “Would you say that for most of the time, people... a) attempt to be helpful? b) or only act in their own interests?”. Out of those questions I construct two binary outcome variables, labeled Fair and

Coop, such that the value 1 indicates a higher level of perceived fairness and willingness to help and act cooperatively, respectively. As Guiso, Sapienza, and Zingales (2012) show for the WVS, answers from social value survey questions concerning trust reﬂect the beliefs of the sender about the trustworthiness of the receiver. Survey measures are therefore well suited to elicit the updating and persistence of beliefs following a exogenous legal change.

The geographical level at which individuals can be observed is the district (county or Kreis). There are more than 400 districts in Germany, and one district is equivalent to the NUTS 3 level of the Statisti- cal Units of the European Union.28 To account for other inﬂuences of social capital on the individual level, all regressions control for personal characteristics of the respondent, such as age and its square, as well as a dummy variables for gender. To control for the legacy of the former Communist regime in Germany, I include a dummy variable that takes on the value 1 if the respondent lived in the former

East of Germany before 1990. Furthermore, I control for ﬁve categories of religious afﬁliation which are

Protestant, Catholic, any other Christian religion, a different religion that is not Christian or none. The personal characteristics, as well as the location before 1990 and the religious denomination are plausible exogenous to the Napoleonic treatment and will serve as my baseline controls. Additionally, differences in educational attainment are controlled for using six different indicator variables for schooling that disringuish between having no school degree at all, having attended school for a total of 9 years and

27Using a principal component instead of the mean produces almost identical results, as both measures are strongly correlated (0.99). 28Because of the conﬁdentiality requirements of the data on this detailed regional level, it was accessed via a remote access.

12 obtained a degree from the middle school (Hauptschule), 10 year middle school (Realschule), 12 year secondary school (Fachhochschulreife) or having obtained the highest secondary school degree after 13 years (Abitur). To capture the economic situation of the respondent, I use a dummy measuring unemployed during the last year, as well as the household income in the last month. Since educational and economic outcomes might have been affected by the Napoleonic institutions, they are added separately to the baseline controls.29 Table 1 provides summary statistics for all individual variables from the SOEP.

[Table 1 about here]

The current districts of residence of the observed individuals are matched with 16 historical territories for which historical territory characteristics are available (urbanization rate in 1800 and 1850, the share of

Protestants in 1800, the longitude and latitude of the territories’ capital and its distance to Paris, as well as the coding of different reforms in 19th century Germany). The deﬁnition of the territories follows

Acemoglu et al. (2011), from where I also take the historical territory data. Historical borders match remarkably well with current district borders and cut only through very few counties. Current counties that straddle a historical border are assigned to a historical region depending of the historical location of the counties’ current capital. The matched sample contains more than 17000 individual observations included in the SOEP. In addition to historical territory characteristics, I control for geographic differences across districts, i.e. the distance of its centroid to a major European river, the distance to the closest coast, the slope of the terrain, and the suitability of the soil for growing wheat. Slope and wheat suitability are taken from the FAO-GAEZ database. Table 2 displays summary statistics for the historical and geographic controls.

[Table 2 about here]

3.2 Results for the Complete Sample

I begin by estimating the effect of the Code Civil on current trust in the sample of 16 historical regions.

The treatment variable is the number of years the Code Civil was in place in a region before 1900. Figure

3 displays the spatial distribution of the treatment.30

29As my focus of attention lies on the intergenerational transfer of individuals’ beliefs from their ancestors that have lived in the particular historical territory many years ago, ideally I would like to consider only individuals for which I can identify the historical territory of their ancestors residence during the 19th century. Except for the location of individuals before the fall of the Berlin Wall, the SOEP data does not provide information about the location of ancestors. 30Table D1 in the Appendix gives the name and definition of the territories included in the sample and the treatment definition. The inclusion and definition of territories follows Acemoglu et al. (2011). Figure D.1 in the Appendix shows a map of the historical territories.

13 [Figure 3 about here]

The following base model is regressed:

0 0 0 yi,d,h,m = α + βcivilh + Xi δ + geodγ + histhω + ηm + ui,d,h,m (1)

where yi,d,h,m measures one of the social capital related outcomes of individual i living in district d, belonging to historical territory h and macro-region m. civilh measures the number of years the Code

Civil was applied in historical region h before the year 1900. The matrix Xi contains a list of individual controls, speciﬁed above. ηm denote ﬁxed effects for German macro-regions. geod contains geographical controls evaluated at the district level and histh denotes historical controls, both pre- and post-treatment, measured at the historical territory. Standard errors are clustered at the level of a district.

Results of the regressions are displayed in Table 3.31 Panel A shows the basic correlation between the

Code Civil and the different measures for social trust, controlling only for macro-region dummies of the four macro-regions north, south, east and west, as well as the baseline individual controls (age, gender, religion, location before 1990). I estimate a positive and signiﬁcant coefﬁcient for all three outcome variables. Thus, individual trust and cooperation increases in the number of years the Code Civil was used in a historical region. Panel B adds potential pre-treatment correlates of the Code Civil evaluated at the territory level, such as the rate of urbanization in 1800, the share of Protestants in 1800, the longitude and latitude of the capital and its distance to Paris. Urbanization and Protestantism are both proxies of pre-existing differences in economic development, and the latter has been also shown to be related to higher levels of literacy (Becker and Woessmann, 2009). In addition, religion as a measure of

“culture” can also explain differences in norms and beliefs. Protestantism has been associated with high levels of social cooperation and trust (Alesina and La Ferrara, 2002; Tabellini, 2008). Including those controls increases the effect of the Code Civil on all three outcome variables considerably, see Panel B of Table 3. The coefﬁcient for Trust for example doubles in magnitude and gains signiﬁcance. Historical characteristics of the territories are thus important covariates of both the Napoleonic treatment and social trust today.

[Table 3 about here]

31Additional regressions examining different deﬁnitions of the treatment and functional forms are shown in the Appendix, Table B2.

14 Panel C of Table 3 adds another set of plausible exogenous controls, which relate to geographic characteristics of districts. Adding geographic controls increases the magnitudes of the coefﬁcients further.

Finally, in Panel D I include a set of post-treatment variables, both at the individual and historical territory, that were potentially affected by the Napoleonic treatment (Acemoglu et al., 2011). More precisely,

Panel D adds educational attainment, household income and unemployment status of the individual, as well as the urbanization rate of the territory in 1850. Estimated coefﬁcients stay relatively stable in magnitude when post-treatment controls are included and they retain their signiﬁcance.32

3.3 Robustness of the Complete Sample

The robustness of the results in the complete sample is assessed further in Table 4, using a specification with the entire set of pre-treatment controls similar to Panel C of Table 3. In the first specification the historical territories Baden and Saxony are dropped, which chose to adopt a civil code that guaranteed legal equality - although this choice was severely constrained in the case of Baden. The finding of a positive, significant effect is robust to this specification, as column 1 in Table 4 shows. Next, within Prussia, i.e. holding Prussian kingdom characteristics constant, the positive association holds true. Prussia is defined in either the borders of 1815 (column 2) or 1866 (column 3).

[Table 4 about here]

Finally, I use self-reported life satisfaction and interest in politics as placebo outcomes, on which the

Code Civil should not have had any long-run inﬂuence. Columns 4 and 5 indeed display estimated coefﬁcients which are not statistically different from zero.

3.4 Code Civil versus Further Napoleonic Reforms

The package of French reforms did not only consist of the introduction of the legal code, but it also included the abolition of guilds, the abolition of serfdom and land reforms. The potentially highly correlated Code Civil treatment variable could pick up the effects of these reforms, instead of having an independent impact. I test this hypothesis by using these alternative reforms as treatment in the econo- metric speciﬁcation of the complete sample, controlling for pre-treatment and geographic controls similar to Table 3 Panel C. The measures for the abolition of serfdom, the agrarian reform and the abolition of

32Table B3 in the Appendix shows the estimated coefﬁcients for all variables included in Panel D. As expected, having lived in the GDR before 1990 decreases trust, more educated individuals have more trust, unemployed individuals trust less, and income is positively associated with higher trust. For the variables on the territory level only the share of Protestants in 1800 shows signiﬁcant association with higher cooperation.

15 guilds are the time span between the years of enactment of the respective reform before 1900. The reform index is a combined measure of all reforms - including the Code Civil - evaluated as of the year

1900.33 Table 5 displays estimation results. In Panel A, each alternative reform is included separately as treatment variable. As expected, the reform index that itself contains the Code Civil shows a positive and significant association with Trust, and Coop (see the first row, columns 1 and 2). Regarding each reform independently, the only one that is significantly correlated with the outcomes of interest is the abolition of serfdom, while all other coefficients turn out not to be different from zero.

[Table 5 about here]

More interesting than each reform by itself is the robustness of the main treatment when each of the alternative reforms is controlled for. In Panel B I run such “horse races” between the Code Civil and each alternative reform, by including, one at a time, the latter to the estimation of Trust on the Code

Civil. As reported in columns 1 - 4, the estimated coefﬁcient for the Code Civil treatment variable is robust and stays positive, and highly signiﬁcant. It moves between 0.11 and 0.15 in magnitude (Table 5,

Panel C). All other estimated coefficients turn out to be insignificant. Even when including all reforms at the same time, the the duration of application of the Code Civil is the only significant predictor of trust today (column 5). These results indicate that the Napoleonic Code has an independent effect on cooperation and does not simply proxy for other institutional reforms.34

3.5 Identiﬁcation from Neighboring Districts

So far, the section has established a robust positive association between the application of the Napoleonic legal code throughout the 19th century and social capital today. However, this association could be driven by (potentially unobserved) territory characteristics that have not been included in the regressions

(omitted variable bias), or by initial differences in social capital that are correlated with the reception of the Code Civil (reverse causality). In the following, I use an empirical strategy that exploits the geographic spread of the Code Civil and the circumstance that in the Western parts of Germany nearby districts differ greatly in the number of years the Code Civil was used. This strategy, which is similar to a geographic discontinuity design, allows me to focus on close-by districts and thereby to reduce heterogeneity in fundamental pre-treatment characteristics. The comparison of neighboring districts

33The deﬁnitions of the reforms and their duration are taken from Acemoglu et al. (2011). Table D1 in the Appendix displays the variable for each territory, and Table D2 the correlation of each measure with the duration of the Code Civil. 34Further reforms that Napoleon undertook concerned the educational system and the administration. However, in the case of education the reforms turned out to be a failure, see Schmenk (2008), or were adjusted after the Prussian takeover, as in the case of the administration system, see Rowe (2003). Reassuringly, the results of both the complete and border sample hold true within Prussia, that is holding education, the administrative system and political rights constant.

16 follows the idea that these localities were supposedly very similar in terms of geography, economic potential, and culture, already prior to the Napoleonic treatment. For this empirical strategy, I restrict the sample to districts that are located not more than 50 km around a boundary that separates areas that used the Code Civil for at least 90 years from those that used it a maximum of 13 years.35 The map in Figure

4 displays the counties that are located within the 50 km bandwidth around the Code Civil discontinuity.

Summary statistics for the border sample are displayed in Table 1.

[Figure 4 about here]

The discontinuity is derived from the number of years that a territory applied the Code Civil, but does not overlap with a geographical obstacle, or a long-existing common political border.36 A threat to this identification strategy is therefore that the assignment variable (number of years the Code Civil was used in a region) was potentially manipulated by the inhabitants of the historical territories.37 However, I claim that the discontinuity can offer plausible exogenous variation at the local level, even if one acknowledges that the duration of application of the Code Civil in a territory was to some extent influenced by region- specific characteristics. Whether a district located along the border used the Code Civil in the 19th century, while its neighbor did not, is as good as randomly assigned when comparing nearby districts.

The assignment of treatment status stems from a number of historical accidents, including the precise geographic extent of French occupation (e.g. until its “natural border”, the river Rhine), the reallocation of territories at the congress of Vienna and the subsequent state-wide policies adopted by new rulers, and from other local idiosyncrasies. Some of these idiosyncrasies are linked to the historical formation and reshufﬂing of political borders that determined the afﬁliation of localities to historical states, either prior or during the Napoleonic period. This historical process can be illustrated by the many changes that occurred to the city of Duisburg located at the discontinuity on the right bank of the river Rhine. From the

13th century onwards, Duisburg belonged to the territory of Cleve. In the 17th century Cleve fell in the hands of Prussia, which made Duisburg a part of the Prussian Empire. After the Peace of Basel in 1795 the left bank of Cleve was given to France. Duisburg remained Prussian and used its Civil Code until the right bank came under Napoleonic occupation in 1805. The entire territory of Cleve subsequently used the Code Civil until 1815, when it was integrated into the Prussian Rhine province (Rhineland).

35If the institutional and legal reform had an effect on culture, a slowly moving and hard to change variable, it is more likely that the effect of the Code Civil needed several years to disperse. Indeed, Guiso, Sapienza, and Zingales (2008b) estimate, that a shock requires about 75 years (2 to 3 generations) to have a long-lasting effect on social capital. The time span of the longest adoption of the Code in the sample, more than 90 years, ﬁts very well to this estimation. 36However, the Code Civil border is in fact largely equivalent to the borders of the historical German territories, many of which were created in the reorganization of Germany during the Napoleonic Wars. In addition, the border segment in the North follows the course of the river Rhine. 37Precise manipulation by the agents makes a regression discontinuity design invalid (Lee and Lemieux, 2010).

17 However, contrary to other Rheinish districts and even to neighboring areas of former Cleve located on the left bank of the Rhine, Duisburg (and few other Rheinish districts) had to abolish the Code Civil.

This followed the Prussian agenda to (re-) install Prussian law in territories that belonged to Prussia and that had already used the Prussian Civil Code before 1800 (Heymann, 1914). That Duisburg ended up not using the Code Civil is therefore not linked to any underlying city-speciﬁc characteristic, but results from a number of arbitrary historical events. Therefore, at the local level the treatment assignment can be plausibly assumed to be as good as random.

My empirical strategy consists primarily of estimating local linear regressions with a ﬁxed bandwidth of

50 km. I show robustness of the results to the estimation of a standard regression discontinuity design that controls for location of the district, and a neighbor-pair ﬁxed effect estimation in the spirit of Acemoglu,

García-Jimeno, and Robinson (2012). In addition, following the regression discontinuity literature, I conduct two falsiﬁcation tests in section 3.7. The ﬁrst test consists of a placebo treatment generated by moving the border. The second exercise examines empirically whether districts that received treatment were systematically different from those that did not.

Local Linear Regressions The ﬁrst approach to estimate differences in social capital at the discontinuity are local linear regressions with ﬁxed bandwidth of 50 km.

The following model is estimated:

0 yi,d,h = α + βborder + Xi,dδ + ui,d,h (2)

where border = 1 if Code Civil ≥ 90, and 0 otherwise. In all speciﬁcations standard errors are clustered at the district level.

Table 6 shows the estimation results.

[Table 6 about here]

Using only base controls in row a), I estimate a coefﬁcient of the treatment border that is positive and signiﬁcant for all three outcome variables. Individuals living in the treated regions have, on average, a

0.06 higher level of trust. Row b) adds Federal States ﬁxed effects. Being a federal republic, Germany has transferred some of its administrative duties to the sixteen federal states (Länder) that are, for example, in charge of the education, the police and court system. It is possible that current differences in

18 the functioning of local institutions influence social capital, and thus potentially bias the results. Esti- mated coefficients controlling for Federal States dummies increase in both magnitudes and significance levels. Results are robust when the full set of individual controls is included in row c), and when the sample of territories is restricted to those that belonged to Prussia in row e). The sample restriction follows the same reasoning as in the complete sample analysis, as Baden, which did not directly adopt the

Code Civil through French occupation, is dropped. This specification estimates effects within Prussia, as the Rhineland, Mark and Westphalia were part of Prussia after 1815. In other words, holding all other common Prussian characteristics constant, the estimation produces a positive and significant coefficient, albeit the effects are somewhat weaker. The inclusion of geographic controls in row e) increases coef-

ficients in magnitude and significance.38 Results are very similar in magnitude and significance if the dichotomous treatment variable is replaced by the number of years the Code Civil was applied in row f). Finally, average trust levels in the districts used around the Code Civil discontinuity are displayed graphically in Figure 5.

[Figure 5 about here]

Spatial Regression Discontinuity Design Alternatively, I use a spatial regression discontinuity design

(RDD) within the same 50 km corridor that controls for the two-dimensional location (longitude and latitude) of the district:

0 yi,d,h = α + βborder + f (locationd) + Xi,d ∗ δ + ui,d,h (3)

Equation (3) adds to equation (2) the function f (locationd) of either the ﬁrst, second or third order polynomial of the district’s two-dimensional location, following the work by Dell (2010).39

[Table 7 about here]

As Panel A of Table 7 shows, estimated coefﬁcients are robust to including the longitude – latitude polynomial, and very similar compared to those obtained without locational controls in Table 6, row e).

38Table B4 in the Appendix shows that the results are robust to including each geographic control separately. 39An important assumption of any RDD is the smoothness of the covariates at the discontinuity. Table B5 in the Appendix regresses the covariates on the border dummy. There exist few signiﬁcant differences in individual characteristics at the border with the exception for religious variables (Catholics, Other Christian religion).

19 Neighbor-Pair Fixed Effects Estimator Finally, I implement a neighbor-pair ﬁxed estimator, similar to a matching estimator, that compares pairs of treated and untreated districts that share a common border

(following Acemoglu, García-Jimeno, and Robinson, 2012). I create couples (neighbor-pairs) of adjacent districts that either used the Code Civil for at least 90 years, or that used it for less than 90 years. The main advantage of the neighbor-pair estimator is that it allows to include neighbor-pair ﬁxed effects that take into account all unobservable characteristics that are shared by a pair of districts. Adjacent districts are most likely very similar across geographic and economic conditions, contemporary institutions, as well as any other unobservable. The identiﬁcation strategy assumes that the number of years the Code

Civil was used is the only exogenous source of variation within each neighbor-pair. As argued above, whether the French legal system reached a district was as good as randomly allocated within a pair of neighbors.

Estimation results are provided in Panel B of Table 7. The estimated coefﬁcients from the neighbor-pair

fixed effects model are positive and highly significant for Trust and Coop, and small and statistically equal to zero for Fair. Coefficient magnitudes for Trust and Coop are larger than the comparable ones obtained from local linear regressions. Keeping in mind the smaller sample size, this could indicate that not taking into account unobservables across adjacent municipalities biases the estimation downwards.40

3.6 Falsiﬁcation Tests

Moving the Border The first falsification exercise tests whether the treatment effect found in the previous subsection reflects a general geographic pattern, or a genuine treatment effect.41 To generate a false treatment the border is moved twice. The first time it is moved outwards, that is to the North-East. This allows to generate a placebo treatment comparing individuals living in regions unaffected by the true treatment on both sides of the border. Second, the border is moved inwards, that is to the South-West, so to compare regions that have been treated on each side of the border. I construct two new borderlines which are the joint outline of the regions included in the original border sample. For the inward shift this is the common outline of the treated, for the outward shift this is the common outline of the untreated group. Treatment is assigned to regions on the “left" to the border. See Figure 4 for a graphical repre- sentation of the regions included in the different samples. If the results found in the previous section are

40A disadvantage of the neighbor-pair ﬁxed effects estimator is that it needs a sufﬁciently large number of neighboring districts. With historical and city-level data the number of cross-sectional units that share a border is even lower. Therefore, in the remainder of the paper I will estimate local linear regressions along the border. Local linear regressions with bandwidth 50km have the additional advantage that they are less sensitive to measurement error in the border assignment (the “true” border cuts through several modern districts), by using additional observations that are not located directly at the border. 41This exercise is inspired by the paper Becker et al. (2015).

20 due to the true treatment, moving the border and performing the same regressions as above should not display an effect of the ﬁctitious treatment on social capital outcomes in both cases. For the outwards shift because there was no true long-term Napoleonic treatment. For the inward shift because regions that actually were treated should be similar with respect to trust levels.

[Table 8 about here]

Results from this falsification test are shown in Table 8. Panel A display the results from moving the border outwards, Panel B respectively the results from a moving the border inwards. Panel A shows that moving the border outwards produces statistically insignificant coefficients. The estimated coefficients are very small and tend towards zero, either using only base individual controls or the full set of individual controls. A similar picture appears in Panel B for the inwards shift of the border. The coefficient for

Trust is insignificant and about one third of the magnitude of the coefficient obtained in the “true” border specification. The only significant, but negative, difference exists when using the dependent variable

Coop. Thus, only in the correct border specification, significant and positive treatment effects for regions on the left of the border can be found. Hence, the falsification tests and the original results suggest a peculiarity at the border, that I argue can be attributed to the Napoleonic institutions.

Differences in Pre-Treatment Characteristics A further threat to my empirical strategy comes from the possibility that variation in initial characteristics are responsible for the observed differences in social capital. The border estimation offers only a valid comparison if districts were similar prior to receiving treatment. Table 9 tests for pre-existing discontinuities in observables at the border. In Panel A, I regress geographic characteristics on the border indicator. The only signiﬁcant difference appears with respect to the proximity to a river (column 1), i.e. treated regions are on average located closer to major rivers. This is problematic if the presence of a river is associated with more trust. Districts are similar with respect to the slope of the terrain, the suitability of the soil for growing wheat, or the distance to the coast. As reported above, all estimation results are robust to the inclusion of all four geographic characteristics, in both the complete and the border sample.

[Table 9 about here]

Border regions could have differed with respect to socio-political outcomes prior to 1802. To proxy for early socio-political development, I use information about characteristics of medieval cities taken from

Jacob (2010). In particular I investigate whether contemporary districts contained a Free city, a Hanseatic

21 city, a city with a Bishop or with a University.42 Results are shown in Panel B of Table 9. I proxy for early political institutions by the presence of a medieval Free city. As Guiso, Sapienza, and Zingales (2008a) have shown for Italy and Jacob (2010) for Germany, the presence of Free cities goes along with higher levels of social capital. However, districts that used the Code Civil are on average not significantly more likely to have hosted a Free city. Another dimension that matters for cooperation is trade and economic exchange. I proxy for the importance of trade by city membership in the Hanseatic League, a network of trade cities that existed from about the 13th to the 17th century. Jacob (2010), for example, finds that former Hanseatic cities have more social capital today. Again, I do not find a statistically significant difference along the border regarding city membership in the Hanseatic trade network (column 2). In addition, districts located at the discontinuity are also balanced with respect to the presence of a Bishop or a university (columns 3 and 4).

Finally, I analyze city sizes along the border prior to 1800 to test for pre-existing differences in local economic activity. The data is taken from Bairoch, Batou, and Chevre (1988). I use the log population of cities along the border discontinuity at different points in time. As columns 1 - 4 of Panel C show, city population in 1500, 1600, 1700 and 1800 is balanced across cities that adopted the Code Civil and those that did not. In conclusion, the results suggest that important pre-existing historical socio-economic characteristics, which are potential determinants of social cooperation, were balanced along the border.

3.7 Magnitudes

How large are the above documented effects of the Napoleonic Code on contemporary social capital?

Given that one expects the Code Civil to have an inﬂuence only after a substantially long application, the relevant comparison is a very long duration (above 90 years) to very short duration of usage. In the complete sample the magnitudes vary between an increase of 0.09 to 0.24 of a standard deviation in Trust

(about 0.54), between 0.17 and 0.3 of a standard deviation in Coop (about 0.48), and between 0.06 and

0.07 of a standard deviation in Fair (about 0.5), depending on the specification. In the border sample differences in social capital range between a 0.1 to 0.2 of a standard deviation in Trust, and between 0.15 and 0.19 of a standard deviation in Coop. Magnitudes in the spatial regression discontinuity design, and those obtained from the neighbor-pair fixed effects estimation are very similar. Putting the magnitudes into relation, having lived in East Germany before 1990 reduces trust by about 0.14 of a standard deviation, being unemployed by around 0.23 of a standard deviation and individuals that finished a 13-year

42To assign treatment to cities, I match city locations to current county borders as of 2003 that are included in the border sample.

22 secondary school report 0.28 of a standard deviation more trust than individuals without any school degree. Compared to similar studies, Nunn and Wantchekon (2011) for example report “beta” coefﬁcients of between -0.10 and -0.16 of the impact of the slave trade on trust, and Guiso, Sapienza, and Zingales

(2008a) ﬁnd that Italian free cities are associated with an increase of 0.15 of a standard deviation in associational membership. The ﬁndings are therefore economically meaningful and comparable to other historical events.

4 Potential Channels

The previous section uncovered a robust, positive association between the historical application of the

Napoleonic Code Civil and contemporary norms of cooperation. This section investigates the underlying mechanisms of the relationship between the legal code and trust today. It ﬁrst analyzes whether the documented effects are a direct consequence of the legal system or mediated through its effect on economic development. And second, it investigates alternative possible mechanisms, in particular the distribution of wealth, the emergence of generalized morality and effects on early social cooperation.

4.1 Short and Long-Term Economic Development

Acemoglu et al. (2011) document a positive effect of the Napoleonic institutions on urbanization in

German regions, especially from mid-19th century onward. That income and social trust are positively correlated on the individual and aggregate level is a ﬁrm result in the trust literature, although there exists only limited causal evidence that income affects trust (see e.g. Algan and Cahuc, 2013). The positive effects on income of the Napoleonic institutions could be a potential explanation why trust is elevated today. I examine in the following whether economic development is indeed the main channel at work in the samples that I consider. Starting with the complete sample, Panel A of Table 10 regresses log city population as a measure for historical economic development and contemporaneous income per capita on the Code Civil treatment.

[Table 10 about here]

Based on these cross-sectional estimates, I find that regions that used the Code Civil do not have larger cities in 1850 (column 1) or 1900 (column 2), and cities did not grow faster on average during the 19th century (column 3). I also estimate non-significant coefficients for city sizes in the years 1875 and 1910, as well as regarding city growth between 1850 and 1900, see Appendix Table B6. Estimated coefficients for the years 1800 to 1910 and 95 % confidence intervals are plotted in the Appendix Figure C.2 a).

23 However, districts today seem to have slightly higher levels of per capita incomes (column 4), and this effect is signiﬁcant at the 5 % level. According to this estimation, a 90 year application of the Code Civil goes along with a 3.9 % higher income per capita today.

Considering only observations around the border, the cross-sectional estimates reported in Panel B of

Table 10 suggest that post-treatment differences in economic development are small and non-signiﬁcant.

Economic outcomes measured in the 19th century (columns 1 - 3), as well as contemporary economic development (column 4) is balanced around the border. The magnitude of the non-significant estimate on contemporary income is small, too: treated areas have only about 0.8% higher incomes per capita. I also estimate non-significant coefficients for city sizes in the years 1875 and 1910, as well as regarding population growth from 1850 to 1900, see Appendix Table B6. Estimated coefficients for the years

1800 to 1910 and 95 % confidence intervals are plotted in the Appendix Figure C.2 b), while those per capita income are plotted in Appendix Figure C.3. Appendix Table B7 reports similar results from an estimation that controls for a quadratic function in longitude and latitude, and from the border movement falsification tests. Coefficients from the falsification tests are oftentimes of similar magnitude, sometimes even larger (see columns 1 and 2 of Panel B), than in the original border specification.

In Panel C of Table 10, I exploit the panel structure of the city population data from 1700 to 1910 and estimate difference-in-differences speciﬁcations for cities along the border. Column 1 runs a simple difference-in-difference estimation and reports that treated cities did not grow faster after 1800. This

finding is robust to including city and time dummies in column 2, or when considering the number of years the Code Civil was in place instead of the border dummy in column 3. The fully flexible estimation that interacts the border treatment with year fixed effects in column 4 confirms that city growth around the border did not differ prior to or post 1800. The parallel evolution of city sizes can be seen graphically in the Appendix Figure C.5 a).43 Estimated coefficients from the fully flexible estimation and 95 % confidence intervals are plotted in the Appendix Figure C.4.

Table 11 analyzes further outcomes of 19th century socio-economic development taken from the ifo

Prussian History Database.44 Unfortunately, the data is only available for Prussian territories, i.e. the upper part of the border. The ﬁrst set of outcomes closely follows Becker and Woessmann (2009) and

43Figure C.5 b) in the Appendix plots the evolution of urbanization rates in the territories close to the border. In this reduced sample, most of the difference in urbanization rates between treated and untreated regions appear in the ﬁrst half of the 19th century as untreated regions stagnate (compare to Acemoglu et al. (2011) Figures 2A and 2B). In the ﬁrst decade after 1850 both groups grow equally fast, and at the end of the century territories in the control group catch up and close the gap that existed in 1850 by 1900. 44Since counties in 1871 have different borders as today, to assign treatment I match the centroids of the historical counties to contemporaneous county borders of counties included in the border sample.

24 takes the log average wage of teachers (1886), the tax income per person (1877) and occupational shares in manufacturing and service or agriculture (all for 1882) as proxies for late 19th century development.

Counties on both sides of the border are balanced regarding these outcomes, see Panel A of Table 11.

Additional socio-economic outcomes that I consider are religion and literacy rates in 1871. Columns 1 and 2 of Panel B show that the border is not a cultural border separating religious groups.45 Individuals on the treated side of the border were, however, signiﬁcantly less literate in 1871 (see columns 3 and

4).46

[Table 11 about here]

Overall, the results do not give much support for economic development - around the border - as the dominant channel explaining the positive association between the Code Civil and trust. While the results diverge locally from those in Acemoglu et al. (2011), I still ﬁnd some evidence that the Code Civil has had a positive effect on city growth in the complete sample using a difference-in-differences estimation (see

Appendix Table B8.), and a small effect on incomes per capita today. However, the complete sample compares very heterogeneous areas. Once a more demanding speciﬁcation is used and the sample is narrowed to more similar neighboring areas, economic differences become negligable.47 Therefore, economic differences cannot explain the differences in social capital observed around the Code Civil discontinuity.

4.2 Redistribution and Inequality

As a manifestation of the ideas of the French Revolution, one of the central topics in the Code Civil was the promotion of equality of economic incomes and wealth (Lyons, 1994). Through the removal of primogeniture, the inheritance right of the ﬁrstborn, and introduction of equal inheritance between siblings, the creators of the Code Civil sought to end the concentration of wealth in the hands of few. The equal partition of estates was argued to have been successful and to have led to the division of land into small parcels (Mokyr, 2003). The resulting change in the wealth distribution could have promoted social

45Religious denomination was potentially quite stable since the Reformation, so one might expect this result to hold in the pre-treatment period, too. However, individual data from the SOEP shows differences in Catholicism at the border, see Table B5 in the Appendix. 46This conﬁrms the historical literature arguing that the Napoleonic education reform was a failure, as a shortage of teachers and funds led to the decay of the school system. Schmenk (2008, p. 232) for example estimates that in the Rhineland three fourths of the population could not read and write in 1814 and regular schooling was an exception. 47Keller and Shiue (2015) show that city growth in 19th century was much more affected by market integration than by institutional reforms, using the French rule in Germany as instrumental variable for the latter. They argue that market integration is a mechanism from institutions to growth. It may be that increased economic integration and trade is an additional channel that links institutions with social trust and cooperation. Data limitations does not allow me to test this channel empirically.

25 cooperation if mitigating class distinctions based on economic fortune eases inter-class cooperation. To test whether economic inequality was linked to the Code Civil, I use a measure of land concentration, e.g. a land Gini index, taken from Ziblatt (2009), for the entire German Empire as of 1898.48 Results are shown in Panel A of Table 12. Using the complete sample, I estimate a positive and non-significant coefficient without controls (column 1) and a negative and marginally significant coefficient in the model using all controls (column 2). Around the border I find small and non-significant coefficients that suggest that differences in wealth inequality were not important.

[Table 12 about here]

4.3 Generalized Morality and Social Cooperation

Cheating and Elite Behavior A theoretical mechanism through which legal institutions could increase social capital is through a reduction in cheating and an increase in moral behavior towards everybody in a society (what Tabellini (2008) calls generalized morality). By enforcing legal equality, the Code

Napoleon might have discouraged exploitative behavior, cheating and the exposition of negative exter- nalities on others. “Cheating” has been used as an outcome measure of civic behavior, e.g. for example by Fisman and Miguel (2007) who study violation of parking tickets by diplomats, or by Guiso, Sapienza, and Zingales (2012) who use data on the share of pupils that cheat in school exams. I apply this idea in the context of 19th century Germany, and use data on electoral fraud in elections to the Reichstag between

1871 and 1912, assembled by Ziblatt (2009). Electoral fraud can be defined as an illegal interference with an election, i.e. a “violation of the law” (Lehoucq, 2003), and it includes “violence, coercion, influence, voting-buying, or procedural manipulations” (Ziblatt, 2009). Electoral fraud thus undermines the fairness of an election and is a form of cheating for private gain on the society as a whole. The data lists the number of disputed elections be constituency and year. Disputed elections encompass manipulations on the election day, local government intervention into election campaigns, influence of private individuals, for example agrarian or industrial employers, or vote buying. I aggregate the number of disputed elections per district prior to 1900 as the legal framework was equalized afterwards. Results are robust to adding post-1900 years. While the data has the advantage of coming from elections with a uniform electoral institutional framework, it has several caveats. First, it only measures the degree of cheating for political elites. Second, as all eligible voters were able to file a complaint against an election, high cheating could reflect differences in the willingness of voters to complain. I account for the latter by

48As above, the original data is measured on the level of a historical administrative unit, the electoral constituency. I therefore match the centroids of historical constituencies with county borders in 2003.

26 controlling for voter mobilization with voter turnout.

As Panel B of 12 shows, the number of contested elections decreases signiﬁcantly with the years the

Code Civil was used (column 1), controlling for population, turnout, electoral competition and the share of Catholics. This difference goes away, however, in the more demanding specification using the full set of geographic and historical controls (column 2). Considering the border sample, treated districts have a significantly lower number of contested elections (column 3) conditional on population, turnout, and the share of Catholics. Controlling for geographic characteristics of the district reduces the magnitude of the coefficient, but it stays significant at the 5% significance level. The magnitude of the coefficient is sizable, about -0.6, given the mean (1.70) and standard deviation (1.64) of the dependent variable.

Although, the evidence is restricted to political elites, the lower levels of cheating could be interpreted as the existence of a norm guided behavior that complies more often with the “rule of law”, and a greater willingness to behave civic. Consistent with a political economy mechanism in which elites act unselﬁsh, because cheating is costly under the modern impartial legal system that does not differentiate anymore between elites and the rest of the society, and elite behavior might feed back on social cooperation.49

Moreover, in a society with strong norms of generalized morality, people are likely to demand higher standards of behavior from their politicians.50,51

Social Capital and Associations Next, I investigate in a more direct fashion the effects of the Code

Civil on social cooperation in an intermediate period. One way to measure the degree of social cooperation in historical societies and in the absence of survey data is to rely on outcome measures, such as participation in voluntary associations (Putnam, 1994; Guiso, Sapienza, and Zingales, 2008a). A comprehensive dataset of associations in cities in Weimar Germany has been collected by Satyanath,

Voigtländer, and Voth (2016). For each city, the data contains information about the number of clubs for a wide range of types, from animal breeders to music and military clubs. The type of club is particularly valuable for my purpose, since it allows to divide associations into “bonding” and “bridging” types of social capital, following Putnam (2001). Bonding social capital characterizes cooperation between people with homogeneous backgrounds, and strengthens divisions between groups in a society. In contrast,

49In the same vein but in a different context, Nunn and Wantchekon (2011) argue that the weak legal system was a potential factor for the persistence of mistrust in regions affected by the slave trade. They argue that unconstrained political leaders were not forced to behave trustworthy, and could misbehave without being held responsible. 50Nannicini et al. (2013) show that social capital is associated with lower levels of political misbehavior in contemporary Italy. 51The effects can also be reconciled with the theory and empirical ﬁndings described in Aghion et al. (2010), where low regulation goes along with high trust. Indeed, the Code helped to remove entry barriers and to establish a new elite, characterized by economic success.

27 bridging social capital refers to interactions between people with socially heterogeneous backgrounds.

Translated into categories of social trust, bonding social capital goes along with limited morality and a small radius of trust between individuals that belong to well-deﬁned groups. Bridging social capital, however, widens the radius of trust to outsiders. It is associated with strong trust in strangers and generalized morality. Following the classiﬁcation in Satyanath, Voigtländer, and Voth (2016), I characterize each club as either bonding or bridging.52 I then construct a measure of association density, that divides the total number of bridging and bonding associations by population, and a second measure that gives the share of bridging associations over total associations to assess the composition of social capital.

[Table 13 about here]

Table 13 reports the estimation results for the complete sample in columns 1 - 3, and for the border specification in columns 4 - 6. As column 1 shows, cities that used the Code Civil during the 19th century where not significantly different with respect to the density of associations in the 1920’s. Considering the composition of associations, however, cities where the Code Civil was used show a significantly higher share of bridging associations (see column 2). This finding is robust to a large set of controls (column 3).

A 90 year application of the Code Civil goes along with an increase in the share of bridging social capital associations, of about 0.5 of a standard deviation. Results for the border sample reported in columns 4

- 6 are similar. There is no effect on association density, but a positive effect on the share of bridging associations. In terms of magnitudes, coefficients range between a 0.4 to a 0.5 of a standard deviation increase in the share of clubs that fall into the bridging social capital category. Overall, the finding of a positive association between the Napoleonic legal system and bridging social capital measured in an intermediate period are in line with the results on contemporary social trust in strangers documented above. The findings support the interpretation that the removal of barriers to cooperation through legal equality also led to more social interactions between diverse societal groups. Complementarities between the types of social associations created and norms of cooperation could be one mechanism through which differences in social cooperation were perpetuated.

Cross-sectional estimations at a given point in time are informative, ideally, however, one would like to trace the evolution of associations over time. This is possible for one type of associations, the shooting associations (Schützenvereine), as information on the number of clubs founded per county, from the Mid-

52See the Appendix for the classiﬁcation of clubs.

28 dle Ages to 1939 at a yearly level, in the two Prussian provinces Rhineland and Westphalia exists.53,54

I use the years from 1700 to 1900. The data has the advantage to track the historical evolution of one particular association over time. However, shooting associations are an ambiguous measure of social capital as they were neither clearly bridging nor necessarily bonding associations.55 Furthermore, shooting associations played a role in 19th century political life. Keeping these caveats in mind, I use the foundation dates of shooting clubs available in a panel of 128 counties and an annual time period from

1700 to 1900. I aggregate the data on the county level into four different periods composed of 50 years, i.e. 1700 – 1750, 1750 – 1800, 1800 – 1850 and 1850 – 1900. I estimate a set of binary regressions

- equivalent to simple t-tests - comparing the mean number of clubs founded per province per period.

Results are shown in Table 14 and the graphical illustration of the evolution of clubs is provided in Figure

C.6 in the Appendix.

[Table 14 about here]

The results indicate that both provinces did not differ signiﬁcantly in the average number of clubs founded during the period from 1700 to 1750. However, starting from the second period there appears a signiﬁcant difference between the two provinces. From 1750 to 1800 on average an addition of 0.5 shooting clubs per county were founded in Westphalia. The negative difference aggravates in the period between 1800 and 1850. Only from 1850 onwards, a the sign of the mean difference changes and gets positive. From

1850, more clubs were founded in the Rhineland using the Code Civil then in the neighboring province of Westphalia. The same pattern shows up when the lagged cumulative number of clubs per county is controlled for in Panel B. Generalizing the results is difﬁcult, as the type of association is very particular.

However, the data shows roughly the expected pattern. While before 1800 there was a negative difference, this trend turns in the second half of the 19th century. Why the increase in associations did not start right after 1800 can only be answered speculatively. One possibility could relate to the ban of shooting associations by Napoleon in the occupied areas in 1808. The disruption of participation in shooting associations might have been aggravated in areas that were occupied longer and more intensively, such as the Rhineland. Second, in the years from 1815 to the March revolution in 1848 (Vormärz), shooting

53The source of the data is Plett (1991), who gathered this information by directly surveying current day shooting clubs about their year of foundation. In the data, a club is reported to exist in year t, if its date of foundation falls in this year. See section A.1 in the Appendix for further information. 54In 19th century Prussia the major types of associations were the singing clubs, the gymnasts, and the shooters (often jointly referred to as the “cloverleaf”) (Klenke, 1998, p. 20). These clubs were engaged in all kinds of community life, especially important for organizing big folk festivals (Klenke, 1998, p. 20). The marksmen’s festival (Schützenfest), organized by the shooters sometimes attracted hundreds of people. While the shooters started out as a para-military organization in the middle ages, from the 16th century on they changed their focus from defending cities and villages to a place of social gathering and celebrations. 55They are therefore excluded in the above analysis of social capital in Weimar Germany.

29 associations where locus of political opposition against rulers. Different levels of political discontent and participation in revolutionary activities might be another reason for the diverge in the ﬁrst half of the

19th century. A further possibility is that it took norms of generalized morality several decades to diffuse before they manifested themselves in increased associational engagement. Finally, the empirical results cannot discard the possibility that differences in the rate of urbanization in late 19th century played an important role in the evolution of shooting associations.

5 Conclusion

The Code Napoleon was created to serve as the backbone of a new social order in European societies.

It modernized legal institutions radically in parts of 19th century Germany. This paper documents that its application goes along with desirable social outcomes in the long-run. My empirical results suggest a strong complementarity between having applied the Code Civil and contemporaneous norms of trust and cooperation. Identiﬁcation from individuals living in close-by districts separated by the historical application of the Code Civil, a falsiﬁcation test consisting in moving the Code Civil border, and the comparison of pre-treatment characteristics, point towards a causal legacy of the French legal system.

Disentangling direct from indirect effects and identifying mechanisms of persistence is an important, but difﬁcult task faced by the persistence literature. I analyze a variety of intermediate outcomes to better understand the relationship between past legal institutions and trust today. The post-treatment evolution of economic development and the distribution of wealth cannot explain this association. In contrast, I

ﬁnd evidence that the Code Napoleon fostered generalized morality, evident for 19th century political elites who were signiﬁcantly less likely to commit electoral fraud. Its application also went along with a higher share of social associations emphasizing intra-group cooperation, and a general increase in the formation of social associations.

My findings suggest that impartial legal institution can remove barriers to cooperation and shift a society’s equilibrium to high cooperation for generations, even when those institutions were imposed. Strong norms towards cooperation persisted in the long-run, albeit institutions converged. Thus, persistence of norms can be expected to be exacerbated in situations where formal institutions remain weak. My results also relate to the negative long-term effects attributed to civil law systems (La Porta, López-de-Silanes, and Shleifer, 2008). In contrast, my findings imply that a civil law system is not inherently bad, but can create beneficial outcomes relative to other legals systems that do not guarantee equivalent rights.

30 References

ACEMOGLU, D., S. JOHNSON, and J. ROBINSON (2001). “The Colonial Origins of Comparative Devel-

opment: An Empirical Investigation”. The American Economic Review 91.5, pp. 1369–1401.

ACEMOGLU, D., D. CANTONI, S. JOHNSON, and J. ROBINSON (2011). “The Consequences of Radical

Reform: The French Revolution”. The American Economic Review 101.7, pp. 3286–3307.

ACEMOGLU, D., C. GARCÍA-JIMENO, and J. A. ROBINSON (2012). “Finding Eldorado: Slavery and

long-run development in Colombia”. Journal of Comparative Economics 40.4, pp. 534–564.

AGHION, P., Y. ALGAN, P. CAHUC, and A. SHLEIFER (2010). “Regulation and Distrust”. The Quarterly

Journal of Economics 125.3, pp. 1015–1049.

ALESINA, A. and E. LA FERRARA (2002). “Who Trusts Others?” Journal of Public Economics 85.2,

pp. 207–234.

ALGAN, Y. and P. CAHUC (2013). “Trust and Human Development: Overview and Policy Implications”.

In: Handbook of Economic Growth. Vol. 1A-2013. Aghion, P. and Durlauf, S.

ARVIND, T. and L. STIRTON (2010). “Explaining the Reception of the Code Napoleon in Germany: A

Fuzzy-Set Qualitative Comparative Analysis”. Legal Studies 30.1, pp. 1–29.

BAIROCH, P., J. BATOU, and P. CHEVRE (1988). The Population of European Cities. Data Bank and

Short Summary of Results: 800-1850. Geneva (CH): Geneva University, Centre of International Eco-

nomic History, 1988.

BANFIELD, E. C. (1967). The Moral Basis of a Backward Society. Free Press.

BECKER, H. (1985). “Das Rheinische Recht und seine Bedeutung für die Rechtsentwicklung in Deutsch-

land im 19. Jahrhundert”. Juristische Schulung 25, pp. 338–345.

BECKER, S. and L. WOESSMANN (2009). “Was Weber Wrong? A Human Capital Theory of Protestant

Economic History”. The Quarterly Journal of Economics 124.2, pp. 531–596.

BECKER, S., K. BOECKH, C. HAINZ, and L. WOESSMANN (2015). “The Empire Is Dead, Long Live the

Empire! Long-Run Persistence of Trust and Corruption in the Bureaucracy”. The Economic Journal

forthcoming.

BISIN, A. and T. VERDIER (2001). “The Economics of Cultural Transmission and the Dynamics of

Preferences”. Journal of Economic Theory 97.2, pp. 298–319.

CANTONI, D. and N. YUCHTMAN (2014). “Medieval Universities, Legal Institutions, and the Commer-

cial Revolution”. The Quarterly Journal of Economics 129.2, pp. 823–887.

31 CASSAR, A., G. D’ADDA, and P. GROSJEAN (2013). “Institutional Quality, Culture, and Norms of Co-

operation: Experimental Evidence from Italy and Kosovo”. Journal of Law and Economics 57.3,

pp. 821–863.

COLEMAN, J. S. (1988). “Social Capital in the Creation of Human Capital”. American Journal of Soci-

ology 94, pp. 95–120.

DE LONG, J. B. and A. SHLEIFER (1993). “Princes and Merchants: European City Growth before the

Industrial Revolution”. Journal of Law and Economics 36.2, pp. 671–702.

DELL, M. (2010). “The Persistent Effects of Peru’s Mining Mita”. Econometrica 78.6, pp. 1863–1903.

DIDEROT (1992). “Continuation of the Dialogue between A and B.” In: Political Writings. Ed. by J. H.

MASON and R. WOKLER. Cambridge University Press.

DOHMEN, T., A. FALK, D. HUFFMAN, and U. SUNDE (2012). “The Intergenerational Transmission of

Risk and Trust Attitudes”. The Review of Economic Studies 79.2, pp. 645–677.

DOYLE, W. (1990). The Oxford History of the French Revolution. Oxford University Press (Oxford and

New York).

FEHR, E. and S. GÄCHTER (2000). “Cooperation and Punishment in Public Goods Experiments”. The

American Economic Review 90.4, pp. 980–994.

FEHRENBACH, E. (2008). Vom Ancien Régime zum Wiener Kongress. Vol. 12. Oldenbourg Wis-

senschaftsverlag.

FISMAN, R. and E. MIGUEL (2007). “Corruption, Norms, and Legal Enforcement: Evidence from Diplo-

matic Parking Tickets”. Journal of Political Economy 115.6, pp. 1020–1048.

GENNAIOLI, N., R. LA PORTA, F. L. DE SILANES, and A. SHLEIFER (2014). “Growth in Regions”.

Journal of Economic Growth 19.3, pp. 259–309.

GLAESER, E. L., R. LA PORTA, F. LOPEZ-DE SILANES, and A. SHLEIFER (2004). “Do Institutions

Cause Growth?” Journal of Economic Growth 9.3, pp. 271–303.

GRAB, A. (2003). Napoleon and the Transformation of Europe. Palgrave Macmillan.

GUISO, L., P. SAPIENZA, and L. ZINGALES (2008a). “Long Term Persistence”. NBER Working Paper

14278.

— (2008b). “Social Capital as Good Culture”. Journal of the European Economic Association 6.2-3,

pp. 295–320.

— (2012). “Civic Capital as the Missing Link”. In: The Handbook of Social Economics. Jess Benhabib,

Alberto Bisin, and Matthew O. Jackson, pp. 417–480.

32 HEYMANN, E. (1914). “Landsberg, Ernst, Die Gutachten der Rheinischen Immediat-Justiz-Kommission

und der Kampf um die Rheinische Rechts- und Gerichtsverfassung 1814-1819”. Zeitschrift der

Savigny-Stiftung für Rechtsgeschichte, Germanistische Abteilung 35, pp. 622 –631.

JACOB, M. (2010). “Long-Term Persistence: The Free and Imperial City Experience in Germany”.

mimeo.

JACOB, M. and M. TYRELL (2010). “The Legacy of Surveillance: An Explanation for Social Capital

Erosion and the Persistent Economic Disparity between East and West Germany”. mimeo.

KELLER, W. and C. H. SHIUE (2015). “Market Integration as a Mechanism of Growth”.

KLEINHEYER, G. (1979). “Aspekte der Gleichheit in den Aufklärungskodiﬁkationen und den Konstitu-

tionen des Vormärz”. In: Von der ständischen Gesellschaft zur bürgerlichen Gleichheit. Duncker und

Humboldt.

KLENKE, D. (1998). Der singende deutsche Mann: Gesangsvereine und deutsches Nationalbewusstsein

von Napoleon bis Hitler. Muenster: Waxmann Verlag.

LA PORTA, R., F. LÓPEZ-DE-SILANES, and A. SHLEIFER (2008). “The Economic Consequences of

Legal Origins”. Journal of Economic Literature 46.2, pp. 285–332.

LEE, D. S. and T. LEMIEUX (2010). “Regression Discontinuity Designs in Economics”. Journal of Eco-

nomic Literature 48, pp. 281–355.

LEHOUCQ, F. (2003). “Electoral Fraud: Causes, Types, and Consequences”. Annual Review of Political

Science 6.1, pp. 233–256.

LÓPEZ-DE-SILANES, F., R. LA PORTA, A. SHLEIFER, and R. VISHNY (1998). “Law and Finance”.

Journal of Political Economy 106, pp. 1113–1155.

LYONS, M. (1994). Napoleon Bonaparte and the Legacy of the French Revolution. Vol. 1. Palgrave

Macmillan.

MILGROM, P., D. NORTH, and B. WEINGAST (1990). “The Role of Institutions in the Revival of Trade:

the Law Merchant, Private Judges, and the Champagne Fairs”. Economics and Politics, pp. 1–23.

MOKYR, J. (2003). The Oxford Encyclopedia of Economic History. Vol. 1. Oxford University Press.

NANNICINI, T., A. STELLA, G. TABELLINI, and U. TROIANO (2013). “Social Capital and Political

Accountability”. American Economic Journal: Economic Policy 5.2, pp. 222–250.

NIPPERDEY, D. (1983). Deutsche Geschichte 1800-1866: Bürgerwelt und starker Staat, Munich. Mu-

nich: C.H. Beck.

NUNN, N. (2012). “Culture and the Historical Process”. Economic History of Developing Regions 27,

pp. 108–126.

33 NUNN, N. and L. WANTCHEKON (2011). “The Slave Trade and the Origins of Mistrust in Africa”. The

American Economic Review 101.7, pp. 3221–3252.

PLETT, W. (1991). The Riﬂe Clubs in the Rhineland and in Westphalia 1789-1939. Vol. ZA8112 Data

ﬁle Version 1.0.0. GESIS Data Archive, Cologne.

PUTNAM, R. D. (1994). Making Democracy Work: Civic Traditions in Modern Italy. Princeton Univer-

sity Press.

— (2001). Bowling Alone: The Collapse and Revival of American Community. Simon and Schuster.

ROWE, M. (2003). From Reich to State: the Rhineland in the Revolutionary Age, 1780-1830. Cambridge

University Press.

SATYANATH, S., N. VOIGTLÄNDER, and H.-J. VOTH (2016). “Bowling for Fascism: Social Capital and

the Rise of the Nazi Party”. Journal of Political Economy, forthcoming.

SCHLOSSER, H. (1995). “Kodiﬁkationen im Umfeld des Preussischen Allgemeinen Landrechts: der

Französische Code Civil (1804) und das Österreichische Allgemeine Bürgerliche Gesetzbuch

(1811)”. In: Kodiﬁkation gestern und heute: zum 200. Geburtstag des Allgemeinen Landrechts für

die Preussischen Staaten.

SCHMENK, H. (2008). Xanten im 19. Jahrhundert. Böhlau Verlag Köln.

SCHMON, S. E.-M. (2007). Machtspruch und Gesetzesherrschaft. Böhlau-Verlag.

SIEBELS, V. (2011). Ernst Landsberg (1860-1927): ein jüdischer Gelehrter im Kaiserreich. Vol. 68.

Mohr Siebeck.

SMITH, A. (1776). “The Wealth of Nations (1937)”. New York: Modern Library 11937.

SPERBER, J. (1992). Rhineland Radicals: The Democratic Movement and the Revolution of 1848-1849.

Princeton University Press.

STEIN, W. H. (2004). Die Bedeutung des Code Civil für Deutschland - Revolutionäres Recht - Imperiales

Recht - Liberales Recht. Speech Manuscript.

STURM, F. (2011). 200 Jahre Badisches Landrecht. Schriftenreihe des Rechtshistorischen Museums

Karlsruhe. Verlag der Gesellschaft für Kulturhistorische Dokumentation.

TABELLINI, G. (2008). “The Scope of Cooperation: Values and Incentives”. The Quarterly Journal of

Economics 123.3, pp. 905–950.

— (2010). “Culture and Institutions: Economic Development in the Regions of Europe”. Journal of the

European Economic Association 8.4, pp. 677–716.

TOCQUEVILLE, A. DE (2015). The Old Regime and the Revolution, Volume I: The Complete Text (Volume

1). University of Chicago Press.

34 VOIGTLÄNDER, N. and H. VOTH (2012). “Persecution Perpetuated: The Medieval Origins of Anti-

Semitic Violence in Nazi Germany”. The Quarterly Journal of Economics 127.3, pp. 1339–1392.

WAGNER, G., J. FRICK, and J. SCHUPP (2007). “The German Socioeconomic Panel Study (SOEP):

Scope, Evolution and Enhancements”. Schmollers Jahrbuch 127, pp. 139–169.

ZIBLATT, D. (2009). “Shaping Democratic Practice and the Causes of Electoral Fraud: The Case of

Nineteenth-Century Germany”. American Political Science Review 103.01, pp. 1–21.

35 A Tables

Table 1: Summary Statistics SOEP Data

Full Sample Border = 0 Border = 1

N Mean StDev Min Max N Mean StDev N Mean StDev

Treatment

Code Civil 17664 24.85 37.90 0 98 3242 3.31 4.45 3117 95.73 3.61 Border 3242 0 0 3117 1 0

Social Capital

Trust 17664 2.31 0.54 1 4 3242 2.30 0.56 3117 2.36 0.56 Coop 17664 0.36 0.48 0 1 3242 0.36 0.48 3117 0.44 0.50 Fair 17664 0.53 0.5 0 1 3242 0.54 0.50 3117 0.57 0.49

Control Variables

Personal Characteristics Age 17664 47.01 17.13 17 100 3242 46.43 16.99 3117 46.82 16.96 Age sq. 17664 2502.96 1697.27 289 10000 3242 2444.60 1672.42 3117 2479.56 1701.50 Male 17664 0.49 0.5 0 1 3242 0.49 0.50 3117 0.48 0.50 GDR before 1990 17664 0.25 0.43 0 1 3242 0.03 0.18 3117 0.03 0.16

Religion Protestant 17664 0.33 0.47 0 1 3242 0.36 0.48 3117 0.32 0.47 Catholic 17664 0.32 0.46 0 1 3242 0.36 0.48 3117 0.45 0.50 Other Christian 17664 0.02 0.15 0 1 3242 0.04 0.21 3117 0.02 0.14 Other Non-Christian 17664 0.04 0.2 0 1 3242 0.07 0.26 3117 0.05 0.22 None 17664 0.29 0.46 0 1 3242 0.17 0.37 3117 0.16 0.36

Education Secondary School (13 years) 17664 0.2 0.4 0 1 3242 0.19 0.39 3117 0.22 0.42 Secondary School (12 years) 17664 0.05 0.21 0 1 3242 0.05 0.23 3117 0.06 0.23 Middle School (10 years) 17664 0.29 0.45 0 1 3242 0.23 0.42 3117 0.21 0.41 Middle School (9 years) 17664 0.35 0.48 0 1 3242 0.38 0.48 3117 0.37 0.48 Other degree 17664 0.07 0.25 0 1 3242 0.09 0.28 3117 0.09 0.28 No degree 17664 0.05 0.22 0 1 3242 0.07 0.25 3117 0.06 0.23

Economic Controls HH Income 17664 2811.49 2091.13 0 85000 3242 3018.66 1987.66 3117 3005.77 1829.61 Unemployment 17664 0.07 0.25 0 1 3242 0.05 0.22 3117 0.05 0.21

36 Table 2: Summary Statistics of Historical and Geographic Controls

Panel A: SOEP Sample N Mean StDev Min Max

Historical Controls (evaluated at the historical territory) Urbanization Rate 1800 17664 11.19 5.23 3.39 22.50 Urbanization Rate 1850 17664 15.20 7.84 5.44 33.29 Share of Protestants 1800 17664 0.57 0.36 0.05 1 Latitude 17664 50.66 1.73 48.13 54.52 Longitude 17664 10.05 2.09 7.60 13.73 Distance to Paris 17664 615.74 157.96 413.06 851.01 Reform Index 17664 64.56 19.47 37.25 100.25 Abolition of Serfdom 17664 89.97 11.88 68.00 117.00 Agrarian Reform 17664 78.71 11.77 38.00 96.00 Abolition of Guilds 17664 62.59 30.69 31.00 105.00 Geographic Controls (evaluated at the district) Distance Major River (in km) 17664 36.39 25.37 0.19 143.43 Slope Index 17664 8350.80 1225.84 2920.14 9965 Wheat Suitability 17664 4.58 0.81 1 6.24 Distance Coast (in km) 17664 255.17 126.05 0.99 483.12

Panel B: Other Dependent Variables N Mean StDev Min Max

City Population (ln) City Population 1500 122 1.51 0.83 0.00 3.81 (ln) City Population 1600 153 1.54 0.82 0.00 3.81 (ln) City Population 1700 171 1.47 0.75 0.00 4.01 (ln) City Population 1800 214 1.86 0.70 0.00 5.15 (ln) City Population 1850 209 2.30 0.76 0.00 6.08 (ln) City Population 1875 211 2.70 0.90 0.94 6.87 (ln) City Population 1900 211 3.17 1.07 1.17 7.54 (ln) City Population 1910 211 3.37 1.15 1.17 7.64 City Population Growth 1800-1900 211 1.29 0.75 -0.56 4.14 City Characteristics Free City 223 0.19 0.39 0.00 1.00 Hanseatic City 223 0.30 0.46 0.00 1.00 Bishop City 223 0.12 0.32 0.00 1.00 University City 223 0.11 0.32 0.00 1.00 Inequality and Electoral Fraud Land Gini 1898 241 0.70 0.12 0.46 0.95 No of Contested Elections 1871-1900 241 1.70 1.64 0.00 8.00 Post-Treatment Development, Education and Religion Share of Protestants 1871 62 0.31 0.31 0.00 0.95 Share of Jews 1871 62 0.01 0.01 0.00 0.03 % Literate 1871 62 0.92 0.03 0.86 0.99 % Illiterate 1871 62 0.06 0.03 0.01 0.12 (ln) Teacher Wage 1886 62 7.03 0.21 6.69 7.52 Income Tax p.c. 1878 43 2.14 0.59 0.88 3.49 Share Manuf. & Service 1882 62 0.19 0.07 0.07 0.34 Share Agriculture 1882 62 0.15 0.09 0.00 0.29 (ln) GDP p.c. 2005 352 9.76 0.11 9.51 10.25 Post-Treatment Social Capital Assoc Density 1920’s 188 2.93 5.59 0.00 72.73 Share “Bridging” Assoc 1920’s 187 0.81 0.12 0.42 1.00 Number of Shooting Clubs founded 512 2.81 4.84 0.00 31.00

Notes: See section A.1 in the Appendix for further information.

37 Table 3: Complete Sample

Panel A: Macroregion FE and Base Controls Panel B: Adding Pre-Treatment Controls

(1) (2) (3) (4) (5) (6) Dependent variables: Trust Coop Fair Trust Coop Fair

Code Civil (x100) 0.057∗∗ 0.091∗∗∗ 0.031∗ 0.120∗∗∗ 0.154∗∗∗ 0.040 (0.027) (0.027) (0.018) (0.027) (0.027) (0.018) Observations 17664 17664 17664 17664 17664 17664 R-squared 0.02 0.04 0.02 0.02 0.04 0.02 No. of Clusters 349 349 349 349 349 349

Panel C: Adding Geographic Controls Panel D: Adding Post-Treatment Controls

(1) (2) (3) (4) (5) (6) Dependent variables: Trust Coop Fair Trust Coop Fair

Code Civil (x100) 0.144∗∗∗ 0.161∗∗∗ 0.042 0.126∗∗∗ 0.163∗∗∗ 0.041 (0.036) (0.034) (0.026) (0.038) (0.039) (0.028) Observations 17664 17664 17664 17664 17664 17664 R-squared 0.02 0.04 0.02 0.07 0.05 0.04 No. of Clusters 349 349 349 349 349 349 Notes: OLS regressions. The unit of observation is the individual. See main text and section A.1 in the Appendix for a description of the variables. Panel A controls for macro-region dummies and base controls (age, age sq., gender, religion, location before 1990). Panel B adds pre-treatment controls measured at the historical territory (urbanization rate 1800, share of Protestants 1800, longitude and latitude of the territory, distance to Paris). Panel C adds geographic controls measured at the district (suitability for growing wheat, distance to coast, distance to river, slope). Panel D adds post-treatment controls on the individual level (education, household income, unemployment indicator) and territory level (urbanization rate 1850). Standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

38 Table 4: Complete Sample: Robustness

Without Baden/Saxony Prussia 1815 Prussia 1866 Placebo Outcomes

(1) (2) (3) (4) (5) Dependent variables Trust Life Satisfaction Political Interest

Code Civil (x100) 0.097∗∗ 0.122∗∗ 0.121∗∗ -0.097 -0.002 (0.038) (0.057) (0.058) (0.092) (0.051) Observations 15340 7470 9823 17643 17623 R-squared 0.02 0.03 0.03 0.04 0.10 No. of Clusters 319 109 166 349 349 Notes: OLS regressions. The unit of observation is the individual. See main text and section A.1 in the Appendix for a description of the variables. All regressions control for the full set of controls at the individual level, as well as geographic and historical controls as in Table 3, Panel C. Standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

39 Table 5: Further Dimensions of the Napoleonic Reforms and Social Capital

Panel A: Alternative Reforms, one at a time. (1) (2) (3) Dependent variables: Trust Coop Fair

Reform Index (x100) 0.416∗∗∗ 0.503∗∗∗ 0.095 (0.110) (0.111) (0.083)

Abolition of Serfdom (x100) 0.386∗∗∗ 0.256∗ 0.038 (0.135) (0.154) (0.095)

Agrarian Reform (x100) 0.212∗ 0.088 0.026 (0.108) (0.108) (0.087)

Abolition of Guilds (x100) -0.034 0.009 -0.023 (0.074) (0.076) (0.057) Observations 17664 17664 17664 No. of Clusters 349 349 349

Panel B: Horse Race (1) (2) (3) (4) (5) Dependent variable: Trust

Code Civil (x100) 0.113∗∗∗ 0.122∗∗∗ 0.137∗∗∗ 0.154∗∗∗ 0.131∗∗∗ (0.043) (0.036) (0.036) (0.035) (0.036)

Reform Index (x100) 0.162 (0.129)

Abolition of Serfdom (x100) 0.211 0.250 (0.133) (0.183)

Agrarian Reform (x100) 0.126 -0.066 (0.104) (0.150)

Abolition of Guilds (x100) 0.070 0.059 (0.065) (0.071)

Observations 17664 17664 17664 17664 17664 No. of Clusters 349 349 349 349 349 Notes: OLS regressions, controlling for macro-region ﬁxed effects, as well as geographic and historical controls as in Table 3, Panel C. The unit of observation is an individual. See main text and section A.1 in the Appendix for a description of the variables. Panel A reports results for estimations using alternative institutional reforms one at a time. Panel B adds alternative reforms to the regression using the duration of the Code Civil as the independent variable. Abolition of Serfdom, Agrarian Reform and Abolition of Guilds measure the years of enactment of the respective reform, as of 1900. The reform index is a combined measure of all reforms - including the Code Civil - evaluated as of the year 1900. Reform variables are taken from Acemoglu et al. (2011). Standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

40 Table 6: Border Estimations: Local Linear Regressions, 50km

(1) (2) (3) Dependent variables: Trust Coop Fair

a) Base controls Border 0.059∗ 0.083∗∗∗ 0.034∗ (0.030) (0.027) (0.020) Observations 6359 6359 6359 R-squared 0.01 0.04 0.01

b) Adding State FE Border 0.087∗∗∗ 0.074∗∗ 0.044∗∗ (0.032) (0.030) (0.022) Observations 6359 6359 6359 R-squared 0.02 0.04 0.01

c) Adding full set of individual controls Border 0.072∗∗ 0.072∗∗ 0.035∗ (0.030) (0.029) (0.020) Observations 6359 6359 6359 R-squared 0.07 0.05 0.04

d) Only Rhineland, Westphalia and Mark Border 0.050∗ 0.070∗∗∗ 0.033 (0.028) (0.024) (0.026) Observations 3,368 3,368 3,368 R-squared 0.08 0.03 0.04

e) Adding geographic controls Border 0.100∗∗∗ 0.078∗∗ 0.049∗ (0.038) (0.034) (0.029) Observations 6359 6359 6359 R-squared 0.08 0.06 0.04

f) Duration of application of Code Civil Code Civil (x100) 0.113∗∗∗ 0.089∗∗ 0.058∗ (0.042) (0.037) (0.033) Observations 6359 6359 6359 R-squared 0.08 0.05 0.04 No. of Clusters 110 110 110

Notes: OLS regressions. The unit of observation is the individual. Base controls include age, age sq., gender, religion and location before 1990. The full set of individual controls adds household income, unemployment and educational categories. Geographic controls are distance to a river, distance to coast, wheat suitability of the soil and the average slope. See main text and section A.1 in the Appendix for a description of the variables. Robust standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

41 Table 7: Border Estimations: Alternative Speciﬁcations

Panel A: Spatial Regression Discontinuity Design, < 50 km (1) (2) (3) Dependent variables: Trust Coop Fair

i) f(location): Linear in Long/Lat Border 0.109∗∗∗ 0.084∗∗ 0.047 (0.040) (0.033) (0.028)

ii) f(location): Quadratic in Long/Lat Border 0.106∗∗ 0.067∗ 0.031 (0.050) (0.034) (0.036)

iii) f(location): Cubic in Long/Lat Border 0.102∗∗ 0.057∗ 0.027 (0.046) (0.029) (0.034)

Observations 6359 6359 6359 State FE Yes Yes Yes No. of Clusters 110 110 110

Panel B: Neighbor Pair Fixed Effects (1) (2) (3) Dependent variables: Trust Coop Fair

a) dichotomous treatment Border 0.098∗∗ 0.095∗∗ 0.041 (0.037) (0.038) (0.035)

b) duration of application of Code Civil Code Civil (x100) 0.110∗∗∗ 0.108∗∗ 0.047 (0.041) (0.041) (0.039)

Neighbor-Pair FE Yes Yes Yes Observations 3046 3046 3046 No. of Clusters 48 48 48

Notes: OLS regressions. The unit of observation is the individual. Panel A reports estimates from a spatial regression discontinuity design controlling for polynomials in longitude and latitude. All regressions control for the full set of individual controls and geographic controls. Panel B reports estimates from neighbor-pair ﬁxed effect estimation that include ﬁxed effects for couples of adjacent districts. All regressions control for the full set of individual controls. See main text and section A.1 in the Appendix for a description of the variables. Robust standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

42 Table 8: Placebo Estimates: Moving the Border

Panel A: Moving the Border Outwards (1) (2) (3) (4) (5) (6) Dependent variables: Trust Coop Fair Trust Coop Fair

Border -0.008 0.032 -0.022 -0.026 0.022 -0.036 (0.028) (0.022) (0.024) (0.032) (0.021) (0.024) Observations 4768 4768 4768 4768 4768 4768 R-squared 0.02 0.05 0.02 0.08 0.06 0.04 No. of Clusters 94 94 94 94 94 94 Base Controls Yes Yes Yes Yes Yes Yes Full Controls No No No Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes

Panel B: Moving the Border Inwards (1) (2) (3) (4) (5) (6) Dependent variables: Trust Coop Fair Trust Coop Fair

Border 0.036 -0.110∗∗ -0.027 0.023 -0.081∗ -0.037 (0.046) (0.053) (0.034) (0.044) (0.042) (0.038) Observations 3679 3679 3679 3679 3679 3679 R-squared 0.01 0.03 0.01 0.07 0.07 0.04 No. of Clusters 74 74 74 74 74 74 Base Controls Yes Yes Yes Yes Yes Yes Full Controls No No No Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Notes: OLS regressions. The unit of observation is an individual. Base controls include age, age sq., gender, religion and location before 1990. Full controls add household income, unemployment and educational categories, as well as geographic controls. See main text and section A.1 in the Appendix for a description of the variables. Standard errors clustered at the district level in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

43 Table 9: Balancedness of Pre-Treatment Characteristics at the Border

Panel A: Geography (1) (2) (3) (4) Dependent variables: Dist. River Slope Wheat Suitability Dist. Coast

Border -13.218*** 97.337 -0.229 -23.612 (3.801) (111.244) (0.156) (22.553) Observations 110 110 110 110 R-squared 0.10 0.01 0.02 0.01

Panel B: City Characteristics (1) (2) (3) (4) Dependent variables: Free City Hanseatic City Bishop City University

Border 0.028 -0.135 0.074 -0.009 (0.101) (0.101) (0.057) (0.065) Observations 74 74 74 74 R-squared 0.00 0.02 0.02 0.00

Panel C: Economic Development (1) (2) (3) (4) Dependent variables: (ln) City Pop. 1500 (ln) City Pop. 1600 (ln) City Pop. 1700 (ln) City Pop. 1800

Border 0.079 -0.238 0.094 0.070 (0.299) (0.207) (0.208) (0.151) Observations 33 45 54 68 R-squared 0.00 0.03 0.00 0.00 Notes: This table examines balancedness of geographic characteristics (Panel A), city characteristics (Panel B) and city size (Panel C) around the border. The unit of observation is the district in columns (1), (2) and the city in column (3). See main text and section A.1 in the Appendix for a description of the variables. Heteroscedastic-robust standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

44 Table 10: Short and Long-Run Development Outcomes

Panel A: Cross-Sectional Estimation in the Complete Sample (1) (2) (3) (4) Dependent variables: (ln) City Population City Population Growth (ln) Gdp p.c. 2005 1850 1900 1800-1900

Code Civil (x100) 0.021 -0.031 -0.078 0.045** (0.192) (0.140) (0.209) (0.020) (ln) Population 1800 0.165** (0.069) Macro-Region Fe Yes Yes Yes Yes Geographic Controls Yes Yes Yes Yes Historical Controls Yes Yes Yes Yes Observations 209 211 211 352 R-squared 0.09 0.08 0.13 0.50

Panel B: Cross-Sectional Estimation in the Border Sample (1) (2) (3) (4) Dependent variables: (ln) City Population City Population Growth (ln) Gdp p.c. 2005 1850 1900 1800-1900

Border 0.176 0.190 0.178 0.008 (0.183) (0.311) (0.267) (0.016) (ln) Population 1800 0.123 (0.183) Geographic Controls Yes Yes Yes Yes Observations 65 67 67 110 R-squared 0.07 0.08 0.10 0.17

Panel C: Difference-in-Differences Estimation at the Border (1700-1910) (1) (2) (3) (4) Dependent variable: (ln) City Population Post 1800 x Border 0.076 0.188 (0.215) (0.206) Border 0.132 (0.155) Post 1800 1.323*** (0.165) Post 1800 x Code Civil 0.002 (0.002) Border x Year 1750 0.061 (0.109) Border x Year 1800 0.132 (0.184) Border x Year 1850 0.268 (0.212) Border x Year 1875 0.205 (0.252) Border x Year 1900 0.330 (0.289) Border x Year 1910 0.232 (0.319) Observations 444 444 444 444 R-squared 0.33 0.74 0.74 0.74 City FE No Yes Yes Yes Year FE No Yes Yes Yes No. of Clusters 68 68 68 68 Notes: This table reports the following estimation results: Cross-sectional regressions in the complete sample in Panel A, controlling for macro-region ﬁxed effects, as well as geographic and historical controls as in Table 3, Panel C. The unit of observation is the city in columns (1) - (3), and the district in column (4). Standard errors clustered at the historical territory reported in parentheses. Panel B reports cross-sectional regressions in the border sample, controlling for geographical characteristics. The unit of observation is the city in columns (1) - (3), and the district in column (4). Heteroscedastic-robust standard errors in parentheses. Panel C reports difference-in-differences estimation in the border sample for the period 1700 to 1910. The unit of observation is the city. Standard errors clustered at the city reported in parentheses. See main text and section A.1 in the Appendix for a description of the variables. * p < 0.10, ** p < 0.05, *** p < 0.01. 45 Table 11: Additional Post-Treatment Development Outcomes at the Border

Panel A: Economic Outcomes (1) (2) (3) (4) Dependent variables: (ln) Teacher Wage Income Tax p.c. Share Manuf. and Service Share Agriculture 1886 1878 1882 1882

Border 0.017 -0.197 -0.002 0.022 (0.060) (0.214) (0.019) (0.026) Observations 62 43 62 62 R-squared 0.00 0.02 0.00 0.01

Panel B: Religion & Education (1) (2) (3) (4) Dependent variables: % Protestants % Jews % Literate % Illiterate 1871 1871 1871 1871

Border -0.107 0.002 -0.025*** 0.023*** (0.087) (0.002) (0.008) (0.007) Observations 62 62 62 62 R-squared 0.03 0.02 0.15 0.13 Notes: This table reports regressions using the border sample. The unit of observation is the district. See main text and section A.1 in the Appendix for a description of the variables. Heteroscedastic-robust standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

46 Table 12: Land Inequality and Electoral Fraud

Panel A: Inequality Complete Sample Border Sample

(1) (2) (3) (4) Dependent variable Land Gini 1898

Code Civil (x100) 0.017 -0.026* (0.059) (0.013) Border 0.016 -0.007 (0.022) (0.015) (ln) Population 0.069** 0.046* 0.268*** 0.097*** (0.030) (0.024) (0.027) (0.025) Share Catholics -0.002*** -0.001* 0.001*** -0.000 (0.000) (0.000) (0.000) (0.000) Macro-Region Fe Yes Yes Geographic Controls Yes Yes Historical Controls Yes Observations 241 241 73 73 R-squared 0.28 0.72 0.46 0.82

Panel B: Electoral Fraud Complete Sample Border Sample

(1) (2) (3) (4) Dependent variable: Number of Contested Elections 1871-1900

Code Civil (x100) -0.684** -0.063 (0.255) (0.301) Border -0.740** -0.608** (0.304) (0.299) (ln) Population 1.383*** 1.418*** 2.982*** 2.711*** (0.306) (0.331) (0.754) (0.782) Turnout -0.787 -0.197 6.432* 6.260* (2.534) (2.608) (3.241) (3.446) Competition -2.659*** -2.626*** -0.953 -1.733 (0.808) (0.803) (0.735) (1.053) Share Catholics -0.003 -0.002 0.009 0.016* (0.004) (0.005) (0.006) (0.009) Macro-Region Fe Yes Yes Geographic Controls Yes Yes Historical Controls Yes Observations 241 241 73 73 R-squared 0.36 0.39 0.51 0.54 Notes: This table reports regressions with land Gini in 1898 as dependent variable in Panel A, and the number of contested elections between 1871 and 1900 in Panel B. The unit of observation is the district. See main text and section A.1 in the Appendix for a description of the variables. Heteroscedastic-robust standard errors in parentheses, clustered at the historical territory level in columns (1) and (2). * p < 0.10, ** p < 0.05, *** p < 0.01.

47 Table 13: Social Capital in the 1920s

Complete Sample Border Sample

(1) (2) (3) (4) (5) (6) Dependent variables: Assoc Density Share “Bridging” Assoc Assoc Density Share “Bridging” Assoc

Code Civil (x100) -1.321 0.074*** 0.072** (0.867) (0.025) (0.028) Border 0.118 0.047* 0.061* (0.202) (0.025) (0.036) Share Catholics 4.802 0.010 0.053 -0.613* 0.037 0.024 (5.344) (0.055) (0.044) (0.362) (0.045) (0.058) (ln) Population -1.315*** -0.012 -0.010 -0.629*** -0.003 -0.006 (0.433) (0.009) (0.008) (0.071) (0.009) (0.011) Macro-Region Fe Yes Yes Yes Geographic Controls Yes Yes Territory Controls Yes Observations 188 187 187 81 81 81 R-squared 0.16 0.17 0.23 0.46 0.07 0.08 Notes: The unit of observation is the city. Assoc Density is the total number of associations per 1,000 city inhabitants. Share “Bridging” Assoc is the share of all associations that can be deﬁned as bridging (versus bonding) following the classiﬁcation in Satyanath, Voigtlän- der, and Voth, 2016. See main text and section A.1 in the Appendix for a description of the variables. Heteroscedastic-robust standard errors in parentheses, clustered at the historical territory level in columns (1) - (3). * p < 0.10, ** p < 0.05, *** p < 0.01.

48 Table 14: Shooting Clubs

Panel A (1) (2) (3) (4) Dependent variable: Number of Clubs founded in 1700 – 1750 1750 – 1800 1800 – 1850 1850 – 1900

Code Civil -0.147 -0.500** -2.782*** 1.954* (0.288) (0.236) (0.959) (1.100) Observations 128 128 128 128

Panel B (1) (2) (3) (4) Dependent variable: Number of Clubs founded in 1700 – 1750 1750 – 1800 1800 – 1850 1850 – 1900

Code Civil -0.704*** -3.310*** 2.180** (0.207) (0.842) (0.934) Lagged No Clubs 0.082*** 0.265*** 0.284*** (0.014) (0.076) (0.063) Observations 128 128 128 Notes: OLS regressions. The dependent variable is the number of clubs founded per county. See main text and section A.1 in the Appendix for a description of the variables. Code Civil equals 1 for counties belonging to the province Rhineland, and 0 for counties belonging to Westphalia. Source: Plett, W. (1991). Robust standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01

49 B Figures

Figure 1: Law and Trust

(a) Unconditional Correlation

(b) Conditional Correlation

Notes: The ﬁgure shows cross-country correlation between legal quality and social trust. Countries with greater protection of individual property, and in which courts and judges are independent and impartial display higher levels of interpersonal trust. This does not merely mirror an income effect, since the positive association is robust to controlling for population, income per capita, education and ethnic fractionalization as shown in Figure 1 b)). Sources: World Values Survey, Economic Freedom of the World Index 2010.

50 Figure 2: Legal Systems in 19th Century German

Notes: This map illustrates the distribution of legal systems in 19th century Germany. See the Appendix for the original map that is taken from Stein (2004).

51 Figure 3: Treatment in the Complete Sample

Notes: This map shows the distribution of the number of years the Code Civil was in place in different parts of Germany. Polygons represent contemporary districts. White areas represent those historical territories that are not included in the sample (the sample deﬁnition follows Acemoglu et al. (2011)).

52 Figure 4: Border Treatment

Notes: This map shows the border sample. Circles represent the centroids of the counties included in the original border sample with distances smaller than 50 km to the borderline. Triangles and squares represent centroids of the counties that are added in the falsiﬁcation exercise. Region borders represent historical territories.

53 Figure 5: Trust at the Border

Notes: This map shows average levels of trust for the 110 districts included in the border sample.